Abstract
With solar photovoltaic (PV)-based power generation becoming popular, research is being carried out in multiple facets of the PV system. The electrical aspect is one such area where objectives are extracting maximum power, developing power management strategies, etc. Hence, while experimenting with new strategies it is desirable to have controllable irradiation and temperature conditions. This ensures that external factors are constant for comparing the effectiveness of different strategies. Thus, PV emulators are a suitable solution to this. A deep learning-based PV emulator is proposed where the proposed PV emulator is capable of emulating a PV panel under uniform and non-uniform irradiance conditions. A deep learning network is modeled with irradiance, temperature, ratio of partial shading, PV current as inputs, & PV voltage as output. Then, a buck converter is controlled to follow the reference voltage, thereby mimicking the desired I-V characteristics. The deep learning model was trained and built in Python, and simulations to validate the performance were carried out in Simulink.
TopIntroduction
With increasing awareness of the consequences of climate change, corrective actions are being implemented in sectors such as electrical power, transportation, etc. The fundamental sources of carbon emission in the power sector are the fossil fuel-based generating stations which are being replaced with green sources such as solar power, wind power, etc. Solar energy is a consequence of the nuclear fusion reactions in the core of the sun which maintains the temperature of the surface of the sun at 5778 K. Then through black-body radiation, this energy reaches the earth in the form of electromagnetic waves. The infrared spectrum of the emitted electromagnetic radiation can be tapped as heat energy through solar thermal collectors. This tapped heat energy can be used for producing steam to spin electric generators or for space heating applications thereby replacing other methods that have a higher carbon footprint. On the other hand, solar photovoltaic (PV) panels absorb the visible spectrum of the electromagnetic waves to generate electricity directly. These solar PV panels consist of series and/or parallel connections of multiple p-n junction diodes termed as solar cells. The energy of the photon in the visible spectrum is used to knock electrons out of their bonds in solar cells (generation of electron-hole pairs) which then flow through external circuit as electric current.
The output power from the solar PV panels depends on the incident solar irradiance and the temperature of panel. Since the aforementioned factors are stochastic in nature, it is essential to develop adaptive power management strategies that extract maximum possible power from the solar PV panels thereby increasing the utilization factor of the solar PV-based generation systems. However, to validate and compare the effectiveness of such strategies, an environment where the incident solar irradiance and the temperature of the panel are controllable is necessary. Solar PV emulators are suitable solutions to this problem as they are essentially controllable sources whose output voltage and current mimic the characteristics of a typical solar PV panel.
Azharuddin et al.(2014) have proposed a PV emulator that employs a buck converter controlled using a PI controller. The desired I-V characteristics is solved using MATLAB/Simulink where PV current is the input to the model and PV voltage is the output of the model and the generated reference PV voltage is tracked using a PI controller through a dSPACE controller. A similar PV emulator is proposed by Moussa et al. (2017) where the power converter is a buck converter and the PV model is implemented in a FPGA controller that is programmed using Matlab/XSG toolbox. This PV emulator’s PV model calculates the reference current based on the sensed output voltage and the desired input values for irradiance and temperature. Thus the buck converter’s controller calculates its duty cycle to track the reference current. Ayop & Tan(2018)have focussed on improving the PV model by developing a current-resistance (I-R) model from the conventional single-diode model of PV. To solve this, I-R model for mimicking the desired I-V characteristics, binary search method is proposed. Additionally, a resistance feedback controller is also proposed. This PV emulator also computes current as a reference signal thereby controlling the power converter in current control mode. A PV emulator developed by Ullah et al.(2020) adopts a two-diode PV model and improves the controller by employing fractional order sliding mode controller.
Key Terms in this Chapter
Diffusion Length: It is the average distance a charge carrier travels after electron-hole pair generation but before recombination.
Solar Azimuth Angle: It is the angle on the horizontal plane between ‘the line due south’ and ‘the projection of direct radiation on horizontal plane’.
Global Radiation: It is the sum of direct and diffuse solar radiation.
Solar Irradiation: It is the amount of solar energy received per unit area. Hence it is basically mathematical integration of solar irradiance over a given period of time. Its SI unit is J/m2. For convenience, it can also be expressed in watt-hour (Wh) or kWh, where 1 Wh = 3600 joule. It is also called solar insolation.
AM0: This means no atmosphere at all and hence signifies extra-terrestrial radiation. The irradiance associated with AM0 spectrum is 1367 W/m 2 .
Iteration: Due to memory constraint, the training data is usually split into smaller batches. One iteration covers a single batch. Thus for one epoch (full training dataset), there may be several iterations based on the batch size.
AM1.5: This means the solar radiation passes through the atmosphere 1.5 times longer than their shortest path. The solar spectrum associated with AM1.5 is useful in temperate latitude regions. The solar irradiance associated with the AM1.5 spectrum is approximately 1000 W/m 2 .
Charge Carrier Lifetime: It is the time duration during which the carrier is free from bond before recombination occurs.
Pyrheliometer: It is a device used to measure the direct or beam irradiance.
Solar Irradiance: It is the amount of solar power received per unit area. Its SI unit is W/m 2 .
Direct Radiation: It is the solar radiation reaching the earth’s surface in a straight line without any scattering. It is also called beam radiation.
Epoch: It is the number of times the artificial neural network goes through the entire training dataset while training.
Diffuse Radiation: It is the solar radiation that reaches the earth’s surface after being scattered from atmosphere, clouds, etc.
Altitude Angle: It is the angle between the ‘direct sun rays’ and ‘its projection on the horizontal plane’
Air Mass (AM): It is the ratio of ‘path length of direct radiation from the sun at any point in the sky’ to ‘the path length of direct radiation from the sun, if the sun were directly overhead at zenith’. It signifies the distance travelled by the solar radiation in the atmosphere. From the perspective of solar rays, it signifies the thickness of the atmosphere. The solar power reaching the earth’s surface is affected by the thickness of the atmosphere. The longer the radiation travel in the atmosphere, more the absorption is (less irradiance on earth’s surface). Thus the solar spectrum received on the earth’s surface varies with the thickness of the atmosphere i.e. AM.
Pyranometer: It is device used to measure the global solar irradiance.
Zenith Angle: It is the angle between the ‘sun’s rays’ and ‘a line perpendicular to the horizontal plane through the point.’
Artificial Neuron: It is fundamental building block of a neural network and is analogous to biological neurons.
AM1: If the air mass is 1 (AM 1), then the sun is directly overhead. The solar spectrum associated with AM1 is useful in equatorial regions.
Declination Angle: It is the angle between the ‘line joining the centres of the earth and the sun’ and ‘the projection of that line on the earth’s equatorial plane’
Perceptron: It is one of the early artificial neural networks with just a single neuron which was used for classification purposes.