Available online at www.sciencedirect.com
ScienceDirect Procedia Manufacturing 2 (2015) 307 – 312
2nd International Materials, Industrial, and Manufacturing Engineering Conference, MIMEC2015, 4-6 February 2015, Bali Indonesia
Compensation control of dissolved oxygen in an activated sludge system via hybrid neuro fuzzy technique M.S Gaya*a, N.A Wahabb, Amir Baturec,Ukashatu Abubakara, I.S Madugua, M.L Kauranginid, L. Baballe Ilac b
a Department of Electrical Engineering, Kano University of Science & Technology, Wudil, Nigeria Department of Control & Mechatronics Engineering, Universiti Teknologi Malaysia, Skudai, Malaysia c Department of Electrical Engineering, Bayero University Kano, Kano, Nigeria d Department of Mathematics, Kano University of Science and Technology, Wudil, Nigeria
Abstract Effective treatment of wastewater relies on a dissolved oxygen (DO) concentration level. Insufficient supply of the DO worsen the quality of the effluent (treated water) whereas too much high dissolved oxygen level maximizes energy consumption and cost. Therefore, DO control is highly crucial for economic, performance improvement as well as safety reasons. However, nonlinearities, influent fluctuations together with the time varying parameters of the system make DO control very difficult using classical control techniques. This paper presents compensation control of dissolved oxygen in the system based on hybrid neuro fuzzy technique. The performance of the proposed control algorithm is demonstrated by set-point tracking. For comparison a PID controller tuned using Zeigler-Nicholas method was used. The simulation results indicate that the proposed strategy is quite effective by tracking well the DO set-point trajectories. This kind of control strategy can be applied to real wastewater treatment system. Keywords: Wastewater; Inverse model; Fuzzy inference system; Feedback controller
1. Introduction Activated sludge system is the main component within a wastewater treatment plant. The success of the system banks upon the establishment of a consortium of microorganisms capable of treating the wastewater effectively. Dissolved oxygen is highly essential for the growth and survival of the microorganisms in the system. However, influent variations, nonlinearities and time varying parameters of the system make the DO control quite hard and often impossible using the classical approach. Literature reveals that varieties of advanced control strategies were proposed [1],[2],[3],[4] and have demonstrated reliable control, which in turns led to maximizing efficiency of the system. Nevertheless, the main inconveniences * Corresponding author. Tel.: +234 802 306 3442 E-mail address:
[email protected]
2351-9789 © 2015 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Selection and Peer-review under responsibility of the Scientific Committee of MIMEC2015 doi:10.1016/j.promfg.2015.07.054
308
M.S. Gaya et al. / Procedia Manufacturing 2 (2015) 307 – 312
with most of the above mentioned advanced control schemes include tuning complexity, requirements of high expertise and detailed knowledge of the biological/biochemical processes involved which may not adequately be guaranteed. An attractive and simple algorithm is the proportional integral derivative (PID) controller based, but its performance is affected by system nonlinearities. In order to establish an efficient and reliable control algorithm the PID controller can be combined with an inverse model of the plant. The nonlinearities of the system can be handled by the inverse model whereas the PID deals with modelling errors. The main challenge in this technique is to obtain the inverse model of the plant since analytically is quite cumbersome and often impossible. Adaptive neuro fuzzy inference system (ANFIS) introduced by [5] which belongs to a class of hybrid neuro fuzzy system has been proven to be an efficient tool for nonlinear approximation and quite adaptive. Therefore, it is the objective of this paper to purvey a straightforward and effective control algorithm for the DO. The system, state equations and the proposed control technique are briefly discussed. 2. The System Description The raw wastewater introduced to the system is mixed with microorganisms in the aeration tank where adequate oxygen is supplied. The sludge (solids) is separated from the treated water (effluent) in the settler by gravity sedimentation. The mathematical representation of the system is expressed as:
dX dt
½ ° ° ° SS SO Q P ° X Sin S * ¾ YSH K S SS KOH SO V ° ° 1 ff Y SS SO S SH * * X f S kd X u ° P fYSH K S SS KOH SO °¿
P
dSS dt dSO dt
SS SO Q Q X kd X X i W CX * K S SS KOH SO V V
(1)
where X is the biomass (microorganisms) concentration, SS is the substrate concentration; SO is the dissolved oxygen concentration. Full definition of the parameters could be found in [3]. However, in reality (practice) the parameter KS SS and KOH SO .Based on typical characteristics of the real wastewater treatment plant, the dynamic operation ranges of the parameters of the process were defined [1] as: CQ QSin Q >6.05,10.05@ ; >0.012,0.142@ ; P kd W > 2.9495,5.9495@ ; P YSH >9.0634,18.1296@ ; V V V
P 1 ff S YSH ff S YSH kd fYSH
> 4.936,9.943@
Arbitrary selection of the values within the given range yields: ª xx t º « » ª 5.411 0 1 º ª x t º ª0º ª0º «x » « « » « » » « » « ss t » «13.562 6.32 0 » « ss t » «0» u t «1» w t « x » « 8.156 0 1¼» «¬ so t »¼ ¬«1¼» ¬«0¼» « so t » ¬ ¬ ¼
(2)
ª x t º 0 0 1 > @ «« ss t »» «¬ so t »¼ The realized process model is controllable but unstable. To obtain a stable model, the poles are placed at y t
M.S. Gaya et al. / Procedia Manufacturing 2 (2015) 307 – 312
p
>6.4987
309
1 0.25i 1 0.25i @ .
Fig. 1. The schematic of the activated sludge system
3. Hybrid Neuro Fuzzy The integration of fuzzy logic and neural network formed neuro fuzzy system. Neuro fuzzy systems are classified as i) concurrent ii) cooperative and iii) hybrid neuro fuzzy [6]. In hybrid neuro fuzzy, the fuzzy system and neural network are one fully merged entity. The main idea behind hybrid neuro fuzzy system is the interpretation of a fuzzy system in a neural network structure [7]. ANFIS belongs to the class of hybrid neuro fuzzy systems. ANFIS is an adaptive network of Sugeono fuzzy model type. ANFIS consists of layers and nodes as depicted in Fig. 2. The circular nodes are fixed whereas the square nodes contained parameters in them that are updated during learning process [8],[9]. In ANFIS mapping, once the input-output data of a given function to be approximated is presented. The nodes in the layers perform certain function based on the incoming signals (input) and parameters associated with nodes. In layer 1- the membership grades of the input variables are generated, which could be bellshaped or Gaussian or trapezoid or triangular, and the layer contained the premise parameters. Layer 2- computes the firing strength of the rule through multiplication of the incoming signal from layer 1. Layer 3- normalizes the rules' firing strengths. Layer 4- calculates the rule outputs according to the consequent parameters. Layer 5- calculates the overall output as the sum of contribution from each rule. The error between the ANFIS model and the desired model is minimized by updating the parameters via the hybrid learning algorithm until the stopping criterion is met. The hybrid learning algorithm uses the least square method to optimize the consequent parameters, and gradient descent to update the premise parameters.
Fig. 2. The ANFIS structure
310
M.S. Gaya et al. / Procedia Manufacturing 2 (2015) 307 – 312
4. Compensation Control Scheme Fig. 3 illustrates the compensation scheme where the feedback controller only sees the inverse plant model and the plant itself [10]. The inverse plant model handles the nonlinearity of the plant which mostly worsen the performance of the PID as the feedback controller, and the feedback controller may compensate for errors caused by model inaccuracies. The ability of ANFIS to map complex nonlinear system makes it useful for realizing an inverse model of nonlinear plants. The inverse model is realized through ANFIS inverse learning. In the learning stage, a training data set is obtained by generating inputs
u k
arbitrarily and observing the corresponding outputs
y k
yielded by the plant. The ANFIS is then utilized to learn the inverse model of the plant by fitting the data pairs
ª¬ y k , y k 1 ; u k º¼ .
The structure of the fuzzy inference system (FIS) was determined using the Matlab
function “genfis1”, which generated a first-order Sugeno FIS. Two (2) Gaussian2 membership functions were assigned to each input variable, which resulted in four (4) fuzzy rules and each rule generates one rule output. The aggregate of the rule outputs yielded the final single output. As the FIS structure is now made available, ANFIS utilizes the hybrid learning algorithm to tune (optimize) the premise (nonlinear) and consequent (linear) parameters of the FIS via learning from the training data set and minimizing the error in order to realize the desired ANFIS inverse model. The realized inverse plant model is utilized for the compensation control. Zeigler Nicholas tuning method is used to obtained the parameters of the PID as the feedback controller, the parameters are : K p 1.84 , Ki 1.26 and Kd 0.32 .
r
Feedback Controller
Inverse Plant Model
Plant
y
Compensation Controller
Fig. 3. The schematic of the compensation control 5. Result and Discussion The tracking capability of the proposed control scheme is explored through simulation. The simulation scenarios considered are similar to the usual practice of DO control in real (experimental) wastewater treatment plant, that is constant and variable DO set-points. For comparison, proportional integral derivative (PID) controller with the same parameters as that of the PID controller of the compensation scheme is used. For the constant set-point, the DO concentration is maintained at 1.9mg/l whereas for the variable set point, the DO set-point is varied from 2.5mg/l to 2mg/l, the results are shown in Fig. 4 and Fig. 5 respectively. Since the usual practice is to keep the DO concentration at 1.5mg/l to 4mg/l, but 2mg/l is widely used in order to save energy and increase effluent quality.. It is essential to test the effectiveness of the controller since simulation gives greater flexibility to compare the performances. Table 1 illustrate the performances of the controllers as the DO set-point changes immediately from 2.5mg/l to 2mg/l which looks like a typical load disturbance. The performances were evaluated based on the measures commonly used such rise time tr and settling time ts . Although both controllers were able to track well the DO set-point trajectory, but the proposed controller exhibited faster response and better settling-time compared to the PID controller. Table 1. Performance Measure Controller Compensation Controller PID Controller
Rise time
tr
0.5275 0.9717
Settling time 1.9118 1.9812
ts
311
M.S. Gaya et al. / Procedia Manufacturing 2 (2015) 307 – 312
It can be seen from the Fig. 4 and Fig. 5 that the proposed control technique effectively controlled the DO level at the desired set-points and also indicating that the proposed controller requires a small amount of time to initiate the control effort. Real-time application of this proposed control strategy is quite possible with some modification and when the scheme is properly implemented on real wastewater treatment plant could results in improving performance, reduce energy consumption and cost as well as reject disturbances. 3
1.5
Dissolved oxygen concentration (mg/l)
Dissolved oxygen concentration (mg/l)
2
Set-point Compensation Controller PID Controller
1 0.5
2.5 2 Set-point
1.5
Compensation Controller
1 0.5
PID Controller
0
0 0
20
Time (h)
40
Fig. 4 DO constant set-point control
60
0
20
40
60
Time (h) Fig. 5 DO set-point variation control
6. Conclusion The paper has presented ANFIS based compensation control of dissolved oxygen. The results indicate that the proposed controller is effective in tracking well the dissolved oxygen point. An enticing feature of the proposed scheme is the capability of handling nonlinearities of the system. The proposed control strategy may serves as valuable scheme for wastewater treatment plant. Acknowledgements The authors wish to thank Kano University of Science & Technology, Wudil, Universiti Teknologi Malaysia and Bayero University Kano for their financial support. The support is gratefully acknowledged. References [1] [2] [3] [4] [5] [6] [7] [8]
P. Y. Feng, Yuzhao, Long Teng-rui, Guo jing-song, Optimal robustness control method of activated sludge system based on uncertain parameters, China Water and Wastewater, 19 (2003) 14-19. M. Yong, P. Yongzhen, and W. Shuying, Feedforward-feedback control of dissolved oxygen concentration in a predenitrification system., Bioprocess Biosyst. Eng., 27 (2005) 223-228. H.G. Han and J.F. Qiao, Adaptive dissolved oxygen control based on dynamic structure neural network, Appl. Soft Comput., 11 (2011) 3812–3820. R. Aguilar-López, Robust generic model control for dissolved oxygen in activated sludge wastewater plant, Chem. Biochem. Eng. Q., 22 (2008) 71–79. J. Jang, ANFIS: Adaptive-network-based fuzzy inference system, Syst. Man Cybern. IEEE Trans., 23 (1993) 665–685. S. Sumathi and S. Paneerselvam, Computational intelligence paradigms: theory & applications using MATLAB. 2010. A. Azar, Adaptive neuro-fuzzy systems, Fuzzy Syst., INTECH, Rijeka, Crotia, 2010. J.R Jang and C. Sun, Neuro-Fuzzy Modeling and Control, Proc. IEEE, 83 (1995) 378-406.
312
[9] [10]
M.S. Gaya et al. / Procedia Manufacturing 2 (2015) 307 – 312
J. Jang, C. Sun, and E. Mizutani, Neuro-fuzzy and soft computing: A Computational Approach to Learning and Machine Intelligence, Prentice Hall, New Jersey 1997. G. Looye, M. Thümmel, M. Kurze, M. Otter, and J. Bals, Nonlinear Inverse Models for Control, Proceedings of the 4th Modelica Conference, 2005, pp. 267–279.
