Photovoltaic maximum power point tracking method and system fusing deep learning and particle swarm

Abstract

The invention discloses a photovoltaic maximum power point tracking method and system fusing deep learning and a particle swarm, and relates to the field of photovoltaic power generation, and the method comprises the steps: collecting photovoltaic data through a sensor, and carrying out the preprocessing; a deep learning model is constructed, and the maximum power is output through preprocessed data training; optimizing a particle swarm algorithm by adopting a Levy flight method; and inputting the maximum power output by the deep learning model into the optimized particle swarm algorithm to calculate the photovoltaic maximum power value. The initial maximum power value of the photovoltaic system is predicted in real time through the deep learning model, the particle swarm optimization algorithm further optimizes the maximum power point by using the Levy flight strategy, and the dual optimization mechanism not only improves the adaptive capacity and tracking precision of the photovoltaic system under the rapidly changing environmental condition, but also improves the tracking precision of the photovoltaic system under the rapidly changing environmental condition. And moreover, the defect that the traditional method is easy to fall into local optimum is avoided, and the energy conversion efficiency and stability of the photovoltaic system are effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of photovoltaic power generation, and specifically to a photovoltaic maximum power point tracking method and system integrating deep learning and particle swarm. Background Art

[0002] Photovoltaic systems will produce different output powers under different light intensities and temperature conditions. In order to maximize the energy output of photovoltaic systems, maximum power point tracking (MPPT) algorithms are widely used. Traditional MPPT algorithms, such as the perturbation and observation method (P&O) and the incremental conductance method (IncCond), have problems of slow response speed and low accuracy under rapidly changing environmental conditions. With the development of deep learning technology, using it to optimize the MPPT algorithm has become a new trend. However, when deep learning is used alone in the maximum power tracking process, it is easy to fall into local optimality when encountering environmental mutations, resulting in a decrease in tracking accuracy. To this end, this patent uses deep learning as a first-order optimization method to make it approach the maximum power point of the system, and uses the improved particle swarm algorithm as a secondary optimization method to accurately track the maximum power point, thereby solving the problem of falling into local extreme points. Summary of the invention

[0003] In view of the above-mentioned problems, the present invention is proposed.

[0004] Therefore, the technical problem solved by the present invention is: how to achieve more accurate and efficient maximum power point tracking in photovoltaic systems to address the shortcomings and inefficiencies of traditional methods under dynamic environmental changes.

[0005] In order to solve the above technical problems, the present invention provides the following technical solution: a photovoltaic maximum power point tracking method integrating deep learning and particle swarm, which comprises the following steps:

[0006] Use sensors to collect photovoltaic data and perform pre-processing;

[0007] Build a deep learning model and output maximum power through preprocessed data training;

[0008] The particle swarm algorithm is optimized using the Levy flight method;

[0009] The maximum power output by the deep learning model is input into the optimized particle swarm algorithm to calculate the maximum power value of the photovoltaic.

[0010] As a preferred solution of the photovoltaic maximum power point tracking method integrating deep learning and particle swarm described in the present invention, wherein: the photovoltaic data collected by using sensors includes ambient temperature, light intensity and corresponding power history data;

[0011] The preprocessing is to denoise the ambient temperature, light intensity and power history data, normalize the data, perform time series segmentation and feature extraction on the data, retain the dynamic change characteristics under different environmental conditions, and divide the processed data into a training set, a verification set and a test set.

[0012] As a preferred solution of the photovoltaic maximum power point tracking method integrating deep learning and particle swarm described in the present invention, wherein: the construction of the deep learning model is to take the light intensity and ambient temperature as input, and the corresponding power history data as output, train the deep learning model, and use the cross entropy loss function as the loss value between the calculated predicted value and the true value. When the loss function value is less than the preset value, the final trained deep learning model is obtained to determine the relationship between the light intensity, ambient temperature and power;

[0013] The light intensity and ambient temperature are input into the trained deep learning model, and the initial maximum power of the output layer is calculated.

[0014] As a preferred solution of the photovoltaic maximum power point tracking method integrating deep learning and particle swarm described in the present invention, the deep learning model includes the connection of several neurons, and simulates the nonlinear relationship of photovoltaics through a multi-layer neural network. In the network learning process, the training sample set is X, the expected value set is D, and the actual output is Y. The input expression of the jth neuron corresponding to the hidden layer is:

[0015]

[0016] Where n is the number of input neurons, ω ij is the connection weight from the i-th neuron in the input layer to the j-th neuron in the hidden layer, is the threshold of the jth neuron in the hidden layer, x j is the corresponding j-th sample in the training sample set;

[0017] The output expression of the jth neuron in the hidden layer is:

[0018]

[0019] Among them, x i is the jth sample in the training sample set, and f is the activation function of the hidden layer.

[0020] As a preferred solution of the photovoltaic maximum power point tracking method integrating deep learning and particle swarm described in the present invention, the photovoltaic characteristic model is expressed as:

[0021]

[0022] Among them, Ipv is the photovoltaic current, I L is the equivalent inductor current, I0 is the photocurrent, V pv is the photovoltaic voltage, R s Photovoltaic equivalent series resistance, R sh is the photovoltaic equivalent parallel resistance, n is a fixed parameter factor, K is the Boltzmann constant, and T is the absolute temperature of the photovoltaic cell;

[0023] Use particle swarm algorithm to calculate the photovoltaic characteristic model and solve the maximum power value;

[0024] The expression of particle swarm algorithm is:

[0025]

[0026] Where ω is the inertia weight.

[0027] As a preferred solution of the photovoltaic maximum power point tracking method integrating deep learning and particle swarm described in the present invention, wherein: the maximum power value of photovoltaic is calculated by taking the predicted maximum power value output by the deep learning model as the initial input value, applying the particle swarm optimization algorithm to calculate the actual maximum power of photovoltaic, in each iteration, calculating the fitness of each particle, and finding the individual optimum and the global optimum, if the position of the particle is not updated during the iteration process, then using the Levy flight method to update the position;

[0028] Determine whether the termination condition is met at the current moment. If so, the optimization is terminated, and the final maximum photovoltaic power value is calculated and output according to the voltage at the current position;

[0029] If the termination condition is not met at the current moment, the fitness calculation and particle position update will continue until the termination condition is met;

[0030] The termination conditions include when the number of iterations reaches a preset value and the particle position stops updating after using the Levy flight method.

[0031] As a preferred solution of the photovoltaic maximum power point tracking method integrating deep learning and particle swarm described in the present invention, the Levy flight method is to improve the inertia weight in the particle swarm algorithm by using the Levy flight method and adjust the particle position solution formula, which is expressed as:

[0032]

[0033] Among them, i max is the maximum number of iterations of the particle swarm algorithm, i is the number of iterations of the current particle swarm algorithm, m is the control coefficient, the smoothness of the overall curve between the weight and the number of changes, Represents X id The k-th generation position of represents point-to-point multiplication, α represents a factor that can control the step size, and Levy(λ) is a function that follows the path search.

[0034] Another purpose of the present invention is to provide a photovoltaic maximum power point tracking system that integrates deep learning and particle swarm, which can collect environmental data of the photovoltaic system in real time, use the deep learning model to predict the initial maximum power value, and then further accurately optimize the maximum power point through the optimized particle swarm algorithm, solving the problem that the existing technology is difficult to accurately track the maximum power point of the photovoltaic system in real time under complex environmental conditions. It shows stronger adaptability and global search capabilities in dynamic environments, avoids the defect that traditional methods are prone to fall into local optimality, and improves the overall energy conversion efficiency of the photovoltaic system.

[0035] In order to solve the above technical problems, the present invention provides the following technical solutions: A photovoltaic maximum power point tracking system integrating deep learning and particle swarm, comprising: a data collection and preprocessing module, a deep learning model training module, a particle swarm optimization algorithm module, a Levy flight optimization module, and a result evaluation and termination judgment module;

[0036] The data collection and preprocessing module collects photovoltaic data in real time through sensors, and performs denoising, normalization, time series segmentation and feature extraction on the collected data;

[0037] The deep learning model training module builds a deep learning model based on the preprocessed data set, defines the input, output and network structure, uses the training data set to train the deep learning model, learns the relationship between ambient temperature and light intensity and power, and predicts the initial maximum photovoltaic power value based on the new input data after the training is completed;

[0038] The particle swarm optimization algorithm module initializes the population parameters of the particle swarm, sets the individual optimal value and the global optimal value, calculates the fitness of each particle using the particle swarm optimization algorithm according to the initial prediction value of the deep learning model, updates the individual optimal value and the global optimal value, and optimizes the power output value by updating the speed and position of the particles to find the global optimal solution;

[0039] The Levy flight optimization module adds a large-step jump search strategy to the particle swarm optimization algorithm, and in each iteration, the position of the particle is updated using the Levy flight distribution;

[0040] The result evaluation and termination judgment module checks whether the current optimization result meets the preset termination condition at the end of each iteration, evaluates the state of the current particle swarm, and determines whether the optimal power point of the photovoltaic system has been found. If the condition is met, the algorithm is terminated and the final photovoltaic maximum power value is output.

[0041] A computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the photovoltaic maximum power point tracking method integrating deep learning and particle swarm are implemented as described above.

[0042] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the photovoltaic maximum power point tracking method integrating deep learning and particle swarm as described above.

[0043] The beneficial effects of the present invention are as follows: the present invention predicts the initial maximum power value of the photovoltaic system in real time through a deep learning model, and the particle swarm optimization algorithm further optimizes the maximum power point using the Levy flight strategy. This dual optimization mechanism not only improves the adaptability and tracking accuracy of the photovoltaic system under rapidly changing environmental conditions, but also avoids the defect of traditional methods that are prone to falling into local optimality, thereby effectively improving the energy conversion efficiency and stability of the photovoltaic system. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0045] Figure 1 An overall flow chart of a photovoltaic maximum power point tracking method integrating deep learning and particle swarm provided in the first embodiment of the present invention;

[0046] Figure 2 A diagram of the deep learning network structure in the photovoltaic maximum power point tracking method integrating deep learning and particle swarm provided in the first embodiment of the present invention;

[0047] Figure 3 A flowchart of a particle optimization algorithm in a photovoltaic maximum power point tracking method integrating deep learning and particle swarm provided in the first embodiment of the present invention. DETAILED DESCRIPTION

[0048] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0049] Example 1, reference Figure 1 to Figure 3As an embodiment of the present invention, a photovoltaic maximum power point tracking method integrating deep learning and particle swarm is provided, comprising:

[0050] Step 1: Collect the PV system ambient temperature, light intensity and corresponding power as a data set.

[0051] The ambient temperature, light intensity and power historical data are denoised, the data are normalized, the data are segmented into time series and feature extracted, the dynamic change characteristics under different environmental conditions are retained, and the processed data are divided into training set, validation set and test set.

[0052] Step 2: If Figure 2 A deep learning model is constructed and trained with the collected data, with light intensity and ambient temperature as input and power as output. During the training process, the cross entropy loss function is used to calculate the loss between the predicted value and the true value. Through continuous training, the loss function value is reduced to the minimum possible, thereby obtaining a model with the best training performance and determining the relationship between light intensity, ambient temperature and the environment. By inputting light intensity and ambient temperature into the trained model, the initial maximum power of the output layer can be obtained.

[0053] It is further explained that the construction of the deep learning network mainly includes the connection of multiple neurons. Through the multi-layer neural network system, the photovoltaic nonlinear system can be accurately simulated. In the network learning process, the training sample set is X, the expected value set is D, the actual output is Y, and the input of the jth neuron in the corresponding hidden layer is:

[0054]

[0055] Where n is the number of input neurons; ω ij is the connection weight from the i-th neuron in the input layer to the j-th neuron in the hidden layer; is the threshold of the jth neuron in the hidden layer. The output of the jth neuron in the hidden layer is:

[0056]

[0057] In the formula, f is the activation function of the hidden layer. This patent uses ReLU as the activation function. The established model is trained by collecting data, and the network model with the best training performance is obtained through analysis. It is assumed that this patent determines that the network has the best training performance when the number of hidden layers is 10 after training.

[0058] Step three: Use Levy flight and propose a new inertia weight value change function to improve the particle swarm algorithm and obtain an improved new algorithm.

[0059] The relationship between individual particles and the group in the particle swarm algorithm is:

[0060]

[0061] In the formula, ω is the inertia weight, which has a very important impact on the overall search process of the algorithm. In the process of maximum power optimization, ω changes according to a certain rule. In the initial stage of the algorithm, the weight value needs to be increased to improve the global search ability; in the final stage, the weight value needs to be reduced to improve the final local search ability, that is, the change process of ω presents a nonlinear decreasing process. Since the calculation of the weights of the conventional particle swarm algorithm cannot meet the requirements, it is easy for the maximum power tracking to fall into the local optimal solution. Based on this, a weight value change function is proposed:

[0062]

[0063] Where m is the control coefficient, which indicates the smoothness of the overall curve between the control weight and the number of changes. Using this weight change function, in the initial stage of the algorithm, a larger inertia weight value can enhance the global search capability of the algorithm and weaken the local search capability; in the final stage of the algorithm, a smaller inertia weight value can increase the local search capability and weaken the global search capability. This can not only ensure that it is not easy to fall into the local optimum during the maximum power tracking process, but also shorten the time of the entire optimization process and improve the optimization accuracy. In order to ensure that the curve can achieve the above effect, the value of m should be between 4 and 10 to obtain better results.

[0064] Levy flight is a random search behavior that follows the Levy distance distribution. In Levy flight, individual movements are more flexible. Under the premise of taking the same number of steps or travel, the activity range of particles performing Levy flight is much larger than that of particles performing ordinary free movement, and they can also search farther. Levy flight is widely used in the range search optimization of algorithms due to its advantage of being able to perform long-distance displacement. Therefore, adding Levy flight to the traditional particle swarm on the basis of particle movement can expand the original search range of the algorithm, making it easier to jump out of local advantages and achieve maximum tracking of the target.

[0065] The update formula of Levi's flight position is:

[0066]

[0067] In the formula, Represents X id The k-th generation position of; represents point-to-point multiplication; α represents a factor that can control the step size; Levy(λ) is the function of path search.

[0068] In the process of particle swarm optimization algorithm, as the number of iterations increases, the position of the particle is also updated. However, during the iteration process, the particle cannot reach the position of the optimal solution due to its small running range, resulting in the phenomenon that the particle position is not updated after iteration and still maintains the previous "optimal solution". In this way, the "optimal solution" iterated out is still not close to the actual optimal result. Or more iterations are required to approach the optimal result, which increases the complexity of the original calculation. By improving the algorithm by using Levy flight, the search range of the particle can be expanded. According to whether the particle position in the algorithm is in a state of frequent update with the increase in the number of iterations, if it remains in the updated state, the next iteration will be carried out directly, otherwise the particle position will be changed through Levy flight. This particle positioning update method increases the search range of the particle and improves the accuracy of the calculation.

[0069] Construct a photovoltaic characteristic model, the expression is:

[0070]

[0071] The particle swarm algorithm is used to calculate the photovoltaic characteristic model and solve the maximum power value.

[0072] It is further explained that Figure 3 First, randomly initialize the population, initialize the current population parameters and boundary conditions, and if the boundary conditions are exceeded, take the boundary value to obtain the current population parameters.

[0073] Calculate the fitness value of each individual and save it. Each particle will track two extreme values ​​during the search process: individual optimal and global optimal. Individual optimal is the optimal solution found by the particle itself, and global optimal is the optimal solution found by the entire particle group.

[0074] Update the particle position and velocity according to the following three equations. The particle's velocity and position are updated according to the individual optimal solution and the global optimal solution.

[0075]

[0076] Among them, i max is the maximum number of iterations of the particle swarm algorithm, i is the number of iterations of the current particle swarm algorithm, m is the control coefficient, the smoothness of the overall curve between the weight and the number of changes, Represents X id The k-th generation position of represents point-to-point multiplication, α represents a factor that can control the step size, and Levy(λ) is a function that follows the path search.

[0077] Update individual optimal and global optimal After each iteration, the individual optimum and global optimum are updated, and the optimal value is selected to update the data through the position optimization degree before and after the iteration. As the number of iterations increases, the algorithm needs to judge whether to perform Levy flight to expand the particle search range and prevent the algorithm from falling into the local optimal solution.

[0078] Step 4: Input the value approximating the maximum power of the system obtained by deep learning into the improved particle swarm algorithm to obtain the maximum power value of the photovoltaic system and improve the output efficiency of the system.

[0079] It should be further explained that the present invention collects the ambient temperature and light intensity parameters of the photovoltaic system, and inputs the collected data set into the established deep learning network for training to obtain and save the model with the best training effect. Subsequently, different parameters are input into the trained model to obtain the approximate maximum power value as the first optimization method.

[0080] This value is then input into the improved particle swarm algorithm by changing the inertia weight and adopting Levy flight to obtain the maximum power value of the photovoltaic system, and thus input into the photovoltaic system.

[0081] By applying this patent to photovoltaic systems, as the environment continues to change, the maximum photovoltaic power value can be obtained more quickly and accurately, thereby effectively reducing the power loss of the photovoltaic system and improving output efficiency.

[0082] Embodiment 2 is an embodiment of the present invention, which provides a system for photovoltaic maximum power point tracking method integrating deep learning and particle swarm, including: a data collection and preprocessing module, a deep learning model training module, a particle swarm optimization algorithm module, a Levy flight optimization module, and a result evaluation and termination judgment module;

[0083] The data collection and preprocessing module collects photovoltaic data in real time through sensors, and performs denoising, normalization, time series segmentation and feature extraction on the collected data;

[0084] The deep learning model training module builds a deep learning model based on the preprocessed data set, defines the input, output and network structure, and uses the training data set to train the deep learning model to learn the relationship between ambient temperature, light intensity and power. After the training is completed, the initial maximum power value of the photovoltaic is predicted based on the new input data;

[0085] The particle swarm optimization algorithm module initializes the population parameters of the particle swarm, sets the individual optimal value and the global optimal value, calculates the fitness of each particle using the particle swarm optimization algorithm according to the initial prediction value of the deep learning model, updates the individual optimal value and the global optimal value, and optimizes the power output value by updating the speed and position of the particles to find the global optimal solution;

[0086] The Levy flight optimization module adds a large-step jump search strategy to the particle swarm optimization algorithm. In each iteration, the position of the particle is updated using the Levy flight distribution.

[0087] The result evaluation and termination judgment module checks whether the current optimization result meets the preset termination condition at the end of each iteration, evaluates the state of the current particle swarm, and determines whether the optimal power point of the photovoltaic system has been found. If the condition is met, the algorithm is terminated and the final photovoltaic maximum power value is output.

[0088] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program code.

[0089] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0090] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0091] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0092] Embodiment 3: In this embodiment, in order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments. This embodiment experiments the existing traditional method and the method of this embodiment respectively.

[0093] In this simulation experiment, we built a simulation platform for a photovoltaic system. The simulation platform runs on a server with high-performance computing capabilities. The server is configured with a 16-core CPU, 64GB RAM, and equipped with GPU acceleration (such as NVIDIA TeslaV100), and the operating system is Ubuntu 20.04. The simulation platform contains a virtual photovoltaic system model that can simulate photovoltaic output characteristics under different environmental conditions (such as light intensity, ambient temperature, etc.). In order to ensure the accuracy and repeatability of the experiment, all photovoltaic data and environmental parameters are derived from public photovoltaic data sets and laboratory historical records.

[0094] In this simulation experiment, we selected two traditional maximum power point tracking (MPPT) methods as comparison benchmarks:

[0095] The first method is the perturbation and observation method, which continuously perturbs the working voltage of the photovoltaic system, observes the change of the output power, and then gradually adjusts the operating point to make it approach the maximum power point. However, this method performs poorly under rapidly changing environmental conditions and is prone to power oscillation and tracking lag.

[0096] The second method is the incremental conductance method, which determines the position of the current operating point relative to the maximum power point by calculating the ratio of the conductance of the photovoltaic system and the incremental conductance, and adjusts the operating voltage accordingly. Compared with the P&O method, the incremental conductance method has a faster response speed when dealing with changes in light, but it still has certain limitations in complex environments.

[0097] During the experiment, we simulated the output of the photovoltaic system under different environmental conditions, used the method of the present invention and the traditional methods 1 and 2 to perform MPPT control, and recorded their respective performance data. The following is the comparison data:

[0098] Table 1 Effect comparison table

[0099]

[0100] It can be seen from the above data tables that the output power of the method of the present invention under various environmental conditions is higher than that of traditional methods one and two, especially in the case of large changes in light intensity (such as cloudy and rapidly changing light scenes), the output power advantage of the present invention is more obvious. This shows that the present invention can adjust and optimize the operating point of the photovoltaic system more quickly when dealing with rapid changes in environmental conditions, ensuring higher power output. In addition, the method of the present invention also shows significant advantages in optimization time. In all test scenarios, the optimization time is shorter than that of methods one and two, especially in the case of frequent light fluctuations, the optimization speed is more than twice as fast. These results fully reflect the beneficial effects of the present invention, that is, by combining deep learning and particle swarm optimization algorithms, the present invention can not only achieve higher photovoltaic system output power under different environmental conditions, but also significantly shorten the optimization time and improve the overall efficiency and stability of the system.

[0101] Table 2 Data comparison table

[0102]

[0103]

[0104] It can be seen from Table 2 above that the method of the present invention performs well in multiple key performance indicators, fully demonstrating its beneficial effects. In terms of energy conversion efficiency, the method of the present invention achieves higher conversion efficiency under all test environments. For example, under conditions of stable lighting on a sunny day, the conversion efficiency of the present invention is as high as 94%, which is significantly higher than the 89% of the traditional P&O method and the 92% of the INC method. In cases where lighting conditions change frequently, such as cloudy and uncertain lighting scenes, the response time of the method of the present invention is only 210 milliseconds, while method one and method two require 500 milliseconds and 430 milliseconds, respectively.

[0105] The present invention can adjust the operating point of the photovoltaic system more quickly, improve the efficiency of maximum power point tracking, and especially can respond to and adapt to changes more quickly in a dynamically changing environment.

[0106] In terms of output power stability, the method of the present invention also shows great advantages. Under different lighting conditions, the power fluctuation amplitude of the method of the present invention is significantly smaller than that of the traditional method. For example, under cloudy lighting conditions, the power fluctuation amplitude of the P&O method is 30W, the INC method is 25W, and the method of the present invention is only 10W. In summary, through the comparison of energy conversion efficiency, response speed, power fluctuation amplitude and other aspects, the method of the present invention not only improves the performance of the photovoltaic system, but also significantly enhances the stability and adaptability of the system, and can better cope with changes under different environmental conditions, and achieve more efficient and reliable photovoltaic power generation.

[0107] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A photovoltaic maximum power point tracking method integrating deep learning and particle swarm is characterized by: include: Use sensors to collect photovoltaic data and perform pre-processing; Build a deep learning model and output maximum power through preprocessed data training; Construct a photovoltaic characteristic model and use the particle optimization algorithm to solve the maximum power value; The particle swarm algorithm is optimized using the Levy flight method; The maximum power output by the deep learning model is input into the optimized particle swarm algorithm to calculate the maximum power value of the photovoltaic.

2. The photovoltaic maximum power point tracking method integrating deep learning and particle swarm as claimed in claim 1, characterized in that: The photovoltaic data collected by the sensor include ambient temperature, light intensity and corresponding power history data; The preprocessing is to denoise the ambient temperature, light intensity and power history data, normalize the data, perform time series segmentation and feature extraction on the data, retain the dynamic change characteristics under different environmental conditions, and divide the processed data into a training set, a verification set and a test set.

3. The photovoltaic maximum power point tracking method integrating deep learning and particle swarm as claimed in claim 2, characterized in that: The deep learning model is constructed by taking the light intensity and the ambient temperature as inputs and the corresponding power history data as outputs, training the deep learning model, and using the cross entropy loss function as a loss value between the calculated predicted value and the true value. When the loss function value is less than a preset value, the final trained deep learning model is obtained to determine the relationship between the light intensity, the ambient temperature and the power. The light intensity and ambient temperature are input into the trained deep learning model, and the initial maximum power of the output layer is calculated.

4. The photovoltaic maximum power point tracking method integrating deep learning and particle swarm as claimed in claim 3, characterized in that: The deep learning model includes the connection of several neurons, and simulates the photovoltaic nonlinear relationship through a multi-layer neural network. In the network learning process, the training sample set is X, the expected value set is D, the actual output is Y, and the input expression of the jth neuron corresponding to the hidden layer is: Where n is the number of input neurons, ω ij is the connection weight from the i-th neuron in the input layer to the j-th neuron in the hidden layer, is the threshold of the jth neuron in the hidden layer, x j is the corresponding j-th sample in the training sample set; The output expression of the jth neuron in the hidden layer is: Among them, x i is the jth sample in the training sample set, and f is the activation function of the hidden layer.

5. The photovoltaic maximum power point tracking method integrating deep learning and particle swarm as claimed in claim 4, characterized in that: The photovoltaic characteristic model is expressed as: Among them, I pv is the photovoltaic current, I L is the equivalent inductor current, I0 is the photocurrent, V pv is the photovoltaic voltage, R s Photovoltaic equivalent series resistance, R sh is the photovoltaic equivalent parallel resistance, n is a fixed parameter factor, K is the Boltzmann constant, and T is the absolute temperature of the photovoltaic cell; Use particle swarm algorithm to calculate the photovoltaic characteristic model and solve the maximum power value; The expression of particle swarm algorithm is: Where ω is the inertia weight.

6. The photovoltaic maximum power point tracking method integrating deep learning and particle swarm as claimed in claim 5, characterized in that: The maximum photovoltaic power value is calculated by taking the predicted maximum power value output by the deep learning model as the initial input value, applying the particle swarm optimization algorithm to calculate the actual maximum photovoltaic power, and calculating the fitness of each particle in each iteration, and finding the individual optimum and the global optimum. If the position of the particle is not updated during the iteration process, the Levy flight method is used to update the position; Determine whether the termination condition is met at the current moment. If so, the optimization is terminated, and the final maximum photovoltaic power value is calculated and output according to the voltage at the current position; If the termination condition is not met at the current moment, the fitness calculation and particle position update will continue until the termination condition is met; The termination conditions include when the number of iterations reaches a preset value and the particle position stops updating after using the Levy flight method.

7. The photovoltaic maximum power point tracking method integrating deep learning and particle swarm as claimed in claim 6, characterized in that: The Levy flight method uses the Levy flight method to improve the inertia weight in the particle swarm algorithm and adjust the particle position solution formula, which is expressed as: Among them, i max is the maximum number of iterations of the particle swarm algorithm, i is the number of iterations of the current particle swarm algorithm, m is the control coefficient, the smoothness of the overall curve between the weight and the number of changes, Represents X id The k-th generation position of represents point-to-point multiplication, α represents a factor that can control the step size, and Levy(λ) is a function that follows the path search.

8. A system using the photovoltaic maximum power point tracking method integrating deep learning and particle swarm as claimed in any one of claims 1 to 7, characterized in that: It includes data collection and preprocessing module, deep learning model training module, particle swarm optimization algorithm module, Levy flight optimization module and result evaluation and termination judgment module; The data collection and preprocessing module collects photovoltaic data in real time through sensors, and performs denoising, normalization, time series segmentation and feature extraction on the collected data; The deep learning model training module builds a deep learning model based on the preprocessed data set, defines the input, output and network structure, uses the training data set to train the deep learning model, learns the relationship between ambient temperature and light intensity and power, and predicts the initial maximum photovoltaic power value based on the new input data after the training is completed; The particle swarm optimization algorithm module initializes the population parameters of the particle swarm, sets the individual optimal value and the global optimal value, calculates the fitness of each particle using the particle swarm optimization algorithm according to the initial prediction value of the deep learning model, updates the individual optimal value and the global optimal value, and optimizes the power output value by updating the speed and position of the particles to find the global optimal solution; The Levy flight optimization module adds a large-step jump search strategy to the particle swarm optimization algorithm, and in each iteration, the position of the particle is updated using the Levy flight distribution; The result evaluation and termination judgment module checks whether the current optimization result meets the preset termination condition at the end of each iteration, evaluates the state of the current particle swarm, and determines whether the optimal power point of the photovoltaic system has been found. If the condition is met, the algorithm is terminated and the final photovoltaic maximum power value is output.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the photovoltaic maximum power point tracking method integrating deep learning and particle swarm are implemented as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the photovoltaic maximum power point tracking method integrating deep learning and particle swarm are implemented as described in any one of claims 1 to 7.

Images