Photovoltaic array ash deposition prediction method based on improved SSA-BP

Abstract

The invention discloses an improved SSA-BP-based photovoltaic array dust deposition prediction method, which adopts a method of combining a sparrow search algorithm and a BP neural network, and introduces Levy flight optimization SSA to enhance global search capability and local development capability and improve the problems of excessive gathering, diversity reduction and easy falling into local optimum when approaching an optimal position. The convergence precision is improved, and the optimal weight and threshold of the BP neural network are obtained by adopting the algorithm; and meanwhile, an electrical detection method principle is utilized, a simulation model is established in MATLAB / Simulink to analyze output PV and IV curves of the photovoltaic array in different fault states, characteristic quantities are extracted, sample data are acquired, and the algorithm is utilized to perform classification prediction on the data, so that the photovoltaic dust deposition degree and the shadow shielding condition are conveniently and accurately identified, and the accuracy of the detection result is improved. The optimal cleaning period is determined, the operation loss is reduced, and the economic benefit is improved.

Description

Technical Field

[0001] The invention belongs to the technical field of photovoltaic power generation, and mainly relates to a photovoltaic array dust accumulation prediction method based on improved SSA-BP. Background Art

[0002] As the photovoltaic industry develops rapidly, its safety and reliability also need to be guaranteed. In field operation, photovoltaic arrays are prone to panel light transmittance reduction due to dust, sand, dirt and other dust accumulation problems, and local hot spots may be caused by shadows from clouds, leaves, rain and fog. Whether it is dust accumulation or shadows, it will lead to a decline in the performance of photovoltaic panels, greatly affecting the power generation efficiency of photovoltaic panels and reducing economic benefits.

[0003] Accurately identifying the dust accumulation state and shadow occlusion type, and finding the characteristic criteria and identification methods are very important for the problem of photovoltaic panel power loss. Currently, there are three main ways to conduct detection, namely: traditional electrical measurement method, image detection method and intelligent system detection method.

[0004] In the traditional electrical measurement method, Gyu Gwang Kim ("GG Kim, W. Lee, BG Bang, JH Choi and H.-K Ahn,"Fault Detection for Photovoltaic Systems Using Multivariate Analysis With Electrical and Environmental Variables," in IEEE Journal of Photovoltaics, vol. 11, no. 1, pp. 202-212, Jan. 2021") detects faults in photovoltaic systems based on power ratio (PR), voltage ratio (VR) and current ratio (IR). By setting the upper and lower thresholds of electrical quantities, photovoltaic system faults including dust accumulation can be accurately identified to ensure the reliability of system operation. However, this method requires accurate electrical parameter settings. If the parameters are not set properly, it may cause false alarms or missed faults. Ma Mingyao et al. ("Ma Mingyao, Wang Haisong, Ma Wenting, Zhang Zhixiang, Zhang Xing. Diagnosis of shadow shading fault of crystalline silicon photovoltaic modules based on SV characteristic analysis [J]. Acta Energiae Solaris Sinica, 2022, 43(09): 64-72.") proposed an online diagnosis method for photovoltaic module shadow shading fault based on the SV curve characteristics of the IV curve inflection point, step, and integral area (S) decline characteristics. This method utilizes the characteristics of the IV curve and can effectively detect the shadow shading fault of the photovoltaic module. It has high detection accuracy, but it has high requirements for the collection and analysis of the IV curve, and requires high-precision measurement equipment and complex analysis algorithms.

[0005] In the image recognition method, the temperature change of the bottom plate of the photovoltaic panel caused by the hot spot effect can be used to identify the fault type and dust accumulation degree through infrared recognition combined with image detection. Liao Kun ("Liao Kun. Recognition and analysis of dust accumulation status of photovoltaic panels based on deep learning [D]. Northeast Electric Power University, 2021.") proposed a photovoltaic panel dust accumulation status image prediction model based on deep learning, which can identify photovoltaic panels and quantitatively analyze dust accumulation status through on-site images, with high degree of automation and fast detection speed. However, this method relies on a large amount of high-quality training data, the training model process is complex, the image quality requirements are high, and it is easily affected by the external environment. Lin Yi ("Lin Yi. Research on hot spot fault detection of photovoltaic modules [D]. Anhui University of Science and Technology, 2023.") took the infrared image of photovoltaic hot spots as the research object, and realized the fault detection of photovoltaic hot spots by denoising and segmenting the infrared images. The method is simple and robust, but the quality of infrared images is easily affected by the external environment, the segmentation accuracy needs to be improved, and it is highly dependent on hardware equipment.

[0006] Hu Yihua et al. (“Hu Yihua, Chen Hao, Xu Ruidong, et al. Output characteristics of photovoltaic panels under shadow influence [J]. Transactions of China Electrotechnical Society, 2011, 26(01): 123-128+134.”) used the system detection method to monitor photovoltaic modules. The sensors detect the output current and voltage in real time, and can timely judge whether the photovoltaic modules are faulty. It has real-time monitoring capability, but the sensor layout is complex, and the sensor accuracy and stability requirements are high, and the maintenance cost is high. Wang Huachao (“Wang Huachao. Design of high-precision photovoltaic panel dust scattering detection module [D]. Xi'an University of Technology, 2023.”) studied the photovoltaic panel dust scattering light intensity measurement method based on scattering theory, and designed a photovoltaic panel dust scattering light intensity measurement device, which helps to reduce the maintenance investment of photovoltaic power generation equipment and improve the power generation quality. However, the scattered light intensity measurement is sensitive to external lighting conditions, the equipment design and manufacturing are complex, and the practicality needs to be further verified. Wu Jianming et al. ("Wu Jianming, Yang Peihong, Zhang Jihong. Design of fault diagnosis for photovoltaic power generation system based on Internet of Things [J]. Renewable Energy, 2019, 37(1): 65-70.") designed a photovoltaic data remote acquisition system that can integrate and process the collected electrical data, has remote monitoring and data analysis capabilities, and is convenient for fault diagnosis. However, no specific fault judgment method is given, which relies on the stability of data transmission and the reliability of the acquisition system. Summary of the invention

[0007] The present invention is aimed at the problems in the prior art that the dust accumulation and shadow problems of photovoltaic panels lead to the decline of photovoltaic panel performance, affect the power generation efficiency, reduce economic benefits, and the image recognition method is greatly affected by the environment and has a high cost; the system detection method has complex installation and configuration and high reliability requirements. A photovoltaic array dust accumulation prediction method based on improved SSA-BP is provided, which adopts a method combining the sparrow search algorithm with the BP neural network, introduces the Levy flight optimization SSA to enhance the global search capability and local development capability, improve the problem of excessive aggregation when approaching the optimal position, reduced diversity, and easy to fall into the local optimum, improves the convergence accuracy, and uses the algorithm to obtain the optimal weights and thresholds of the BP neural network; at the same time, the principle of electrical detection method is used to establish a simulation model in MATLAB / Simulink to analyze the output PV and IV curves of the photovoltaic array in different fault states, extract feature quantities, obtain sample data, and use the algorithm to classify and predict the data, so as to accurately identify the degree of photovoltaic dust accumulation and shadow occlusion, so as to determine the optimal cleaning cycle, reduce operating losses, and improve economic benefits.

[0008] In order to achieve the above-mentioned purpose, the technical scheme adopted by the present invention is: based on the photovoltaic array dust accumulation prediction method of improved SSA-BP, the Levy flying optimized sparrow search algorithm (SSA) is introduced to obtain the optimal weights and thresholds of the BP neural network, and at the same time, the principle of electrical detection method is used to establish a simulation model in MATLAB / Simulink to analyze the output PV and IV curves of the photovoltaic array in different fault states, extract feature quantities, obtain sample data, and classify and predict the data.

[0009] As an improvement of the present invention, a photovoltaic array dust accumulation prediction method based on improved SSA-BP includes the following steps:

[0010] S1: Establish a photovoltaic array model, simulate different fault states of the photovoltaic array and analyze the performance of the short-circuit current, open-circuit voltage, maximum power point current and maximum power point voltage in the output PV and IV curves under different faults, extract the main features that can distinguish photovoltaic dust accumulation and shadow, obtain sample data by controlling the change of feature quantities and perform normalization processing; the input features of the model sample data are factors affecting dust accumulation, and the output target is the actual dust accumulation situation;

[0011] S2: Initialize the BP neural network parameters, including determining the number of nodes in each layer of the BP neural network, initializing the weights and thresholds of the BP neural network;

[0012] S3: Initialize SSA parameters, including but not limited to the population size, population location, number of iterations, ratio of discoverers to alerters, and determine the dimension, upper and lower limits of the problem;

[0013] S4: Evaluate the BP neural network corresponding to each individual sparrow in the population and calculate its fitness value;

[0014] S5: Enter the iterative optimization process and update the positions of the discoverer, joiner, and vigilant;

[0015] S6: Calculate the inertia weight factor, use roulette wheel selection, determine the mutation probability according to the fitness value of the sparrow individual, and perform Levy flight mutation on the selected sparrow individual;

[0016] S7: If the difference between the current optimal fitness value and the theoretical optimal fitness value is less than 5% of the theoretical optimal fitness value, the BP neural network parameters are output and the iteration is exited, otherwise the condition for continuing to execute step S4 is entered;

[0017] S8: Obtain optimized BP neural network parameters, and obtain the corresponding BP neural network weight and threshold according to the optimal sparrow individual finally obtained;

[0018] S9: Use the optimized BP neural network for prediction, input the test set data into the BP neural network with the optimal weight and threshold, and obtain the prediction result of dust accumulation.

[0019] As an improvement of the present invention, the photovoltaic array simulation model in step S1 is specifically:

[0020] The maximum power current I of the battery under standard lighting conditions given by the battery manufacturer mref , short circuit current I scref , Maximum power voltage V mref and open circuit voltage V ocref , the mathematical model constructed with these parameters can simulate a UI curve similar to the output characteristics of photovoltaic cells. In the actual simulation process under any environment, the above four parameters will change according to the rules. Therefore, it is necessary to design corresponding compensation coefficients to approximately calculate the values ​​of these four parameters under any light intensity and temperature conditions. The designed parameter expression is:

[0021] AT=TT ref

[0022]

[0023] V oc =V ocref (1-cΔT)In(1+bΔS)

[0024]

[0025] V m =V mref (1-cΔT)In(1+bΔS)

[0026] Where: T is the actual temperature, T ref is the reference temperature; S is the actual light intensity, S ref is the reference light intensity; ΔS is the difference between the actual light intensity and the reference value; ΔT is the difference between the actual temperature value and the reference value; V m ,I m 、V oc ,I sc is the actual output maximum power point voltage, maximum power point current, open circuit voltage, and short circuit current; V mref ,I mref 、V ocref ,I scref are the maximum power point voltage, maximum power point current, open circuit voltage and short circuit current under standard test conditions; a, b, c are I m ,I sc 、V m 、V oc The compensation factor.

[0027] As another improvement of the present invention, in step S2, the number of nodes in each layer of the BP neural network is determined by comparing the mean square error of the training sets with different numbers of hidden layer nodes. The comparison method is specifically as follows:

[0028]

[0029] Where: Q is the number of hidden layer nodes, M is the number of input nodes, L is the number of output nodes, α is a constant between 1 and 10, which is used to adjust the number of hidden layer nodes; N MSE is the mean square error, X i is the theoretical value; x i is the predicted value, and N is the number of samples.

[0030] As another improvement of the present invention, in step S3, the initialization of the sparrow population position is specifically:

[0031] X=lb+rand*(ub-lb)

[0032] Among them, X represents the position of an individual in the population, and ub and lb represent the upper and lower boundaries of the population position respectively.

[0033] As another improvement of the present invention, the method for calculating the individual fitness value in step S4 is:

[0034] F fitness =(MSE(input_train)+MSE(input_test)) / 2

[0035] Among them, F fitnessis the individual fitness value, MSE(input_train) is the mean square error of the input training set, and MSE(input_test) is the mean square error of the input test set.

[0036] As another improvement of the present invention, in step S5, the location update method of the discoverer is:

[0037]

[0038] Where: i represents the serial number of the sparrow, j represents the position of the sparrow; t and T are the current and maximum update times of the algorithm respectively, X t i，j is the j-th dimension position of the ith individual in the tth generation; α is a random number whose value range is (0, 1); Q is a random number that follows a normal distribution; L is a row short matrix whose elements are all 1; R2 is a warning value greater than 0 and less than 1, and ST is a safety value greater than 0.5 and less than 1;

[0039] The method for updating the location of the joiner is: if i>n / 2, then

[0040]

[0041] Otherwise, the position update formula is:

[0042]

[0043] Where: Q is a random number that follows a normal distribution; X t i，j is the j-th dimension position of the ith individual in the tth generation, X t worst is the worst position in the current population; X p (t+1) The best position after position update among all discoverers; A + =A T (AA T ) -1 , A is a row matrix whose elements are randomly 1 or -1; L is a row matrix whose elements are all 1; n is the number of sparrows in the population;

[0044] The position update formula of the sentinel is:

[0045]

[0046] Where: X t best is the best position in the current population, X t i，j is the j-th dimension position of the ith individual in the tth generation; β is a random number and follows the standard normal distribution; fi ,f g ,f w are the optimal and worst fitness values ​​of the current individual and the current population respectively; K takes a random value in the range of (-1, 1), and ε is a constant.

[0047] As another improvement of the present invention, the calculation method of the inertia weight factor in step S6 is:

[0048]

[0049] Where: t is the current iteration number, T is the maximum iteration number.

[0050] As a further improvement of the present invention, in step S6, the calculation method of the Levy flight path is as follows using the roulette wheel selection:

[0051]

[0052] Among them, the parameter β is the stability index of Levy flight, and its value range is 0<β<2; the parameters u and v are normal distribution random numbers.

[0053] Compared with the prior art, the present invention has the following beneficial effects: the present invention discloses a photovoltaic array dust accumulation prediction method based on improved SSA-BP, obtains sample data by establishing a photovoltaic array fault condition simulation model, extracts characteristic quantities, adopts Levy flight optimization SSA to enhance global search capability, improves falling into local optimum, improves convergence accuracy, improves SSA search capability, avoids falling into local optimum, improves the algorithm's global optimization capability, and combines with BP neural network to improve the problems of long search time, slow convergence speed, easy to fall into local optimum, poor generalization ability, etc. of BP neural network. Using this method, the photovoltaic array dust accumulation type and shadow occlusion are classified and identified, the prediction accuracy is improved, it is convenient for timely cleaning, and the system operation loss is reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 The photovoltaic cell single diode equivalent circuit diagram when the photovoltaic array simulation model is established in step S1 of the present invention;

[0055] Figure 2 It is a flowchart of the improved SSA-BP in the embodiment of the present invention;

[0056] Figure 3 It is the confusion matrix diagram of the SSA-BP test set and training set in the test example of the present invention;

[0057] Figure 3 (a) is the confusion matrix diagram of the SSA-BP test set;

[0058] Figure 3(b) is the confusion matrix diagram of the SSA-BP training set;

[0059] Figure 4 is the SSA-BP convergence curve in the test example of the present invention;

[0060] Figure 5 It is a graph of the prediction results of the SSA-BP test set and the training set in the test example of the present invention;

[0061] Figure 5 (a) is the prediction result diagram of the SSA-BP test set;

[0062] Figure 5 (b) is the prediction result diagram of the SSA-BP training set;

[0063] Figure 6 The improved SSA-BP test set and training set confusion matrix diagram in the test example of the present invention;

[0064] Figure 6 (a) Confusion matrix diagram of the improved SSA-BP test set;

[0065] Figure 6 (b) Confusion matrix diagram of the improved SSA-BP training set;

[0066] Figure 7 It is the improved SSA-BP convergence curve in the test example of the present invention;

[0067] Figure 8 It is a prediction result diagram of the improved SSA-BP test set and training set in the test example of the present invention;

[0068] Figure 8 (a) is the prediction result diagram of the improved SSA-BP test set;

[0069] Figure 8 (b) is the prediction result diagram of the improved SSA-BP training set. DETAILED DESCRIPTION

[0070] The present invention will be further explained below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention.

[0071] Example 1

[0072] The photovoltaic array dust accumulation prediction method based on the improved SSA-BP, in which the improved SSA-BP neural network flow chart is as follows Figure 2 As shown, the following steps are included:

[0073] S1: Establish a photovoltaic array model, simulate different fault states of the photovoltaic array and analyze the performance of short-circuit current, open-circuit voltage, maximum power point current and maximum power point voltage in the output PV and IV curves under different faults, extract the main features that can distinguish photovoltaic dust accumulation and shadow, obtain sample data by controlling the change of feature quantities and perform normalization processing. The input features of the model sample data are the factors affecting dust accumulation, and the output target is the actual dust accumulation situation:

[0074] The maximum power current I of the battery under standard lighting conditions given by the battery manufacturer mref , short circuit current I scref , Maximum power voltage V mref and open circuit voltage V ocref , the mathematical model constructed with these parameters can simulate a UI curve similar to the output characteristics of photovoltaic cells. In the actual simulation process under any environment, the above four parameters will change according to the rules. Therefore, it is necessary to design corresponding compensation coefficients to approximately calculate the values ​​of these four parameters under any light intensity and temperature conditions. Figure 1 As shown, through the single diode equivalent circuit analysis of photovoltaic cells, the design parameter expressions are (1)-(6) to establish the simulation model of photovoltaic cells: Since the single photovoltaic panel required in the actual experiment is composed of 3*10 photovoltaic cells, the simulated single cell model is connected in series and parallel to obtain the actual photovoltaic panel model, and finally the two photovoltaic panel models are connected in series to obtain the corresponding model of the actual photovoltaic array. In MATLAB / SIMULINK, the output IV and PV curves of different dust accumulation conditions and shadow shading types are analyzed, and the selected characteristic quantities are: short-circuit current I sc , the current at the maximum power point I ma x, maximum power point voltage U max , characteristic inflection point quantity N, light intensity G.

[0075] △T=TT ref (1)

[0076]

[0077] V oc =V ocref (1-c△T)In(1+b△S)(4)

[0078]

[0079] V m =V mref (1-c△T)In(1+b△S)(6)

[0080] Where: T is the actual temperature, T refis the reference temperature; S is the actual light intensity, S ref is the reference light intensity; ΔS is the difference between the actual light intensity and the reference value; ΔT is the difference between the actual temperature value and the reference value; V m ,I m 、V oc ,I sc is the actual output maximum power point voltage, maximum power point current, open circuit voltage, and short circuit current; V mref ,I mref 、V ocref ,I scref are the maximum power point voltage, maximum power point current, open circuit voltage and short circuit current under standard test conditions. a, b, c are I m ,I sc 、V m 、V oc The compensation coefficient is: a = 0.0025; b = 0.5; c = 0.00288;

[0081] S2: Initialize the BP neural network parameters, including determining the number of nodes in each layer of the BP neural network, initializing the weights and thresholds of the BP neural network;

[0082] The number of nodes in the hidden layer of the BP neural network plays a decisive role in the performance of the BP neural network. Too few nodes will reduce the time to obtain the information regularity of the training samples, making it difficult to ensure the correctness of the information obtained, and the generalization ability is poor; too many nodes will prolong the time to obtain the information regularity of the training samples, and even lead to overfitting. According to formulas (7)-(8), the mean square error (MSE) of the training set with different numbers of hidden layer nodes is compared to select the optimal number of hidden layer nodes:

[0083]

[0084] Where: Q is the number of hidden layer nodes, M is the number of input nodes, L is the number of output nodes, α is a constant between 1 and 10, which is used to adjust the number of hidden layer nodes; N MSE is the mean square error, X i is the theoretical value; x i is the predicted value, and N is the number of samples.

[0085] S3: Initialize SSA parameters, including but not limited to the population size, population location, number of iterations, ratio of discoverers to alerters, and determine the dimension, upper and lower limits of the problem;

[0086] The sparrow search algorithm (SSA) is a swarm intelligence optimization algorithm, which is mainly inspired by the foraging and anti-predation behaviors of sparrows. It finds the optimal solution by simulating the foraging and anti-predation behaviors of sparrows. The algorithm has good global search ability and convergence speed, is simple to implement and has fewer parameters, and has relatively little dependence on the initial population. By continuously iteratively updating the positions of sparrows (candidates for solutions), the algorithm gradually approaches the optimal solution. The sparrow search algorithm generates the initial population using a random method and initializes the position of the sparrow population according to formula (9):

[0087] X=lb+rand*(ub-lb)(9)

[0088] Among them, X represents the position of an individual in the population, and ub and lb represent the upper and lower boundaries of the population position respectively.

[0089] S4: Evaluate the BP neural network corresponding to each individual sparrow in the population and calculate its fitness value;

[0090] The choice of fitness function often depends on the specific problem and optimization goal. Generally speaking, the smaller the fitness function value, the higher the training accuracy and the better the prediction accuracy of the model. In terms of the selection of fitness function, the mean square error and root mean square error are commonly used. Among them, the mean square error focuses on minimizing the overall error, while the root mean square error focuses on the overall distribution of errors and the sensitivity to large errors. In order to consider the generalization performance of the model, the overall mean square error of the training set and the test set of the input data is selected as the fitness function, and the individual fitness of the sparrow is calculated according to formula (10):

[0091] F fitness =(MSE(input_train)+MSE(input_test)) / 2(10)

[0092] Among them, F fitness is the individual fitness value, MSE(input_train) is the mean square error of the input training set, and MSE(input_test) is the mean square error of the input test set.

[0093] S5: Enter the iterative optimization process and update the positions of the discoverer, joiner, and vigilant;

[0094] In the sparrow search algorithm, discoverers are an important part of the population, usually accounting for 10%-20%. They are responsible for actively exploring the search space, providing foraging direction and location information for the population, and have good fitness values. Their positions are updated according to the algorithm rules and search status, involving factors such as random numbers. When the algorithm is running, the identity of the discoverer may change dynamically. This dynamicity maintains population diversity and search vitality, balances exploration and utilization, and avoids falling into local optimality too early. For each discoverer, the position is updated according to formula (11):

[0095]

[0096] Where: i represents the serial number of the sparrow, j represents the position of the sparrow; t and T are the current and maximum update times of the algorithm respectively, X t i，j is the j-th dimension position of the ith individual in the tth generation; α is a random number whose value range is (0, 1); Q is a random number that follows a normal distribution; L is a row short matrix whose elements are all 1; R2 is a warning value greater than 0 and less than 1, and ST is a safety value greater than 0.5 and less than 1;

[0097] Joiners are an important part of the sparrow search algorithm. They account for a large proportion of the population, mainly follow the discoverer to obtain food, monitor and compete, and have different behaviors with different fitness. Joiners increase the diversity of the population, collaborate with the discoverer to find the best solution, and avoid local optimality. For each joiner, the position is updated according to different situations. If i>n / 2 (where n is the number of sparrows in the population), it means that the joiner has not obtained food and is in a hungry state. The position update formula is:

[0098]

[0099] Otherwise, the position update formula is:

[0100]

[0101] Where: Q is a random number that follows a normal distribution; X t i，j is the j-th dimension position of the ith individual in the tth generation, X t worst is the worst position in the current population; X p (t+1) The best position after position update among all discoverers; A + =A T (AA T ) -1 , A is a row matrix whose elements are randomly 1 or -1; L is a row matrix whose elements are all 1; n is the number of sparrows in the population;

[0102] The sentinels are an important part of the sparrow search algorithm. They monitor the environment and detect danger. They judge danger according to the rules, issue alarms to guide the population to adjust strategies or change positions, increase the security and stability of the algorithm, and avoid unfavorable situations. The position update formula is:

[0103]

[0104] Where: X t best is the best position in the current population, X t i，j is the j-th dimension position of the ith individual in the tth generation; β is a random number and follows the standard normal distribution; f i ,f g ,f w are the optimal and worst fitness values ​​of the current individual and the current population respectively; K takes a random value in the range of (-1, 1), and ε is a constant.

[0105] S6: Calculate the inertia weight factor, use roulette wheel selection, determine the mutation probability according to the fitness value of the sparrow individual, and perform Levy flight mutation on the selected sparrow individual;

[0106] Levy flight is a search pattern of animals in nature, which is characterized by short-distance exploratory jumps and occasional longer-distance walks. This pattern has good exploration capabilities and can achieve a good balance between local detailed search and large-scale search. Introducing Levy flight in the sparrow search algorithm helps to increase its search capability and diversity.

[0107] The inertia weight factor is usually used to control the inertia of the algorithm during the search process. Its function is to balance the global search and local search capabilities of the algorithm. The inertia weight factor is calculated according to formula (15):

[0108]

[0109] Where: t is the current iteration number, T is the maximum iteration number.

[0110] Roulette selection is a selection mechanism used to select individuals from a sparrow population for mutation operations. Using roulette selection, if the random number rand>f, the selected sparrow individual is subjected to Levy flight mutation, a random step length is generated according to the Levy distribution, and the position of the selected sparrow individual is updated according to the generated step length. The Levy flight path is calculated by the following formula (16):

[0111]

[0112] Among them, parameter β is the stability index of levy flight, and its value range is usually 0<β<2, and β=1.5 is generally taken; parameters u and v are normal distribution random numbers, which obey the following normal distribution, see formula (17):

[0113]

[0114] The corresponding normal distribution standard deviation σ u , σ v The value of satisfies formula (18):

[0115]

[0116] Where γ is the gamma function.

[0117] S7: If the difference between the current optimal fitness value and the theoretical optimal fitness value is less than 5% of the theoretical optimal fitness value, the BP neural network parameters are output and the iteration is exited. Otherwise, the process proceeds to step S4 to continue the execution.

[0118] S8: Obtain optimized BP neural network parameters, and obtain the corresponding BP neural network weights and thresholds according to the optimal sparrow individual finally obtained.

[0119] S9: Use the optimized BP neural network for prediction, input the test set data into the BP neural network with the optimal weight and threshold, and obtain the prediction result of dust accumulation.

[0120] Test Case

[0121] The test environment of this case is MATLAB R2022b, and the test data comes from the simulated data generated by different fault models of photovoltaic array simulation. By comparing the Levy flight optimization SSA-BP algorithm with the SSA-BP algorithm, the effectiveness and performance of the Levy flight optimization SSA-BP in the photovoltaic dust accumulation prediction problem are verified. Figures 3 to 8 As shown in Figure 2, the prediction performance of the model is quantitatively analyzed using the confusion matrix, convergence curve and prediction results of the test set and training set of the output model. The two are compared from the following three aspects. First, the prediction error of the confusion matrix is ​​compared, as shown in Figure 2. Figure 3 , Figure 5 The numbers 1 to 7 shown in the figure represent the classification of dust accumulation and faults into normal conditions, 60% to 80% dust accumulation, 80% and above dust accumulation, small-scale irregular dust accumulation such as bird droppings and leaves, uneven dust accumulation due to wind and sand, and shadow occlusion. The confusion matrix of the SSA-BP test set and training set is shown in Figure 3 (a) Figure 3(b) As can be seen from the figure, SSA-BP can achieve 100% complete recognition of 40% to 60% dust accumulation and shadow occlusion, normal conditions and wind-blown uneven dust accumulation can reach more than 99% recognition, and 80% and above dust accumulation can also reach more than 98% recognition. The recognition accuracy of small-scale irregular dust accumulation such as bird droppings and leaves and 60% to 80% dust accumulation is relatively low, the former is 95.3%, and the latter is only about 90%. For the improved SSA-BP, the confusion matrix of its test set and training set is shown in Figure 6 (a) Figure 6 (b) Compared with the unimproved SSA-BP, it is found that: with other accuracies remaining basically unchanged, the dust recognition accuracy of 60% to 80% is improved from about 90% to 97.7%, and the overall accuracy has increased, reflecting good dust prediction performance.

[0122] Secondly, the convergence speed is compared. The SSA-BP convergence curve is as follows: Figure 4 As shown in the figure, the number of iterations of the sparrow population reaches the minimum value at the 39th generation, and the fitness value is 0.007 at this time, while the number of iterations of the improved SSA-BP reaches the minimum value of 0.01 at the 31st generation. Relatively speaking, the convergence performance is better, but the fitness value will be relatively worse, such as Figure 7 shown.

[0123] Finally, the prediction accuracy is compared. The prediction results of the SSA-BP test set and training set are shown in the following figure. Figure 5 As shown in (a) and (b), the prediction accuracy of the test set and training set are 97.5806% and 98.2727% respectively, while the prediction results of the improved SSA-BP test set and training set are as follows Figure 8 (a) Figure 8 As shown in (b), the prediction accuracy can reach 98.6559% and 98.5455% respectively, and the accuracy of both the test set and the training set has been significantly improved. This shows that the improved dust accumulation prediction model has high reliability and has sufficient advantages in result accuracy.

[0124] In summary, this case adopts the principle of electrical detection method to simulate the photovoltaic array under different dust accumulation and shadow conditions and analyze its output characteristic curve, extract fault characteristics, and propose a photovoltaic array dust accumulation prediction method based on the improved sparrow search algorithm to optimize the BP neural network. The Levy flying optimization SSA is introduced to enhance the global search capability, improve the local optimum, and improve the convergence accuracy. The algorithm is used to optimize the BP neural network to find the optimal weights and thresholds for it, and improve the problems of long search time, slow convergence speed, easy to fall into local optimum, and poor generalization ability caused by random selection of weights and thresholds. Finally, the algorithm is used to predict fault data to improve the recognition accuracy of dust accumulation and shadow faults in photovoltaic arrays.

[0125] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0126] It should be noted that the above content only illustrates the technical idea of ​​the present invention and cannot be used to limit the protection scope of the present invention. For ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications all fall within the protection scope of the claims of the present invention.

Claims

1. A photovoltaic array dust accumulation prediction method based on improved SSA-BP is characterized by: The Levy flying optimized sparrow search algorithm (SSA) is introduced to obtain the optimal weights and thresholds of the BP neural network. At the same time, the principle of electrical detection method is used to establish a simulation model in MATLAB / Simul ink to analyze the output PV and IV curves of the photovoltaic array under different fault states, extract feature quantities, obtain sample data, and classify and predict the data.

2. The photovoltaic array dust accumulation prediction method based on improved SSA-BP according to claim 1 is characterized in that: The steps include: S1: Establish a photovoltaic array model, simulate different fault states of the photovoltaic array and analyze the performance of the short-circuit current, open-circuit voltage, maximum power point current and maximum power point voltage in the output PV and IV curves under different faults, extract the main features that can distinguish photovoltaic dust accumulation and shadow, obtain sample data by controlling the change of feature quantities and perform normalization processing; the input features of the model sample data are factors affecting dust accumulation, and the output target is the actual dust accumulation situation; S2: Initialize the BP neural network parameters, including determining the number of nodes in each layer of the BP neural network, initializing the weights and thresholds of the BP neural network; S3: Initialize SSA parameters, including but not limited to the population size, population location, number of iterations, ratio of discoverers to alerters, and determine the dimension, upper and lower limits of the problem; S4: Evaluate the BP neural network corresponding to each individual sparrow in the population and calculate its fitness value; S5: Enter the iterative optimization process and update the positions of the discoverer, joiner, and vigilant; S6: Calculate the inertia weight factor, use roulette wheel selection, determine the mutation probability according to the fitness value of the sparrow individual, and perform Levy flight mutation on the selected sparrow individual; S7: If the difference between the current optimal fitness value and the theoretical optimal fitness value is less than 5% of the theoretical optimal fitness value, the BP neural network parameters are output and the iteration is exited, otherwise the condition for continuing to execute step S4 is entered; S8: Obtain optimized BP neural network parameters, and obtain the corresponding BP neural network weight and threshold according to the optimal sparrow individual finally obtained; S9: Use the optimized BP neural network for prediction, input the test set data into the BP neural network with the optimal weight and threshold, and obtain the prediction result of dust accumulation.

3. The photovoltaic array dust accumulation prediction method based on improved SSA-BP according to claim 2 is characterized in that: The photovoltaic array simulation model in step S1 is specifically: △T=T-T ref In oc =V ocref (1-c△T)In(1+b△S) In m =V mref (1-c△T)In(1+b△S) Where: T is the actual temperature, T ref is the reference temperature; S is the actual light intensity, S ref is the reference light intensity; ΔS is the difference between the actual light intensity and the reference value; ΔT is the difference between the actual temperature value and the reference value; V m ,I m 、V oc ,I sc is the actual output maximum power point voltage, maximum power point current, open circuit voltage, and short circuit current; V mref ,I mref 、V ocref ,I scref are the maximum power point voltage, maximum power point current, open circuit voltage and short circuit current under standard test conditions; a, b, c are I m ,I sc 、V m 、V oc The compensation factor.

4. The photovoltaic array dust accumulation prediction method based on improved SSA-BP according to claim 3 is characterized by: In step S2, the number of nodes in each layer of the BP neural network is determined by comparing the mean square errors of training sets with different numbers of hidden layer nodes. The comparison method is specifically as follows: Where: Q is the number of hidden layer nodes, M is the number of input nodes, L is the number of output nodes, α is a constant between 1 and 10; N MSE is the mean square error, X i is the theoretical value; x i is the predicted value, and N is the number of samples.

5. The photovoltaic array dust accumulation prediction method based on improved SSA-BP as claimed in claim 4, characterized in that: In step S3, the position of the sparrow population is initialized as follows: X=lb+rand*(ub-lb) Among them, X represents the position of an individual in the population, and ub and lb represent the upper and lower boundaries of the population position respectively.

6. The photovoltaic array dust accumulation prediction method based on improved SSA-BP as claimed in claim 5, characterized in that: The calculation method of the individual fitness value in step S4 is: F fitness =(MSE(input_train)+MSE(input_test)) / 2 Among them, F fitness is the individual fitness value, MSE(input_train) is the mean square error of the input training set, and MSE(input_test) is the mean square error of the input test set.

7. The photovoltaic array dust accumulation prediction method based on improved SSA-BP as claimed in claim 6, characterized in that: In step S5, the location update method of the discoverer is: Where: i represents the serial number of the sparrow, j represents the position of the sparrow; t and T are the current and maximum update times of the algorithm respectively, X t i，j is the j-th dimension position of the ith individual in the tth generation; α is a random number whose value range is (0, 1); Q is a random number that follows a normal distribution; L is a row short matrix whose elements are all 1; R2 is a warning value greater than 0 and less than 1, and ST is a safety value greater than 0.5 and less than 1; The method for updating the location of the joiner is: if i>n / 2, then Otherwise, the position update formula is: Where: Q is a random number that follows a normal distribution; X t i，j is the j-th dimension position of the ith individual in the tth generation, X t worst is the worst position in the current population; X p (t+1) The best position after position update among all discoverers; A + =A T (AA T ) -1 , A is a row matrix whose elements are randomly 1 or -1; L is a row matrix whose elements are all 1; n is the number of sparrows in the population; The position update formula of the sentinel is: Where: X t best is the best position in the current population, X t i，j is the j-th dimension position of the ith individual in the tth generation; β is a random number and follows the standard normal distribution; f i ,f g ,f w are the optimal and worst fitness values ​​of the current individual and the current population respectively; K takes a random value in the range of (-1, 1), and ε is a constant.

8. The photovoltaic array dust accumulation prediction method based on improved SSA-BP as claimed in claim 7, characterized in that: The calculation method of the inertia weight factor in step S6 is: Where: t is the current iteration number, T is the maximum iteration number.

9. The photovoltaic array dust accumulation prediction method based on improved SSA-BP as claimed in claim 7, characterized in that: In step S6, the calculation method of the Levy flight path is as follows using the roulette wheel selection: Among them, the parameter β is the stability index of Levy flight, and its value range is 0<β<2; the parameters u and v are normal distribution random numbers.

Images