Photovoltaic array fault diagnosis method based on Transform-BiLSTM model

Abstract

A photovoltaic array fault diagnosis method based on a Transform-BiLSTM model relates to a photovoltaic array fault diagnosis method, and comprises the following steps: S1, inputting photovoltaic array sample data under normal and different fault working conditions, carrying out standardized conversion on the sample data, and dividing a training set and a test set through VMD modal component decomposition; s2, initializing parameters of a Transform-BiLSTM model and parameters of an IIVYA algorithm, and setting the number of hedera helix populations, the upper and lower limits of independent variables, the maximum number of iterations and the dimension size; s3, initializing a hedera helix population; s4, calculating the fitness value of the new position of the hedera helix according to the position updating formula of each hedera helix population; s5, judging whether the number of iterations reaches the maximum or not, comparing the optimal fitness values stored in all iterations, and outputting optimal parameters; and S6, constructing an optimal Transform-BiLSTM model by using the optimal parameter value obtained by the IIVYA algorithm, and carrying out diagnosis test on the photovoltaic array fault.

Description

Technical Field

[0001] The invention relates to a photovoltaic array fault diagnosis method, and in particular to a photovoltaic array fault diagnosis method based on a Transformer-BiLSTM model. Background Art

[0002] Photovoltaic arrays are the main equipment for photovoltaic power generation. In outdoor places with complex environments, various faults are prone to occur, which will reduce power generation efficiency and affect the safe operation of the system. Therefore, it is very important to accurately and efficiently diagnose the faults of photovoltaic arrays.

[0003] According to the differences in fault type characteristics and detection methods, the diagnostic technical methods are classified into two categories: visual and imaging diagnostic methods and electrical characteristic parameter diagnostic methods. Visual and imaging diagnostic methods include infrared thermal imaging, visible light imaging, electroluminescence, photoluminescence and computer vision. For faults such as cracks, photoluminescence detection technology is used to timely detect defects in production, thereby improving the average efficiency of photovoltaic cells. Methods for photovoltaic module classification and defect detection using infrared imaging technology. In addition, in order to realize the fault diagnosis of photovoltaic modules, deep learning algorithms are introduced, and convolutional neural networks (CNN) and their improved versions can also be applied to the diagnosis of photovoltaic faults. Electrical characteristic parameter diagnostic methods include IV curve method, physical detection method, power loss analysis method, voltage-current measurement method and machine learning classification method. By processing voltage and current information and establishing a semi-supervised ladder network, open circuit faults, line-to-line faults and shadow faults can be diagnosed. Photovoltaic array fault diagnosis methods based on supervised learning include artificial neural networks, probabilistic neural networks, etc. Judging the fault of photovoltaic arrays through electrical characteristic parameter diagnosis is timely, efficient and accurate.

[0004] In view of the fact that the BP neural network algorithm is prone to fall into the local optimum, the improved BP neural network algorithm is used to optimize the initial connection weights and thresholds in the neural network model to avoid falling into the local optimum, thereby ensuring the comprehensiveness of photovoltaic cell fault diagnosis and improving the accuracy of fault diagnosis; based on the long short-term memory neural network algorithm, the characteristic values ​​of the photovoltaic array under different fault modes are used as training data sets, and its fault diagnosis accuracy is also improved compared with the traditional neural network. Compared with LSTM, BiLSTM can obtain more comprehensive sequence information, and can often achieve better results in many tasks that need to consider contextual bidirectional information. Transformer can effectively capture long-distance dependencies in the input sequence, which is very important for fault diagnosis. In complex systems, the manifestation of a fault may be far away from the cause of the fault in time or space. It can process information of different time scales at the same time, taking both short-term and long-term changes into consideration.

[0005] With the rapid development of intelligent algorithms, algorithm improvements have also been widely used in the field of photovoltaic faults. Based on an improved northern goshawk algorithm (INGO), the INGO-HKELM fault diagnosis model is built by combining the Gaussian detection mechanism and the combined kernel extreme learning machine (HKELM), and applied to photovoltaic array fault diagnosis. A photovoltaic array fault diagnosis method based on variational mode decomposition VMD (variational mode decomposition) and improved dung beetle algorithm IDBO (improved dung beetle optimizer) to optimize the long short-term memory LSTM (long short-term memory) network. A photovoltaic array multi-type composite fault diagnosis method based on the Harris Hawk (HHO) algorithm to optimize the extreme learning machine (ELM).

[0006] However, most of the above studies are relatively sufficient for the diagnosis of single faults in photovoltaic arrays, but the diagnosis of compound faults is relatively limited. Photovoltaic systems may have multiple faults at the same time, making it difficult for researchers to establish accurate mechanism models and extract reasonable fault characteristics; at the same time, the above methods do not consider in detail the impact of parameter selection on the performance of their algorithms, and parameter selection is crucial to the impact of algorithm diagnosis effects.

[0007] Therefore, in the prior art, the traditional photovoltaic array fault diagnosis method cannot accurately diagnose multiple fault conditions at the same time. Summary of the invention

[0008] The purpose of the present invention is to provide a photovoltaic array fault diagnosis method based on the Transformer-BiLSTM model, the method adopts IIVYA refers to an improved ivy optimization algorithm, and Transformer-BiLSTM refers to a combined model of a Transformer encoding part and a bidirectional long short-term memory network. The present invention can accurately diagnose different types of photovoltaic array fault conditions based on the photovoltaic array fault diagnosis method based on the IIVYA-Transformer-BiLSTM combined model.

[0009] The present invention adopts the following technical solutions:

[0010] A photovoltaic array fault diagnosis method based on a Transformer-BiLSTM model, the method comprising the following steps:

[0011] S1, inputting the sample data of the photovoltaic array under normal working conditions and different fault conditions, and performing standardization transformation on the sample data of the photovoltaic array and then performing VMD modal component decomposition to divide the data into a training set and a test set;

[0012] S2, initialize the parameters of the Transformer-BiLSTM model and the parameters of the IIVYA algorithm, and set the number of ivy populations, upper and lower limits of independent variables, maximum number of iterations, and dimension size;

[0013] S3, initialize the ivy population;

[0014] S4, calculating the fitness value of the new position of the ivy according to the position update formula of each ivy population;

[0015] S5, determine whether the number of iterations has reached the maximum, if so, end the iteration, compare the optimal fitness values ​​stored in all iterations and output the optimal parameters; if not, return to step S3;

[0016] S6, construct the optimal Transformer-BiLSTM model with the optimal parameter values ​​found by the IIVYA algorithm, and perform diagnostic tests on photovoltaic array faults.

[0017] In the photovoltaic array fault diagnosis method based on the Transformer-BiLSTM model, in step S1, the fault conditions include short circuit fault, open circuit fault, partial shading and aging fault.

[0018] In the photovoltaic array fault diagnosis method based on the Transformer-BiLSTM model, in step S1, the standardized conversion adopts the following formula:

[0019]

[0020] Among them, G and T are the irradiance and temperature under normal conditions, G STC and T STC is the irradiance and temperature at STC, β i is the short-circuit current temperature coefficient, β v is the open circuit voltage temperature coefficient, K g is the irradiance correction factor for voltage, I STC and V STC Iout and Vout are the current and voltage under normal conditions; STC refers to standard test conditions.

[0021] In the photovoltaic array fault diagnosis method based on the Transformer-BiLSTM model, in step S1, the standardized transformation is then decomposed by VMD modal components using the following formula:

[0022]

[0023] The photovoltaic array fault diagnosis method based on the Transformer-BiLSTM model, step S2 includes the following steps:

[0024] S21, using chaotic numbers to replace the random variables in the original IIVYA algorithm, specifically adding Sine-Tent-Cosine chaotic mapping, the formula is as follows:

[0025]

[0026] In the formula, x(i+1) represents the value of the i+1th generation chaotic sequence, x(i) represents the value of the ith generation chaotic sequence, i is the current iteration number, r∈[0,1], and r=0.7;

[0027] S22, in Ivy League Member I i Utilize I ii A variable spiral position update strategy is introduced during the climbing and moving phase along the direction of the light source to ensure that there are diverse search paths to effectively update the position. The position update formula is as follows:

[0028]

[0029] In the formula, t represents the current iteration number, T represents the total iteration number, l represents a random number between [0,1], k is 5, The updated location information of the i-th ivy, I i The current location information of the i-th ivy, is the dot product;

[0030] S23, introduce the Levi flight strategy to update the position, in member Ii Globally roam through the search space to the nearest neighbor I ii After the stage, member I i Try to directly follow the best member of the entire population member I Best , which is equivalent to member I Best Find the optimal solution around to avoid falling into the local optimal solution;

[0031] Its position update formula is as follows:

[0032] In the formula, μ and v are variables that obey the normal distribution, that is, Γ is the standard gamma function, s is a Levy distribution with parameter β, 0<β<2, α is a scaling factor of 0.01, and r is a random number in [0,1]. is the dot product;

[0033]

[0034] In the photovoltaic array fault diagnosis method based on the Transformer-BiLSTM model, in step S3, the ivy population is initialized according to the Sine-Tent-Cosine chaotic mapping formula in step S21.

[0035] The photovoltaic array fault diagnosis method based on the Transformer-BiLSTM model, step S4 includes:

[0036] Determine whether the fitness value of the current ivy population is better than the stored best fitness value. If so, replace the stored best fitness value with the current fitness value and save the corresponding value; otherwise, continue to store the previous best fitness value;

[0037] The formula for calculating the fitness value is:

[0038] fitness=1-acc;

[0039] Among them, fitness is the fitness value. The smaller the fitness, the more accurate the model diagnosis. Acc is the accuracy of the hyperparameter model after optimization.

[0040] In the photovoltaic array fault diagnosis method based on the Transformer-BiLSTM model, in step S6, the optimized parameter values ​​include the number of hidden layer neurons, the learning rate, and the regularization coefficient of the Transformer-BiLSTM model.

[0041] The photovoltaic array fault diagnosis method based on the Transformer-BiLSTM model, the diagnostic system of the method includes:

[0042] A data processing module is used to input the sample data of the photovoltaic array under normal working conditions and different fault conditions, and perform standardization conversion on the sample data of the photovoltaic array, and then divide the data into a training set and a test set through VMD modal component decomposition;

[0043] The parameter setting module is used to initialize the parameters of the Transformer-BiLSTM model and the parameters of the IIVYA algorithm, and to set the number of ivy populations, upper and lower limits of independent variables, maximum number of iterations, and dimension size;

[0044] Initialization module, used to initialize the ivy population;

[0045] A fitness value calculation module is used to calculate the fitness value of the new position of the ivy according to the position update formula of each ivy population;

[0046] The iteration number judgment module is used to judge whether the iteration number has reached the maximum. If so, the iteration is terminated, the optimal fitness values ​​stored in all iterations are compared and the optimal parameters are output; if not, the module returns to the initialization module;

[0047] The model building and testing module is used to build the optimal Transformer-BiLSTM model with the optimal parameter values ​​found by the IIVYA algorithm, and to perform diagnostic tests on photovoltaic array faults.

[0048] The beneficial effects of the present invention are:

[0049] (1) The present invention improves the IVYA algorithm in three aspects: first, the position of the ivy population is initialized using the Sine-Tent-Cosine chaotic map; second, the position of the ivy member I is initialized using the Sine-Tent-Cosine chaotic map; i Utilize I ii In the climbing and moving phase along the direction of the light source, a variable spiral position update strategy is introduced to ensure that there are diverse search paths to effectively update the position; thirdly, the Levy flight strategy is introduced for position update. i Globally roam through the search space to the nearest neighbor I ii After the stage, member I i Try to directly follow the best member of the entire population member I Best , which is equivalent to member I Best Find the optimal solution around to avoid falling into the local optimal solution;

[0050] (2) The present invention proposes a photovoltaic array modeling based on multiple working conditions, and uses the fault features extracted from the IV characteristic curve as the model input and the labels representing different working conditions as the model output, and constructs a photovoltaic fault diagnosis method based on IIVYA to optimize the Transformer-BiLSTM parameters;

[0051] (3) The present invention introduces the improved ivy optimization algorithm to optimize the parameters of Transformer-BiLSTM, and then establishes a fault diagnosis model based on IIVYA-Transformer-BiLSTM, which can more accurately identify normal, aging, shadow, short circuit and open circuit faults as well as composite faults occurring simultaneously under the above conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 A flow chart of a photovoltaic array fault diagnosis method based on the IIVYA-Transformer-BiLSTM combined model in the present invention;

[0053] Figure 2 It is a schematic diagram of an IU curve under different irradiances and temperatures under normal working conditions of a photovoltaic array;

[0054] Figure 3 Schematic diagram of IU curve and PU curve under different fault conditions;

[0055] Figure 4 Schematic diagram of IU curve and PU curve under different local shadow faults;

[0056] Figure 5 Schematic diagram of IU curve and PU curve under different short circuit faults;

[0057] Figure 6 Schematic diagram of LSTM and BiLSTM network structure;

[0058] Figure 7 This is the principle structure diagram of Transformer self-attention mechanism;

[0059] Figure 8 A schematic diagram of a photovoltaic array fault simulation model in the present invention;

[0060] Fig. 9 Schematic diagram of optimization curves of the four models of IIVYA algorithm, IVYA algorithm, SSA algorithm and WOA algorithm for test functions f1 and f2 respectively in the present invention;

[0061] Fig.10 It is a schematic diagram of the function graph of the test functions f1 and f2;

[0062] Fig.11Diagnosis results of IIVYA-Transformr-BiLSTM. DETAILED DESCRIPTION

[0063] In order to more clearly illustrate the embodiments of the present invention, the specific implementation methods of the present invention will be described below with reference to the accompanying drawings. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, other accompanying drawings and other implementation methods can be obtained based on these accompanying drawings without creative work.

[0064] Example:

[0065] See Figure 1 The present invention provides a photovoltaic array fault diagnosis method based on the IIVYA-Transformer-BiLSTM combined model, comprising the following steps:

[0066] S1, inputting the sample data of the photovoltaic array under normal working conditions and different fault conditions, and performing standardization transformation on the sample data of the photovoltaic array and then performing VMD modal component decomposition to divide the data into a training set and a test set;

[0067] S2, initialize the parameters of the Transformer-BiLSTM model and the parameters of the IIVYA algorithm, and set the number of ivy populations, upper and lower limits of independent variables, maximum number of iterations, and dimension size;

[0068] S3, initialize the ivy population;

[0069] S4, calculating the fitness value of the new position of the ivy according to the position update formula of each ivy population;

[0070] S5, determine whether the number of iterations has reached the maximum, if so, end the iteration, compare the optimal fitness values ​​stored in all iterations and output the optimal parameters; if not, return to step S3;

[0071] S6, construct the optimal Transformer-BiLSTM model with the optimal parameter values ​​found by the IIVYA algorithm, and perform diagnostic tests on photovoltaic array faults.

[0072] The present invention constructs a 5×4 photovoltaic array simulation model considering the configuration of bypass diodes and blocking diodes in the Matlab / Simulink environment. Figure 8 , and simulate and analyze different external environments and fault conditions to extract variables that can reflect fault characteristics. The main parameters of photovoltaic modules are as follows: short-circuit current I sc is 7.84A, open circuit voltage U oc 36.3V, maximum power P mis 213.15V, and the maximum power point current is I m 7.35A, maximum power point voltage U m The open circuit voltage temperature coefficient is 29V, the open circuit voltage temperature coefficient is -0.361, and the short circuit current temperature coefficient is 0.102. In the actual environment, multiple faults often occur simultaneously in the photovoltaic array. Therefore, according to the actual possible fault types, 15 operating conditions are selected as the simulation research objects, and the output of the fault diagnosis model is defined as shown in Table 1.

[0073] Table 1 lists the fault types

[0074]

[0075]

[0076] The selection of photovoltaic array fault characteristics has a significant impact on whether the model can accurately and efficiently identify faults. Figure 2 The external environmental factors temperature T and irradiance S have a great influence on the IU curve of the photovoltaic array. Figure 2 As can be seen from a, as T increases, the open circuit voltage U oc Reduce the short-circuit current I sc Slightly increased; see Figure 2 b shows that when the irradiance S increases, I sc and U oc See Figure 3 It can be seen that when a short circuit occurs in the photovoltaic array, the maximum power point voltage U m , P m Reduce; when an open circuit fault occurs in the photovoltaic array, the maximum power point current I m ,I sc Reduce; when the photovoltaic array is abnormally aged, the maximum power P m See Figure 4 It can be seen that when the photovoltaic array is in a partial shadow fault, the output curve will show the phenomenon of "multiple peaks" and "multiple knees". m ,I m , U m When the PV module is in full shadow failure, the U oc and I sc The output curve is similar to the normal working curve, but the P at the maximum power point is m ,I m , U m There have also been significant changes. Figure 5 It can be seen that when the photovoltaic module is in a short circuit fault, the output characteristic curve U oc ,I scremains basically unchanged, but it can be seen that P at the maximum power point m ,I m , U m Due to the existence of blocking diodes and bypass diodes, when a short circuit occurs in the photovoltaic array, the fault branch current remains positive, and its volt-ampere characteristic curve presents a "multi-peak" state, that is, there are multiple local maximum power points, which is similar to a local shadow fault.

[0077] The present invention selects temperature T, irradiance S, open circuit voltage U oc , short circuit current I sc , maximum power P m , Maximum power point voltage U m , Maximum power point current I m , filling factor FF and 12 variables of F1, F2, F3 and F4 are used as input variables of the fault diagnosis model.

[0078]

[0079] The IVYA algorithm simulates the coordinated and orderly growth and spread evolution of an ivy plant population based on the growth pattern of ivy plants. The growth rate of ivy plants is modeled using differential equations and a data-intensive experimental process. The knowledge of nearby ivy plants is fully utilized to determine the direction of growth. The different life stages of ivy, including growth, rise and spread, are shown in the ivy plant population. The IVYA algorithm is a population-based optimization method that operates through an iterative process, utilizing the collective search capabilities of population members.

[0080] In order to change the situation that the traditional IVYA algorithm is not good at finding the optimal solution, chaotic numbers are used to replace the random variables in the original IIVYA algorithm. Specifically, the Sine-Tent-Cosine chaotic mapping is added. The formula is as follows:

[0081]

[0082] In the formula, x(i+1) represents the value of the chaotic sequence of the i+1th generation, x(i) represents the value of the chaotic sequence of the ith generation, i is the current iteration number, r∈[0,1], and r=0.7;

[0083] Member of Ivy League i Utilize I ii A variable spiral position update strategy is introduced during the climbing and moving phase along the direction of the light source to ensure that there are diverse search paths to effectively update the position. The position update formula is as follows:

[0084]

[0085] In the formula, t represents the current iteration number, T represents the total iteration number, l represents a random number between [0,1], k is 5, The updated location information of the i-th ivy, I i The current location information of the i-th ivy, is the dot product;

[0086] S23, introduce the Levi flight strategy to update the position, in member I i Globally roam through the search space to the nearest neighbor I ii After the stage, member I i Try to directly follow the best member of the entire population member I Best , which is equivalent to member I Best Find the optimal solution around and avoid falling into the local optimal solution.

[0087] Its position update formula is as follows:

[0088] In the formula, μ and v are variables that obey the normal distribution, that is, Γ is the standard gamma function, s is a Levy distribution with parameter β, 0<β<2, α is a scaling factor of 0.01, and r is a random number in [0,1]. is the dot product;

[0089]

[0090] The bidirectional LSTM network (BiLSTM) model can learn bidirectional time features. On the basis of ordinary LSTM, by further combining the reverse information transmitted by the sequence, it can achieve better prediction performance and improve the accuracy of model prediction. Therefore, it is proposed as an optimization of the traditional LSTM network model. Due to the addition of reverse LSTM, BiLSTM can fully learn the sequence data of the past and future moments, making up for the defect that the traditional unidirectional LSTM cannot capture information from the back to the front, so it is very suitable for modeling time series data.

[0091] Transformer can effectively capture long-distance dependencies in the input sequence, which is very important for fault diagnosis. In complex systems, the manifestation of a fault may be far away from the cause of the fault in time or space. It can process information at different time scales simultaneously, taking into account both short-term and long-term changes. In terms of the decoder, Transformer can use self-attention to establish effective relationships between features, which is conducive to the use of multi-scale information. The multi-head mechanism is used in the encoding module to speed up the calculation and save time by processing multiple heads in parallel at the same time. The model introduces a multi-head attention mechanism, which can capture local information on multiple heads in parallel, see Figure 7 .

[0092] Q=XW q

[0093] K=XW k

[0094] V=XW v

[0095] Where: Q (Query) is the query matrix; K (Key) is the key matrix; V (Value) is the value matrix; W q , W k , W v are the weight matrices corresponding to Q, K, and V respectively.

[0096] The calculation formula of the self-attention mechanism is as follows:

[0097]

[0098] Where: d k The dimensions of Q and K are obtained by performing dot product operation on matrix Q and matrix K to obtain the weight matrix; divided by d k The dot product is scaled to achieve accelerated convergence of the model. The Softmax(·) function globally activates the result. The similarity weight matrix obtained after activation is multiplied by the matrix V.

[0099] This design uses the Matlab / Simulink simulation platform to build a 4×5 series-parallel photovoltaic array. For the 15 different fault conditions in Table 1, the corresponding operating conditions are simulated. 2 and T∈[15,45]℃, respectively, with a step size of 20W / m 2 and 3℃, a total of 6504 sets of fault data were collected, 5205 sets of data were randomly selected as training samples, and the remaining 1299 sets of data were used as test samples.

[0100] The improved algorithm IIVAY is compared with the IVAY algorithm, the whale optimization algorithm (WOA) and the sparrow search algorithm (SSA) on two benchmark test functions. The number of iterations, population size and dimension of each algorithm are 200, 30 and 30 respectively. In order to reduce the contingency of the experiment, each algorithm conducts 30 independent experiments on the two test functions to obtain the optimal value, average value and standard deviation of the function, and draws the average convergence curve. The average convergence curve of each test function is shown in Fig. 9 .

[0101] The SSA and WOA algorithms performed poorly in terms of fitness and showed problems of premature convergence and falling into local extreme values. Although the IVYA algorithm can jump out of the local extreme value, the IIVYA algorithm is superior to the IVYA algorithm in terms of convergence speed and optimization effect. This proves that the strategy adopted by the IIVYA algorithm not only speeds up the convergence speed of the algorithm, but also improves the optimization efficiency.

[0102] This design proposes a photovoltaic array fault diagnosis method based on 12-dimensional feature quantity and IIVYA-Transformer-BiLSTM, which effectively identifies the photovoltaic array in open circuit, short circuit, partial shadow, aging and multiple fault states. Fig.11 , the accuracy rate is as high as 97.15%. The main conclusions of the present invention are as follows:

[0103] (1) Select T, S, or U oc ,I sc , P m , U m ,I m , FF, F1, F2, F3 and F4 are used as the input variables of the fault diagnosis model. Through experimental verification, these fault characteristic quantities can effectively identify fault characteristics.

[0104] (2) Compared with traditional SSA, WOA and IVYA algorithms, the Sine-Tent-Cosine chaotic mapping, spiral search strategy and levy flight improved IVYA algorithm proposed in the present invention show better optimization ability, stability and faster convergence speed.

[0105] (3) The IIVYA-Transformer-BiLSTM fault diagnosis model was proposed, which performed better in terms of diagnostic accuracy for various types of faults and could quickly and accurately distinguish between various single faults and multiple faults.

[0106] The above description is only a detailed description of the preferred embodiments and principles of the present invention. For ordinary technicians in this field, according to the ideas provided by the present invention, there will be changes in the specific implementation methods, and these changes should also be regarded as the protection scope of the present invention.

Claims

1. A photovoltaic array fault diagnosis method based on Transformer-BiLSTM model, characterized in that: The method comprises the following steps: S1, inputting the sample data of the photovoltaic array under normal working conditions and different fault conditions, and performing standardization transformation on the sample data of the photovoltaic array and then performing VMD modal component decomposition to divide the data into a training set and a test set; S2, initialize the parameters of the Transformer-BiLSTM model and the parameters of the IIVYA algorithm, and set the number of ivy populations, upper and lower limits of independent variables, maximum number of iterations, and dimension size; S3, initialize the ivy population; S4, calculating the fitness value of the new position of the ivy according to the position update formula of each ivy population; S5, determine whether the number of iterations has reached the maximum, if so, end the iteration, compare the optimal fitness values ​​stored in all iterations and output the optimal parameters; if not, return to step S3; S6, construct the optimal Transformer-BiLSTM model with the optimal parameter values ​​found by the IIVYA algorithm, and perform diagnostic tests on photovoltaic array faults.

2. A photovoltaic array fault diagnosis method based on a Transformer-BiLSTM model according to claim 1, characterized in that: In step S1, the fault conditions include short circuit fault, open circuit fault, partial shading and aging fault.

3. A photovoltaic array fault diagnosis method based on Transformer-BiLSTM model according to claim 1, characterized in that: In step S1, the standardization conversion adopts the following formula: Among them, G and T are the irradiance and temperature under normal conditions, G STC and T STC is the irradiance and temperature at STC, β i is the short-circuit current temperature coefficient, β v is the open circuit voltage temperature coefficient, K g is the irradiance correction factor for voltage, I STC and V STC Iout and Vout are the current and voltage under normal conditions; STC refers to standard test conditions.

4. A photovoltaic array fault diagnosis method based on Transformer-BiLSTM model according to claim 1, characterized in that: In step S1, the standardized transformation is then decomposed into VMD modal components using the following formula:

5. The photovoltaic array fault diagnosis method based on the Transformer-BiLSTM model according to claim 1, characterized in that: The step S2 includes the following steps: S21, using chaotic numbers to replace the random variables in the original IIVYA algorithm, specifically adding Sine-Tent-Cosine chaotic mapping, the formula is as follows: In the formula, x(i+1) represents the value of the i+1th generation chaotic sequence, x(i) represents the value of the ith generation chaotic sequence, i is the current iteration number, r∈[0,1], and r=0.7; S22, in Ivy League Member I i Utilize I ii A variable spiral position update strategy is introduced during the climbing and moving phase along the direction of the light source to ensure that there are diverse search paths to effectively update the position; the position update formula is as follows: In the formula, t represents the current iteration number, T represents the total iteration number, l represents a random number between [0,1], k is 5, The updated location information of the i-th ivy, I i The current location information of the i-th ivy, is the dot product; S23, introduce the Levi flight strategy to update the position, in member I i Globally roam through the search space to the nearest neighbor I ii After the stage, member I i Try to directly follow the best member of the entire population member I Best , which is equivalent to member I Best Find the optimal solution around to avoid falling into the local optimal solution; Its position update formula is as follows: In the formula, μ and v are variables that obey the normal distribution, that is, Γ is the standard gamma function, s is a Levy distribution with parameter β, 0<β<2, α is a scaling factor of 0.01, and r is a random number in [0,1]. is the dot product; 6. A photovoltaic array fault diagnosis method based on Transformer-BiLSTM model according to claim 1, characterized in that: In the step S3, the ivy population is initialized according to the Sine-Tent-Cosine chaotic mapping formula in step S21.

7. The photovoltaic array fault diagnosis method based on the Transformer-BiLSTM model according to claim 1, characterized in that: The step S4 includes: Determine whether the fitness value of the current ivy population is better than the stored best fitness value. If so, replace the stored best fitness value with the current fitness value and save the corresponding value; otherwise, continue to store the previous best fitness value; The formula for calculating the fitness value is: fitness=1-acc; Among them, fitness is the fitness value. The smaller the fitness, the more accurate the model diagnosis. Acc is the accuracy of the hyperparameter model after optimization.

8. The photovoltaic array fault diagnosis method based on the Transformer-BiLSTM model according to claim 1, characterized in that: In step S6, the optimized parameter values ​​include the number of hidden layer neurons, the learning rate, and the regularization coefficient of the Transformer-BiLSTM model.

9. The photovoltaic array fault diagnosis method based on the Transformer-BiLSTM model according to claim 1, characterized in that: The diagnostic system of the method comprises: A data processing module is used to input the sample data of the photovoltaic array under normal working conditions and different fault conditions, and perform standardization conversion on the sample data of the photovoltaic array, and then divide the data into a training set and a test set through VMD modal component decomposition; The parameter setting module is used to initialize the parameters of the Transformer-BiLSTM model and the parameters of the IIVYA algorithm, and to set the number of ivy populations, upper and lower limits of independent variables, maximum number of iterations, and dimension size; Initialization module, used to initialize the ivy population; A fitness value calculation module is used to calculate the fitness value of the new position of the ivy according to the position update formula of each ivy population; The iteration number judgment module is used to judge whether the iteration number has reached the maximum. If so, the iteration is terminated, the optimal fitness values ​​stored in all iterations are compared and the optimal parameters are output; if not, the module returns to the initialization module; The model building and testing module is used to build the optimal Transformer-BiLSTM model with the optimal parameter values ​​found by the IIVYA algorithm, and to perform diagnostic tests on photovoltaic array faults.

Images