Photovoltaic array fault diagnosis method and system based on distributed evolutionary algorithm

Abstract

The invention provides a photovoltaic array fault diagnosis method and system based on a distributed evolutionary algorithm. The method comprises the following steps: obtaining typical feature data of a photovoltaic array in various operation states; calculating the fitness of each individual through the residual fusion diagnosis model corresponding to the initial hyper-parameter based on the typical feature data; taking the initial population as a parent original population, and selecting a parent elite population based on the fitness of each individual; based on the parent original population and the parent elite population, using a distributed evolutionary algorithm to carry out repeated iteration updating, and calculating to obtain an optimal hyper-parameter of the residual fusion diagnosis model; based on the typical feature data, training a residual fusion diagnosis model corresponding to the optimal hyper-parameter to obtain an optimal residual fusion diagnosis model; according to the method, the optimal hyper-parameter of the residual fusion diagnosis model is subjected to collaborative updating solution combining global search and local search through the distributed evolutionary algorithm, so that the construction efficiency of the model and the performance of the model are improved.

Description

Technical Field

[0001] The present invention belongs to the field of photovoltaic power generation, and in particular relates to a photovoltaic array fault diagnosis method and system based on a distributed evolutionary algorithm. Background Art

[0002] In recent years, photovoltaic power generation has been widely used as a renewable energy technology. However, since photovoltaic systems are exposed to harsh natural environments all year round, problems such as component failures, inverter failures, and connection failures occur from time to time, greatly affecting the power generation efficiency and safe operation of photovoltaic arrays. Therefore, fault diagnosis technology for photovoltaic arrays has become an important means to ensure the reliability and economy of photovoltaic systems. Traditional fault diagnosis methods rely on empirical rules, physical models, or simple statistical methods, which are difficult to cope with complex and changeable actual fault conditions, especially in small sample environments where data is scarce or incomplete. The diagnostic performance of these traditional methods is usually severely restricted, and it is difficult to accurately capture subtle differences in complex fault modes.

[0003] At present, there are three main problems in the existing technology: first, the PV array data is scarce and incomplete, which makes it difficult for the model to effectively learn representative fault features, resulting in poor generalization performance, limiting the application potential of the model in actual PV array fault diagnosis; second, in the existing technology, fixed neural network (Convolutional Neural Network, CNN) hyperparameters are usually used or hyperparameters are repeatedly adjusted manually to optimize model performance. This is not only time-consuming and inefficient, but also easily causes the model to fall into local optimality and cannot fully adapt to complex fault features; third, the traditional neural network model may have problems with gradient vanishing or gradient explosion when the depth increases, affecting the model performance and resulting in low accuracy of model fault diagnosis. Summary of the invention

[0004] In order to overcome the above-mentioned deficiencies of the prior art, the present invention proposes a method for constructing a residual fusion diagnosis model based on a distributed evolutionary algorithm, comprising:

[0005] Obtain typical characteristic data of photovoltaic arrays under various operating conditions;

[0006] Initialize the parameters of the distributed evolutionary algorithm, wherein each individual of the initial population corresponds to a set of initial hyperparameters; based on the typical feature data, calculate the fitness of each individual through the residual fusion diagnostic model corresponding to the initial hyperparameters; use the initial population as the parent original population, and select the parent elite population based on the fitness of each individual; perform global search optimization and update on the parent original population to obtain the offspring original population, and perform local search optimization and update on the parent elite population to obtain the offspring elite population; based on the offspring original population and the offspring elite population, use the distributed evolutionary algorithm to perform repeated iterative updates until the maximum number of iterations is reached, and use a set of hyperparameters represented by the individual with the largest fitness value in the current offspring original population and the current offspring elite population as the optimal hyperparameters of the residual fusion diagnostic model;

[0007] Based on the typical feature data, the residual fusion diagnosis model corresponding to the optimal hyperparameters is trained to obtain the optimal residual fusion diagnosis model.

[0008] Preferably, the method of performing global search optimization and updating on the parent generation original population to obtain the offspring generation original population, and performing local search optimization and updating on the parent generation elite population to obtain the offspring generation elite population, comprises:

[0009] Using a genetic algorithm, a global search optimization update is performed on the parent generation original population to obtain a child generation original population with better fitness;

[0010] A variable neighborhood search algorithm is used to perform local search optimization and update on the individuals in the parent elite population to obtain a child elite population with better fitness.

[0011] Preferably, the variable neighborhood search algorithm is used to perform local search optimization and update on individuals in the parent elite population to obtain a child elite population with better fitness, including:

[0012] Based on the individuals in the parent elite population, determining the neighborhood structure corresponding to each individual;

[0013] Searching for individuals in the neighborhood structure, and determining whether the fitness value of the individuals in the neighborhood structure is greater than the fitness value of the individuals in the parent elite population;

[0014] If yes, the individuals in the neighborhood structure are substituted for the individuals in the parent elite population to obtain a child elite population updated by local search;

[0015] Otherwise, continue to perform local search. If the fitness value of the individuals in the neighborhood structure is not greater than the fitness value of the individuals in the parent elite population and the maximum number of iterations is reached, the parent elite population is used as the child elite population.

[0016] Preferably, the distributed evolutionary algorithm is used to repeatedly iterate and update based on the offspring original population and the offspring elite population until the maximum number of iterations is reached, and a set of hyperparameters represented by the individuals with the largest fitness values ​​in the current offspring original population and the current offspring elite population are used as the optimal hyperparameters of the residual fusion diagnostic model, including:

[0017] The individual information of the random individuals in the offspring elite population is replaced with the individual information of the individual with the smallest fitness value in the offspring original population to obtain an optimized offspring original population;

[0018] Based on the optimized offspring original population, the dominant individuals in the optimized offspring original population are screened out, and the dominant individuals are used as the new parent original population;

[0019] Selecting a new parent generation elite population based on the fitness of each individual in the new parent generation original population;

[0020] Performing global search optimization and updating on the new parent generation original population to obtain a new offspring generation original population, and performing local search optimization and updating on the new parent generation elite population to obtain a new offspring generation elite population;

[0021] Determine whether the maximum number of iterations has been reached. If so, output a set of hyperparameters represented by the individual with the largest fitness value in the current offspring elite population and the offspring original population as the optimal hyperparameters of the residual fusion diagnostic model; otherwise, based on the current offspring elite population and the offspring original population, continue to iterate until the maximum number of iterations is reached.

[0022] Preferably, the obtaining of typical characteristic data of the photovoltaic array under various operating conditions includes:

[0023] Based on the circuit structure of the photovoltaic array, typical characteristic data under various operating conditions are obtained through the simulation platform;

[0024] The multiple operating states include one or more of the following: normal working state, short circuit fault state, open circuit fault state, aging fault state, partial shading state, aging shading fault state, short circuit shading fault state and open circuit shading fault state; the typical characteristic data include one or more of the following: open circuit voltage, short circuit current, maximum power point voltage, maximum power, fill factor, temperature and light intensity.

[0025] Based on the same inventive concept, the present invention also provides a residual fusion diagnosis model construction system based on a distributed evolutionary algorithm, comprising:

[0026] Data acquisition module: used to obtain typical characteristic data of photovoltaic arrays under various operating conditions;

[0027] Optimal hyperparameter solution module: used to initialize the parameters of the distributed evolutionary algorithm, in which each individual of the initial population corresponds to a set of initial hyperparameters; based on the typical feature data, the fitness of each individual is calculated through the residual fusion diagnostic model corresponding to the initial hyperparameters; the initial population is used as the parent original population, and the parent elite population is selected based on the fitness of each individual; the parent original population is globally searched and optimized to obtain the offspring original population, and the parent elite population is locally searched and optimized to obtain the offspring elite population; based on the offspring original population and the offspring elite population, the distributed evolutionary algorithm is used to perform repeated iterative updates until the maximum number of iterations is reached, and a set of hyperparameters represented by the individual with the largest fitness value in the current offspring original population and the current offspring elite population is used as the optimal hyperparameters of the residual fusion diagnostic model;

[0028] Model building module: used to train the residual fusion diagnosis model corresponding to the optimal hyperparameters based on the typical feature data to obtain the optimal residual fusion diagnosis model.

[0029] Preferably, the optimal hyperparameter solving module includes:

[0030] A global search optimization updating unit: used to use a genetic algorithm to perform global search optimization updating on the parent generation original population to obtain a child generation original population with better fitness;

[0031] Local search optimization update unit: used for adopting variable neighborhood search algorithm to perform local search optimization update on individuals in the parent elite population to obtain a child elite population with better fitness.

[0032] Preferably, the local search optimization update unit includes:

[0033] A neighborhood structure determination subunit: used for determining the neighborhood structure corresponding to each individual based on the individuals in the parent elite population;

[0034] Individual judgment subunit: used to search for individuals in the neighborhood structure, and judge whether the fitness value of the individuals in the neighborhood structure is greater than the fitness value of the individuals in the parent elite population;

[0035] Individual replacement subunit: when the fitness value of the individual in the neighborhood structure is greater than the fitness value of the individual in the parent elite population, the individual in the neighborhood structure is replaced by the individual in the parent elite population to obtain a local search updated child elite population;

[0036] The offspring population determination subunit is used to use the parent elite population as the offspring elite population when the fitness value of the individuals in the neighborhood structure is not greater than the fitness value of the individuals in the parent elite population and the maximum number of iterations is reached.

[0037] Preferably, the optimal hyperparameter solving module further includes:

[0038] Knowledge transfer unit: used to replace the individual information of the individual with the smallest fitness value in the original population of the offspring with the individual information of the random individuals in the elite population of the offspring, so as to obtain the optimized original population of the offspring;

[0039] A dominant individual retaining unit is used to screen out dominant individuals from the optimized offspring original population based on the optimized offspring original population, and use the dominant individuals as new parent original population;

[0040] A parent generation elite population selection unit: used for selecting a new parent generation elite population based on the fitness of each individual in the new parent generation original population;

[0041] Distributed optimization and updating unit: used for performing global search optimization and updating on the new parent generation original population to obtain a new child generation original population, and performing local search optimization and updating on the new parent generation elite population to obtain a new child generation elite population;

[0042] Iteration judgment unit: used to judge whether the maximum number of iterations has been reached; if so, a set of hyperparameters represented by the individual with the largest fitness value in the current offspring elite population and the offspring original population is output as the optimal hyperparameters of the residual fusion diagnosis model; otherwise, based on the current offspring elite population and the offspring original population, iterate until the maximum number of iterations is reached.

[0043] Preferably, the data acquisition module is specifically used to: based on the circuit structure of the photovoltaic array, obtain typical characteristic data under various operating conditions through a simulation platform; the various operating conditions include one or more of the following: normal working state, short circuit fault state, open circuit fault state, aging fault state, partial shading state, aging shading fault state, short circuit shading fault state and open circuit shading fault state; the typical characteristic data include one or more of the following: open circuit voltage, short circuit current, maximum power point voltage, maximum power, fill factor, temperature and light intensity.

[0044] Based on the same inventive concept, the present invention also provides a photovoltaic array fault diagnosis method, comprising:

[0045] Obtaining the photovoltaic array data to be diagnosed;

[0046] Inputting the photovoltaic array data to be diagnosed into the optimal residual fusion diagnosis model for fault diagnosis to obtain a photovoltaic array fault diagnosis result;

[0047] Among them, the optimal residual fusion diagnosis model is constructed based on any one of the above-mentioned residual fusion diagnosis model construction methods based on a distributed evolutionary algorithm.

[0048] Preferably, the optimal residual fusion diagnostic model is constructed based on a convolutional neural network combined with a residual network structure.

[0049] Preferably, the convolutional neural network includes: an input layer, a convolutional layer, a pooling layer, a fully connected layer and an output layer; the residual network structure includes: a first residual block and a second residual block; the optimal residual fusion diagnostic model includes:

[0050] The input of the input layer is the input of the optimal residual fusion diagnosis model, which is used to input the photovoltaic array data to be diagnosed;

[0051] The convolution layer is used to capture features of the photovoltaic array data to be diagnosed, and output the captured feature data to the first residual block;

[0052] The first residual block is used to transform the captured feature data through an identity mapping to obtain first residual transformed feature data, and output it to the pooling layer;

[0053] The pooling layer is used to perform a pooling operation on the first residual transformation feature data, and input the pooled feature data into the second residual block;

[0054] The second residual block is used to transform the pooled feature data through an identity mapping to obtain second residual transformed feature data, and output it to the fully connected layer;

[0055] The fully connected layer is used to integrate the second residual transformation feature data and output the photovoltaic array fault diagnosis result through the output layer. The output of the output layer is the output of the optimal residual fusion diagnosis model.

[0056] Preferably, the expression of the constant mapping transformation is as follows:

[0057] y=F(x,{W i})+x

[0058] Where y is the output of the first residual block or the second residual block; W i is the weight parameter of the convolutional layer; F(x,{W i}) represents the nonlinear transformation function learned by the convolutional layer; x is the input of the first residual block or the second residual block.

[0059] Based on the same inventive concept, the present invention also provides a photovoltaic array fault diagnosis system, comprising:

[0060] Diagnosed data acquisition module: used to acquire the PV array data to be diagnosed;

[0061] Data diagnosis module: used for inputting the photovoltaic array data to be diagnosed into the optimal residual fusion diagnosis model for fault diagnosis to obtain the photovoltaic array fault diagnosis result;

[0062] Among them, the optimal residual fusion diagnosis model is constructed based on any one of the above-mentioned residual fusion diagnosis model construction methods based on a distributed evolutionary algorithm.

[0063] Preferably, the optimal residual fusion diagnostic model is constructed based on a convolutional neural network combined with a residual network structure.

[0064] Preferably, the convolutional neural network includes: an input layer, a convolutional layer, a pooling layer, a fully connected layer and an output layer; the residual network structure includes: a first residual block and a second residual block; the optimal residual fusion diagnostic model includes:

[0065] The input of the input layer is the input of the optimal residual fusion diagnosis model, which is used to input the photovoltaic array data to be diagnosed;

[0066] The convolution layer is used to capture features of the photovoltaic array data to be diagnosed, and output the captured feature data to the first residual block;

[0067] The first residual block is used to transform the captured feature data through an identity mapping to obtain first residual transformed feature data, and output it to the pooling layer;

[0068] The pooling layer is used to perform a pooling operation on the first residual transformation feature data, and input the pooled feature data into the second residual block;

[0069] The second residual block is used to transform the pooled feature data through an identity mapping to obtain second residual transformed feature data, and output it to the fully connected layer;

[0070] The fully connected layer is used to integrate the second residual transformation feature data and output the photovoltaic array fault diagnosis result through the output layer. The output of the output layer is the output of the optimal residual fusion diagnosis model.

[0071] Preferably, the expression of the constant mapping transformation is as follows:

[0072] y=F(x,{W i})+x

[0073] Where y is the output of the first residual block or the second residual block; W i is the weight parameter of the convolutional layer; F(x,{W i}) represents the nonlinear transformation function learned by the convolutional layer; x is the input of the first residual block or the second residual block.

[0074] Based on the same inventive concept, the present invention also provides an electronic device, characterized in that it comprises: at least one processor and a memory; the memory and the processor are connected via a bus;

[0075] The memory is used to store one or more programs;

[0076] When the one or more programs are executed by the at least one processor, a residual fusion diagnosis model construction method based on a distributed evolutionary algorithm as described in any one of the above items, or a photovoltaic array fault diagnosis method as described in any one of the above items is implemented.

[0077] Based on the same inventive concept, the present invention also proposes a readable storage medium, characterized in that an execution program is stored thereon, and when the execution program is executed, a residual fusion diagnosis model construction method based on a distributed evolutionary algorithm as described in any of the above items, or a photovoltaic array fault diagnosis method as described in any of the above items is implemented.

[0078] Compared with the closest prior art, the present invention has the following beneficial effects:

[0079] The present invention provides a method and system for constructing a residual fusion diagnostic model based on a distributed evolutionary algorithm, comprising: obtaining typical characteristic data of a photovoltaic array under various operating conditions; initializing the parameters of the distributed evolutionary algorithm, wherein each individual of the initial population corresponds to a set of initial hyperparameters; based on the typical characteristic data, calculating the fitness of each individual through the residual fusion diagnostic model corresponding to the initial hyperparameters; using the initial population as the parent original population, and selecting the parent elite population based on the fitness of each individual; performing global search optimization and updating on the parent original population to obtain the offspring original population, and performing local search optimization and updating on the parent elite population to obtain the offspring elite population; based on the offspring original population and the offspring elite population, The population is repeatedly iterated and updated using the distributed evolutionary algorithm until the maximum number of iterations is reached, and a set of hyperparameters represented by the individuals with the largest fitness values ​​in the current offspring original population and the current offspring elite population are used as the optimal hyperparameters of the residual fusion diagnosis model; based on the typical feature data, the residual fusion diagnosis model corresponding to the optimal hyperparameters is trained to obtain the optimal residual fusion diagnosis model; the present invention generates typical feature data under various operating conditions through a simulation platform, provides high-quality photovoltaic array data support for the construction of the model, ensures that the constructed model can cover various fault types and environmental conditions, and solves the problem of low diagnostic accuracy of the constructed model due to the scarcity and incompleteness of photovoltaic array data in the prior art. In addition, the present invention uses a distributed evolutionary algorithm to globally search for the optimal solution of the hyperparameters of the residual fusion diagnosis model, combines the synergy of global search and local search, accelerates the training process of the model, improves the construction efficiency of the model, and improves the performance of the model.

[0080] The present invention provides a photovoltaic array fault diagnosis method and system, comprising: obtaining photovoltaic array data to be diagnosed; inputting the photovoltaic array data to be diagnosed into the optimal residual fusion diagnosis model for fault diagnosis, and obtaining a photovoltaic array fault diagnosis result; the model architecture of the optimal residual fusion diagnosis model in the present invention is to introduce a residual network as an extension module in a traditional convolutional neural network, and to solve the problem of gradient vanishing or gradient exploding in the process of increasing the depth of the model to further extract advanced features through the skip connection between the layers of the residual network in the prior art, thereby improving the stability of the residual fusion diagnosis model and the accuracy of photovoltaic fault feature diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 A schematic diagram of a residual fusion diagnosis model construction method based on a distributed evolutionary algorithm provided by the present invention;

[0082] Figure 2 A schematic diagram of a photovoltaic array fault simulation provided by the present invention;

[0083] Figure 3 A schematic diagram of a distributed evolutionary algorithm framework according to the present invention;

[0084] Figure 4 A schematic diagram of the structure of a residual fusion diagnosis model construction system based on a distributed evolutionary algorithm provided by the present invention;

[0085] Figure 5 A schematic diagram of a photovoltaic array fault diagnosis method provided by the present invention;

[0086] Figure 6 A schematic diagram of a photovoltaic array fault diagnosis decision flow chart provided by the present invention;

[0087] Figure 7 A current-voltage characteristic curve under different temperatures and light intensities provided by the present invention;

[0088] Figure 8 A power-voltage characteristic curve under different temperatures and light intensities provided by the present invention;

[0089] Fig. 9 A current-voltage characteristic curve under different faults provided by the present invention;

[0090] Fig.10 A power-voltage characteristic curve under different faults provided by the present invention;

[0091] Fig.11 A schematic diagram of the structure of a photovoltaic array fault diagnosis system provided by the present invention;

[0092] Fig.12 A schematic diagram of an electronic device provided by the present invention. DETAILED DESCRIPTION

[0093] The specific implementation modes of the present invention are further described in detail below with reference to the accompanying drawings.

[0094] Embodiment 1:

[0095] The invention provides a method for constructing a residual fusion diagnosis model based on a distributed evolutionary algorithm. The specific process is as follows: Figure 1 As shown, including:

[0096] Step 11: Obtain typical characteristic data of the photovoltaic array under various operating conditions;

[0097] Step 12: Initialize the parameters of the distributed evolutionary algorithm, wherein each individual of the initial population corresponds to a set of initial hyperparameters; based on the typical feature data, calculate the fitness of each individual through the residual fusion diagnostic model corresponding to the initial hyperparameters; use the initial population as the parent original population, and select the parent elite population based on the fitness of each individual; perform global search optimization and update on the parent original population to obtain the offspring original population, and perform local search optimization and update on the parent elite population to obtain the offspring elite population; based on the offspring original population and the offspring elite population, use the distributed evolutionary algorithm to perform repeated iterative updates until the maximum number of iterations is reached, and use a set of hyperparameters represented by the individual with the largest fitness value in the current offspring original population and the current offspring elite population as the optimal hyperparameters of the residual fusion diagnostic model;

[0098] Step 13: Based on the typical feature data, the residual fusion diagnosis model corresponding to the optimal hyperparameter is trained to obtain the optimal residual fusion diagnosis model.

[0099] Specifically, in step 11, it is considered that the existing technology has the problem that the model is difficult to effectively learn representative fault features due to insufficient data, and then the generalization performance is poor, which limits its application potential in actual photovoltaic array fault diagnosis. The present invention simulates data in various real situations in reality through a simulation platform for subsequent model training and construction. The data generated by the simulation platform has the advantages of strong integrity and availability, which is conducive to the model to capture the typical characteristics of the data, thereby improving the performance of the model. Specifically, in the present invention, based on the circuit structure of the photovoltaic array, the simulation platform is used to obtain typical characteristic data under various operating states, wherein the various operating states include one or more of the following: normal working state, short circuit fault state, open circuit fault state, aging fault state, partial shading state, aging shading fault state, short circuit shading fault state and open circuit shading fault state; the typical characteristic data includes one or more of the following: open circuit voltage, short circuit current, maximum power point voltage, maximum power, fill factor, temperature and light intensity.

[0100] In a specific embodiment of the present invention, Figure 2The schematic diagram of photovoltaic array fault simulation includes a power supply, a photovoltaic array with an open circuit fault, a photovoltaic array with a short circuit fault, a photovoltaic array with an aging fault, and a photovoltaic array with a shading fault. Usually under standard test conditions, typical single faults of a photovoltaic array include short circuit, open circuit, aging, and partial shading. In a specific embodiment of the present invention, an open circuit fault is simulated by connecting an infinitely large resistor in series between two parallel photovoltaic cell branches. A short circuit fault is simulated by connecting an infinitesimal resistor close to 0 in parallel across a photovoltaic cell. An aging fault simulates the aging effect by connecting a 5Ω resistor in series between two parallel photovoltaic cell branches. A partial shading fault simulates the effect of partial shading by adjusting the light intensity of the photovoltaic module.

[0101] There are multiple types of faults that photovoltaic arrays may experience simultaneously during actual operation. For example, multiple composite faults such as aging and shading, short circuit and shading, open circuit and shading occur simultaneously. Composite faults increase the complexity of photovoltaic array operation and the difficulty of fault diagnosis. In order to conduct in-depth analysis, in a specific embodiment of the present invention, three composite fault conditions, namely aging shading, short circuit shading and open circuit shading, are selected for analysis and simulation. In this way, the operating characteristics and parameter changes of photovoltaic arrays under different composite fault conditions can be more comprehensively understood, thereby improving the accuracy and efficiency of fault diagnosis.

[0102] In the present invention, typical characteristic data include one or more of the following: open circuit voltage, short circuit current, maximum power point voltage, maximum power, fill factor, temperature and light intensity. Considering the use of models to diagnose photovoltaic array faults, the selection of fault characteristic data is also crucial. The reason for feature selection is mainly to ensure that the data input to the model can fully reflect the operating status of the photovoltaic array components, thereby improving the accuracy of fault diagnosis. The following are the expressions of several key features in the above typical characteristic data and the reasons for their selection:

[0103] (1) Open circuit voltage (Uoc) is the maximum voltage value of a photovoltaic module under no-load conditions, which can reflect the health of the module. When the module ages or has an open circuit failure, the open circuit voltage will change significantly. The open circuit voltage can usually be described by the following formula:

[0104]

[0105] Among them: U oc is the open circuit voltage of the photovoltaic module; n is the ideal factor of the diode, k is the Boltzmann constant, T is the absolute temperature, q is the electron charge, I ph is the photogenerated current, which varies with light intensity, and I0 is the reverse saturation current.

[0106] (2) Short-circuit current (Isc) is the maximum current value of the component when there is no voltage. It is sensitive to changes in light intensity. When a short circuit or shading occurs, the short-circuit current will change significantly, so it is an important feature for diagnosing abnormal current faults. The short-circuit current can usually be described by the following formula:

[0107] I sc ≈I ph (2)

[0108] Where: I sc is the short-circuit current of the photovoltaic module, I ph is the photogenerated current, which varies with light intensity.

[0109] (3) Maximum power point voltage (Um) and maximum power point current (Im) represent the operating voltage and current of the PV module at maximum power output, which are directly related to the output performance of the module. In the event of a fault, the parameters of the maximum power point tend to deviate significantly from the normal values. The maximum power point voltage and current are the voltage and current of the PV module when it outputs maximum power, which are usually found through the IV curve of the PV array.

[0110] (4) Maximum power (Pm) is the core indicator of the output power of photovoltaic modules, reflecting the actual working efficiency of the modules. By monitoring the changes in maximum power, the performance degradation or failure of the modules can be quickly detected. At this time, the power Pm is the maximum and meets the following conditions:

[0111] P m =U m ×I m (3)

[0112] Among them, P m is the maximum power of the photovoltaic module; U m is the maximum power point voltage of the photovoltaic module; I m is the maximum power point current of the PV module.

[0113] (5) Fill factor (FF) is one of the performance parameters of the module, reflecting the nonlinear relationship between voltage and current. The change of fill factor can reveal problems such as increased internal loss of the module and material aging. The higher the fill factor, the better the performance of the photovoltaic module. The formula of FF is as follows:

[0114]

[0115] Where: FF is the filling factor of the photovoltaic module; P m is the maximum power of the photovoltaic module; U oc is the open circuit voltage of the photovoltaic module; I sc is the short-circuit current of the PV module;

[0116] (6) The output performance of photovoltaic modules is very sensitive to environmental factors, especially temperature and light intensity. These two parameters directly affect electrical characteristics such as open circuit voltage and short circuit current, so changes in environmental conditions must be considered in order to distinguish between normal fluctuations and fault signals.

[0117] U oc (T) = U oc (T0)+α(T-T0) (5)

[0118] I sc (T) = I sc (G0)+β(G-G0) (6)

[0119] Among them: U oc (T) is the open circuit voltage of the photovoltaic module at the current temperature; U oc (T0) is the open circuit voltage of the photovoltaic module at the reference temperature; I sc (T) is the short-circuit current of the photovoltaic module at the current temperature; I sc (G0) is the short-circuit current of the photovoltaic module under the reference light intensity; α is the temperature coefficient of voltage changing with temperature, β is the light coefficient of current changing with light intensity, T is the current temperature, T0 is the reference temperature, G is the current light intensity, and G0 is the reference light intensity.

[0120] In the current field of photovoltaic array fault diagnosis, the problem that the accuracy and reliability of the model's fault diagnosis in a small sample environment due to insufficient sample data needs to be improved. The present invention simulates the typical characteristics of various fault conditions through a simulation platform, and comprehensively considers the above key features. During the model training process, the diagnostic model can fully understand the operating status of the photovoltaic array and reduce the possibility of false alarms and missed alarms. The selection of these features can also enhance the generalization ability of the model, especially in a small sample environment, to ensure that the diagnostic model can learn effective fault modes from limited data.

[0121] Specifically, in step 12, the residual fusion diagnosis model is constructed based on the convolutional neural network (CNN) combined with the residual network structure. Considering that the traditional deep convolutional network is prone to gradient vanishing or gradient explosion when the network depth is increased, which makes the model difficult to converge, further, how to improve the accuracy and reliability of fault diagnosis in a small sample environment has become a key issue to be solved in the current research field. With the development of machine learning and deep learning technology, the application of convolutional neural network (CNN) and other machine learning methods in photovoltaic array fault diagnosis has significantly improved the automation and accuracy of diagnosis. Through the automatic feature extraction and powerful pattern recognition capabilities of CNN, the limitations of traditional methods in complex fault modes are overcome, and noisy data and signal incompleteness can be effectively processed. In practical applications, these methods not only improve the accuracy and robustness of photovoltaic array fault diagnosis, but also provide important support for the intelligent operation and maintenance and fault management of photovoltaic systems, and promote the reliability and economic development of photovoltaic power generation systems.

[0122] The present invention solves the above technical problems by introducing a residual network structure as an extension module in a convolutional neural network (CNN). The residual network structure is usually composed of one or more residual blocks, and the skip connection function realized by the residual block improves the performance of the neural network model. In a specific embodiment of the present invention, the specific structure of the residual fusion diagnosis model includes: a convolutional neural network and a residual network, wherein the convolutional neural network includes: an input layer, a convolutional layer, a pooling layer, a fully connected layer and an output layer; the residual network includes one or more residual blocks, and in a specific embodiment of the present invention, the residual network is a first residual block and a second residual block. The present invention introduces a residual network structure as an extension module of CNN. The residual network adds skip connections between layers so that information can be directly transmitted between different network layers, thereby effectively alleviating the gradient vanishing problem in the deep network, while retaining the ability of the deep network to extract complex fault features. Through this improvement, CNN can not only maintain its powerful feature learning ability under small sample conditions, but also further improve the recognition accuracy of photovoltaic fault features and the stability of the model through the introduction of the residual module.

[0123] In summary, the present invention integrates the residual network into the convolutional neural network and constructs the architecture of the residual fusion diagnostic model. The hyperparameters in the specific residual fusion diagnostic model need to calculate the optimal hyperparameter combination through the distributed evolutionary algorithm proposed in the present invention, and finally determine the corresponding optimal residual fusion diagnostic model based on the optimal hyperparameters.

[0124] The present invention adopts the distributed algorithm framework and the Memetic algorithm to form a new distributed algorithm framework, which integrates the evolutionary algorithm on the basis of this framework. The evolutionary algorithm is used as the global search algorithm, and the local search adopts the variable neighborhood search, which not only ensures the global search capability, but also improves the local search capability of the algorithm, avoiding the result from falling into the local optimum. Specifically, Figure 3 It is an overall framework diagram of the distributed evolutionary algorithm, which reflects the main idea of ​​the distributed evolutionary algorithm proposed in the present invention. First, the original population POP is initialized, and then excellent individuals are selected to obtain the initialized elite population EAR. Different optimization and update algorithms are used for the two populations. A global search for the best individual is performed on the original population, and a local search for the best individual is performed on the elite population. After exchanging individuals in the original population and the elite population, the global search and local search are continued.

[0125] Based on the aforementioned typical feature data, the present invention solves the optimal hyperparameters of the residual fusion diagnosis model through a distributed evolutionary algorithm, specifically, including:

[0126] Step 12.1: Initialize the original population POP, including: initialize the parameters of the distributed evolutionary algorithm, wherein each individual in the initial population corresponds to a set of initial hyperparameters. In a specific embodiment of the present invention, in order to ensure that the individuals in the original population are full of diversity, the present invention uses random initialization to generate individuals, and each individual gene in the initial population corresponds to a set of initial hyperparameters. Each gene in the individual gene represents a hyperparameter, and each hyperparameter has its lower and upper bounds, and the values ​​are continuous. The present invention uses real number coding to represent each individual. Among them, a set of hyperparameters includes one or more of the following: convolution kernel size, number of convolution kernels, pooling layer size, learning rate, number of iterations, etc.

[0127] Assuming that the genetic dimension of each individual is d, where each genetic dimension represents a hyperparameter in a set of hyperparameters, the lower bound of each genetic dimension is lb, the upper bound is ub, and the initial population size is pop_size, then a random matrix of [pop_size×d] is generated, where each row represents an individual and each column represents the value of the individual in a genetic dimension. For each genetic dimension of each individual, a random number between the upper bound ub and the lower bound lb is generated. The specific formula is as follows:

[0128] X i,j =lb+(ub-lb)×rand() (7)

[0129] Where: X i,jrepresents the number of the jth genetic dimension of the ith individual, lb represents the lower bound of the individual dimension, ub represents the upper bound of the individual dimension, and rand() is a function that generates uniformly distributed random numbers between [0,1]. This ensures that each dimension of each individual in the population is randomly distributed within the corresponding upper and lower bounds, thereby ensuring the diversity of the population.

[0130] Step 12.2: Evaluate the fitness of individuals in the initial population, including: based on the typical feature data, through the residual fusion diagnostic model corresponding to the initial hyperparameters, calculate the fitness of each individual. Specifically, in the present invention, the fitness of each individual in the initial population is determined by the accuracy of the residual fusion diagnostic model corresponding to a set of hyperparameters corresponding to each individual on the validation set. Specifically, based on the typical feature data generated by the simulation platform in step 11, the fitness of each individual is calculated through the residual fusion diagnostic model corresponding to the initial hyperparameters.

[0131] Step 12.3: Selecting the parent original population and the parent elite population, including: using the initial population as the parent original population, or selecting individuals whose fitness values ​​meet the preset values ​​from the initial population as the parent original population. Afterwards, the parent elite population is selected from the parent original population based on the fitness of each individual. In a specific embodiment of the present invention, in order to ensure the diversity of the elite population, the spatial distribution can be considered when selecting the optimal individuals to avoid selecting individuals that are too close to each other. If the distance between individuals is too close, the individuals with higher rankings but too close can be skipped, and the individuals with slightly lower rankings but farther distance can be selected.

[0132] The initialization method of the parent elite population is designed to ensure the high quality of its members, so as to carry out targeted local development of surrounding potential areas and effectively guide the evolution of the entire original population (POP). In the subsequent evolutionary process, the elite population will be continuously updated through local search. At each iteration, if the newly generated individuals are of better quality than those in the parent elite population, the new individuals will replace the individuals of poor quality in the parent elite population. This ensures that the elite population always maintains high-quality elite individuals, thereby effectively guiding the evolution of the entire population.

[0133] Step 12.4: Global search, including: using a genetic algorithm to perform global search optimization and update on the parent original population to obtain a child original population with better fitness. Throughout the process, the elite retention strategy is used to save the current best hyperparameter combination to prevent the loss of the optimal solution. In the present invention, the global search uses a genetic algorithm (GA) to optimize the hyperparameters of the residual fusion diagnostic model. The genetic algorithm iteratively optimizes the individuals in the parent original population by simulating the natural evolution process, including operations such as selection, crossover and mutation. Specifically, in a specific embodiment of the present invention, the fitness of each individual in the original population is first evaluated, and the fitness is determined by the accuracy of the residual fusion diagnostic model corresponding to each set of hyperparameters on the validation set. Then, individuals with high fitness are selected as the parent original population for crossover and mutation operations to generate new individuals called the child original population. Through global update iterations, the optimal hyperparameter combination is gradually searched to ensure that the performance of the residual fusion diagnostic model corresponding to the screened hyperparameter combination is better. Among them, the main steps and related formulas of the genetic algorithm are as follows, including selection, crossover and mutation:

[0134] (1) Selection: The parent individuals are selected according to the fitness values ​​of the individuals. The selection strategy in the present invention can be: roulette selection or tournament selection.

[0135] For roulette wheel selection: the probability of each individual being selected is proportional to its fitness. The expression of roulette wheel selection probability is as follows:

[0136]

[0137] Where: P(x i ) represents the roulette wheel selection probability of the i-th individual; N is the population size, x i represents the i-th individual, f(x i ) represents the fitness of the i-th individual in the population, f(x j ) represents the fitness of the jth individual in the population.

[0138] For tournament selection: Several individuals are randomly selected from the population for comparison, and the individual with the highest fitness is selected as the parent.

[0139] (2) Crossover: The crossover operation generates a new offspring individual by exchanging part of the genetic information of two parent individuals. In the present invention, the crossover method can be: single-point crossover, double-point crossover or uniform crossover. In a specific embodiment of the present invention, assume that the first parent individual is x p =[x p,1 ,……,x p,n ], the second parent individual is x q =[x q,1 ,……,x q,n], the crossover point is position k, and the formula for generating offspring individuals is:

[0140] x c =[x p,1 ,…,x p,k ,x q,k+1 ,…,x q,n ] (9)

[0141] Among them, x c is the gene of the offspring individual, x p,k is the gene of the crossover point k in the offspring individual gene, and the genes from 1 to k are the genes of the first parent individual, x q,k+1 is the gene after the crossover point k in the offspring individual gene, and the genes from k+1 to n are the genes of the second parent individual, x q,n is the nth gene of the second parent individual, x p,n is the nth gene of the first parent individual.

[0142] (3) Mutation: Mutation is to maintain the diversity of the population and prevent it from falling into a local optimal solution. The mutation operation randomly perturbs the genes of individuals. In a specific embodiment of the present invention, the mutation method adopts uniform mutation:

[0143] X i,j =x i,j +σ·N(0,1) (10)

[0144] Where: X i,j is the jth gene of the ith individual after mutation, x i,j is the jth gene of the ith individual; σ is the variation intensity; N(0,1) is the standard normal distribution.

[0145] Step 12.5: Local search.

[0146] Variable Neighborhood Search (VNS) is a heuristic optimization algorithm. Its main idea is to find the global optimal solution by constantly changing the neighborhood structure. The so-called neighborhood refers to other solutions around the current solution, that is, mutation is performed on the basis of the current solution. It uses new solutions generated by different mutations to expand the local search range, thereby improving the local search capability.

[0147] In the present invention, a variable neighborhood search algorithm is used to perform local search optimization and update on individuals in the parent elite population to obtain a child elite population with better fitness. Specifically, based on the individuals in the parent elite population, the neighborhood structure corresponding to each individual is determined; the individuals are searched in the neighborhood structure to determine whether the fitness value of the individuals in the neighborhood structure is greater than the fitness value of the individuals in the parent elite population; if so, the individuals in the neighborhood structure are replaced by the individuals in the parent elite population to obtain a child elite population updated by local search; otherwise, the local search is continued, and when the maximum number of iterations is reached, the parent elite population is used as the child elite population.

[0148] In a specific embodiment of the present invention, variable neighborhood search is performed by the following steps: Step 12.4.1: Initialization. Select individuals from the elite population; Step 12.4.2: Construct a neighborhood structure. Determine the number and type of neighborhood structure. Usually the number of neighborhoods is 1-3, and the types of neighborhood structures usually include mutation, crossover, and small adjustments. Step 12.4.3: Search for the best individual in the neighborhood. Search in each neighborhood of the solution represented by the current individual. For each neighborhood, apply the corresponding neighborhood operation to generate a new individual. Step 12.4.4: Select the updated individual. If the fitness value of the new individual found is better than the current individual, the currently found individual is updated to the new individual. If there is no individual with a better fitness value, keep the current individual unchanged and continue searching in other neighborhoods. Step 12.4.5: Loop iteration. Repeat the above steps until the termination condition is met. Specifically, in the present invention, the termination condition can be that the maximum number of iterations is reached, or the fitness value of the new individual is better than the current individual.

[0149] In the present invention, the neighborhood structure is constructed based on the variable neighborhood search algorithm and the representation of the solution. In a specific embodiment of the present invention, real number coding is adopted, and the genes of each individual represent a set of hyperparameters of the residual fusion diagnosis model. The neighborhood of this individual can be mutated on the basis of the original parameters to generate new individuals, or add or subtract a random number to mutate, fine-tune, etc. The corresponding pseudo code of the variable neighborhood search is shown in Table 1, where x k _fit represents the new solution in the neighborhood structure, and x_fit represents the current solution in the neighborhood:

[0150]

[0151]

[0152] Table 1

[0153] Step 12.6: Perform repeated iterative updates using a distributed evolutionary algorithm, including: performing repeated iterative updates based on the offspring original population and the offspring elite population using the distributed evolutionary algorithm until a maximum number of iterations is reached, and taking a set of hyperparameters represented by the individuals with the largest fitness values ​​in the current offspring original population and the current offspring elite population as the optimal hyperparameters of the residual fusion diagnostic model.

[0154] Specifically, in a specific embodiment of the present invention, first, the individuals in the offspring elite population and the offspring original population are exchanged. This process is usually called knowledge transfer. There are two main ways, one is to directly exchange individuals, and the other is to exchange certain characteristics of individuals. The exchange method can be flexibly selected as needed. The main purpose of knowledge transfer is to generate new solutions by exchanging individuals or partial characteristics to achieve more extensive search and gene transfer. High-quality individuals in the global population can be used to improve the current individuals, thereby improving the overall performance of the algorithm.

[0155] In the present invention, the individual information of the individual with the smallest fitness value in the offspring original population is replaced by the individual information of the random individual in the offspring elite population to obtain the optimized offspring original population; in a specific embodiment of the present invention, one individual is randomly selected from the offspring elite population to optimize the worst individual in the offspring original population.

[0156] Assume that there are two individuals X1 and X2 for exchange, and the eigenvector of each individual is:

[0157] X1=[x 11 ,x 12 …x 1d ] (11)

[0158] X2=[x 21 ,x 22 …x 2d ] (12)

[0159] Among them, x 11 is the gene of the first dimension of the first individual; x 12 is the gene of the second dimension of the first individual; x 1d is the gene of the dth dimension of the first individual; x 21 is the gene of the first dimension of the second individual; x 22 is the gene of the second dimension of the second individual; x 2d is the gene of the dth dimension of the second individual;

[0160] Usually, the individual exchange methods include direct exchange and feature exchange. Direct exchange means replacing all the features of one individual with all the features of another individual, that is, complete replacement. The individuals after exchange are X1' and X2'. The corresponding expression is:

[0161] X1'=[x 21 ,x 22 …x 2d ] (13)

[0162] X2'=[x 11 ,x 12 …x 1d ] (14)

[0163] Feature exchange means exchanging some features between two individuals, that is, selecting a specific dimension for exchange instead of the entire individual. Assuming that the kth feature is selected for exchange, the mathematical formula for feature exchange is:

[0164] X1'=[x 11 ,…,x 1(k-1) ,x 2k ,x 1(k+1) …x 1d ] (15)

[0165] X2'=[x 21 ,…,x 2(k-1) ,x 1k ,x 2(k+1) …x 2d ] (16)

[0166] Among them, x 11 is the gene of the first dimension of the first individual; x 1(k-1) is the gene of the k-1th dimension of the first individual; x 2k is the gene of the kth dimension of the second individual; x 1(k+1) is the gene of the k+1th dimension of the first individual; x 1d is the gene of the dth dimension of the first individual; x 21 is the gene of the first dimension of the second individual; x 2(k-1) is the gene of the k-1th dimension of the second individual; x 1k is the gene of the kth dimension of the first individual; x 2(k+1) is the gene of the k+1th dimension of the second individual; x 2dis the gene of the dth dimension of the second individual; since both methods have their own advantages, in a specific implementation, different knowledge transfer methods can be selected according to specific problems, or both methods can be adopted, and each knowledge transfer method is determined according to the fitness of the newly generated individual. Afterwards, based on the optimized offspring original population obtained after the knowledge transfer exchange optimization, the dominant individuals are screened out and the dominant individuals are used as the new parent original population; in a specific implementation of the present invention, an elite retention strategy is used to save the best hyperparameter combination in the current population to prevent the loss of the optimal solution. Finally, a new parent elite population is selected based on the fitness of each individual in the new parent original population; a new child original population is obtained by performing global search optimization and update on the new parent elite population, and a new child elite population is obtained by performing local search optimization and update on the new parent elite population; it is determined whether the maximum number of iterations has been reached, and if so, a set of hyperparameters represented by the individual with the largest fitness value in the current child elite population and the child original population is output as the optimal hyperparameters of the residual fusion diagnostic model; otherwise, based on the current child elite population and the child original population, iteration is continued until the maximum number of iterations is reached.

[0167] The present invention introduces a distributed evolutionary algorithm to globally optimize the relevant hyperparameters of the residual fusion diagnosis model. The distributed evolutionary algorithm effectively avoids the local optimal problem and accelerates the convergence of model parameters by parallelizing the global search and information exchange of sub-populations. This method can improve the generalization performance and diagnostic accuracy of the model under the condition of data scarcity, thereby providing strong support for the efficient and reliable operation and maintenance of photovoltaic systems.

[0168] In the prior art, fixed model hyperparameters are usually used or the model hyperparameters are manually and repeatedly adjusted to optimize model performance. This is not only time-consuming and inefficient, but also easily causes the model to fall into local optimality and cannot fully adapt to complex fault characteristics. The present invention uses a distributed evolutionary algorithm to perform global search and local optimization on the hyperparameters of the residual fusion diagnosis model, namely: convolution kernel size, learning rate, number of iterations, etc., and finally obtains the optimal hyperparameters. This method can automatically find the optimal hyperparameter combination in a relatively short time, significantly improving the diagnostic performance of the model. The automated hyperparameter optimization process reduces the complexity of manual adjustment, improves optimization efficiency, and avoids the problem that traditional methods are prone to falling into local optimality.

[0169] In step 13, the corresponding residual fusion diagnostic model is determined based on the optimal hyperparameters obtained in step 12, and then the residual fusion diagnostic model corresponding to the optimal hyperparameters is trained based on the typical feature data to obtain the optimal residual fusion diagnostic model. Among them, the model architecture of the optimal residual fusion diagnostic model is based on a convolutional neural network and is constructed in combination with a residual network structure. The hyperparameters in the optimal residual fusion diagnostic model are calculated using a distributed evolutionary algorithm. The optimal hyperparameters are model parameters that can optimize the overall performance of the residual fusion diagnostic model.

[0170] The present invention trains the model through typical feature data generated by a simulation platform, and utilizes the advantages of strong integrity and high quality of the data generated by the simulation platform to enable the residual fusion diagnosis model to better capture the features in the data, further improving the diagnostic performance of the residual fusion diagnosis model, and solving the problem in the prior art that the operating state of the photovoltaic array is affected by various environmental factors, resulting in more data noise and low quality of training data, which in turn affects the model training effect.

[0171] The present invention aims at the problems existing in the existing photovoltaic array fault diagnosis methods and provides a fault diagnosis method based on distributed evolutionary algorithm and convolutional neural network hyperparameter optimization. In the prior art, fixed CNN hyperparameters are usually used or hyperparameters are repeatedly adjusted manually to optimize model performance. This is not only time-consuming and inefficient, but also easily causes the model to fall into local optimality and cannot fully adapt to complex fault characteristics. In addition, the operating state of the photovoltaic array is affected by a variety of environmental factors, such as light intensity and temperature changes. Traditional diagnostic methods are difficult to respond to these changes in real time and accurately, which in turn affects the accuracy of fault diagnosis.

[0172] In order to solve the above problems, the present invention uses a distributed evolutionary algorithm to perform global search and local optimization on the hyperparameters of CNN (including convolution kernel size, learning rate, number of iterations, etc.). This method can automatically find the optimal hyperparameter combination in a short time, significantly improving the diagnostic performance of the model. At the same time, combined with the Matlab / Simulink simulation platform, multiple fault states of the photovoltaic array are simulated, typical fault feature data are generated, and data support is provided for the training and testing of the model to ensure that the diagnostic model can cover various fault types and environmental condition changes. Through the method of the present invention, the automated hyperparameter optimization process reduces the complexity of manual adjustment, improves the optimization efficiency, and avoids the problem that traditional methods are prone to fall into local optimality. At the same time, the present invention enhances the adaptability and robustness of the CNN model to photovoltaic array fault diagnosis, can quickly and accurately identify different faults, and ensures the accuracy and efficiency of the diagnostic process.

[0173] In summary, the present invention proposes an innovative method for the fault diagnosis problem of photovoltaic arrays in a small sample environment. This method combines the mechanisms of distributed evolutionary algorithm and convolutional neural network, deeply explores the field of data analysis and machine learning, and designs an efficient and adaptable solution for common faults in photovoltaic energy systems.

[0174] Embodiment 2:

[0175] Based on the same inventive concept, the present invention also provides a residual fusion diagnosis model construction system based on a distributed evolutionary algorithm. The specific system structure is as follows: Figure 4 As shown, including:

[0176] Data acquisition module: used to obtain typical characteristic data of photovoltaic arrays under various operating conditions;

[0177] Optimal hyperparameter solution module: used to initialize the parameters of the distributed evolutionary algorithm, in which each individual of the initial population corresponds to a set of initial hyperparameters; based on the typical feature data, the fitness of each individual is calculated through the residual fusion diagnostic model corresponding to the initial hyperparameters; the initial population is used as the parent original population, and the parent elite population is selected based on the fitness of each individual; the parent original population is globally searched and optimized to obtain the offspring original population, and the parent elite population is locally searched and optimized to obtain the offspring elite population; based on the offspring original population and the offspring elite population, the distributed evolutionary algorithm is used to perform repeated iterative updates until the maximum number of iterations is reached, and a set of hyperparameters represented by the individual with the largest fitness value in the current offspring original population and the current offspring elite population is used as the optimal hyperparameters of the residual fusion diagnostic model;

[0178] Model building module: used to train the residual fusion diagnosis model corresponding to the optimal hyperparameters based on the typical feature data to obtain the optimal residual fusion diagnosis model.

[0179] Preferably, the optimal hyperparameter solution module includes: a global search optimization update unit: used to use a genetic algorithm to perform a global search optimization update on the parent original population to obtain a offspring original population with better fitness; a local search optimization update unit: used to use a variable neighborhood search algorithm to perform a local search optimization update on the individuals in the parent elite population to obtain a offspring elite population with better fitness.

[0180] Preferably, the local search optimization update unit includes: a neighborhood structure determination subunit: used to determine the neighborhood structure corresponding to each individual based on the individuals in the parent elite population; an individual judgment subunit: used to search for individuals in the neighborhood structure and judge whether the fitness value of the individuals in the neighborhood structure is greater than the fitness value of the individuals in the parent elite population; an individual replacement subunit: used to replace the individuals in the parent elite population with the individuals in the neighborhood structure when the fitness value of the individuals in the neighborhood structure is greater than the fitness value of the individuals in the parent elite population, so as to obtain a locally search updated offspring elite population; a offspring population determination subunit: used to use the parent elite population as the offspring elite population when the fitness value of the individuals in the neighborhood structure is not greater than the fitness value of the individuals in the parent elite population and the maximum number of iterations is reached.

[0181] Preferably, the optimal hyperparameter solution module also includes: a knowledge transfer unit: used to replace the individual information of the individual with the smallest fitness value in the offspring original population with the individual information of the random individuals in the offspring elite population, so as to obtain an optimized offspring original population; a dominant individual retention unit: used to screen out the dominant individuals in the optimized offspring original population based on the optimized offspring original population, and use the dominant individuals as the new parent original population; a parent elite population selection unit: used to select a new parent elite population based on the fitness of each individual in the new parent original population; distributed optimization An updating unit is used to perform global search optimization and update on the new parent original population to obtain a new offspring original population, and to perform local search optimization and update on the new parent elite population to obtain a new offspring elite population; an iteration judgment unit is used to judge whether the maximum number of iterations has been reached; if so, a set of hyperparameters represented by the individual with the largest fitness value in the current offspring elite population and the offspring original population is output as the optimal hyperparameters of the residual fusion diagnostic model; otherwise, based on the current offspring elite population and the offspring original population, the iteration continues until the maximum number of iterations is reached.

[0182] Preferably, the data acquisition module is specifically used to: based on the circuit structure of the photovoltaic array, obtain typical characteristic data under various operating conditions through a simulation platform; the various operating conditions include one or more of the following: normal working state, short circuit fault state, open circuit fault state, aging fault state, partial shading state, aging shading fault state, short circuit shading fault state and open circuit shading fault state; the typical characteristic data include one or more of the following: open circuit voltage, short circuit current, maximum power point voltage, maximum power, fill factor, temperature and light intensity.

[0183] The present invention provides a residual fusion diagnostic model construction system based on a distributed evolutionary algorithm. The system can obtain high-quality data covering various fault types and changes in environmental conditions through a data acquisition module, thereby providing data support for model construction. In addition, through an optimal hyperparameter solving module, a distributed evolutionary algorithm can be used to globally search for the optimal solution of the hyperparameters of the residual fusion diagnostic model. The synergistic effect of global search and local search is combined to accelerate the model training process and improve the model construction efficiency.

[0184] Embodiment 3:

[0185] Based on the same inventive concept, the present invention also provides a photovoltaic array fault diagnosis method, the specific process is as follows: Figure 5 As shown, including:

[0186] Step 21: Acquire the photovoltaic array data to be diagnosed;

[0187] Step 22: inputting the photovoltaic array data to be diagnosed into the optimal residual fusion diagnosis model for fault diagnosis to obtain a photovoltaic array fault diagnosis result;

[0188] Among them, the optimal residual fusion diagnosis model is constructed based on the residual fusion diagnosis model construction method based on a distributed evolutionary algorithm described in Example 1.

[0189] The optimal residual fusion diagnosis model in the present invention mainly solves the problems in the prior art by introducing the residual network structure as an extension module in the convolutional neural network (CNN). In the convolutional neural network, the fusion of residual blocks can significantly improve the training effect and performance of the network. When the traditional deep convolutional network increases the network depth, it is easy to have the problem of gradient vanishing or gradient explosion, which makes the model difficult to converge. The introduction of residual blocks can effectively solve this problem through skip connections, thereby improving the performance of the neural network. The core idea of ​​the residual block is to allow the input to bypass the middle convolution layer directly through identity mapping, thereby avoiding information loss. By fusing residual blocks in CNN, the network can further extract advanced features by increasing the depth without increasing the difficulty of training due to the gradient vanishing problem. Finally, the residual network structure greatly improves the training effect of the deep network and shows strong performance in various tasks.

[0190] Specifically, in a specific embodiment of the present invention, the specific structure of the optimal residual fusion diagnosis model includes: an input layer, a convolution layer, a pooling layer, a fully connected layer, an output layer, a first residual block and a second residual block; the input of the input layer is the input of the residual fusion diagnosis model; the convolution layer is used to capture the features of the photovoltaic array data to be diagnosed input through the input layer, and output the captured typical feature data to the first residual block; the first residual block is used to transform the captured feature data through an identity mapping to obtain first residual transformation feature data, and output it to the pooling layer; the pooling layer is used to perform a pooling operation on the first residual transformation feature data, and input the pooled feature data to the second residual block; the second residual block is used to transform the pooled feature data through an identity mapping to obtain second residual transformation feature data, and output it to the fully connected layer; the fully connected layer is used to integrate the second residual transformation feature data, and output the photovoltaic array fault diagnosis result through the output layer, and the output of the output layer is the output of the residual fusion diagnosis model. Among them, identity mapping bypasses the middle convolutional layer and adds skip connections between layers, so that information can be directly transmitted between different network layers, thereby avoiding information loss. Specifically, assuming the input is x, the output of the residual block is:

[0191] y=F(x,{W i})+x

[0192] Where F(x,{W i}) represents the nonlinear transformation function learned by the convolutional layer, W i is the weight parameter of the convolutional layer, x is the input, and y is the output of the residual block. The advantage of this structure is that even if the output F(x) of the convolutional layer is small or close to zero, the network can still retain the input information through direct skip connections, thereby alleviating the gradient vanishing problem in deep networks.

[0193] In a specific embodiment of the present invention, the decision-making process from the construction of the residual fusion diagnosis model to the fault diagnosis is as follows: Figure 6As shown, it specifically includes: step S1: using Matlab / Simulink simulation platform to simulate the photovoltaic array, simulating various common photovoltaic array faults through the platform and outputting fault feature data under various fault states for training the model; step S2: initializing hyperparameters and the constraint range of each hyperparameter, including the lower bound lb and upper bound ub of parameters such as convolution kernel size kernal_size, learning rate learning_rate, number of iterations epochs, etc. Step S3: using distributed algorithm to optimize hyperparameters, and calculating the fitness value of individuals, iterating according to the algorithm stop condition, if it meets the preset maximum number of iterations, then output the optimal hyperparameters, otherwise continue to iterate. Step S4: output the optimized optimal hyperparameters, and apply the optimal hyperparameters to the CNN model, construct the DA-CNN model, and use the training data generated in step S1 for training, and then save the trained model to obtain the optimal residual fusion diagnosis model. Step S5: randomly generate test data within a reasonable light intensity and temperature range through the photovoltaic array simulation platform, and evaluate the data with the optimal residual fusion diagnosis model generated in step S4, so as to output the diagnosis result.

[0194] In view of the problem of photovoltaic array fault diagnosis in a small sample environment existing in the prior art, the present invention proposes a diagnostic method combining a distributed evolutionary algorithm and an improved convolutional neural network. The present invention uses a distributed evolutionary algorithm to globally optimize the relevant hyperparameters of the residual fusion diagnostic model, and through the global search, local search and information exchange of parallel sub-populations, effectively avoids the problem of local optimality, accelerates the convergence of model parameters, obtains the hyperparameter combination that makes the model performance optimal, and then constructs the optimal residual fusion diagnostic model. The optimal residual fusion diagnostic model constructed in the present invention integrates the residual network structure, so that the model can maintain high generalization performance and diagnostic accuracy under the condition of data scarcity, thereby providing strong support for the efficient and reliable operation and maintenance of photovoltaic systems. At the same time, the present invention enhances the adaptability and robustness of the residual fusion diagnostic model constructed based on the neural network (CNN) model to the diagnosis of photovoltaic array faults, can quickly and accurately identify different faults, and ensure the accuracy and efficiency of the diagnostic process.

[0195] Embodiment 4:

[0196] The present invention also discloses a photovoltaic array fault diagnosis method based on an improved convolutional neural network (CNN) and a distributed evolutionary algorithm. In order to verify the effect of the method in practical applications, the present invention uses the Matlab / Simulink simulation platform to build a 3×3 photovoltaic array model and simulates the output characteristics of the photovoltaic array under various fault conditions to generate fault feature data for training and testing the neural network model. Figure 7is the current-voltage characteristic curve under different temperatures and light intensities. Figure 8 The power-voltage characteristic curves under different temperatures and light intensities are shown in Figure 7 and Figure 8 T stands for temperature, G stands for light intensity. Fig. 9 is the current-voltage characteristic curve under different operating conditions, Fig.10 is the power voltage characteristic curve under different operating conditions, Fig. 9 and Fig.10 They are all photovoltaic characteristic curves, where different operating states include: short circuit fault state, normal operating state, open circuit fault state, shadow fault state, aging fault state, short circuit shadow fault state, open circuit shadow fault state and aging shadow fault state.

[0197] Step 4.1: Build a photovoltaic array simulation platform: Using the Matlab / Simulink simulation platform, a 3×3 photovoltaic array model was established. This model simulates the operating status of the photovoltaic array under different working conditions, including typical faults such as component aging, short circuit, open circuit, and shading. The operating parameters of the photovoltaic array, such as light intensity (G) and temperature (T), are set within a reasonable range. In order to simulate the working status of the photovoltaic array under different environmental conditions, the light intensity is set between 300 and 1000W / m 2 Between, every 20W / m 2 The step length is 15 to 45°C, and the temperature is set between 15 and 45°C, with a step length of 3°C, to generate array output data under different environmental conditions. As shown in Table 2, the present invention generates 8 photovoltaic array states through the simulation platform, including 7 fault states of normal working state, short circuit, short circuit shading, open circuit, open circuit shading, aging, aging shading and shading, and each working state includes typical characteristic data of open circuit voltage, short circuit current, maximum power point voltage, maximum power, fill factor, temperature and light intensity.

[0198]

[0199]

[0200] Table 2

[0201] Step 4.2: Data generation and preprocessing: The simulation generated 3168 sets of data for training and validating the model. The training data contains fault data under different combinations of light intensity and temperature within the above range, covering the working characteristics of the photovoltaic array under various fault conditions. The data is divided into training set and validation set, and the ratio is 8:2 to ensure the generalization ability of the model. The data is preprocessed by standardization technology to eliminate the dimensional differences between different features and improve the convergence speed and prediction accuracy of the model.

[0202] Step 4.3: Model construction and parameter optimization: The present invention uses a convolutional neural network (CNN) for fault diagnosis, and automatically extracts the fault features of the photovoltaic array through the hierarchical structure of CNN. In view of the overfitting problem that may occur in CNN under small sample environments, the present invention introduces a residual network (ResNet) structure to improve the stability and accuracy of the model. The residual network alleviates the gradient vanishing problem of the deep network by adding jump connections, ensuring the effective transmission of fault features in the deep network. In order to further improve the diagnostic performance of the model, the present invention uses a distributed evolutionary algorithm to globally optimize the hyperparameters of CNN (including learning rate, convolution kernel size, number of iterations, etc.). The distributed evolutionary algorithm avoids falling into local optimality and accelerates the training process of the model through parallel search and information exchange of multiple sub-populations.

[0203] Step 4.4: Testing and Validation: To evaluate the diagnostic capability of the model, the test data was used under reasonable light intensity (300 to 1000 W / m 2 ) and temperature (10 to 45°C), generating 100 sets of test data for each type of fault state, for a total of 800 sets of data. During the test, the trained CNN model was used to diagnose each set of test data and evaluate its accuracy in identifying PV array faults. In order to verify the robustness and stability of the model, the light intensity and temperature of the test data were randomly changed in different ranges to simulate the real PV array operating environment.

[0204] Step 4.5: Experimental results: In the experiment of photovoltaic array fault diagnosis, the present invention compares the performance of LSTM, CNN and GA-CNN models. 2) and temperature (10 to 45 ° C) range for evaluation, generating 100 sets of data for each type of fault state, a total of 800 sets of data. The LSTM model has advantages in processing time series data. The fault diagnosis accuracy in this experiment is 90%, and it can better capture the features under different fault states. The CNN model extracts local features through the convolution layer, and its diagnostic accuracy is 87%, with a certain overfitting phenomenon under small sample data. In contrast, the GA-CNN model combines the genetic algorithm to optimize the hyperparameters of CNN, significantly improving the generalization ability of the model, and the fault diagnosis accuracy reaches 98.75%, which is the best performance among all models. The experimental results show that the GA-CNN model can effectively improve the accuracy of fault identification under small sample conditions, showing superior diagnostic performance and robustness. The key point of the present invention is to provide a convolutional neural network hyperparameter optimization method based on a distributed evolutionary algorithm for fault diagnosis of photovoltaic arrays. Through the synergistic effect of the distributed evolutionary algorithm combined with the global search and the local search, the hyperparameters of the residual fusion diagnosis model, such as the convolution kernel size, the learning rate and the number of iterations, can be automatically optimized, thereby improving the performance of the fault diagnosis model. In addition, the present invention also generates a variety of photovoltaic array fault data through simulation for training and building models, thereby further enhancing the diagnostic accuracy and robustness of the model.

[0205] In summary, the present invention has a positive effect on the prior art in three aspects: 1. Method flow: The present invention designs a complete optimization process. First, a simulation platform is used to simulate various fault states of the photovoltaic array to generate simulation data; then the hyperparameter range of the CNN model is initialized, and these hyperparameters are searched globally and locally through a distributed evolutionary algorithm; finally, the optimal hyperparameter combination is selected according to the fitness function to construct an optimized CNN model. 2. Uniqueness of the method: The core innovation of the present invention lies in the intelligent optimization of hyperparameters through a distributed evolutionary algorithm, which avoids the tediousness and inefficiency of traditional manual adjustment and can quickly converge to the global optimal solution. In addition, the elite retention mechanism ensures that the optimal solution will not be lost during the local search process. 3. Optimization results: The implementation of the present invention can significantly improve the accuracy and efficiency of photovoltaic array fault diagnosis. The fault recognition rate after model optimization is greatly improved, the risk of misdiagnosis and missed diagnosis is reduced, and the stability and reliability of the photovoltaic power generation system are guaranteed, which has significant practical value.

[0206] Embodiment 5:

[0207] Based on the same inventive concept, the present invention also proposes a photovoltaic array fault diagnosis system, the structure of which is as follows: Fig.11 As shown, including:

[0208] Diagnosed data acquisition module: used to acquire the PV array data to be diagnosed;

[0209] Data diagnosis module: used for inputting the photovoltaic array data to be diagnosed into the optimal residual fusion diagnosis model for fault diagnosis to obtain the photovoltaic array fault diagnosis result;

[0210] Among them, the optimal residual fusion diagnosis model is constructed based on a residual fusion diagnosis model construction method based on a distributed evolutionary algorithm as described in any of the previous items.

[0211] Preferably, the optimal residual fusion diagnostic model is constructed based on a convolutional neural network combined with a residual network structure.

[0212] Preferably, the convolutional neural network includes: an input layer, a convolutional layer, a pooling layer, a fully connected layer and an output layer, and the residual network structure includes: a first residual block and a second residual block; the optimal residual fusion diagnosis model includes: the input of the input layer is the input of the optimal residual fusion diagnosis model, which is used to input the photovoltaic array data to be diagnosed; the convolutional layer is used to capture features of the photovoltaic array data to be diagnosed, and output the captured feature data to the first residual block; the first residual block is used to transform the captured feature data through an identity mapping to obtain a first The residual transformation feature data is obtained by the step of integrating the residual transformation feature data and outputting it to the pooling layer; the pooling layer is used to perform a pooling operation on the first residual transformation feature data, and input the pooled feature data into the second residual block; the second residual block is used to transform the pooled feature data through an identity mapping to obtain second residual transformation feature data, and output it to the fully connected layer; the fully connected layer is used to integrate the second residual transformation feature data, and output the photovoltaic array fault diagnosis result through the output layer, and the output of the output layer is the output of the optimal residual fusion diagnosis model.

[0213] Preferably, the expression of the constant mapping transformation is as follows:

[0214] y=F(x,{(W i})+x

[0215] Where y is the output of the first residual block or the second residual block; W i is the weight parameter of the convolutional layer; F(x,{W i}) represents the nonlinear transformation function learned by the convolution layer; x is the input of the first residual block or the second residual block. The present invention provides a photovoltaic array fault diagnosis system, which uses the optimal residual diagnosis model in the data diagnosis module to diagnose the data with high accuracy, and provides a photovoltaic fault diagnosis system with strong model stability and high data diagnosis accuracy in a small sample scenario.

[0216] Example 6

[0217] like Fig.12 As shown, the present invention also provides an electronic device, which may be a computer device, a single-chip device, an intelligent mobile device, etc. The electronic device in this embodiment may include a processor, a memory, a transceiver component, etc. The memory, the processor, and the transceiver component are connected via a bus; the memory may be used to store an execution program, and an exemplary execution program may include instructions; the processor is used to execute the instructions stored in the memory. The memory may also be used to store data, which may be called and / or modified when the instructions are executed.

[0218] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in a storage medium to implement the corresponding method flow or corresponding function, so as to realize a residual fusion diagnosis model construction method based on a distributed evolutionary algorithm in the above-mentioned embodiment, or the steps of a photovoltaic array fault diagnosis method.

[0219] Example 7

[0220] Based on the same inventive concept, the present invention also provides a readable storage medium, specifically an electronic device readable storage medium (Memory), which is a memory device in the electronic device for storing programs and data. It can be understood that the storage medium here can include both the built-in storage medium in the electronic device and the extended storage medium supported by the electronic device. The storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and these instructions can be one or more execution programs (including program codes). It should be noted that the storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk storage. The processor loads and executes one or more instructions stored in the storage medium, which can implement a residual fusion diagnosis model construction method based on a distributed evolutionary algorithm in the above-mentioned embodiment, or a step of a photovoltaic array fault diagnosis method.

[0221] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0222] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0223] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0224] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0225] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit its protection scope. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that after reading the present invention, those skilled in the art can still make various changes, modifications or equivalent substitutions to the specific implementation methods of the application, but these changes, modifications or equivalent substitutions are all within the protection scope of the claims to be approved.

Claims

1. A method for constructing a residual fusion diagnosis model based on a distributed evolutionary algorithm, characterized in that: include: Obtain typical characteristic data of photovoltaic arrays under various operating conditions; Initialize the parameters of the distributed evolutionary algorithm, wherein each individual of the initial population corresponds to a set of initial hyperparameters; based on the typical feature data, calculate the fitness of each individual through the residual fusion diagnostic model corresponding to the initial hyperparameters; use the initial population as the parent original population, and select the parent elite population based on the fitness of each individual; perform global search optimization and update on the parent original population to obtain the offspring original population, and perform local search optimization and update on the parent elite population to obtain the offspring elite population; based on the offspring original population and the offspring elite population, use the distributed evolutionary algorithm to perform repeated iterative updates until the maximum number of iterations is reached, and use a set of hyperparameters represented by the individual with the largest fitness value in the current offspring original population and the current offspring elite population as the optimal hyperparameters of the residual fusion diagnostic model; Based on the typical feature data, the residual fusion diagnosis model corresponding to the optimal hyperparameters is trained to obtain the optimal residual fusion diagnosis model.

2. The method according to claim 1, characterized in that The method of performing global search optimization and updating on the parent generation original population to obtain the offspring generation original population, and performing local search optimization and updating on the parent generation elite population to obtain the offspring generation elite population, includes: Using a genetic algorithm, a global search optimization update is performed on the parent generation original population to obtain a child generation original population with better fitness; A variable neighborhood search algorithm is used to perform local search optimization and update on the individuals in the parent elite population to obtain a child elite population with better fitness.

3. The method according to claim 2, characterized in that The variable neighborhood search algorithm is used to perform local search optimization and update on individuals in the parent elite population to obtain a child elite population with better fitness, including: Based on the individuals in the parent elite population, determining the neighborhood structure corresponding to each individual; Searching for individuals in the neighborhood structure, and determining whether the fitness value of the individuals in the neighborhood structure is greater than the fitness value of the individuals in the parent elite population; If yes, the individuals in the neighborhood structure are substituted for the individuals in the parent elite population to obtain a child elite population updated by local search; Otherwise, continue to perform local search. If the fitness value of the individuals in the neighborhood structure is not greater than the fitness value of the individuals in the parent elite population and the maximum number of iterations is reached, the parent elite population is used as the child elite population.

4. The method according to claim 1, characterized in that: The method of repeatedly iterating and updating based on the offspring original population and the offspring elite population using the distributed evolutionary algorithm until the maximum number of iterations is reached, and taking a set of hyperparameters represented by the individuals with the largest fitness values ​​in the current offspring original population and the current offspring elite population as the optimal hyperparameters of the residual fusion diagnostic model, including: The individual information of the random individuals in the offspring elite population is replaced with the individual information of the individual with the smallest fitness value in the offspring original population to obtain an optimized offspring original population; Based on the optimized offspring original population, the dominant individuals in the optimized offspring original population are screened out, and the dominant individuals are used as the new parent original population; Selecting a new parent generation elite population based on the fitness of each individual in the new parent generation original population; Performing global search optimization and updating on the new parent generation original population to obtain a new offspring generation original population, and performing local search optimization and updating on the new parent generation elite population to obtain a new offspring generation elite population; Determine whether the maximum number of iterations has been reached. If so, output a set of hyperparameters represented by the individual with the largest fitness value in the current offspring elite population and the offspring original population as the optimal hyperparameters of the residual fusion diagnostic model; otherwise, based on the current offspring elite population and the offspring original population, continue to iterate until the maximum number of iterations is reached.

5. The method according to any one of claims 1 to 4, characterized in that: The typical characteristic data of the photovoltaic array under various operating conditions are obtained, including: Based on the circuit structure of the photovoltaic array, typical characteristic data under various operating conditions are obtained through the simulation platform; The multiple operating states include one or more of the following: normal working state, short circuit fault state, open circuit fault state, aging fault state, partial shading state, aging shading fault state, short circuit shading fault state and open circuit shading fault state; the typical characteristic data include one or more of the following: open circuit voltage, short circuit current, maximum power point voltage, maximum power, fill factor, temperature and light intensity.

6. A residual fusion diagnosis model construction system based on distributed evolutionary algorithm, characterized in that: include: Data acquisition module: used to obtain typical characteristic data of photovoltaic arrays under various operating conditions; Optimal hyperparameter solution module: used to initialize the parameters of the distributed evolutionary algorithm, in which each individual of the initial population corresponds to a set of initial hyperparameters; based on the typical feature data, the fitness of each individual is calculated through the residual fusion diagnostic model corresponding to the initial hyperparameters; the initial population is used as the parent original population, and the parent elite population is selected based on the fitness of each individual; the parent original population is globally searched and optimized to obtain the offspring original population, and the parent elite population is locally searched and optimized to obtain the offspring elite population; based on the offspring original population and the offspring elite population, the distributed evolutionary algorithm is used to perform repeated iterative updates until the maximum number of iterations is reached, and a set of hyperparameters represented by the individual with the largest fitness value in the current offspring original population and the current offspring elite population is used as the optimal hyperparameters of the residual fusion diagnostic model; Model building module: used to train the residual fusion diagnosis model corresponding to the optimal hyperparameters based on the typical feature data to obtain the optimal residual fusion diagnosis model.

7. The system according to claim 6, characterized in that The optimal hyperparameter solution module includes: A global search optimization updating unit: used to use a genetic algorithm to perform global search optimization updating on the parent generation original population to obtain a child generation original population with better fitness; Local search optimization update unit: used for adopting variable neighborhood search algorithm to perform local search optimization update on individuals in the parent elite population to obtain a child elite population with better fitness.

8. The system according to claim 7, characterized in that The local search optimization updating unit comprises: A neighborhood structure determination subunit: used for determining the neighborhood structure corresponding to each individual based on the individuals in the parent elite population; Individual judgment subunit: used to search for individuals in the neighborhood structure, and judge whether the fitness value of the individuals in the neighborhood structure is greater than the fitness value of the individuals in the parent elite population; Individual replacement subunit: when the fitness value of the individual in the neighborhood structure is greater than the fitness value of the individual in the parent elite population, the individual in the neighborhood structure is replaced by the individual in the parent elite population to obtain a local search updated child elite population; The offspring population determination subunit is used to use the parent elite population as the offspring elite population when the fitness value of the individuals in the neighborhood structure is not greater than the fitness value of the individuals in the parent elite population and the maximum number of iterations is reached.

9. The system according to claim 6, characterized in that The optimal hyperparameter solution module also includes: Knowledge transfer unit: used to replace the individual information of the individual with the smallest fitness value in the original population of the offspring with the individual information of the random individuals in the elite population of the offspring, so as to obtain the optimized original population of the offspring; A dominant individual retaining unit is used to screen out dominant individuals from the optimized offspring original population based on the optimized offspring original population, and use the dominant individuals as new parent original population; A parent generation elite population selection unit: used for selecting a new parent generation elite population based on the fitness of each individual in the new parent generation original population; Distributed optimization and updating unit: used for performing global search optimization and updating on the new parent generation original population to obtain a new child generation original population, and performing local search optimization and updating on the new parent generation elite population to obtain a new child generation elite population; Iteration judgment unit: used to judge whether the maximum number of iterations has been reached; if so, a set of hyperparameters represented by the individual with the largest fitness value in the current offspring elite population and the offspring original population is output as the optimal hyperparameters of the residual fusion diagnosis model; otherwise, based on the current offspring elite population and the offspring original population, iterate until the maximum number of iterations is reached.

10. The system according to any one of claims 6 to 9, characterized in that: The data acquisition module is specifically used to: based on the circuit structure of the photovoltaic array, obtain typical characteristic data under various operating conditions through a simulation platform; the various operating conditions include one or more of the following: normal working state, short circuit fault state, open circuit fault state, aging fault state, partial shading state, aging shading fault state, short circuit shading fault state and open circuit shading fault state; the typical characteristic data include one or more of the following: open circuit voltage, short circuit current, maximum power point voltage, maximum power, fill factor, temperature and light intensity.

11. A photovoltaic array fault diagnosis method, characterized in that: include: Obtaining the photovoltaic array data to be diagnosed; Inputting the photovoltaic array data to be diagnosed into the optimal residual fusion diagnosis model for fault diagnosis to obtain a photovoltaic array fault diagnosis result; The optimal residual fusion diagnosis model is constructed based on a residual fusion diagnosis model construction method based on a distributed evolutionary algorithm according to any one of claims 1 to 5.

12. The method according to claim 11, characterized in that The optimal residual fusion diagnosis model is constructed based on a convolutional neural network combined with a residual network structure.

13. The method according to claim 12, characterized in that The convolutional neural network includes: an input layer, a convolutional layer, a pooling layer, a fully connected layer and an output layer; the residual network structure includes: a first residual block and a second residual block; the optimal residual fusion diagnosis model includes: The input of the input layer is the input of the optimal residual fusion diagnosis model, which is used to input the photovoltaic array data to be diagnosed; The convolution layer is used to capture features of the photovoltaic array data to be diagnosed, and output the captured feature data to the first residual block; The first residual block is used to transform the captured feature data through an identity mapping to obtain first residual transformed feature data, and output it to the pooling layer; The pooling layer is used to perform a pooling operation on the first residual transformation feature data, and input the pooled feature data into the second residual block; The second residual block is used to transform the pooled feature data through an identity mapping to obtain second residual transformed feature data, and output it to the fully connected layer; The fully connected layer is used to integrate the second residual transformation feature data and output the photovoltaic array fault diagnosis result through the output layer. The output of the output layer is the output of the optimal residual fusion diagnosis model.

14. The method according to claim 13, characterized in that The expression of the constant mapping transformation is as follows: y=F(x,{W i })+x Where y is the output of the first residual block or the second residual block; W i is the weight parameter of the convolutional layer; F(x,{W i }) represents the nonlinear transformation function learned by the convolutional layer; x is the input of the first residual block or the second residual block.

15. A photovoltaic array fault diagnosis system, characterized in that: include: Diagnosed data acquisition module: used to acquire the PV array data to be diagnosed; Data diagnosis module: used for inputting the photovoltaic array data to be diagnosed into the optimal residual fusion diagnosis model for fault diagnosis to obtain the photovoltaic array fault diagnosis result; The optimal residual fusion diagnosis model is constructed based on a residual fusion diagnosis model construction method based on a distributed evolutionary algorithm according to any one of claims 1 to 5.

16. The system according to claim 15, characterized in that The optimal residual fusion diagnostic model is constructed based on a convolutional neural network combined with a residual network structure.

17. The system according to claim 16, characterized in that The convolutional neural network includes: an input layer, a convolutional layer, a pooling layer, a fully connected layer and an output layer; the residual network structure includes: a first residual block and a second residual block; the optimal residual fusion diagnosis model includes: The input of the input layer is the input of the optimal residual fusion diagnosis model, which is used to input the photovoltaic array data to be diagnosed; The convolution layer is used to capture features of the photovoltaic array data to be diagnosed, and output the captured feature data to the first residual block; The first residual block is used to transform the captured feature data through an identity mapping to obtain first residual transformed feature data, and output it to the pooling layer; The pooling layer is used to perform a pooling operation on the first residual transformation feature data, and input the pooled feature data into the second residual block; The second residual block is used to transform the pooled feature data through an identity mapping to obtain second residual transformed feature data, and output it to the fully connected layer; The fully connected layer is used to integrate the second residual transformation feature data and output the photovoltaic array fault diagnosis result through the output layer. The output of the output layer is the output of the optimal residual fusion diagnosis model.

18. The system according to claim 17, characterized in that The expression of the constant mapping transformation is as follows: y=F(x,{W i })+x Where y is the output of the first residual block or the second residual block; W i is the weight parameter of the convolutional layer; F(x,{W i }) represents the nonlinear transformation function learned by the convolutional layer; x is the input of the first residual block or the second residual block.

19. An electronic device, characterized in that: include: at least one processor and memory; The memory and the processor are connected via a bus; The memory is used to store one or more programs; When the one or more programs are executed by the at least one processor, a residual fusion diagnosis model construction method based on a distributed evolutionary algorithm as described in any one of claims 1 to 5, or a photovoltaic array fault diagnosis method as described in any one of claims 11 to 14 is implemented.

20. A readable storage medium, characterized in that: An execution program is stored thereon, and when the execution program is executed, a residual fusion diagnosis model construction method based on a distributed evolutionary algorithm as described in any one of claims 1 to 5, or a photovoltaic array fault diagnosis method as described in any one of claims 11 to 14 is implemented.

Images