Two-step method for photovoltaic array fault severity assessment based on fault parameter estimation and fine tuning

Abstract

The two-step photovoltaic array fault degree evaluation method based on fault parameter estimation and fine adjustment of the application belongs to photovoltaic power generation technology, and comprises the following steps: S1: obtaining the measured I-V characteristic curve of the photovoltaic array under different environments and faults; S2: obtaining the estimated fault parameter after training the back propagation neural network according to the measured I-V characteristic curve; S3: establishing a photovoltaic array I-V characteristic curve simulation model to obtain a simulation I-V characteristic curve; S4: minimizing the root mean square error between the measured I-V characteristic curve and the simulation I-V characteristic curve according to the particle swarm optimization algorithm to obtain the optimized fault parameter; and S5: calculating the fault degree diagnosis quantitative precision according to the optimized fault parameter. The method can calculate the quantitative precision of the four fault parameters, and the fault research of the photovoltaic array is more thorough through the quantitative precision calculation of the four fault parameters.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of photovoltaic power generation, and particularly relates to a two-step photovoltaic array fault degree evaluation method based on fault parameter estimation and fine adjustment. BACKGROUND

[0002] As a large-scale solar power generation system, photovoltaic arrays are prone to be affected by weather conditions such as wind sand, ultraviolet radiation, hail and the like due to being installed outdoors, resulting in various abnormalities and faults, affecting power output, and even possibly causing fire or electric shock risks. Therefore, in order to ensure the safety and efficient operation of photovoltaic systems, the fault diagnosis of photovoltaic arrays has attracted much attention.

[0003] In recent years, the fault diagnosis methods of photovoltaic arrays mainly include an infrared image-based method and a method based on the measurement of the current-voltage (I-V) characteristics of photovoltaic arrays. The former is not suitable for small photovoltaic power stations due to its high equipment cost. The authorized announcement CN112016260B discloses a hot spot cell temperature estimation method and device based on the I-V curve of a photovoltaic module and a storage medium. In this method, the key information that can represent the fault characteristics of the photovoltaic array is included on the I-V characteristic curve, so the fault diagnosis accuracy can be improved. However, the method of the comparative document can only classify the fault types, and does not study the severity of the faults, and the evaluation of the severity of the faults can quantitatively describe the harm and loss of the faults, which will directly affect the troubleshooting and decision-making of the operation and maintenance personnel, so the research on the faults is not thorough. SUMMARY

[0004] In order to overcome the deficiencies in the prior art, the application provides a two-step photovoltaic array fault degree evaluation method based on fault parameter estimation and fine adjustment, which solves the technical problem of how to increase the research on the severity of the faults.

[0005] In order to achieve the above purpose, the two-step photovoltaic array fault degree evaluation method based on fault parameter estimation and fine adjustment of the application comprises the following steps,

[0006] S1: obtaining the measured I-V characteristic curve of the photovoltaic array under different environments and faults according to the scanning of the inverter on the I-V characteristics of the photovoltaic array;

[0007] S2: obtaining the estimated fault parameters after training the back propagation neural network according to the measured I-V characteristic curve in step S1;

[0008] S3: establishing an I-V characteristic curve simulation model of the photovoltaic array, and obtaining a simulation I-V characteristic curve by configuring the estimated fault parameters;

[0009] S4: According to the particle swarm optimization algorithm, the root mean square error between the measured I-V characteristic curve and the simulation I-V characteristic curve is minimized, the estimated fault parameter is fine-tuned, and the optimized fault parameter is obtained;

[0010] S5: According to the optimized fault parameter, the fault degree diagnosis quantitative precision is calculated;

[0011] In step S2, the noise contained in the measured I-V characteristic curve is de-redundant, smoothed and pre-processed by filtering abnormal data, and the measured I-V characteristic curve is normalized to eliminate temperature, irradiance and uniform interpolation. According to the influence of three typical faults of photovoltaic string short circuit, local shadow shielding and series resistance increase on the measured I-V characteristic curve, 9 characteristic data on the measured I-V characteristic curve are extracted, including: the voltage, current and power at the maximum power point of the measured I-V characteristic curve, i.e. V mpp , I mpp and P mpp ; the open circuit voltage V OC and the maximum value M of the absolute value of the second derivative, the expression of M is:

[0012]

[0013] Wherein V i , I i and dI i are the voltage, current and absolute value of the first derivative of the i-th discrete point on the I-V curve, and n is the number of points on the interpolated measured I-V characteristic curve;

[0014] In addition, the current I M and voltage V M at the maximum value M of the absolute value of the second derivative on the measured I-V characteristic curve are also extracted as features; the last two extracted features are the slope k OC at the open circuit voltage and the slope k SC at the short circuit current, respectively represented as:

[0015]

[0016] After the feature extraction of the measured I-V characteristic curve, the 9 extracted features are combined with the temperature and irradiance corresponding to the measured I-V characteristic curve as 11 input neurons of the back propagation neural network, which are used for training the designed three-layer back propagation neural network; the total number of samples under different environments and faults is shared by 11 neurons, and the number of samples in the training set is 70% of the total number of samples; the learning rate is adjusted to 0.6, tanh and Sigmoid are set as the activation functions of the last two layers, and the average cross-entropy C is set as the loss function, represented as:

[0017]

[0018] where Y b and Y' b are the label and the back propagation neural network estimated value of the bth training sample respectively, and m is the total number of training set samples for the neural network; the preliminary fault diagnosis of the measured I-V characteristic curve is realized on the test set, and the estimated fault parameters output by the back propagation neural network include: the number of short-circuit sub-strings n sc , the number of shadow-shielded sub-strings n ps , the shadow-shielded transmittance d ps , and the increased series resistance r s , a total of four output neurons, corresponding to three typical faults, i.e., n sc corresponding to photovoltaic string short circuit, n ps and d ps corresponding to local shadow shielding, and r s corresponding to increased series resistance.

[0019] The quantification accuracy calculation formula of the four fault parameters is as follows:

[0020]

[0021] where A nsc , A nps , A dps , and A rs are the fault degree diagnosis quantification accuracy of the following four fault parameters, i.e., the number of short-circuit sub-strings n sc , the number of shadow-shielded sub-strings n ps , the shadow-shielded transmittance d ps , and the increased series resistance r s ; n' sc , n' ps , d' ps , and r' s are the optimized fault parameters corresponding to the four fault parameters; the function sgn(x) outputs 1 and -1 when the independent variable is greater than zero and less than zero, respectively, and outputs 0 when the independent variable is 0, and g is the total number of samples in the test set.

[0022] Further, in step S3, the photovoltaic array I-V characteristic curve simulation model is based on a single diode model based on reverse bias, built under the Python framework, and by configuring the temperature and irradiance matrix G, configuring the shadow shielding block and the number of bypass diodes, the simulation I-V characteristic curve is obtained.

[0023] The mathematical expression of the single diode model based on reverse bias can be expressed as:

[0024]

[0025] where I j and V j are the output current and voltage of the photovoltaic module, q, k, T are the electron charge, Boltzmann constant and the temperature of the cell, respectively, and the temperature unit is K; I ph , I s , a, R s , R sh are five parameters of the mathematical model, respectively representing the photo-generated current, the saturation current of the diode, the diode ideality factor, the series resistance and the parallel resistance; n' is the correction factor of avalanche breakdown, V bd is the breakdown voltage, and m' is the breakdown exponent;

[0026] In addition, the photovoltaic array I-V characteristic curve simulation model also includes a shadow blocking block configuration and a sub-string bypass diode state column configuration, and the defined shadow blocking block input matrix B is in the following form:

[0027] B = [[X f , Y f , G f ], …, [X n2 , Y n2 , G n2 ]] (2)

[0028] The shadow block matrix B is a two-dimensional vector representing the number of shadow blocks and the parameters of the shadow blocks, where X f and Y f are the length and width of the fth shadow block, G f is the light transmittance of the fth shadow block, and n2 is the number of shadow blocks; the defined sub-string bypass diode state column D is used as a variable for the photovoltaic array I-V characteristic curve simulation model, and is expressed in the following form:

[0029]

[0030] where d e,1 represents the state of the 1st bypass diode in the e th photovoltaic module, n g is the number of modules in the photovoltaic string; when d = 0, the bypass diode is in normal working state, when d = 1, the bypass diode is in short-circuit fault, and when d = 2, the bypass diode is in open-circuit fault;

[0031] The output voltage V 总 of the photovoltaic array is:

[0032] V 总 = sum (V j x t j ) - I 总 x r s (4)

[0033] where V j is the output voltage of the photovoltaic module, I 总 is the output current of the photovoltaic array, r s is the increased series resistance, and the function sum(x) is used to sum the matrix;

[0034] Thus, the photovoltaic array I-V characteristic curve simulation model is formed by combining the above formulas (1), (2), (3), and (4).

[0035] Further, in step S3, the fault parameter configuration process is as follows: when the shadow blocking matrix in formula (2) is configured, a shadow blocking fault occurs, the shadow block length and width (X f , Y f ) corresponding to the position of the irradiance matrix G and B are increased by the irradiance value increasing coefficient d ps ; at this time, there is a blocking of a component sub-string, n ps is increased by 1, and the new irradiance after blocking directly affects the model parameters a, R s , R sh in formula (1); when no shadow blocking block is configured, a no-shadow blocking fault occurs; when the diode state d = 1 of a sub-string connected to a certain photovoltaic module in formula (3), a short circuit fault occurs, at this time, the module has a short circuit of the corresponding sub-string, n sc is increased by 1, and the output voltage V = 0 in formula (1); when d = 0, a no-short circuit fault occurs; then, all fault conditions are counted, and the number of short-circuit sub-strings n sc and the number of blocked sub-strings n ps are recorded; the output current I j and the output voltage V j of the photovoltaic module in formula (1) are calculated by the Newton-Raphson method combined with variable step scanning, the I j and V j of all photovoltaic modules are superimposed, wherein if a series resistance increase fault occurs, formula (4) needs to be considered, so that the increased series resistance r s participates in the superposition process, if there is no series resistance increase fault, it is not considered, and after superposition, the photovoltaic array output current I 总 and the output voltage V 总 are obtained, finally, the array pair (I 总 , V 总 ) set of the photovoltaic array output current and output voltage is collected, the array pair is a data point on the simulated I-V characteristic curve, and the connection of all data points can form a simulated I-V characteristic curve.

[0036] Further, in step S4, the root mean square error between the measured I-V characteristic curve and the simulated I-V characteristic curve is minimized, that is, the RMSE is set as the objective function, and the expression is as follows:

[0037]

[0038] Wherein, n is the number of points on the I-V curve, I model,i and I meas,i are the simulation value and the measured value of the i-th point in the I-V curve respectively; the initial position P1 of the particle is expressed as follows:

[0039] P1=[n sc ,n ps ,d ps ,r s ]×rand

[0040] Wherein rand is a random number in 0.3-3, and the particle swarm algorithm can be optimized in the range of 0.3-3 coefficient of the estimated fault parameter; when the particle position is updated continuously, the convergence is considered to be achieved when the objective function RMSE is less than the set value 0.0004, and the termination position P2 of the particle is expressed as follows:

[0041] P2=[n′ sc ,n′ ps ,d′ ps ,r s ′]

[0042] Wherein n' sc , n' ps , d' ps , r' s are the optimized fault parameters.

[0043] Beneficial effects: the method first obtains the estimated fault parameters through the back propagation neural network; then the estimated fault parameters are fine-tuned according to the particle swarm optimization algorithm to obtain the optimized fault parameters; and the fault degree diagnosis quantization precision is calculated by using the optimized fault parameters.

[0044] The method can obtain more features on the I-V characteristic curve by using the back propagation neural network; compared with the second derivative feature of the single I-V characteristic curve adopted in the existing patent, the method can reflect the fault type and the fault severity more, and the feature information of the fault is more comprehensive.

[0045] The I-V characteristic curve simulation model of the photovoltaic array is solved by formula (1) (2) (3) (4) to simulate the I-V characteristic curve under each fault condition, and the existing patent can only simulate the I-V characteristic curve under normal conditions.

[0046] The method puts forward the calculation of the quantization accuracy of four fault parameters, and the fault research of the photovoltaic array is more thorough through the quantization accuracy calculation of the four fault parameters. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1 is a flowchart of a specific embodiment of the present application.

[0048] Figure 2 is a specific embodiment of the back propagation neural network topology. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme of the present application will be described clearly and completely below in conjunction with the drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0050] The two-step method for evaluating the fault degree of a photovoltaic array based on fault parameter estimation and fine-tuning includes the following steps,

[0051] S1: According to the inverter scanning photovoltaic array I-V characteristics, the measured I-V characteristic curves of the photovoltaic array under different environments and faults are obtained.

[0052] S2: After training the back propagation neural network according to the measured I-V characteristic curves in step S1, the estimated fault parameters are obtained.

[0053] S3: A photovoltaic array I-V characteristic curve simulation model is established, and the simulation I-V characteristic curve is obtained by configuring the estimated fault parameters.

[0054] S4: According to the particle swarm optimization algorithm, the root mean square error between the measured I-V characteristic curve and the simulation I-V characteristic curve is minimized to fine-tune the estimated fault parameters and obtain the optimized fault parameters.

[0055] S5: The fault degree diagnosis quantization accuracy is calculated according to the optimized fault parameters.

[0056] In step S1, the measured I-V characteristic curves of the photovoltaic array under different environments and faults are obtained, in order to provide the training set and test set under comprehensive conditions for the back propagation neural network in S2.

[0057] In step S2, after preprocessing all measured IV characteristic curves, the normalized extracted curve feature points are trained using the designed backpropagation neural network. After training, the network parameters output the fault parameter prediction result corresponding to each measured IV characteristic curve, i.e., the predicted fault parameters. The predicted fault parameters include: the number of short-circuit substrings n. sc Number of shadow occlusion substrings n ps Shadow occlusion transmittance d ps Increased series resistance r s .

[0058] Specifically, since the measured IV characteristic curve contains a lot of noise, it needs to be processed by removing redundancy, smoothing, and filtering out abnormal data, i.e., preprocessing.

[0059] After normalizing the measured IV characteristic curve to eliminate the effects of temperature, irradiance, and uniform interpolation, nine characteristic data points were extracted from the measured IV characteristic curve based on the impact of three typical faults: photovoltaic string short circuit, local shading, and increased series resistance. These include the voltage, current, and power at the maximum power point of the measured IV characteristic curve, i.e., V0. mpp I mpp and P mpp Open circuit voltage V OC The maximum value M of the absolute value of the second derivative is given by:

[0060]

[0061] Where V i I i and dI i , , and , respectively, represent the absolute values ​​of voltage, current, and first derivative at the i-th discrete point on the IV curve, where n is the number of points on the interpolated measured IV characteristic curve.

[0062] Furthermore, the current I at the maximum absolute value of the second derivative on the measured IV characteristic curve is... M and voltage V M It was also extracted as a feature; the last two extracted features were the slope k at the open-circuit voltage. OC and the slope k at the short-circuit current SC , respectively represented as:

[0063]

[0064] After feature extraction from the measured IV characteristic curve, the nine extracted features are combined with the temperature and irradiance corresponding to the measured IV characteristic curve to serve as the eleven input neurons of the backpropagation neural network. This combination is used for training the designed three-layer backpropagation neural network, as shown in the diagram. Figure 2The total number of samples of 11 neurons in different environments and failures is shown. The number of samples in the training set is 70% of the total number of samples. The learning rate is adjusted to 0.6, tanh and Sigmoid are set as the activation functions of the last two layers, and the average cross-entropy C is set as the loss function, which is represented as:

[0065]

[0066] where Y b and Y' b are the label and the back propagation neural network estimated value of the b-th training sample, and m is the total number of training samples for the neural network.

[0067] The preliminary fault diagnosis of the measured I-V characteristic curve is realized on the test set. The estimated fault parameters output by the back propagation neural network include: the number of short-circuit sub-strings n sc , the number of shadow-shielded sub-strings n ps , the shadow-shielded light transmittance d ps , and the increased series resistance r s . There are four output neurons corresponding to three typical faults, i.e., n sc corresponds to photovoltaic string short circuit, n ps and d ps correspond to local shadow shielding, and r s corresponds to increased series resistance. As the first step of fault parameter estimation and fine tuning, after the estimation of the back propagation neural network, the above-mentioned estimated fault parameters are close in value to the actual fault degree, but the quantitative accuracy needs to be improved. Only has the ability to represent the fault type and the fault severity.

[0068] In step S3, a photovoltaic array I-V characteristic curve simulation model is established based on the single diode model based on reverse bias, and simulation I-V characteristic curves under typical faults of photovoltaic string short circuit, local shadow shielding and increased series resistance are obtained by configuring the estimated fault parameters.

[0069] Specifically, the photovoltaic array I-V characteristic curve simulation model is based on the single diode model based on reverse bias and is built under the Python framework. By configuring the temperature and irradiance matrix G, configuring the shadow shielding block and the number of bypass diodes, and configuring the size of the series resistance, the simulation I-V characteristic curve is obtained.

[0070] The mathematical expression of the single diode model based on reverse bias can be expressed as:

[0071]

[0072] where I j and V jThese are the output current and voltage of the photovoltaic module, respectively, and q, k, and T are the electron charges (1.60217662 × 10⁻⁶). -19 C) Boltzmann constant (1.38064852 × 10⁻⁶) -23 JK -1 The temperature of the solar cells, measured in Kelvin (K); I ph I s a, R s R sh These are the five parameters of the mathematical model, representing the photocurrent, the diode's saturation current, the diode's ideality factor, the series resistance, and the parallel resistance, respectively. The latter three, a and R... s R sh Both have a specific functional relationship with irradiance. n' is the correction factor for avalanche breakdown, and V bd is the breakdown voltage, and m' is the breakdown index.

[0073] In addition, the photovoltaic array IV characteristic curve simulation model also includes the configuration of shading blocks, the configuration of substring bypass diode state columns, and the configuration of series resistance. The shading block input matrix B is defined in the following form:

[0074] B = [[X] f ,Y f G f ],…,[X n2 ,Y n2 G n2 (3)

[0075] The shadow block matrix B is a two-dimensional vector representing the number of shadow blocks and the shadow block parameters, where X f and Y f G represents the length and width of the f-th shaded block. f Let n be the transmittance of the f-th shaded block, and n2 be the number of shaded blocks. The defined substring bypass diode state sequence D is used as a variable in the photovoltaic array IV characteristic curve simulation model, expressed in the following form:

[0076]

[0077] Where d e,1 This represents the state of the first bypass diode in the e-th photovoltaic module, n. g This represents the number of modules in the photovoltaic string. Bypass diodes are defined as follows: d=0 indicates normal operation, d=1 indicates a short-circuit fault, and d=2 indicates an open-circuit fault. The output voltage V of the photovoltaic array is... 总 for:

[0078] V 总 =sum(V j ×t j)-I 总 ×r s (4)

[0079] Among them, V j I is the output voltage of the photovoltaic module. 总 It is the output current of the photovoltaic array, r s To increase the series resistance, the sum(x) function sums the matrix.

[0080] Therefore, the photovoltaic array IV characteristic curve simulation model can only simulate the IV characteristic curve under various fault conditions by combining the above formulas (1), (2), (3), and (4). Otherwise, only the IV characteristic curve under normal conditions can be simulated. The fault parameter configuration process is as follows: When the shading block matrix in formula (2) is configured, it indicates that a shading fault has occurred. The length and width (X) of the shading block at the position corresponding to the irradiance matrix G and B are... f ,Y f The irradiance value increase coefficient d ps At this point, there is an occlusion of a component substring, n ps Increasing the quantity by 1 will directly affect the model parameters a and R in equation (1) after the new irradiance is shaded. s R sh When no shading block is configured, it indicates that no shading fault has occurred; when the diode state d = 1 of a certain substring of a photovoltaic module in equation (3) is, it indicates that a short circuit fault has occurred. At this time, there is a short circuit in the corresponding substring of the module, n sc Increase the quantity by 1, and make the output voltage V = 0 in equation (1); when d = 0, it means that no short circuit fault has occurred; then count all fault conditions, and take the number of short-circuited substrings n in the statistical results. sc And the number of obscured substrings n ps Record; calculate the output current I of the photovoltaic module in formula (1) using Newton's descent method combined with variable step size scanning. j and output voltage V j All photovoltaic modules I j and V j The summation is performed, and if a fault occurs where the series resistance increases, equation (4) also needs to be considered to make the increased series resistance r s If there is no fault of increased series resistance during the superposition process, it is not considered. After superposition, the output current I of the photovoltaic array is obtained. 总 With output voltage V 总 Finally, the output current and output voltage of the photovoltaic array are compared in pairs (I... 总 V 总 The set, array pair, is the data point on the simulated IV characteristic curve. The line connecting all the data points can form the simulated IV characteristic curve.

[0081] In step S4, the root mean square error (RMSE) between the measured IV characteristic curve and the simulated IV characteristic curve is minimized using the particle swarm optimization algorithm. The objective function is defined as follows:

[0082]

[0083] Where n is the number of points on the IV curve, I model,i and I meas,i These are the simulated and measured values ​​at the i-th point in the IV curve, respectively. To ensure the RMSE objective function converges quickly and achieves the goal of fine-tuning the predicted fault parameters, therefore... Divided by I in the formula meas,i .

[0084] The initial position P1 of the particle is expressed as follows:

[0085] P1 = [n sc ,n ps ,d ps ,r s ]×rand

[0086] Where rand is a random number between 0.3 and 3, the particle swarm optimization algorithm can optimize within the range of 0.3-3 for the estimated fault parameters. Convergence is considered achieved when the particle position is iteratively updated until the objective function RMSE is less than the set value of 0.0004. The expression for the particle's termination position P2 at this point is as follows:

[0087] P2=[n′ sc ,n′ ps ,d′ ps ,r s ′]

[0088] Where n' sc 、n' ps ,d' ps 、r' s These are the optimized fault parameters. As the second step of the two-step method of fault parameter prediction and fine-tuning, the predicted fault parameters are used as decision variables. By continuously optimizing the objective function RMSE to make it sufficiently small, the optimized fault parameters can be obtained. The optimized fault parameters are numerically closer to the actual fault degree than the predicted fault parameters and can be used to diagnose the fault type and severity of photovoltaic arrays.

[0089] In step S5, taking a specific photovoltaic array diagnosis scenario as an example, steps S1-S4 are used. The experiment is conducted in a 5.28kW photovoltaic power station. The optimized fault parameters are used as the basis for fault diagnosis. The proportion of fault diagnosis results that are the same as the actual fault conditions for 100 measured IV curve samples of each fault is statistically analyzed.

[0090] As can be seen from the confusion matrix, the diagnostic accuracy of this method for the three typical faults in the statistical sample—short circuit in photovoltaic strings, local shading, and increased series resistance—is 100%, 100%, and 94.7%, respectively. The false diagnosis is small and can meet the accuracy requirements for fault diagnosis of photovoltaic arrays.

[0091] Table 1 presents the quantitative statistical results of fault severity under different fault conditions in specific embodiments. The experiment analyzed the quantitative statistical results of fault severity for three typical faults under all-day temperature and irradiance conditions. The calculation formulas for the quantification accuracy of the four fault parameters are as follows:

[0092]

[0093] Where A nsc A nps A dps A rs These are the four fault parameters, namely the number of short-circuited substrings n. sc Number of shadow occlusion substrings n ps Shadow occlusion transmittance d ps Increased series resistance r s The accuracy of fault degree diagnosis quantification, n' sc 、n' ps ,d' ps 、r' s These are the optimized fault parameters corresponding to the four fault parameters. The sgn(x) function outputs 1 and -1 for independent variables greater than zero and less than zero, respectively, and outputs 0 when the independent variable is 0. g is the total number of samples in the test set. It can be seen that for the diagnosis of actual faults, the average diagnostic quantification accuracy of this method can reach 88.51%. This diagnostic quantification accuracy directly reflects the fault diagnosis effect numerically, indicating that it can meet the requirements for evaluating the actual fault severity level and is applicable to diagnosing the fault type and severity of photovoltaic arrays.

[0094] Table 1 Statistical results of fault severity quantification accuracy under different fault conditions

[0095]

[0096] As can be seen from Table 1, this method has a significant effect on the quantitative accuracy of fault degree diagnosis. The larger the value, the better the quantitative accuracy of fault degree diagnosis.

[0097] Based on the above-described preferred embodiments of the present invention, and through the foregoing description, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A two-step method for assessing the fault severity of photovoltaic arrays based on fault parameter prediction and fine-tuning, characterized in that, Includes the following steps, S1: Based on the inverter scanning of the photovoltaic array IV characteristics, obtain the measured IV characteristic curves of the photovoltaic array under different environments and faults; S2: After training the backpropagation neural network based on the measured IV characteristic curve in step S1, the predicted fault parameters are obtained; S3: Establish a simulation model of the photovoltaic array IV characteristic curve, and obtain the simulated IV characteristic curve by configuring the predicted fault parameters; S4: Based on the particle swarm optimization algorithm, minimize the root mean square error between the measured IV characteristic curve and the simulated IV characteristic curve to achieve fine-tuning of the predicted fault parameters and obtain the optimized fault parameters. S5: Calculate the diagnostic accuracy of the fault degree based on the optimized fault parameters; In step S2, the noise contained in the measured IV characteristic curve is preprocessed by removing redundancy, smoothing, and screening outlier data. After normalizing the measured IV characteristic curve to eliminate temperature, irradiance, and uniform interpolation, nine feature data points are extracted from the measured IV characteristic curve based on the impact of three typical faults: photovoltaic string short circuit, local shading, and increased series resistance. These features include the voltage, current, and power at the maximum power point of the measured IV characteristic curve, i.e., V. mpp I mpp and P mpp Open circuit voltage V OC The maximum value M of the absolute value of the second derivative is given by: Where V i I i and dI i , respectively, are the absolute values ​​of voltage, current and first derivative at the i-th discrete point on the IV curve, and n is the number of points on the interpolated measured IV characteristic curve; Furthermore, the current I at the maximum absolute value of the second derivative on the measured IV characteristic curve is... M and voltage V M It was also extracted as a feature; the last two extracted features were the slope k at the open-circuit voltage. OC and the slope k at the short-circuit current SC , respectively represented as: After feature extraction from the measured IV characteristic curves, the nine extracted features are combined with the temperature and irradiance corresponding to the measured IV characteristic curves to serve as 11 input neurons for the designed three-layer backpropagation neural network. The 11 neurons have a total number of samples under different environments and fault conditions, and the number of samples in the training set is 70% of the total number of samples. The learning rate is adjusted to 0.6, tanh and sigmoid are set as the activation functions for the last two layers, and the average cross-entropy C is set as the loss function, expressed as: Where Y b and Y' b Here, represents the label of the b-th training sample and the predicted value from the backpropagation neural network, respectively, where m is the total number of training samples used for the neural network. Preliminary fault diagnosis of the measured IV characteristic curve is implemented on the test set. The predicted fault parameters output by the backpropagation neural network include: the number of short-circuit substrings n. sc Number of shadow occlusion substrings n ps Shadow occlusion transmittance d ps Increased series resistance r s There are four output neurons in total, corresponding to three typical faults, namely, n sc Corresponding to photovoltaic string short circuit, n ps With d ps Corresponding to local shadow occlusion, r s The corresponding series resistance increases; The formulas for calculating the quantization accuracy of the four fault parameters are as follows: Where A nsc A nps A dps A rs These are the four fault parameters, namely the number of short-circuited substrings n. sc Number of shadow occlusion substrings n ps Shadow occlusion transmittance d ps Increased series resistance r s The accuracy of fault degree diagnosis quantification; n' sc 、n' ps ,d' ps 、r' s These are the optimized fault parameters corresponding to the four fault parameters; the function sgn(x) outputs 1 and -1 for independent variables that are greater than zero and less than zero, respectively, and outputs 0 when the independent variable is 0, and g is the total number of samples in the test set.

2. The two-step photovoltaic array fault severity assessment method based on fault parameter prediction and fine-tuning according to claim 1, characterized in that, In step S3, the simulation model of the photovoltaic array IV characteristic curve is based on the reverse-biased single diode model and built in the Python framework. By configuring the temperature and irradiance matrix G, and configuring the number of shading blocks and bypass diodes, the simulated IV characteristic curve is obtained. The mathematical expression for the reverse-biased single-diode model can be expressed as: Among them, I j and V j These represent the output current and voltage of the photovoltaic module, respectively; q, k, and T represent the electron charge, Boltzmann constant, and temperature of the solar cell, respectively, with temperature measured in Kelvin (K). ph I s a, R s R sh These are the five parameters of the mathematical model, representing the photocurrent, diode saturation current, diode ideality factor, series resistance, and parallel resistance, respectively; n' is the avalanche breakdown correction factor, V bd is the breakdown voltage, and m' is the breakdown exponent; In addition, the photovoltaic array IV characteristic curve simulation model also includes the configuration of shading blocks and the state column configuration of substring bypass diodes. The shading block input matrix B is defined as follows: B=[[X f ,Y f ,G f ],…,[X n2 ,Y n2 ,G n2 ]] (2) The shadow block matrix B is a two-dimensional vector representing the number of shadow blocks and the shadow block parameters, where X f and Y f G represents the length and width of the f-th shaded block. f Let n be the transmittance of the f-th shaded block, and n2 be the number of shaded blocks; the defined substring bypass diode state sequence D is used as a variable in the photovoltaic array IV characteristic curve simulation model, and is expressed in the following form: Where d e,1 This represents the state of the first bypass diode in the e-th photovoltaic module, n. g This represents the number of modules in the photovoltaic string; when d=0, the bypass diode is in normal working condition, when d=1, the bypass diode is in short-circuit fault condition, and when d=2, the bypass diode is in open-circuit fault condition. The output voltage V of the photovoltaic array 总 for: V 总 =sum(V j ×t j )-I 总 ×r s (4) Among them, V j I is the output voltage of the photovoltaic module. 总 It is the output current of the photovoltaic array, r s To increase the series resistance, the sum(x) function sums the matrix. Thus, the simulation model of the photovoltaic array IV characteristic curve is formed by combining the above formulas (1)(2)(3)(4).

3. The two-step photovoltaic array fault severity assessment method based on fault parameter prediction and fine-tuning according to claim 2, characterized in that, In step S3, the fault parameter configuration process is as follows: When the shadow occlusion block matrix in equation (2) is configured, it indicates that a shadow occlusion fault has occurred. The length and width (X) of the shadow block at the corresponding position of the irradiance matrix G and B are determined. f ,Y f The irradiance value increase coefficient d ps At this point, there is an occlusion of a component substring, n ps Increasing the quantity by 1 will directly affect the model parameters a and R in equation (1) after the new irradiance is shaded. s R sh When no shading block is configured, it indicates that no shading fault has occurred; when the diode state d = 1 of a certain substring of a photovoltaic module in equation (3) is, it indicates that a short circuit fault has occurred. At this time, there is a short circuit in the corresponding substring of the module, n sc Increase the quantity by 1, and make the output voltage V = 0 in equation (1); when d = 0, it means that no short circuit fault has occurred; then count all fault conditions, and take the number of short-circuited substrings n in the statistical results. sc And the number of obscured substrings n ps Record; calculate the output current I of the photovoltaic module in formula (1) using Newton's descent method combined with variable step size scanning. j and output voltage V j All photovoltaic modules I j and V j The summation is performed, and if a fault occurs where the series resistance increases, equation (4) also needs to be considered to make the increased series resistance r s If there is no fault of increased series resistance during the superposition process, it is not considered. After superposition, the output current I of the photovoltaic array is obtained. 总 With output voltage V 总 Finally, the output current and output voltage of the photovoltaic array are compared in pairs (I... 总 V 总 The set, array pair, is the data point on the simulated IV characteristic curve. The line connecting all the data points can form the simulated IV characteristic curve.

4. The two-step photovoltaic array fault severity assessment method based on fault parameter prediction and fine-tuning according to claim 3, characterized in that, In step S4, the root mean square error (RMSE) between the measured IV characteristic curve and the simulated IV characteristic curve is minimized, i.e., the RMSE is set as the objective function, and the expression is as follows: Where n is the number of points on the IV curve, I model,i and I meas,i These are the simulated and measured values ​​at the i-th point in the IV curve, respectively; the expression for the initial position P1 of the particle is as follows: P1=[n sc ,n ps ,d ps ,r s ]×rand Where rand is a random number between 0.3 and 3, the particle swarm optimization algorithm can optimize within the range of 0.3-3 for the estimated fault parameters; when the particle position is iteratively updated until the objective function RMSE is less than the set value of 0.0004, it is considered to have converged. At this time, the expression for the particle's termination position P2 is as follows: P2=[n′ sc ,n′ ps ,d′ ps ,r s ′] Where n' sc 、n' ps ,d' ps 、r' s These are the optimized fault parameters.

Images