Using a Backpropagation Artificial Neural Network to Predict ... - MDPI

water Article

Using a Backpropagation Artificial Neural Network to Predict Nutrient Removal in Tidal Flow Constructed Wetlands Wei Li 1,2,3 ID , Lijuan Cui 1,2,3, *, Yaqiong Zhang 1,2,3 , Zhangjie Cai 1,2,3 , Manyin Zhang 1,2,3 , Weigang Xu 1,2,3 ID , Xinsheng Zhao 1,2,3 , Yinru Lei 1,2,3 , Xu Pan 1,2,3 , Jing Li 1,2,3 and Zhiguo Dou 1,2,3 1

2 3

*

Institute of Wetland Research, Chinese Academy of Forestry, Beijing 100091, China; [email protected] (W.L.); [email protected] (Y.Z.); [email protected] (Z.C.); [email protected] (M.Z.); [email protected] (W.X.); [email protected] (X.Z.); [email protected] (Y.L.); [email protected] (X.P.); [email protected] (J.L.); [email protected] (Z.D.) The Beijing Key Laboratory of Wetland Ecological Function and Restoration, Beijing 100091, China Beijing Hanshiqiao National Wetland Ecosystem Research Station, Beijing 101399, China Correspondence: [email protected] or [email protected]; Tel.: +86-10-6282-4151

Received: 3 November 2017; Accepted: 14 January 2018; Published: 18 January 2018

Abstract: Nutrient removal in tidal flow constructed wetlands (TF-CW) is a complex series of nonlinear multi-parameter interactions. We simulated three tidal flow systems and a continuous vertical flow system filled with synthetic wastewater and compared the influent and effluent concentrations to examine (1) nutrient removal in artificial TF-CWs, and (2) the ability of a backpropagation (BP) artificial neural network to predict nutrient removal. The nutrient removal rates were higher under tidal flow when the idle/reaction time was two, and reached 90 ± 3%, 99 ± 1%, and 58 ± 13% for total nitrogen (TN), ammonium nitrogen (NH4 + -N), and total phosphorus (TP), respectively. The main influences on nutrient removal for each scenario were identified by redundancy analysis and were input into the model to train and verify the pollutant effluent concentrations. Comparison of the actual and model-predicted effluent concentrations showed that the model predictions were good. The predicted and actual values were correlated and the margin of error was small. The BP neural network fitted best to TP, with an R2 of 0.90. The R2 values of TN, NH4 + -N, and nitrate nitrogen (NO3 − -N) were 0.67, 0.73, and 0.69, respectively. Keywords: tidal flow; constructed wetlands; nutrients removal; BP neural network

1. Introduction In recent decades, constructed wetlands have been widely used to treat industrial and agricultural wastewater because of their ability to remove nutrients, organic matter, heavy metals, and pathogens from wastewater [1–3]. In addition to their wastewater purification ability, constructed wetlands have low running costs, low energy consumption, and are simple to manage and maintain [4–6]. However, constructed wetlands are also criticized for their low re-oxygenation ability and their vulnerability to low temperatures, so it is therefore necessary to find methods and technologies for improving existing constructed wetlands. A new type of constructed wetland system, a tidal flow constructed wetland (TF-CW), has significantly higher dissolved oxygen (DO) levels than conventional constructed wetlands [7]. The TF-CW uses the interstitial suction created by a change in the bed infiltration surface during tidal operations to suck atmospheric oxygen into the wetland matrix or soil voids [8,9]. Through this mechanism, the oxygen environment in the bed of the artificial wetland can be improved considerably,

Water 2018, 10, 83; doi:10.3390/w10010083

www.mdpi.com/journal/water

Water 2018, 10, 83

2 of 17

such that the plant roots have sufficient oxygen. The TF-CW also provides more efficient removal of ammonium nitrogen (NH4 + -N) and organic matter than traditional constructed wetlands. Nutrient removal in constructed wetlands is linked to many factors. For example, NH4 + -N removal is influenced by, among other factors, the influent concentration, temperature, pH, and DO. The removal of nitrate nitrogen (NO3 − -N) is influenced by total organic carbon (TOC), the influent method, and time [10,11]. Water 2017, 9, 83 + -N and NO − -N are linked, and are influenced, promoted, or restricted 2 of 21 by these The removals of NH 4 3 factors, resulting innitrogen a complex wetland ecosystem. Pollutant removal in TF-CWs is therefore +-N) and organic ammonium (NH4artificial matter than traditional constructed wetlands. Nutrient + a complex process involvingwetlands nonlinear multi-parameter [12]. removal in constructed is linked to many factors. interactions For example, NH 4 -N removal is influenced by, among other factors, the influent concentration, temperature, pH, and DO. The removal of nitrate Rumelhart and McCelland established a multi-layer feed-forward neural network trained by nitrogen (NO3−-N) is algorithm influenced by organic carbon (TOC), the influent method, andnetwork, time [10,11]. an error backpropagation intotal 1986, known as the backpropagation (BP) which later The removals of NH4+-N and NO3−-N are linked, and are influenced, promoted, or restricted by these become one of the most widely used neural network models. These BP networks can learn and store factors, resulting in a complex artificial wetland ecosystem. Pollutant removal in TF-CWs is therefore a many input–output pattern mappings and can predict the complex dynamic systems of nonlinear complex process involving nonlinear multi-parameter interactions [12]. models [13]. Rumelhart The applications of BP neural networks infiltrate into various disciplines; and McCelland established a multi-layer feed-forward neural network trained byfor an example, error backpropagation algorithm in 1986, known as the backpropagation (BP) network, which later they can be used to predict disasters such as typhoon pathways [14] and earthquakes [15], to classify become one of the most widely used neural network models. These BP networks can learn and store remote sensing images [16], identify wireless channels [17], predict the stock market [18], and to predict many input–output pattern mappings and can predict the complex dynamic systems of nonlinear groundwater quality In ecology, neural networks are widely for example, to support models [13]. The[19]. applications of BP BP neural networks infiltrate into variousused, disciplines; for example, ecological and health assessments ecological planning, ecological they canenvironmental be used to predict disasters such as typhoon[20,21], pathways [14] and earthquakes [15], to classify security remote sensing images [16], identifyassessment, wireless channels [17], predict the stockand market and to [22–24]. assessment, eco-environmental quality ecological restoration, pest[18], prediction predict groundwater quality [19]. In ecology, BP neural networks are widely used, for example, However, they have rarely been applied to predict the performance of TF-CWs. In thistostudy, we support ecological and environmental health assessments [20,21], ecological planning, ecological set up a BP network model to predict the removal of pollutants from TF-CWs with four different security assessment, eco-environmental quality assessment, ecological restoration, and pest influent modes. We used However, this model construct the applied artificial neuralthe network topological structure of prediction [22–24]. theytohave rarely been to predict performance of TF-CWs. constructed wetlands, the process ofthe TF-CWs, to validate concentrations of In this study, weto setfurther up a BPsimulate network model to predict removal and of pollutants from the TF-CWs with influent modes. WeThe used thisofmodel to construct artificial network nitrogen four and different phosphorus in effluents. aim this study was tothe predict theneural quality of the effluent topological structure of constructed wetlands, to further simulate the process of TF-CWs, and to discharged from TF-CWs and to optimize the nitrogen and phosphorus removal efficiencies. 2.

validate the concentrations of nitrogen and phosphorus in effluents. The aim of this study was to predict the quality of the effluent discharged from TF-CWs and to optimize the nitrogen and Materials and Methods phosphorus removal efficiencies.

The TF-CW simulator (Beijing, China) used in this study was a cylindrical steel drum, 110 cm 2. Materials and Methods high, with a diameter and volume of 40 cm and 138 L, respectively (Figure 1). The device had 21 The TF-CW simulator (Beijing, China) used in this study was a cylindrical steel drum, 110 cm sampling ports, each with a diameter of 1 cm. These ports were equally spaced around the device in high, with a diameter and volume of 40 cm and 138 L, respectively (Figure 1). The device had 21 seven rows that were arranged at 15-cm intervals from the top to the bottom. The inflow water was sampling ports, each with a diameter of 1 cm. These ports were equally spaced around the device in controlled by rows a water pump that was monitored time (TB-125, LUEABB, seven that were arranged at 15-cm intervals by froma the topcontroller to the bottom. The inflow water wasWenzhou, controlled bywas a water pump that monitored cover by a time controller (TB-125, LUEABB, China). The device covered withwas a sprinkler that had small holes on its Wenzhou, bottom to ensure China). The device was covered with a sprinkler cover that had small holes on its bottom ensure uniform water flow. The effluent water was controlled by electromagnetic valve (2 Wtonormally closed uniform water flow. The effluent water was controlled by electromagnetic valve (2 W normally closed 220 V, LUEABB, Wenzhou, China) and was monitored by a time controller. Zeolite, with a particle 220 V, LUEABB, Wenzhou, China) and was monitored by a time controller. Zeolite, with a particle size of between 4 and 48and mm andand a porosity wasfilled filled the device to a ofheight size of between 8 mm a porosityof of 42%, 42%, was intointo the device to a height 105 cm.of 105 cm. The zeolite planted withwith Typha orientalis. Thewas zeolite was planted Typha orientalis.

Figure 1. Experimental tidal flow constructed wetland (TF-CW) setup.

Figure 1. Experimental tidal flow constructed wetland (TF-CW) setup.

Water 2018, 10, 83

3 of 17

We simulated the sewage of Beijing using ammonium chloride, glucose, and monopotassium phosphate to represent total nitrogen (TN), TOC, and total phosphorus (TP), respectively. The volume of artificial sewage stored was 4 m3 . The concentrations of the main water quality indicators in the simulated sewage are shown in Table 1. Table 1. Water quality of influent wastewater. ρ/(mg/L)

Index TN Range

30~50

NH4

+ -N

pH

NO3 − -N

TOC

TP

0~5

0~20

10~16

30~50

7~9

In this experiment, we studied the effects of nutrient removal and the factors that influenced nutrient removal under different influent modes in four test devices (A–D). We set up three different tidal water inflow models (B, C, and D), and a continuous influent mode as a control (A). The input cycle of the four modes was 24 h. Each of the four installations received the same amount of flow (0.138 m3 /day). Details of the hours of operation are presented in Table 2. We conducted a preliminary experiment from January to March 2014. We then attached the membrane to the substrate and commenced the experiment formally in April 2014. Table 2. Time settings of the different water inputs to the TF-CW simulator. Number

Inflow Mode

Idle Time (h)

Reaction Time (h)

Idle/Reaction Time

A B C D

Continue flow Tidal flow Tidal flow Tidal flow

— 12 8 16

— 12 16 8

— 1:1 1:2 2:1

Note: The idle time is the time used to inject water into the equipment. The reaction time indicates the time for the biochemical reaction to occur in the wetland.

Water samples from the sampling ports at seven treatment depths (15, 30, 45, 60, 75, 90, and 105 cm) in Devices A, B, C, and D and the sewage pool were collected weekly from April 2014 to November 2014 and sent to the laboratory for analysis. Temperature, DO, oxidation reduction potential (ORP), pH, and conductivity (Cond) were measured by a portable multi-parameter water quality analyzer (YSI-EXO, YSI, Yellow Springs, OH, USA). The concentrations of TN, NH4 + -N, nitrite nitrogen (NO2 − -N), NO3 − -N, and TP were determined by an automatic discrete analyzer (SMARTCHEM200, WestCo, Frepillon, France). We used Origin 8.5 (OriginLab, Northampton, MA, USA) for data fitting and SPSS version 19.0 (SPSS, Chicago, IL USA) for statistical analysis. Statistically significant differences were determined at the α = 0.05 significance level. RDA analyses were carried out within Canoco4.5 (Microcomputer Power, Ithaca, NY, USA) and we constructed the plot in CanoDraw4.5 (Microcomputer Power, Ithaca, NY, USA). The BP neural network was simulated in Matlab R2011b (MathWorks, Natick, MA, USA). 3. Results and Discussion 3.1. Nutrient Removal Effect As shown in Figure 2, the TN removal rates in the four different influent conditions in the TF-CW simulators exceeded 65%, and the mean TN removal rate of the continuous flow simulator was 82 ± 5%. When the idle time/reaction time ratios were 1:1 (B), 1:2 (C), and 2:1 (D), the mean TN removal rates were 85 ± 5%, 86 ± 4%, and 90 ± 3%, respectively (Table 3). The removal was optimal for an idle time/reaction time ratio of 2:1 (D) under tidal flow conditions.

Water 2018, 10, 83

4 of 17

Water 2017, 9, 83 Water 2017, 9, 83

4 of 21 4 of 21

Figure 2. Removal rates of total nitrogen (TN) in the four different devices (A–D). Figure2.2.Removal Removalrates rates of of total nitrogen (A–D). Figure nitrogen (TN) (TN)in inthe thefour fourdifferent differentdevices devices (A–D). + + concentrations in the effluent were statistically significantly lower than those in The NH The NH -N concentrations in the effluent were statistically significantly lower than those in the 4 4+-N The NH 4 -N concentrations in the effluent were statistically significantly lower than those in the + -N removal rate was higher than 90% and the removal efficiency theinfluent influent(Figure (Figure The 3).3). The NHNH 4+-N removal rate was higher than 90% and the removal efficiency was influent (Figure 3). The NH4+-N4 removal rate was higher than 90% and the removal efficiency was + -N greater than that ofofTN. The removal efficiency ofofNH 4+-N in in thethe tidal flow constructed wetland was greater than that TN. The removal efficiency NH tidal flow constructed wetland 4 + greater than that of TN. The removal efficiency of NH4 -N in the tidal flow constructed wetland + + exceeded 98%, and the NH 4 -N removal removal rates were significantly different between the four influent exceeded 98%, weresignificantly significantlydifferent differentbetween between four influent exceeded 98%,and andthe theNH NH4 4+-N -N removal rates were thethe four influent conditions (p < 0.05)(Table (Table3). 3). conditions (p < 0.05) conditions (p < 0.05) (Table 3).

Figure 3. Removal rate of NH4+-N in the four different devices (A–D). Figure 3. Removal rate of NH4+-N in the four different devices (A–D). Figure 3. Removal rate of NH4 + -N in the four different devices (A–D). Table 3. Removal efficiency of nutrients in the different TF-CW simulators. Table 3. Removal efficiency of nutrients in the different TF-CW simulators.

Inflow Mode TN Removal Rate(%) NH4++-N Removal Rate (%) TP Removal Rate (%) Inflow Mode Removal Rate(%)of nutrients NH4 -N Removal Rate TF-CW (%) TP Removal Rate (%) TableTN 3. Removal efficiency in the different simulators. A 82 ± 5 a 95 ± 3 a 55 ± 14 a A 82 ± 5 a 95 ± 3 a 55 ± 14 a b b B 85 ± 5 98 ± 1 b 57 ± 16 aa Rate (%) Inflow Mode TN Removal NH4 + -N Rate (%) TP B 85 ± 5 b Rate(%) 98 ±Removal 1 57Removal ± 16 b b C 86 ± 4 b a 99 ± 1 b a 56 ± 19 aa C 86 ± 4 99 ± 1 56 ± 19 A 82 ± 5 95 ± 3 55 ± 14 a D 90 ± 3 cc b 99 ± 1 cc b 58 ± 13 aa D 90 ± 3 99 ± 1 58 ± 13 B 57 ± 16 a 85 ± 5 98 ± 1 Note: Differences between a, b, and c indicate significant differences at the 0.05 level, the average b b C ± 19 a ±± 4 c indicate significant differences 99 ± 1 Note: Differences between a,86b,±and at the 0.05 level, the56 average c c standard deviation. D 90 ± 3 99 ± 1 58 ± 13 a standard deviation. Note: Differences between a, b, and c indicate significant differences at the 0.05 level, the average ± standard The NO3−-N concentrations in the effluent were higher than those in the influent from the four deviation. The NO3−-N concentrations in the effluent were higher than those in the influent from the four

simulations, mainly because of NO3−-N produced during reactions. The removal rates were negative simulations, mainly because of NO3−-N produced during reactions. The removal rates were negative and the difference between the inflow and outflow concentrations in the continuous flow setting was − andThe theNO difference between the inflow outflow concentrations the continuous flow setting was in theand effluent were higher thaninthose in the influent from the four 3 -N concentrations − simulations, mainly because of NO3 -N produced during reactions. The removal rates were negative

Water 2018, 10, 83

5 of 17

Water 2017, 9, 83

5 of 21

and the difference between the inflow and outflow concentrations in the continuous flow setting was greater than the between the concentrations tidal simulation Water 2017, 9, 83difference 5 of 21 greater than the difference between theinflow inflowand and outflow outflow concentrations in in thethe tidal flowflow simulation devices (Figure 4). 4). devices (Figure greater than the difference between the inflow and outflow concentrations in the tidal flow simulation devices (Figure 4).

Figure 4. NO3−-N removal rates in the four different devices (A–D).

Figure 4. NO3 − -N removal rates in the four different devices (A–D). 4. NO 3−-N removal rates in theThe fourresults different In TF-CWs, TP Figure is mainly absorbed by the matrix. in devices Table 3(A–D). show that the TP removal efficiencies theisTF-CW under different influent conditions (A, B,3 C, D) were ± 14%, 57 In TF-CWs,ofTP mainlysimulator absorbed by the matrix. The results in Table show that55 the TP removal In TF-CWs, TP is mainly absorbed by the matrix. The results in Table 3 show that the TP removal ± 16%, 56 ± 19%, and 58 ± 13%, respectively (Figure 5). The TP removal rate did not significantly efficiencies of the TF-CW simulator under different influent conditions (A, B, C, D) were 55 ± 14%, efficiencies of the TF-CW simulator under different influent conditions (A, B, C, D)removal were 55 ± 14%, 57 tidal conditions and the natural zeolite did not of TP 57 ± improve 16%, 56 under ± 19%, andflow 58 ± 13%, respectively (Figure 5). The TP improve removalthe rate did notrate significantly ±from 16%,the 56sewage. ± 19%, and 58 ± 13%, respectively (Figure 5). The TP removal rate did not significantly improve under tidal flow conditions and the natural zeolite did not improve the removal rate of TP improve under tidal flow conditions and the natural zeolite did not improve the removal rate of TP from from the sewage. the sewage.

Figure 5. Total phosphorus (TP) removal rates in the four different devices (A–D). Figure 5. Total phosphorus (TP) removal rates in the four different devices (A–D). 3.2. Building the BP Artificial Neural Network Figure 5. Total phosphorus (TP) removal rates in the four different devices (A–D). The development of a nonlinear function fitting algorithm of a BP neural network can be divided 3.2. Building the BP Artificial Neural Network into three as follows: Construction, 3.2. Building thesteps BP Artificial Neural Networktraining, and prediction [21]. The details of the algorithm development of a6.nonlinear function fitting algorithm of a BP neural network can be divided flow The are shown in Figure This model uses powerful data recognition and simulation capabilities to into three steps as follows: Construction, training, and prediction [21]. The details of the algorithm The development of a nonlinear function fitting algorithm of a BP neural network can be divided solve nonlinear system problems. The input and output data are trained on the BP neural network so flow are shown incan Figure 6. This model uses powerful data recognition andtrained simulation capabilities to into three steps as follows: Construction, training, prediction [21]. The details of the algorithm that the network express the unknown black boxand function and then the BP neural network solve nonlinear system problems. The input and output data are trained on the BP neural network so can shown be used in to Figure predict the system outputs. flow are 6. This model uses powerful data recognition and simulation capabilities to the network can express the unknown black box function and then the trained BP neural network solvethat nonlinear system problems. The input and output data are trained on the BP neural network so can be used to predict the system outputs.

that the network can express the unknown black box function and then the trained BP neural network can be used to predict the system outputs.

Water 2018, 10, 83

6 of 17

Water 2017, 9, 83

6 of 21

Water 2017, 9, 83

6 of 21

Figure 6. Backpropagation (BP) neural network algorithm process.

Figure 6. Backpropagation (BP) neural network algorithm process. It is difficult to control the basic chemical transformations in ecological engineering applications, pollutant wastewater is a highly complicated process. There is considerable variation It isand difficult torecycling control in the basic chemical transformations in ecological engineering applications, Figure 6. Backpropagation (BP) neural network algorithm process. and it is impossible in the many factors that influence the removal efficiency of different pollutants, and pollutant recycling in wastewater is a highly complicated process. There is considerable variation to consider the wide range of potential influences as model inputs. In this study, we used redundancy in the many factors that influence the removal of different pollutants, and it is impossible to It is difficult to control the basic chemicalefficiency transformations in ecological engineering applications, analysis (RDA), based on the actual effluent concentrations, to screen each of the indicators for their andwide pollutant recycling in wastewater is a highlyas complicated process.In There considerable variation consider the range of potential influences model inputs. thisisstudy, we used redundancy influence on the removal efficiency of nutrients in the TF-CW simulator (Figure 7). Redundancy in the many factors that actual influence the removal efficiency of different pollutants, and it is indicators impossible for their analysisanalysis (RDA), based on the effluent concentrations, to screen each of the can extract and summarize the variation in a set of response variables that can be explained to consider the wide range of potential influences as model inputs. In this study, we used redundancy influence theof removal efficiency inconsidered the TF-CW simulator version (Figureof7).principal Redundancy by on a set explanatory variables.ofItnutrients can also be a constrained analysis (RDA), based on the actual effluent concentrations, to screen each of the indicators for their components analysis (PCA), where the canonical axes—built from linear combinations of response analysis can extractonand the variation in in a set response variables can be explained by influence thesummarize removal efficiency of nutrients the of TF-CW simulator (Figurethat 7). Redundancy also be linear combinations of theinexplanatory variables [25]. thatof analysis canvariables. extract and summarize a set response variables can be explained a set of variables—must explanatory It can alsothe bevariation considered a of constrained version principal components by a set of explanatory variables. It can also be considered a constrained version of principal analysis (PCA), where the canonical axes—built from linear combinations of response variables—must components analysis (PCA), where the canonical axes—built from linear combinations of response also be linear combinations of the explanatory variables [25]. variables—must also be linear combinations of the explanatory variables [25].

Figure 7. RDA ordinations for various environmental factors in the TF-CW simulator. Note: TNRE: TN removal rate; NH4RE: NH4+-N removal rate; NO3RE: NO3−-N removal rate; TPRE: TP removal rate; TNIN: TN7.influent concentration; NH4IN: NH4+-N influent NO3IN: NO 3−-NTNRE: influent Figure RDA ordinations for various environmental factorsconcentration; in the TF-CW simulator. Note: Figure 7. RDA ordinations for various environmental factors in the TF-CW simulator. Note: TN removal rate; NH 4RE: NH4+-N removal rate;TOC: NO3RE: NOorganic 3−-N removal rate; TPRE: TP removal rate;TNRE: TN concentration; TPIN: TP influent concentration; total carbon; ORP: oxidation–reduction + − + − removalpotential; rate; NH NH4concentration; -Nhours; removal NO4 3-N RE: NOtime; -NDO: removal rate; TPRE: TP removal rate; TNIN: TN influent NHrate; 4IN: time/emptying NH influent NO 3IN: NO3 -N influent Time: operation RAT: idle dissolved oxygen; Sal: salinity; 4 RE: 3 concentration; + -N concentration; TPIN: TP influent concentration; TOC: total organic carbon; ORP: oxidation–reduction Cond: conductivity; Temp: water temperature; depth ofconcentration; treatment. TNIN: TN influent concentration; NH influent NO3 IN: NO3 − -N influent 4 IN: NH4 Depth: potential; Time: operation hours; RAT: idle time/emptying time; DO: dissolved oxygen; Sal: salinity;

concentration; TPIN: TP influent concentration; TOC: total organic carbon; ORP: oxidation–reduction Cond: conductivity; Temp: water temperature; Depth: depth of treatment. potential; Time: operation hours; RAT: idle time/emptying time; DO: dissolved oxygen; Sal: salinity; Cond: conductivity; Temp: water temperature; Depth: depth of treatment.

Figure 7 shows that the TN and NH4 + -N removal rates were not affected by the operating time, but the removal rate of NO3 − -N was inversely proportional to the operating time. The removal efficiencies

Water 2018, 10, 83

7 of 17

of TN and NH4 + -N were stable, but the removal rate of NO3 − -N decreased gradually. The tidal flow device can effectively purify wastewater with NH4 + -N as the main pollutant [26]. The aging rate was slow, and the removal over time was consistent. When RDA was applied to continuous flow, the idle time/emptying time (RAT) value was denoted by 0. The TN and NH4 + -N removal rates were positively correlated with RAT. The NO3 − -N influent concentration was close to 0, but when coupled with the nitrification of + NH4 -N, the NO3 − -N concentration in the outflow was higher than in the inflow. Therefore, the NO3 − -N removal rates were negative, but were positively correlated with the NO3 − -N influent concentrations. Removal rates of NO3 − -N should be studied in more detail for situations where NO3 − -N is present at higher concentrations and is the main pollutant. The results from RDA showed that RAT was positively correlated with DO, which indicates that TF-CWs had a higher re-oxygenating capacity than continuous flow constructed wetlands. The DO concentrations were positively correlated with the TN and NH4 + -N removal rates, which shows that DO had a strong influence on the nutrient removal efficiency from wastewater when NH4 + -N was the main pollutant [27,28]. Many factors are known to affect NO3 − -N removal, including pH, conductivity, water temperature, and salinity. When the pH is either too high or too low, it has a large effect on the activity of wetland microorganisms. Temperature affects nitrogen removal by denitrification in constructed wetlands, and the rate decreases when the temperature falls below 5 ◦ C. In contrast to other similar studies [29], the NO3 − -N removal rate in this study was not significantly correlated with TOC concentrations. The concentrations of TOC in the four devices were low, and thus the correlation between these two parameters was not accurately reflected. The RDA results showed that the NH4 + -N removal rates were negatively correlated with the TOC concentrations. In this study, the TOC concentrations (between 0 and 20 mg/L) were too low to provide the energy needed for NO3 − -N denitrification. Therefore, we need to carry out further studies to examine nitrogen removal in high TOC concentrations and to find a way to maintain the C/N ratio at a level that is conducive to improved nitrogen removal. The TN removal rates were positively correlated with pH, conductivity, water temperature, and salinity; the NH4 + -N removal rates were positively correlated with depth, and removals of both TN and NH4 + -N were negatively correlated with ORP and TOC (Table 4). The treatment depth had a major influence on TP removal [30], with a correlation coefficient of 0.65. Other factors did not have major roles, which indicates that substrate adsorption and sedimentation played a leading role in TP removal. The RDA results indicated that different factors influenced the removals of TN, NH4 + -N, and NO3 − -N. Therefore, we selected different training samples for the different models, as follows: the data collection times were as close as possible to ensure the data were effective, the samples were as diverse as possible, and significant deviations were removed from the actual data because of their influence on the neural network training effect. The model was divided into training data and validation data [31]. The data were divided into 112 groups of training data and 56 sets of validation data using a randomly generated method. The various indicators of the sample data may not fall within the same order of magnitude, which means that the network cannot converge and thus the time to train the network needs to be increased. To obtain better training results, the index data should be within the same order of magnitude. Before running the model, all the data were mapped to between 0 to 1 and the range of changes of the index were unified, to improve the training efficiency of the network and avoid differences in the order of magnitude that would affect the accuracy of the network recognition. The MATLAB toolbox has many normalization functions, including zscore, premnmx, mapminmax, and postmnmx functions; of these, we used the mapminmax function for the normalization. The output predicted data were the output values of the normalized data, and so the outputs from the network had to be anti-normalized. x∗ =

x − min max − min

(1)

Water 2018, 10, 83

8 of 17

When the output layer of the model was the TN concentration in the effluent, the input layers were DO, RAT, ORP, and TOC. When the NH4 + -N concentration was the output layer, the input layers were DO, RAT, ORP, and Depth. When NO3 − -N was the output layer, Cond, Temp, Sal, and pH were the input layers. The input layers were DO, RAT, Time, and Depth when the output layer was the TP effluent concentration. The number of nodes in the hidden layer is very important. The fitting ability of the network improves as the number of nodes in the hidden layer increases. When the number of hidden nodes is small, the network fitting information cannot meet the demand. If the number of hidden layer nodes is infinite, any nonlinear mapping can be implemented from the input to the output. However, the number and training time are positively related. If time is not considered, there may be problem of overfitting, in which an increase in the test error leads to a decrease in the generalization capacity [32]. Therefore, the method used to determine the number of hidden layer nodes is very important, and is a complex problem. It has a certain relationship between the input and output layers, but it does not provide a good formula for deriving this number. In practice, we can combine empirical formulae to calculate the range of hidden layer nodes, and then use the trial and error method to determine the actual number. The empirical formula most commonly used is [32]: N=

√

n+m+a

(2)

where N is the number of hidden layer nodes, n is the number of input layers, m is the number of output layers, and a is a constant (1 to 10). The empirical formula calculation showed that the number of nodes in the hidden layer should be between 3 and 12, and thus the trial-and-error method was then used to train and verify the number of all the network nodes between 3 and 12, and the numbers were tested in the network. The number of nodes was tested 3 to 5 times, and the average coefficient value derived from the 3 to 5 determination coefficients obtained was used as the determination coefficient of the number of nodes. The optimal number of hidden layer nodes was selected from the coefficient. The relationship between the number of hidden layers and the coefficient of determination (R2 ) is shown in Figure 8. The coefficient of determination is a non-unit parameter that can directly determine the merits of the fitting indicators. In general, the greater the goodness of fit is, the higher the power of the independent variable to explain the dependent variable [33]. Figure 8 shows that the number of nodes increased as the number of hidden layers increased, and there was a clear increase in the R2 of TN when the number of hidden nodes was 9. The training effect was best when TN had 9 hidden layers. The R2 value reached 0.73 ± 0.02, gradually stabilized, and then decreased slightly. For NH4 + -N, when the number of hidden nodes increased from 3 to 4, the training effect improved noticeably, but then the coefficient of the hidden layer started to fluctuate as the hidden layer nodes increased. When the number of hidden nodes reached 11, R2 was maximized and reached 0.63 ± 0.04, which was less than the value for the TN training data. When training the NO3 − -N data, the effect of the training data increased when the number of nodes in the hidden layer increased. We chose to use 12 hidden layer nodes in the NO3 − -N network topology, and achieved an R2 of 0.71 ± 0.06. For TP, there was no obvious change as the number of nodes in the hidden layer increased. The overall R2 was high, and reached 0.93 ± 0.02 when the number of hidden nodes was 9. We therefore used 9 hidden layer nodes when predicting the TP effluent concentrations. Once the input and output layers were determined, the number of nodes in the hidden layer were further selected, and the corresponding topological structure of the artificial neural network model of tidal flow was established. The main influences on TN, NH4 + -N, NO3 − -N, and TP, as indicated by the RDA, were selected as the input layers of the model; there were four input layers and one output layer. Therefore, there were 9, 11, 12, and 9 hidden layer nodes in the input layers of the neural network model, respectively, and one neuron in the output layer.

Water 2018, 10, 83

9 of 17

Table 4. Correlation coefficient matrix. Index

NO3 RE NH4 RE TNRE

NO3 RE NH4 RE TNRE TPRE NO3 IN NH4 IN TNIN TPIN TOC RAT DO Temp Cond Sal pH ORP Time Depth

0.18 −0.03 −0.29 0.52 −0.37 −0.27 0.12 0.07 0.04 0.10 0.25 0.33 0.35 0.46 −0.37 −0.53 −0.38

0.06 −0.17 −0.24 −0.10 −0.13 0.01 −0.35 0.2 0.19 −0.33 −0.30 −0.13 −0.18 −0.47 0.18 0.43

−0.02 −0.36 −0.25 −0.12 −0.67 −0.30 0.12 0.10 −0.18 −0.27 −0.12 −0.26 −0.42 −0.18 0.04

TPRE

0.11 0.05 0.04 0.08 0.03 0.19 0.25 0.15 0.10 0.01 0.12 0.07 0.18 0.65

NO3 IN NH4 IN TNIN

0.75 0.63 0.35 −0.17 0.00 0.01 0.65 0.75 0.17 −0.45 0.07 −0.03 0.00

0.87 0.29 −0.28 0.00 −0.02 0.74 0.61 0.03 −0.54 0.01 0.30 0.00

0.04 0.24 0.00 0.02 0.47 0.39 0.11 0.53 0.05 0.40 0.00

TPIN

TOC

RAT

DO

Temp

Cond

Sal

pH

ORP

Time

0.32 0.00 0.01 0.47 0.47 0.32 0.35 0.07 0.13 0.00

0.16 0.10 0.18 0.13 0.06 0.39 0.07 0.15 0.56

0.92 0.05 0.16 0.11 0.06 0.18 0.00 0.00

0.04 0.16 0.13 0.07 0.22 0.04 0.06

0.76 0.20 0.27 0.12 0.16 0.22

0.52 0.14 0.21 0.30 0.11

0.12 0.15 0.29 0.04

−0.15 −0.52 −0.13

0.32 −0.10

0.00

Depth

Water 2018, 2017, 10, 9, x83 FOR PEER REVIEW

17 10 of 21

Figure 8. Relationship between the coefficients of determination (R2) and the number of hidden layers. Figure 8. Relationship between the coefficients of determination (R2 ) and the number of hidden layers.

3.3. Comparison of the Predicted and Measured Values 3.3. Comparison of the Predicted and Measured Values We collected 168 groups of inflow and outflow data, among which we randomly selected 112 sets We collected 168 groups of inflow and outflow data, among which we randomly selected 112 sets of data as training samples, and other 56 groups as test samples. The model parameters described in of data as training samples, and other 56 groups as test samples. The model parameters described in Section 2 were used to train the network. The predicted and measured values are shown in Figure 9. Section 2 were used to train the network. The predicted and measured values are shown in Figure 9. The results show that the model established by the BP artificial neural network gave reasonable The results show that the model established by the BP artificial neural network gave reasonable predictions of the nutrient concentrations in the effluent. While there were some errors, some of the predictions of the nutrient concentrations in the effluent. While there were some errors, some of the predicted and actual values were significantly correlated, and there was minor deviation between the predicted and actual values were significantly correlated, and there was minor deviation between predicted concentrations in the output water and the measured effluent concentrations. The four the predicted concentrations in the output water and the measured effluent concentrations. The four indicators of the predicted and the measured values showed that there were irregular changes in the indicators of the predicted and the measured values showed that there were irregular changes in the concentrations of the output water, mainly because of how the data were arranged. When the training concentrations of the output water, mainly because of how the data were arranged. When the training model was used to process data for different treatment depths and different water ways, the data model was used to process data for different treatment depths and different water ways, the data were disrupted and sorted randomly. The first 112 groups were extracted as training data, while the were disrupted and sorted randomly. The first 112 groups were extracted as training data, while the last 56 groups were extracted as the validation data. The data were independent and there were no last 56 groups were extracted as the validation data. The data were independent and there were no relationships among them, and thus there were no regular changes in the predicted output water relationships among them, and thus there were no regular changes in the predicted output water concentrations. When analyzing continuous data, the data should not be sorted randomly with the concentrations. When analyzing continuous data, the data should not be sorted randomly with the rand function. rand function.

Water 2018, 10, x; doi: FOR PEER REVIEW

www.mdpi.com/journal/water

Water 2018, 10, 83 Water 2018, 10, x FOR PEER REVIEW

11 of 17 11 of 17

Figure 9. Fitting curve values and model output values. Figure 9. Fitting curve values and model output values.

3.4. Training Errors in the BP Neural Network 3.4. Training Errors in the BP Neural Network The prediction error of the BP network and the percentage prediction error are shown in Figure The prediction error of the BP network and the percentage prediction error are shown in Figure 10. 10. There There were were some some errors errors in in the the effluent effluent concentrations concentrations of of each each index index predicted predicted by by the the BP BP neural neural network. The network prediction error percentage and the prediction error fluctuated significantly network. The network prediction error percentage and the prediction error fluctuated significantly (p By comparing comparing Figures Figures 99 and and 10, 10, we we find find that, when the (p >> 0.05). 0.05). By that, when the measured measured value value was was high, high, the the error error in the predicted value was very obvious and the prediction percentage error was large; when in the predicted value was very obvious and the prediction percentage error was large; when the the measured measured value value was was low, low, the the network network prediction prediction error error and and the the percentage percentage network network prediction prediction error error were When there was no relationship betweenbetween the training and the verification wereboth bothlow. low. When there wasfunctional no functional relationship thedata training data and the data, the BP neural network had a poor predictive ability, and the network only performed well when verification data, the BP neural network had a poor predictive ability, and the network only the data were regular. The network prediction error in the TN effluent concentrations ranged from performed well when the data were regular. The network prediction error in the TN effluent − 15 to 15 mg/L,ranged and, with prediction −5prediction and 5 mg/L, thebetween overall network concentrations from80% −15of tothe 15 mg/L, and,error with between 80% of the error −5 and 5 + -N effluent concentrations ranged from error was small. The error associated with the predicted NH 4 mg/L, the overall network error was small. The error associated with the predicted NH4+-N effluent − 8 to 8 mg/L, and 90%from of the was of between −4 anderror 4 mg/L. The obvious variations concentrations ranged −8predicted to 8 mg/L,error and 90% the predicted was between −4 and 4 mg/L. and the high relative errors resulted from the low effluent concentrations. The errors associated The obvious variations and the high relative errors resulted from the low effluent concentrations.with The − the predicted NOwith effluent concentrations were mostly betweenwere −6 and 6 mg/L. The−6results 3 -Nthe errors associated predicted NO3−-N effluent concentrations mostly between and 6 suggest that the BP neural network wasBP better at predicting effluent TP than the mg/L. The results suggest that the neural networkthe was betterconcentrations at predictingofthe effluent other nutrients since the errors associated with the predicted TP concentrations were small and 90% of concentrations of TP than the other nutrients since the errors associated with the predicted TP the TP prediction errors were −2 and 1 mg/L. errors were between −2 and 1 mg/L. concentrations were small andbetween 90% of the TP prediction

Water 2018, 10, 83 Water 2018, 10, x FOR PEER REVIEW

12 of 17 12 of 17

Figure 10. Prediction error and the percentage prediction error of the BP network. Figure 10. Prediction error and the percentage prediction error of the BP network.

3.5. Network Fitting Ability 3.5. Network Fitting Ability The fitting ability of the network is an indication of how well the constructed BP neural network The fitting ability of the network is an indication of how well the constructed BP neural network fits to the topological structure. The data were automatically divided into training data, validation fits to the topological structure. The data were automatically divided into training data, validation data, and test data during the training and prediction. When calculating the fit between the measured data, and test data during the training and prediction. When calculating the fit between the measured and output values of each data, the R22 represented the fitting effect, the fit line represented the fitting and output values of each data, the R represented the fitting effect, the fit line represented the fitting curve, Y = T represented the fitted line when R22 = 1, and the degree of fitness decreased as the curve, Y = T represented the fitted line when R = 1, and the degree of fitness decreased as the deviation increased. deviation increased. The fitting of the BP neural network to the TN effluent concentrations is shown in Figure 11. The The fitting of the BP neural network to the 2TN effluent concentrations is shown in Figure 11. 2 R value for the training data was 0.70, and the R value for the validation data was 0.61. The fitting The R2 value for the training data was 0.70, and the R2 value for the validation data was 0.61.2 The fitting coefficient of the test data network was 0.67 and the total effect was very good, with an R2 of 0.67. coefficient of the test data network was 0.67 and the total effect was very good, with an R of 0.67.

Water 2018, 10, 83

13 of 17

Water 2018, 10, x FOR PEER REVIEW

13 of 17

Water 2018, 10, x FOR PEER REVIEW

13 of 17

Figure 11. Fitting of the BP neural network to the TN effluent concentrations. Figure 11. Fitting of the BP neural network to the TN effluent concentrations. Figure 11. Fitting of the BP neural network to the TN effluent concentrations.

The fitting of the BP neural network to the NH4+-N effluent concentrations is shown in Figure 2 values The fitting of the BP neural to thevalidation NH4 + -N effluent shown in Figure 12. The Rfitting for fitting thenetwork training and wereconcentrations 0.80 and 0.56, is respectively. The R212. The of the BP neural network to the NH4+-Ndata effluent concentrations is shown in Figure 2 The12. R The values for fitting the training and validation data data were 0.80 and 0.56,0.56, respectively. The The R2 value value ofRthe validation data wastraining relatively because of the lack0.80 of data. The respectively. fitting coefficient 2 values for fitting the andlow validation were and Rof2 2 value of value the validation data was relatively low because of the lack of data. The fitting coefficient of the test data network was 0.68, and the overall effect was very good with an R of 0.73. of the validation data was relatively low because of the lack of data. The fitting coefficient oftest data network 0.68, was and 0.68, the overall was very good with anwith R2 value 0.73.of 0.73. the test datawas network and theeffect overall effect was very good an R2 of value

Figure 12. Fitting of the BP neural network to the NH4+-N effluent concentrations. Figure 12. Fitting of the BP neural network to the NH4+-N effluent concentrations. Figure 12. Fitting of the BP neural network to the NH4 + -N effluent concentrations.

Water 2018, 10, 83

14 of 17

Water 2018, 10, x FOR PEER REVIEW

14 of 17

The fitting of the BP neural network to the NO3 − -N effluent concentrations is shown in Figure 13. The fitting of the BP neural network to the NO3−-N effluent concentrations is shown in Figure 2 of the training data was 0.73 and, with an R2 of 0.74, fitted well with the validation data. The R 13.Water The2018, R2 of training data was 0.73 and, with an R2 of 0.74, fitted well with the validation 10,the x FOR PEER REVIEW 14 of data. 17 2 of 0.53. The overall R2 was 0.69. TheThe fitting ability of the test data was poor with an R 2 2 fitting ability of the test data was poor with an R of 0.53. The overall R was 0.69. The fitting of the BP neural network to the NO3−-N effluent concentrations is shown in Figure 13. The R2 of the training data was 0.73 and, with an R2 of 0.74, fitted well with the validation data. The fitting ability of the test data was poor with an R2 of 0.53. The overall R2 was 0.69.

Figure 13. Fitting of the BP neural network to the NO3−-N effluent concentrations. Figure 13. Fitting of the BP neural network to the NO3 − -N effluent concentrations. Fitting of the BP11–13) neural network to the 3 -N effluent concentrations. The aboveFigure three 13. figures (Figures show that theNO quality of the fit of the BP neural network The above three figures (Figures 11–13) show that the quality fitrelative of the BP neural network differed for each of the three nitrogen forms. Although there wasofa the large error associated Theeach above three figures (Figuresforms. 11–13)Although show that the quality oflarge the fitrelative of the BP neural network with differed for of the three nitrogen there was a error associated + with the NH4 -N fitting, it fitted better to the network than the other two nitrogen forms. Therefore, differed for eachitoffitted the three forms. Although a large relative error associatedwhen + -N fitting, thewhen NH betternitrogen towe theshould network thethere otherwas two nitrogen forms. Therefore, 4 using BP neural networks, notthan just consider the network error or the percentage with the NH4+-N fitting, it fitted better to the network than the other two nitrogen forms. Therefore, using BP neural networks, we should not just consider the network error or the percentage network network error, but we should also consider the network fitting ability. when using BP neural networks, we should not just consider the network error or the percentage 2 error, but should also consider the network fitting ability. Thewe BPerror, neural network fitted well to the TP network but we should also consider theeffluent networkconcentrations fitting ability. (Figure 14). The R of the overall 2 of the 2 The ability BP neural network fitted well to effluent concentrations (Figure The R fitting 0.90 and the well R oftothe training data was 0.91. The validation data had the best fit 2 of The BPreached neural network fitted thethe TP TP effluent concentrations (Figure 14). The R14). the overall 2 2 2 2 the overall fitting reached 0.90 and R training of the training data 0.91. The validation with an Rability ofability 0.92. The overall Rthe ofRthe data was 0.81. The TPwas concentrations in the were fitting reached 0.90 and of test the data was 0.91. The validation data hadeffluent thedata best had fit the 2 the mainly influenced by Depth, effects of theThe other factors were in minimal. Depth was best fit with an R20.92. of 0.92. The overall R the testwas data was 0.81. The TP concentrations in the effluent with an R2 of Thethe overall R2 ofwhile the of test data 0.81. TP concentrations the effluent were therefore the only input layer of the BP neural network for TP, and the fact that the fitting ability of byby the while thethe effects of the other factors were minimal. Depth wasthe were mainly mainlyinfluenced influenced theDepth, Depth, while effects of the other factors were minimal. Depth was TPtherefore effluent concentrations was better than thatnetwork of the other nutrients indicates that networks with only the only input layer theBP BP neural network for fact that the fitting ability of the therefore the only input layer ofof the neural forTP, TP,and andthe the fact that the fitting ability of the TPguiding effluent concentrations was better than thatand of the the other nutrients indicates thatthat networks with with only only factor have awas better fitting ability lower error than other networks. TPone effluent concentrations better than that of other nutrients indicates networks one guiding factor have a better fitting ability and lower error than other networks. one guiding factor have a better fitting ability and lower error than other networks. −

Figure 14. Cont.

Water 2018, 10, 83

15 of 17

Water 2018, 10, x FOR PEER REVIEW

15 of 17

Figure 14. Fitting network to to the the TP TP effluent effluent concentrations. concentrations. Figure 14. Fitting of of the the BP BP neural neural network

4. Conclusions 4. Conclusions In this thisstudy, study,we weexamined examined removal of nutrients in a TF-CW. Wethree usedtidal three tidalmodels inflow In thethe removal of nutrients in a TF-CW. We used inflow models and one continuous flow model to examine the removal rate and the variation of each and one continuous flow model to examine the removal rate and the variation of each pollutant index pollutant at different used select the controls on nutrient removal at differentindex depths. We useddepths. RDA toWe select theRDA maintocontrols onmain nutrient removal and applied themand as applied them as the main factors in the BP neural network for model simulations. the main factors in the BP neural network for model simulations. In constructed constructedwetlands, wetlands,tidal tidaloperation operation can improve removal of nutrients. The tidal In can improve thethe removal raterate of nutrients. The tidal flow flow has a positive effect and refreshes the wetland with fresh air and enhances the dissolved oxygen has a positive effect and refreshes the wetland with fresh air and enhances the dissolved oxygen (DO) (DO) in the system. Such intermittent aeration is morethan effective than aeration, continuous it in the system. Such intermittent aeration is more effective continuous as itaeration, facilitatesasthe facilitates the ofestablishment of aerobic and anaerobic suitable for nitrification and establishment aerobic and anaerobic conditions suitable conditions for nitrification and denitrification [34,35]. denitrification [34,35]. The different nutrients in these four kinds of water treatment methods The different nutrients in these four kinds of water treatment methods behaved differently. The optimal behavedefficiency differently. Theby optimal removal efficiency wetlands of TN by was the tidal flowby constructed removal of TN the tidal flow constructed achieved the influentwetlands method was achieved by the influent with an idle time of from 2:1 (D), was with an idle time/reaction time method of 2:1 (D), which was time/reaction significantly different the which other three significantly different from the other three influent methods (p < 0.05). This treatment also had the + influent methods (p < 0.05). This treatment also had the optimal NH4 -N intake, which reached + − optimal NHThe 4 -N intake, reached 99% ± 1%. The NO 3 -N concentrations in the effluent from the 99% ± 1%. NO3 − -Nwhich concentrations in the effluent from the four simulations were higher than the four simulations were higher than the influent concentrations; removal ratesefficiency were negative, and influent concentrations; the removal rates were negative, and thethe overall removal was better the overall removal efficiency was better in the continuous flow system than in the tidal flow in the continuous flow system than in the tidal flow simulation system. The TP removal rates did not simulation system.between The TP removal rates did(pnot differand significantly between thehigher four devices (p > differ significantly the four devices > 0.05), were not significantly in the tidal 0.05), and were not significantly higher in the tidal flow system. flow system. Our results results show show that that the the ability ability of of the the BP BP artificial artificial neural neural network network model model to to predict predict nutrient nutrient Our concentrations in in the the effluent effluent was was good, good, and and there there were were small small errors errors when when correlating correlating the the predicted predicted concentrations values and the actual values. Of all the indexes, TP had the best fitting ability to the BP neural values and the actual values. Of all the indexes, TP had the best fitting ability to the BP neural 2 2 + − network, with with aa total total RR2 of of 0.90. 0.90. The Theoverall overallRR2 values valuesfor forTN, TN,NH NH+4 -N, -N, and and NO NO3−-N were 0.67, 0.73, network, 4 3 -N were 0.67, 0.73, and 0.69, respectively. Despite the errors, the BP artificial neural network provides a new method for and 0.69, respectively. Despite the errors, the BP artificial neural network provides a new method modeling the effectiveness of TF-CWs. Future studies should explore BP neural networks with for modeling the effectiveness of TF-CWs. Future studies should explore BP neural networks with multiple hidden hidden layers, layers, and and greater greater generalization generalization abilities abilities and and prediction prediction accuracies. accuracies. For For simple simple multiple mapping, a single hidden layer can be selected to achieve greater speed when the network meets the mapping, a single hidden layer can be selected to achieve greater speed when the network meets the precision requirements. For complex mapping, the multiple hidden layer can improve the prediction precision requirements. For complex mapping, the multiple hidden layer can improve the prediction accuracy of the network. We also need to carry out further studies of applications of optimized BP accuracy of the network. We also need to carry out further studies of applications of optimized BP neural networks to TF-CWs based on genetic algorithms. neural networks to TF-CWs based on genetic algorithms. Acknowledgments: This This study study was was supported supported by by the the Fundamental Fundamental Research Research Funds Funds for for the the Central Central Non-Profit Non-Profit Acknowledgments: Research Institution Institutionofof CAF (CAFYBB2014QA029) Lecture and Program Study Program for Outstanding Research CAF (CAFYBB2014QA029) and and “The“The Lecture and Study for Outstanding Scholars from Home andHome Abroad” We thank Alex Boon,Alex PhD,Boon, and Deborah Ballantine, from Liwen Scholars from and(CAFYBB2011007). Abroad” (CAFYBB2011007). We thank PhD, and DeborahPhD, Ballantine, PhD, Bianji, EdanzBianji, GroupEdanz ChinaGroup (www.liwenbianji.cn/ac), for editing thefor English text a draft of this from Liwen China (www.liwenbianji.cn/ac), editing theofEnglish text of manuscript. a draft of this manuscript. Author Contributions: Wei Li and Lijuan Cui conceived and designed the experiments; Yaqiong Zhang and Xinsheng Zhao performed the experiments; Yaqiong Zhang and Manyin Zhang analyzed the data; Weigang Xu, Author Contributions: Wei Li and Lijuanand Cuianalysis conceived andYaqiong designed the experiments; Yaqiong Xu Pan and Jing Li contributed materials tools; Zhang, Wei Li, Zhangjie Cai,Zhang Yinru and Lei, Xinsheng Zhao performed the experiments; Yaqiong Zhang and Manyin Zhang analyzed the data; Weigang Xu, Zhiguo Dou wrote the paper. Xu Pan and Jing Li contributed materials and analysis tools; Yaqiong Zhang, Wei Li, Zhangjie Cai, Yinru Lei, Zhiguo Dou wrote the paper.

Water 2018, 10, 83

16 of 17

Conflicts of Interest: The authors declare no conflict of interest.

References 1. 2. 3.

4. 5.

6. 7. 8. 9. 10. 11. 12.

13. 14. 15. 16. 17. 18. 19. 20. 21. 22.

Behrends, L.; Houke, L.; Bailey, E.; Jansen, P.; Brown, D. Reciprocating constructed wetlands for treating industrial, municipal and agricultural wastewater. Water Sci. Technol. 2001, 44, 399–405. [PubMed] Lee, C.; Fletcher, T.D.; Sun, G. Nitrogen removal in constructed wetland systems. Eng. Life Sci. 2009, 9, 11–22. [CrossRef] Cui, L.J.; Li, W.; Zhang, Y.Q.; Wei, J.M.; Lei, Y.R.; Zhang, M.Y.; Pan, X.; Zhao, X.S.; Li, K.; Ma, W. Nitrogen removal in a horizontal subsurface flow constructed wetland estimated using the first-order kinetic model. Water 2016, 8, 514. [CrossRef] Mitsch, W.J.; Cronk, J.K. Creation and restoration of wetlands: Some design consideration for ecological engineering. Soil Res. 1992, 17, 217–259. Wu, S.; Kuschk, P.; Brix, H.; Vymazal, J.; Dong, R. Development of constructed wetlands in performance intensifications for wastewater treatment: A nitrogen and organic matter targeted review. Water Res. 2014, 57, 40–55. [CrossRef] [PubMed] Zhou, Q.H.; Wang, Y.F.; Wu, Z.B. Progress on study of microorganism in constructed wetlands. Environ. Sci. Technol. 2008, 31, 58–62. Sun, G.; Zhao, Y.; Allen, S. Enhanced removal of organic matter and ammoniacal-nitrogen in a column experiment of tidal flow constructed wetland system. J. Biotechnol. 2005, 115, 189–197. [CrossRef] [PubMed] Zhao, Y.; Sun, G.; Allen, S. Purification capacity of a highly loaded laboratory scale tidal flow reed bed system with effluent recirculation. Sci. Total Environ. 2004, 330, 1–8. [CrossRef] [PubMed] Zhao, Y.; Sun, G.; Lafferty, C.; Allen, S.J. Optimising the performance of a lab-scale tidal flow reed bed system treating agricultural wastewater. Water Sci. Technol. 2004, 50, 65–72. [PubMed] Huang, J.; Reneau, R.B.; Hagedorn, C. Nitrogen removal in constructed wetlands employed to treat domestic wastewater. Water Res. 2000, 34, 2582–2588. [CrossRef] Jing, S.R.; Lin, Y.F. Seasonal effect on ammonia nitrogen removal by constructed wetlands treating polluted river water in southern Taiwan. Environ. Pollut. 2004, 127, 291–301. [CrossRef] Guo, Y.M.; Liu, Y.G.; Zeng, G.M.; Hu, X.J.; Xu, W.H.; Liu, Y.Q. An integrated treatment of domestic wastewater using sequencing batch biofilm reactor combined with vertical flow constructed wetland and its artificial neural network simulation study. Ecol. Eng. 2014, 64, 18–26. [CrossRef] Wang, S.; Zhang, N.; Wu, L.; Wang, Y. Wind speed forecasting based on the hybrid ensemble empirical mode decomposition and GA-BP neural network method. Renew. Energy 2016, 94, 629–636. [CrossRef] Shao, L.M.; Fu, G.; Chao, X.C.; Zhou, J. Application of BP neural network to forecasting typhoon tracks. J. Nat. Disasters 2009, 18, 104–111. Wang, W.; Zhuang, K.; Wu, G.; Zhou, P.; Yang, M.; Li, D.; Jiang, C.; Huang, B. The application of BP neural networks to the new generation expert system for earthquake prediction. Earthquake 1997, 11, 1259–1265. Zhou, Y.; Ruan, R. Application of LM-BP neural networks in remote sensing image classification. Remote Sens. Inf. 2010, 32, 80–86. Li, D.; Wu, G.; Zhao, J.; Niu, W.; Liu, Q. Wireless channel identification algorithm based on feature extraction and BP neural network. J. Inf. Process Syst. 2017, 13, 141–151. [CrossRef] Zhang, Y.; Wu, L. Stock market prediction of S&P 500 via combination of improved BCO approach and BP neural network. Expert Syst. Appl. 2009, 36, 8849–8854. Liu., J.; Jiang, Z.; Wei, J.; Ling, Q. Application of BP neural network model in groundwater quality evaluation. Syst. Eng. Theory Pract. 2000, 20, 124–127. Mo, M.; Wang, X.; Wu, H.; Cai, S.; Zhang, X.; Wang, H. Ecosystem health assessment of Honghu Lake Wetland of China using artificial neural network approach. Chin. Geogr. Sci. 2009, 19, 349–356. [CrossRef] Liang, X.; Chao, F.; Wu, Z.; He, L.; Lu, Y. The application of Neural Network BP to Healthy Assessment of Wetland Ecosystem in Dongting Lake. For. Invent. Plan. 2009, 5, 12. Li, L.; Zhang, H. Assessment model of townlet eco-environmental quality based on BP-artificial neural network. Chin. J. Appl. Ecol. 2008, 19, 2693–2698.

Water 2018, 10, 83

23.

24. 25. 26.

27.

28.

29.

30. 31.

32. 33. 34.

35.

17 of 17

Li, R.; Chen, Z.; Chen, Z.; Zhang, X.; Zheng, L.; Wang, Q. Calculation of ecological recovery based on BP neural network: A case study of Zhuxi Small Watershed in Changting County, Fujian Province. Acta Ecol. Sin. 2015, 35, 1973–1981. He, Y.; He, H.; Meng, C.; Xie, M.; Long, M.; Li, Y. Predicting occurrence degree of rice planthoppers in Guangxi Province based on BP artificial neural network method. Chin. J. Ecol. 2014, 33, 159–168. Csala, A.; Voorbraak, F.P.; Zwinderman, A.H.; Hof, M.H. Sparse Redundancy Analysis of High Dimensional Genetic and Genomic Data. Bioinformatics 2017, 33, 3228–3234. [CrossRef] [PubMed] Hu, Y.S.; Zhao, Y.Q.; Zhao, X.H.; Kumar, J.L.G. Comprehensive analysis of step-feeding strategy to enhance biological nitrogen removal in alum sludge-based tidal flow constructed wetlands. Bioresour. Technol. 2012, 111, 27–35. [CrossRef] [PubMed] Saeed, T.; Sun, G. A review on nitrogen and organics removal mechanisms in subsurface flow constructed wetlands: Dependency on environmental parameters, operating conditions and supporting media. J. Environ. Manag. 2012, 112, 429–448. [CrossRef] [PubMed] Wu, S.; Zhang, D.; Austin, D.; Dong, R.; Pang, C. Evaluation of a lab-scale tidal flow constructed wetland performance: Oxygen transfer capacity, organic matter and ammonium removal. Ecol. Eng. 2011, 37, 1789–1795. [CrossRef] Zhi, W.; Ji, G. Quantitative response relationships between nitrogen transformation rates and nitrogen functional genes in a tidal flow constructed wetland under C/N ratio constraints. Water Res. 2014, 64, 32–41. [CrossRef] [PubMed] Vohla, C.; Kõiv, M.; Bavor, H.J.; Chazarenc, F.; Ülo, M. Filter materials for phosphorus removal from wastewater in treatment wetlands—A review. Ecol. Eng. 2011, 37, 70–89. [CrossRef] Li, W.; Zhang, Y.; Cui, L.J.; Zhang, M.Y.; Wang, Y.F. Modeling total phosphorus removal in an aquatic-environment restoring horizontal subsurface flow constructed wetland based on artificial neural networks. Environ. Sci. Pollut. Res. 2015, 22, 12347–12354. [CrossRef] [PubMed] Shen, H.; Wang, Z.X. Determining the number of BP neural network hidden layer units. J. Tianjin Univ. Tech. 2008, 24, 13–17. Zhang, J. Study on Coefficient of Correlation as Goodness-of-fit Criterion. J. Zhengzhou Univ. Technol. Nat. Sci. Ed. 2001, 22, 72–74. Kizito, S.; Lv, T.; Wu, S.; Ajmal, Z.; Luo, H.; Dong, R. Treatment of anaerobic digested effluent in biochar-packed vertical flow constructed wetland columns: Role of media and tidal operation. Sci. Total Environ. 2017, 592, 197–205. [CrossRef] [PubMed] Ilyas, H.; Masih, I. The performance of the intensified constructed wetlands for organic matter and nitrogen removal: A review. J. Environ. Manag. 2017, 198, 372–383. [CrossRef] [PubMed] © 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).