3.1. Establishment of Photovoltaic System Model
The photovoltaic system is a key component in the field of sustainable energy, with remarkable environmental protection and renewable energy characteristics. With the global pursuit of sustainable development, photovoltaic systems are increasingly widely used in the energy industry. A photovoltaic system takes solar energy as the source and does not produce harmful gases such as carbon dioxide, which conforms to the concept of environmental protection and sustainable development. The photovoltaic system involves many factors such as illumination and temperature and has nonlinear and time-varying properties. Therefore, its control and optimization have certain technical challenges. With the rapid development of the renewable energy market, photovoltaic systems have great potential in the commercial field. By optimizing the control algorithm of the photovoltaic system, the energy output can be increased and the operating cost can be reduced, thus promoting its wide application in commercial applications.
It is very important to understand the current–voltage characteristics of photovoltaic cells when analyzing and designing photovoltaic systems. These characteristics can be described by mathematical models based on physics, among which the most commonly used is the single diode equivalent circuit model. The model considers the nonlinearity and uncertainty of photovoltaic cells and simulates the current generation process of photovoltaic cells under the changes of illumination and temperature through the interaction of circuit elements [
34]. Equation (1) is based on the single diode model, which integrates factors such as photovoltaic current, reverse saturation current, and series resistance, and is used to calculate the current of photovoltaic cells at a specific voltage. Equation (2) describes the relationship between current and illumination intensity, which is very important for understanding the performance of photovoltaic system under different illumination conditions. The current generation process of photovoltaic system can be described by a single diode equivalent circuit model. Considering the nonlinearity and uncertainty of the system, a general single diode equivalent circuit model is adopted, including factors such as photovoltaic current, reverse saturation current, series resistance, etc., as shown in Equation (1):
In the equation,
is the photogenerated current.
is the reverse saturation current.
q is the electron charge,
V is the voltage,
n is the destruction factor,
K is the Boltzmann constant, and
is the series resistance. The relationship between current and light intensity is shown in Equation (2):
In Equation (2), is the output current of the photovoltaic system. is the light intensity factor, which indicates the influence of light intensity on photovoltaic current. The photogenerated current is the current generated by a photovoltaic system under given illumination conditions. is the reverse saturation current, which represents the reverse current under the condition of zero bias. is the natural exponential function. is the electron charge. is the voltage, which represents the output voltage of the photovoltaic system. ideal factor reflects the degree of non-ideal of photovoltaic devices. is the Boltzmann constant. is the temperature. series resistance means the series resistance inside the photovoltaic system.
The photogenerated current represents the current generated by the photovoltaic system under illumination conditions, which is directly related to the illumination intensity and is one of the important outputs of the system. Reverse saturation current describes the reverse current caused by the characteristics of the material itself in the photovoltaic cells. This item is particularly critical in low light conditions, which affects the stability of the system. The series resistance takes into account the influence of the internal resistance of the photovoltaic system, which is an important parameter in the system and affects the response time and stability of the system. By introducing these parameters, people can more accurately capture the influence of illumination, temperature, and other factors on the current in photovoltaic system, which provides an experimental basis for subsequent fuzzy control and neural network control. In order to better understand the model structure,
Figure 1 shows a schematic diagram of the photovoltaic system model:
Figure 2 shows the structural diagram of the photovoltaic system experimental device used in this study. As shown in the figure, the experimental system mainly includes the photovoltaic panel, data acquisition module, environmental condition simulator (used to simulate different lighting and temperature conditions), and control system unit (including fuzzy controller and neural network controller).
The main equipment specifications used in the experiment are shown in
Table 1 below:
3.2. Selection of Fuzzy Control and Neural Network Control Algorithm
In order to cope with the complexity and uncertainty of the photovoltaic system, two advanced algorithms, fuzzy control and neural network control, are selected to improve their practicability in commercial development. Among them, fuzzy control can deal with the fuzzy relationship between the input and output of the system through fuzzy logic. The fuzzy control algorithm is based on the Mamdani model, which aims to deal with the fuzziness and uncertainty of environmental factors, such as light intensity and temperature change. The following steps are followed in the process of establishing the model. Firstly, a set of fuzzy rule bases is established through expert experience and system response requirements. Secondly, the triangular membership function is used to divide the fuzzy set of input variables. Specifically, in the fuzzification stage, each input parameter (such as light intensity and temperature) is transformed into the corresponding membership degree, then fuzzy inference is carried out by using a fuzzy rule base to produce fuzzy output. Finally, in the deblurring stage, the fuzzy output is converted into specific control actions. In this experiment, the fuzzy controller sets 20 fuzzy rules, and the number of membership functions is 3 for each input variable. The algorithm realizes the fuzzy reasoning of the system state by establishing a fuzzy rule base, and adaptively adjusts the system parameters according to the expert experience in the rule base and the dynamic adjustment requirements of the system to optimize the power generation efficiency and stability of the photovoltaic system. Specifically, the fuzzy control algorithm converts the input variables into fuzzy sets through the triangular membership function, and then uses the fuzzy rule base for fuzzy reasoning, and finally uses the center of gravity method to calculate the specific control actions, thus achieving effective control of the photovoltaic system. The fuzzy control algorithm is applied to the photovoltaic system to realize adaptive control of uncertain factors such as light intensity and temperature fluctuation. Specifically, the fuzzy control algorithm establishes the fuzzy rules between the operating parameters of photovoltaic system (such as power generation, current, etc.) and external environmental factors (such as light intensity and temperature). These fuzzy rules are based on expert experience and the need for system dynamic adjustment. For example, when the light intensity is low and the temperature is moderate, the system rules may include the following: if the light intensity is low and the temperature is moderate, increase the power; if the light intensity is moderate and the temperature is high, reduce the power. By considering factors such as illumination and temperature, the fuzzy control algorithm can adaptively adjust the control parameters and optimize the power generation efficiency and stability of photovoltaic system.
In order to realize the guidance of SES, it is inseparable from the key role of the fuzzy controller (structure organization is shown in
Figure 3). Fuzzy controllers can adaptively adjust the working state of the photovoltaic system by considering the fuzzy rules of environmental variables such as illumination and temperature to optimize its power generation efficiency and stability. As an important form of sustainable energy, the performance improvement of photovoltaic systems is very important for the sustainability of the whole energy system.
In
Figure 3, in the process of fuzzy control implementation, fuzzy control interface, fuzzy reasoning, and fuzzy interface are the key steps to building an effective control system. The performance and effect of the fuzzy controller are directly affected by these steps. The fuzzy control interface is responsible for transforming specific input variables (such as light intensity and temperature) into fuzzy membership functions. This process reflects the fuzziness and uncertainty in the real world. Through the carefully designed membership function, the system can capture the fuzzy characteristics of input variables more accurately and provide effective input for subsequent fuzzy reasoning. Fuzzy reasoning is an important process in fuzzy logic. At this stage, the fuzzy controller infers the fuzzy membership of the input according to a series of set fuzzy rules and produces fuzzy output results. The formulation of these rules is usually based on the experience of domain experts and the requirements of system performance. The accuracy of fuzzy reasoning directly affects the stability and adaptability of the control system.
Deblurring interface is the process of transforming the fuzzy output obtained by fuzzy reasoning into specific control actions or parameter adjustment. This step is to clarify the output result of the fuzzy controller, so that it can be directly applied to the actual system control. The design of the deblurring interface needs to consider the actual application scenario of the system and ensure that the influence of the output of the controller on the photovoltaic system is feasible and explainable. The performance of the fuzzy controller also depends on a knowledge base, which includes the definition of fuzzy rules, the selection of membership function and the method of resolving fuzzy. Knowledge base is a key part of fuzzy controller learning and adaptation. Through continuous updating and optimization, the fuzzy control system can better adapt to different working environments and requirements.
The neural network control algorithm is used to learn and simulate the complex nonlinear relationship of the photovoltaic system. By training the neural network, the system can understand the mapping relationship between the output power of photovoltaic cells and input parameters such as illumination and temperature. In real-time operation, the neural network will predict the output of the system and make adjustments according to the predicted results to ensure that the photovoltaic system can operate efficiently and stably under different working conditions.
In the selection of neural network control algorithm, the recurrent neural network (RNN) is adopted to better process the time series data of photovoltaic system. The recurrent neural network (RNN) model is adopted in the neural network control algorithm, aiming at simulating the complex nonlinear relationship of photovoltaic system and realizing the prediction and real-time control of the system state. The algorithm learns the mapping relationship between output power and input parameters (such as light intensity and temperature) of the photovoltaic system by training the neural network to realize the prediction of system state. Specifically, the neural network control algorithm specifies the parameters and hyperparameters of the RNN model, including training rounds, activation function, loss function, and optimization algorithm. Through the back propagation and optimization on the training dataset, the neural network can continuously improve the modeling and prediction ability of the system, thus realizing the real-time control of the photovoltaic system. The RNN introduces cyclic connections into the neural network, which enables it to process sequence data and retain the memory of past information. For a time-dependent photovoltaic system, the RNN can better capture the dynamic relationship between input parameters and improve the accuracy of system modeling. The specific structure is shown in
Figure 4:
In
Figure 4, the input layer receives external environmental parameters such as light intensity and temperature as the input of the network. Each input node represents a specific environment variable. The circulation layer is the core part of RNN, which contains many circulation units. There is a circular connection among these units, which allows the network to retain past information when processing sequence data. Each cycle unit contains multiple neurons. By learning historical data, the network can adaptively adjust the weight and capture the complex relationship between the output power of photovoltaic cells and input parameters such as illumination and temperature. The output layer generates the predicted output power of the photovoltaic system. The number of output nodes corresponds to the system output variables. In order to enhance the expressive ability of the network, several fully connected layers are introduced between the loop layer and the output layer. These layers allow the network to learn the nonlinear relationship between input parameters more flexibly and improve the modeling ability of the network to the dynamic characteristics of the system. Meanwhile, after the output of each neuron, the activation function ReLU is introduced to increase the nonlinear expression ability of the network.
The training process of the neural network is based on a large amount of photovoltaic system operation data. Through supervised learning, the network can learn the mapping relationship between input and output. Finally, the back propagation algorithm and gradient descent method are used to optimize the weight and bias of the neural network continuously, so that it can achieve higher prediction accuracy on the training set. The trained neural network is applied to real-time operation. Real-time input parameters (such as illumination and temperature) of the photovoltaic system are input into the neural network, and the predicted output power is obtained through forward propagation. The system adjusts the control parameters in real time according to the prediction results of the neural network to ensure that the photovoltaic system can operate efficiently and stably under different working conditions.
The details of the implementation of fuzzy control and neural network control algorithm in photovoltaic control system are as follows. Fuzzy controller: Based on expert system, according to two input parameters of illumination and temperature, the running state of photovoltaic system is adjusted through 20 fuzzy rules to optimize power output. The fuzzy member function adopts triangular distribution to ensure that the fuzzification of input parameters can cover a wide range of operating conditions. RNN controller: The structure of circulating neural network (RNN), the input layer receives illumination and temperature as network inputs, and the output layer generates the predicted power value of photovoltaic system through the processing of two circulating layers are adopted. The training of RNN model adopts back propagation algorithm, and the learning rate is set to 0.001. After 100 rounds of training, the model can accurately predict the power output under different environmental conditions.
3.3. Experimental Verification
The purpose of this experiment is to verify the application effect of the proposed fuzzy control and neural network control algorithm in the photovoltaic system to improve the power generation efficiency, stability, and adaptability of the system. On the one hand, it verifies the control effect of fuzzy control algorithm on the photovoltaic system through changes in light intensity and temperature. It is also necessary to evaluate the ability of neural network control algorithm to model and predict the dynamic characteristics of the photovoltaic system. Finally, the system performance of fuzzy control combined with neural network control is analyzed to improve the stability of the system in a complex environment.
In the data acquisition process, professional data acquisition equipment, such as illuminance sensor, temperature sensor, ammeter, and voltmeter, is used to monitor the running state of the photovoltaic system in real time. These sensors provide input parameters (light intensity, temperature) and output parameters (current, voltage, power generation) needed for the experiment. The acquisition frequency occurs once every minute to ensure that the changes of photovoltaic system under different illumination and temperature conditions are fully covered. Data are collected for one week in each season to cover the changes in different weather, seasons, and time periods. In the process of data collection, quality control is carried out, and possible abnormal values or data missing are checked and handled. The quality of the collected data is representative of the experimental results. The experimental dataset includes photovoltaic system data collected at different time periods throughout the year. The specific statistical information is as follows. Total data: 28,800 data points (collected once every minute for 4 weeks). Light intensity range: 100 Lux to 1000 Lux. Temperature range: 10 °C to 30 °C. Main statistical information: average value, standard deviation, minimum value and maximum value for light intensity, the average value is 550 Lux and the standard deviation is 258 Lux. For temperature, the average value is 20 °C and the standard deviation is 5 °C. In data processing, this study uses standardized processing methods to eliminate the dimensional influence of data and make the data more suitable for mathematical analysis. In addition, this study also sets the parameters of fuzzy control and RNN neural network in detail, such as the number of fuzzy rules, the selection of membership function, the number of layers and neurons of circular neural network, learning rate, and so on, to ensure the optimal operation of the algorithm. In this study, in order to evaluate the accuracy of neural network control algorithm in the output power prediction of photovoltaic system, two statistical indicators, root mean square error (RMSE) and determination coefficient (R
2), are adopted. RMSE is a measure of prediction error, which calculates the root of the average square of the deviation between the model prediction value and the actual observation value, and can reflect the accuracy of model prediction [
35]. R
2 measures whether the model fits the data well, and its value is between 0 and 1. The closer the value is to 1, the stronger the explanatory power of the model is and the better the fitting effect is. These two indexes are used to evaluate the prediction ability and fitting degree of neural network model, which provides an important reference for model selection and optimization. The mathematical equations for RMSE and R
2 are shown in Equation (3) and Equation (4), respectively:
represents the predicted value of the model.
indicates the actual observed value, and
represents the number of samples.
represents the average of actual observed values. In addition, in the research, the design of fuzzy controller is aimed at the nonlinear and uncertain factors in photovoltaic system, such as light intensity and temperature change. Fuzzy controller deals with the fuzzy relationship between input and output through fuzzy logic to optimize the power generation efficiency and stability of photovoltaic system. The core components of fuzzy controller include membership function, membership parameters, and fuzzy rules. Membership functions are used to describe the fuzzy characteristics of input variables, and they usually take the form of triangle or Gaussian distribution. In the fuzzy controller, these functions convert the accurate input values into membership degrees in fuzzy sets to reflect the uncertainty and fuzziness of input data. For example, light intensity and temperature can be used as input variables, which can be converted into different levels in fuzzy sets through membership functions, such as “low light”, “medium light”, or “high temperature” and “low temperature”. Membership parameters are the parameters used when defining membership functions, which determine the shape and distribution of fuzzy sets. The setting of these parameters is usually based on expert experience and the actual operation requirements of the system, and it needs to be adjusted through experiments or simulations to obtain the best control effect. Fuzzy rules are a series of logical statements based on expert experience, which defines the relationship between fuzzy sets of input variables and output control actions. For example, the fuzzy rule may include: “If the light intensity is low and the temperature is high, increase the power output of the photovoltaic system”. In a fuzzy controller, these rules are used to infer specific control behaviors from fuzzy sets to adapt to environmental changes and optimize system performance. To sum up, the design of the fuzzy controller needs to comprehensively consider the selection of membership function. The adjustment of membership parameters and the formulation of fuzzy rules also need to be considered to ensure effective control and optimization in the face of uncertainty and nonlinearity in photovoltaic system operation.
The experimental process consists of three parts. The first is the experiment of the fuzzy control algorithm: firstly, the knowledge base and fuzzy rules of the fuzzy control system are established by using the collected data. Then, the operation of the photovoltaic system under different illumination and temperature conditions is simulated on the experimental platform, and real-time control is carried out through the fuzzy control algorithm. Finally, the response data of the system are recorded, including the changing trend of power generation and system stability. The second is the experiment of neural network control algorithm. Firstly, the collected data are divided into a training set and a test set. Then, the RNN neural network is trained by the training set, and the network weights and parameters are optimized. Finally, the prediction accuracy of the neural network on the output power of the photovoltaic system is verified on the test set. The neural network control algorithm is applied in real time on the experimental platform to record the actual response of the system. Thirdly, the experiment of combining fuzzy control with neural network control: the fuzzy control and neural network control algorithm are applied simultaneously on the experimental platform to observe the response of the system in a complex environment. The parameter settings are shown in
Table 2 and
Table 3. RNN model adopts two-layer structure and introduces recurrent connection between hidden layers to capture the dependence of time series data. Rectified linear unit (Relu) is selected as the activation function of the neural network because it can increase the network’s ability to deal with nonlinear problems. Mean squared error (MSE) is used as the loss function and the Adam optimizer is used as the optimization algorithm. In this study, the number of training iterations of the network is 100 rounds and the learning rate is set to 0.001. The proportion of training set and test set is 80% and 20%. This study uses triangular membership function to describe the fuzzy properties of input variables and captures the nonlinear characteristics of input variables through Gaussian membership function. Gaussian membership function is used to describe the uncertainty of input variables such as light intensity and temperature. Gaussian membership function can better reflect the fuzzy properties of input variables when processing smooth data and can effectively reduce the computational complexity. In addition, through expert opinions and system simulation analysis, combined with the actual operation experience of photovoltaic system, people have determined 20 fuzzy rules. These rules consider the dynamic relationships between light intensity, temperature, photovoltaic current, and voltage and can respond to environmental changes more flexibly, thus optimizing the power output of the system. Examples of input and output data are shown in
Table 4: