Maximizing Efficiency of Solar-Powered Systems ... - ACM Digital Library

6.1

Maximizing Efficiency of Solar-Powered Systems by Load Matching Dexin Li and Pai H. Chou Dept. of EECS, University of California, Irvine, CA 92697-2625 USA {dexinl, phchou}@uci.edu

Crystalline Thin films Material Single-Si Poly-Si GaAs a-Si CaTe CuInSe2 Absorption efficiency low low medium high high high Conversion efficiency 15-20% 10-14% 25-30% 5-9% 7% 18% Cost low low high medium medium high

ABSTRACT Solar power is an important source of renewable energy for many low-power systems. Matching the power consumption level with the supply level can make a great difference in the efficiency of power utilization. This paper proposes a source-tracking power management strategy that maximizes the panel’s total energy output under a given solar profile by load matching. The power efficiency was validated by extensive measurement. Compared to a conventional solar powered system, our load matching strategy improves the power utilization by 132% for a portable system performing image processing and wireless communication tasks.

Figure 1: A comparison of PV materials.

Conventional power management algorithms are driven by workload and possibly remaining energy. However, solar powered systems must also consider the output level of the solar panel for power management for several reasons. First, the solar power is not only a non-ideal energy source but also not an energy storage. This implies that the most important consideration is to maximize the utility of the solar power while it is available, instead of always minimizing power by dynamic voltage scaling. Of course, the utility can include recharging the battery, but overcharging the battery can be counterproductive. Another problem that is of particular importance to solar panels is load matching. Solar panels, like batteries and other power sources, have internal resistance Ri . Unlike batteries, whose Ri is around 0.2–0.7Ω [1], solar panels have a much larger Ri value as a function of the solar output and current drawn. Because the Ri forms a voltage divider with the load RL , the effective value of RL directly determines the efficiency of the solar panel. In other words, if the load RL is not balanced with Ri , then much of the power can be wasted. If the load is matched then the system can get the maximum benefit from the available power for longer operation, more precise computation, or higher quality communication. This paper makes the following contributions. First, we present a power source model for solar panels by a mix of analytical and empirical techniques. Second, we develop a solar-aware power management technique that complements today’s workload-driven techniques for maximizing the power efficiency by load matching. We demonstrate the advantages of our technique by measurement on real hardware. Finally, this research work results in a tool that can help designers determine the optimal size and budgets for these solar-powered embedded systems.

Categories and Subject Descriptors C.3.8 [Computer Systems Organization]: Special-Purpose and application-based systems—Real-time and embedded systems

General Terms Design, experimentation

Keywords Solar energy, photovoltaics, power model, solar-aware, power management, load matching

1.

INTRODUCTION

Solar power represents one of the most important sources of renewable energy to complement batteries in portable and autonomous devices. Examples range from satellite systems, emergency telephones, remote sensors, sun-powered radios, and space vehicles. Batteries carry a finite, relatively small amount of energy, and battery technologies have been improving at a slower rate than the increase in demand. Fuel cells may have higher energy density but are expensive. Small nuclear generators are the longest lasting and are found in space probes, but they are unsuitable for many small, disposable systems on earth where cost and safety are obvious concerns. This leaves solar power as a viable, reliable source of energy for extended operations.

Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. ISLPED’04, August 9–11, 2004, Newport Beach, California, USA. Copyright 2004 ACM 1-58113-929-2/04/0008 ...$5.00.

2.

BACKGROUND: PHOTOVOLTAICS

A photovoltaic (PV) cell, also known as solar cell, is a semiconductor device that generates electricity when exposed to light. When light strikes a PV cell, the photons dislodge the electrons from the atoms of the cell. The free electrons then move through the cell, creating and filling in holes in the cell. This movement

162

Voltage vs. Current

of electrons and holes generates electricity. The physical process by which a PV cell converts light into electricity is known as the photovoltaic effect. The major types of materials for building PV cells include crystalline and thin films, which differ in terms of light absorption efficiency, energy conversion efficiency, manufacturing technology, and cost of production [2] (see Fig. 1). Existing electronic models for solar cells characterize properties such as the open circuit voltage (Voc ), the maximum power voltage (Vmp ), and the maximum power current (Imp ) in terms of the short circuit current (Isc ), which is in turn modeled as a function of the beam and diffuse irradiance, air mass (AMa ), incident angle (AOI), and panel temperature (Tc ) [3, 4]. First, the short circuit current is   (Eb · f2 (AOI) + fd · Ediff Isc0 · f1 (AMa ) · E0 (1) Isc = ·(1 + α¯ Isc · (Tc − T0 ))

Voc

I

Figure 2: Solar cell model as a power source.

f (AOI) =

A0 + A1 · AMa + A2 · AMa2 + A3 · AMa3 +A4 · AMa4 B0 + B1 · AOI + B2 · AOI 2 + B3 · AOI 3 +B4 · AOI 4

(2)

3.

=

Isc · (1 + α¯ Isc · (Tc − T0 )) Isc0

(4)

The open-circuit voltage is Voc

3.1

= Voc0 + Ns · δ(Tc ) · ln(Ee ) + βVoc · (Tc − To )

The maximum power is defined as = Vmp · Imp ,

where

(6)

Imp

= Imp0 · (C1 · Ee +C2 · Ee2 ) · (1 + α¯ Imp · (Tc − T0 ))

(7)

Vmp

=

Circuit Model of a Solar-Powered System

Vs = Voc ·

RL RL + R0

(9)

where Voc , Vs are the open-circuit voltage and the observable output voltage of the solar cells, respectively. R0 and RL are the values of the internal resistance of the solar cells and the load resistance, respectively. Assume φ is the sunlight intensity (sunlight power per unit area) that factors in all the environmental parameters like air mass, incident angle, beam irradiance and diffuse irradiance. From Eqn. (5), we know that Voc is a function of the sunlight intensity φ:

Vmp0 +C3 Ns δ(Tc ) · ln(Ee ) +C4 Ns · (δ(Tc ) ln(Ee ))2

(8) +βVmp · (Tc − T0 )

Parameters appearing in Eqns. (1) – (8) are: • Isc0 , short circuit current at reference conditions • Voc0 , open circuit voltage at reference conditions

Voc = h(φ)

• Eb , beam irradiance (W /m2 ) • Ediff , diffuse irradiance

POWER MODELING

The solar cells can be modeled as an ideal power source and an internal resistor, concatenated with a load resistor to form a closed circuit (see Fig. 2). The output voltage of solar cells is :

(5)

Pmp

Figure 3: Output voltage vs. current of a solar panel.

The internal resistance of solar cells is a function of both the sunlight intensity and the current drawn. Unlike batteries, whose internal resistance R0 is about 0.2–0.7Ω, the R0 of solar cells has a much wider range of 10–10kΩ. As a result, some designs include a voltage regulator between the solar cells and the load circuitry. Another alternative is to use solar cells to charge a battery, which then provides the power to the load circuitry with a stable voltage.

(3)

Ee is defined as the ratio of the measured short-circuit current to the short-circuit current at reference conditions. Ee

V

An advantage of the model is that only one variable Isc is needed to derive the others. However, in order to determine Isc , one has to know all the environmental parameters mentioned above, most of which are highly unpredictable and unstable. Even if Isc is determined, the model gives only the upper bound of output power of the solar cells. The actual power output still depends on the load. The following section provides a circuit model that includes the load.

f (AMa ) and f (AOI) are functions of air mass and incident angle, respectively, and are defined as f (AMa ) =

Vs

(10)

Given a RL 6= 0, the closed circuit current I is Vs /RL . From Eqns. (9) – (10), by removing RL we have

(W /m2 )

Vs + R0 · I = h(φ)

• fd , fraction of diffuse irradiance (1 for non-concentrating modules)

(11)

• Tc , T0 : current temperature and reference temperature

Given a constant sunlight intensity, we know from [5] that I and Vs have a non-linear relation defined by the curve similar to the one shown in Fig. 3. Therefore, R0 cannot be a constant, otherwise the Vs and I in Eqn. (11) would have a linear relation, which would contradict the curves shown in Fig. 3.

• δ a function of temperature

3.2

• α¯ Isc , β¯ Voc , α¯ Imp , β¯ Vmp , temperature-related co-efficients

Power Model of Solar Cells

Since R0 is not a constant, it is difficult to derive an analytical model between the output voltage Vs and the closed circuit current I from Eqn. (11). Instead, we take a measurement approach. We

• Ns , number of solar cells • Ai , Bi ,Ci (i = 1, 2, 3, 4), empirical parameters.

163

Power by sunlight intensity and current Voltage vs. Current @ various sunlight brightness 0.25

11200 Lux 14500 Lux 17000 Lux 21800 Lux 31000 Lux 35500 Lux 41500 Lux 43100 Lux 49500 Lux 53000 Lux 55500 Lux 58000 Lux 62500 Lux 65000 Lux

0.2

0.15 I

power (W) 3 2.5 2 1.5 1

0.1

0.5 0

0.05

250 200

0

2

4

6

8

10

12

14

16

18

20

22

10000

24

20000

150

40000 50000 sunlight intensity(Lux)

V

Figure 4: Load current (I) vs. output voltage (Vs ) by sunlight intensity φ.

100

30000

60000

current (mA)

50 700000

Figure 6: Output power P vs. load current I and sunlight intensity φ.

Voltage vs. Power @ various sunlight brightness 3 11200 Lux 14500 Lux 17000 Lux 21800 Lux 31000 Lux 35500 Lux 41500 Lux 43100 Lux 49500 Lux 53000 Lux 55500 Lux 58000 Lux 62500 Lux 65000 Lux

2.5

2

1.5

1

0.5

0

2

4

1

1200 R es is tance (ohm)

P

1400

6

8

10

12

14

16

18

20

22

24

1000 800 600 400

V

200

Figure 5: Output power (P = Vs · I) vs. output voltage Vs by sunlight intensity φ.

0

vary the load resistance RL , and measure Vs at different sunlight intensity φ. We obtain a family of curves in Fig. 4 showing the relationship between Vs and I at different sunlight intensity φ. Note that this model factors out the environmental parameters such as air mass and diffuse irradiance, because the sunlight intensity φ characterizes the actually sunlight power received by the solar panel that already factors in all the environmental parameters. From Fig. 4 we derive Fig. 5 showing the relationship between the output voltage and the output power (P = I · Vs ). The relationship between the power consumed by the load (P) and the load current (I) under different sunlight intensity φ is derived in Fig. 6. At a given sunlight intensity, the power output varies as a function of current drawn. As the load current increases, the power output increases up to a certain current level and then starts to drop. For each given solar intensity, we compute the load resistance that maximizes the output power, as shown in Fig. 7.

0

10000

20000 30000 40000 50000 60000 S olar intens ity (Lux)

80000

Figure 7: Load resistance that maximizes output power.

C ontrol S ignal (0-5V)

B # B ac kend

S # E xtens ion C irc uit

Mic rocontroller (P IC 16F 877)

D/A C onverter (MAX5721)

E thernet (R T L8019)

C urrent S ens or (MAX471)

S # Us er F rontend

S unlight P rofile

4.

70000

DESIGN METHODOLOGY

Amplifier (O P A251)

P ower R egulator (LM317)

P ower S upply (> 30V/1A )

V ariable V oltage (1-24V, >1A)

P ower R egulator (LM2675)

L oad

Solar-powered systems pose new challenges to the design and testing processes. At the design time, without a suitable solar model, a designer may find it difficult to validate the design will full coverage. As a result, they over-design to ensure functional correctness at the expense of losing power saving opportunities. Field tests at

Figure 8: Block diagram of S# Emulator.

164

Mode M0 M1 M2 M3 M4 M5

Config. 206MHz+11Mbps 206MHz+5Mbps 206MHz+2Mbps 206MHz+1Mbps 103MHz+1Mbps 51MHz+1Mbps

Power(mA) 673 593 548 520 405 327

Tcpu 2.0 2.0 2.0 2.0 4.5 10.5

Tnic 0.2 0.25 0.65 1.15 1.15 1.15

Ttotal 2.20 2.25 2.65 3.15 5.65 11.65

Figure 11: Power modes of iPaq 3650+NIC with reduced backlight. Tcpu : compression time; Tnic : transmission time. Figure 9: A photo of the S# systems. to experiment with various conditions that are difficult to reproduce in field tests, as the sunlight profile can vary dramatically. For instance, Fig. 10 shows the same current profile tested under two very different sunlight conditions. Fig. 10(a) shows a sunlight profile collected at a noon, when the solar panel constantly provides 2.8W, sufficient power to the load. Fig. 10(b) shows one collected in an afternoon, when the solar panel outputs around 0.9W, which decreases as the sunlight intensity decreases. This tool is expected to enable the designer to properly budget the power use over a wide range of supply conditions.

runtime are extremely challenging because the testing environment including the sunlight strength, weather conditions, are not easily reproducible. To solve this problem, we developed a solar power emulator named S# for experiments with solar power. This section describes how S# works in design validation. The next section will show how S# can be used to improve power efficiency of a solarpowered system by load matching.

4.1

The S# Solar Panel Emulator

4.3

S# (pronounced “S-sharp”) is a system that emulates the electrical behavior of a solar panel. It extends the capability of the B# battery emulator [1] for a wider dynamic range of output voltage and faster response time for load with higher frequencies. The S# emulator consists of the B# emulator connected to an extension circuit, plus the front-end user interface. Fig. 8 shows the block diagram, and Fig. 9 shows a photo of the real system. The B# back-end reads the load current from the current sensor, looks up the solar model in the memory of the microcontroller and generates a control signal (0–5V) at the output of the D/A converter. By implementing the solar model with lookup tables, we significantly improve the response time over a full-fledged simulation program. The extension circuit uses an op-amp (OPA251) to amplify the control signal to a voltage level in the range of 1–24V as specified by the solar model. This voltage range should match that of a real solar panel and be indistinguishable to the load. The user front-end allows the user to run S# in simulation mode or emulation mode. Simulation mode means the input current and sunlight profiles are read from files, and the output voltage profile is written to a file on the host computer. In emulation mode, the S# supplies power to a real load device in real time according to the solar model downloaded to the microcontroller. The power profile is transferred to the user front-end over Ethernet for profiling purposes.

4.2

Load Matching

Load matching, also called maximum point tracking, is a way to maximize the total energy output of a power supply such as the solar panel by determining the optimal load at runtime. The load can be controlled in the same way that dynamic voltage scaling (DVS) and dynamic power management (DPM) techniques can control the system’s power usage. The key difference is load matching with a solar panel requires that the power manager be aware of the sunlight condition and the efficiency curves of the solar panel. Our approach is to embed a light sensor into the power source and build a solar model and a control unit into the system. Based on the ambient sunlight intensity, the system dynamically adjusts its load to yield the maximum power available for the best system performance.

5.

DESIGN EXAMPLE

We apply load matching to a solar-powered handheld device with a wireless network interface. The system compresses an image (lena, 512x512) and transmits it over an 802.11b wireless interface (CISCO Aironet 350). The load current of the system is determined by setting the components to valid combinations of power modes. The device has six power modes corresponding to valid configurations of the CPU speed and the wireless transmission speed, as listed in Fig. 11. Tcpu and Tnic are the average times to compress an image and to send an image, respectively. Note that the overhead of the operating system and the overhead of network protocol (FTP) are included. Ttotal is the average total time to compress and transmit an image. The sunlight profile is obtained by a field measurement with a light meter (Extech 407026). The solar model used is for the Siemens ST5 thin film solar panel [6]. The iPaq 3650 handheld is powered by the S# emulator through a switching regulator (National LM2675). Fig. 12 shows the three scenarios. In Scenario A the device works in full-on mode – both the CPU and the NIC work at their full speeds (M0). Scenario B assumes the most conservative power mode (M5) to keep the on time of the device as long as possible. In Scenario C the device tracks the sunlight intensity and configures itself to match the potential maximum power from the solar panel. Fig. 15 shows the load profiles in terms of resistance over time

Power Validation

In the early stage of designing a solar-powered system, one may already know an approximate current profile of the load and needs to determine whether the solar panel can power the system both sufficiently and efficiently. The user can use the S# front-end to configure parameters of a solar panel and test the load under various environmental conditions (see Fig. 10(a)). The material type determines the sunlight absorption and conversion rates. The panel size is proportional to the power the solar cells can output. Geographical information (latitude), time of the year, and the weather condition collectively determine the actual sunlight profile received by a solar panel. We used S# to determine the power profile that can be delivered by a given solar panel given a sunlight profile at our location and a current profile representing the load. This setup enables the user

165

(a)

(b)

Figure 10: (a) A current profile working under stronger sunlight. (b) The same current profile working under weaker sunlight. Sun light brightness measured on Jan 29 2004

80000

Accumulated energy provided by the solar panel. 3500

70000

Full on Conservative Load matching

3000

2500

50000

A

C

40000

Energy (mAHr)

Brightness (lux)

60000

B

30000

2000

1500

20000

1000

10000

500

0

7

8

9

10

11

12 13 Time (Hour)

14

15

16

17

0

Figure 12: A sunlight profile and three scenarios. A: full-on, B: conservative, C: load matching.

6

8

10

12 Time (Hr)

14

16

18

Figure 13: Accumulated energy over time.

Accumulated number of pictures taken by the load device. 8000

for three scenarios under the same sunlight profile. The ideal curve in each figure represents the optimal load resistance that maximizes the power output of the solar panel over time. The real curve is the load resistance of the device in each scenario. In full-on mode (a), the load matches the ideal curve well but only for about two hours. In conservative mode (b), the device is on for a longer time but it does not match the ideal curve well. Our load matching method achieves the best solar energy utility and higher system performance. Figs. 13 and 14 show the accumulated energy output and the accumulated performance in terms of the number of images processed, respectively. The handheld obtains more energy from the solar panel by load matching than it does in the full-on and the conservative scenarios. It turns out that in full-on mode, the handheld consumes the least amount of energy over the time of a day due to its short on time (2.16 hours). The handheld processes about twice

Full on Conservative Load matching

7000 6000

# of pictures

5000 4000 3000 2000 1000 0

6

8

10

12 Time (Hr)

14

16

18

Figure 14: Accumulated performance over time.

166

1200

1000

600

R es is tance (ohm)

R es is tance (ohm)

800

800 600

800 600

400

400

400

200

200

200

0

7

8

9

10

11

12 13 T ime (Hour)

14

15

16

17

0

7

real ideal

real ideal

1000

1000 R es is tance (ohm)

1200

1200

real ideal

8

9

10

(a)

11

12 13 T ime (Hour)

14

15

16

17

0

7

8

(b)

9

10

11

12 13 T ime (Hour)

14

15

16

17

(c)

Figure 15: Ideal resistance vs. real resistance in the three scenarios: (a) full-on, (b) conservative, (c) load matching.

Method Full-on Conservative Load-matching Battery

On time (Hour) (%) 2.14 100% 6.22 290% 6.22 290% 0.58 27%

Energy (mAHr) (%) 1440 100% 2034 141% 3337 232% 391 27.1%

Performance (#) (%) 3436 100% 1922 56% 7369 214% 784 22.8%

variation in solar power and do not perform load matching. As a result they waste a consider amount of power, which can be a precious resource. Load matching requires first the characterization of the solar panel’s electrical properties, and second a power management policy that tracks the changes in supply conditions. Our result shows that over 132% of usable power can be reclaimed using our power management strategy using load matching. Future work includes extending load matching to the joint optimization of renewable energy sources along with not only the power consuming devices but also energy storage devices with their own non-ideal characteristics.

Figure 16: Comparison of four power schemes. as the number of images by load matching as in the full-on scenario. The conservative approach gives the worst performance even though its on time is equal to that of the load matching method, mainly because it needs much longer time to process an image. A separate test is performed to first charge a rechargeable battery (NiMH, 1600mAH) using the same solar panel. It takes about three mid-day hours on a sunny day to fully charge the battery at a charge rate of 500mAH. We use the battery as the only power source for the same handheld and continuously perform image compression and transmission. It reaches its cut-off voltage within only 35 minutes and has processed 784 images. This result is about an order of magnitude inferior to that of the load matching method mainly due to the efficiency loss during the energy conversion. Moreover, the battery has rate capacity and rate recovery effects, which are not relevant to the solar panel. Fig. 16 summarizes the three scenarios in terms of on time, energy, and performance. The baseline is the full-on scenario. The conservative scenario uses the lowest power mode whenever possible, and it achieves an on time that is 2.9× that of the full-on scheme. However, low power actually results in high energy and low efficiency, because it consumes 1.4× the energy but is able to process only about half as many images as the full-on strategy. Our load matching scheme achieves the same on time as in the conservative scenario while being able to utilize much more energy from the solar panel. As a result, it doubles the performance of the fullon scheme. Using a rechargeable battery actually yields the worst performance, even though in some scenarios the energy storage capability must be used.

6.

Acknowledgments This research was sponsored in part by DARPA under contract F33615-00-1-1719 and in part by National Science Foundation under grant CCR-0205712. The authors thank Chulsung Park for the fruitful discussions on load matching and for his idea of using a light sensor in the sunlight-aware design. The authors also thank Sungjun Kim and Stacy M. Low for constructing the S# system used in the experiments of this work.

7.

REFERENCES

[1] P.H. Chou, C. Park, J. Park, K. Pham, and J. Liu. B#: a battery emulator and power profiling instrument. In ISLPED, pages 288–293, 2003. [2] Olivia Mah. Fundamentals of photovoltaic materials. In National Solar Power Research Institute (NSPRI) Report, 1998. [3] D.L King, J.A. Kratochvil, and W.E. Boyson. Field experience with a new performance characterization procedure for photovoltaic arrays. In Proc. 2nd World Conference and Exhibition on Photovoltaic Solar Energy Conversion, 1998. [4] M.W. Davis, A. Hunter Fanney, and B.P. Dougherty. Measured versus predicted performance of building integrated photovoltaics. Journal of Solar Energy Engineering Transactions of the ASME, 125:21–7, February 2003. [5] Anna Fay Williams. The Handbook of photovoltaic applications: building applications and system design considerations. Fairmont Press, 1986. [6] Siemens ST5 solar panel data sheet. In http://www.siemenssolar.co.uk/st5.html.

CONCLUSION

Due to the special characteristics of solar panels, solar-powered embedded systems must pay special attention to load matching in order to fully utilize this renewable source of power. Conventional schemes that work for rechargeable batteries do not track the wide

167