Photovoltaic array fault diagnosis method based on photovoltaic power station current signal output

Abstract

The application discloses a photovoltaic array fault diagnosis method based on photovoltaic power station current signal output, collects current signals, and decomposes the denoised current signals into multiple intrinsic mode functions (IMFs) by using an improved AVMD algorithm after denoising; the improved AVMD algorithm introduces an adaptive learning rate adjustment, and combines an improved Adam algorithm to optimize and update bandwidth parameters and frequency parameters; a Lasso regression model is constructed, L1 regularization is introduced, and IMFs with higher importance are screened out as information carriers for diagnosing photovoltaic system faults; wavelet transform is performed on the selected IMFs, low-frequency components and high-frequency components on different frequency bands are obtained, coarse-grained processing is performed on the low-frequency components, multi-scale discrete entropy (MDE) calculation is performed, time scale factors and curves of normalized discrete entropy values are obtained; a threshold is set to judge the health condition of the photovoltaic system, and fault diagnosis is realized. The improved AVMD algorithm is used for decomposing and denoising the current signals, and the method has good robustness and is easy to implement.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of photovoltaic power generation, and relates to a photovoltaic array fault diagnosis method, in particular to a photovoltaic array fault diagnosis method based on current signal output of a photovoltaic power station. BACKGROUND

[0002] With the increasing influence of industrial pollution on the environment, the world has begun to focus on environmental protection, reduce the use of fossil fuels, and turn to renewable clean energy. Among them, solar energy, as one of the highest-yielding renewable clean energy in clean energy, has been widely spread and used. The potential of solar energy is huge. The region has increased photovoltaic power generation and increased the number of solar power plants. With the increasing proportion of intermittent energy sources such as photovoltaics, it undoubtedly increases the risk of dramatic changes in energy production. Fault diagnosis can timely find potential problems of photovoltaic systems, so as to take measures to repair and improve the operation efficiency and power generation of the system.

[0003] Some traditional photovoltaic degradation fault detection methods need to use electroluminescence, infrared thermal imaging, ultraviolet fluorescence and I-V tracer methods, which have high cost, so many manufacturers will only use these methods when the power generation efficiency appears a sharp decline. The DETECT (online diagnosis of electronic system health) research project is carried out in this background, and the purpose is to develop a low-cost FDD (photovoltaic system diagnosis) without additional sensors or expensive equipment. The current application of FDD method is output signal analysis (OSA), which assumes that any fault will affect the output of the photovoltaic system, and the output current or voltage may be affected by numerical decline or signal dynamic change. By fitting analysis of perturbation, the pattern curve of each fault is extracted. However, the diagnosis based on OSA is still sensitive to noise, so how to reduce the influence of noise on detection ability is a difficult problem that needs to be solved in output signal analysis. SUMMARY

[0004] In order to determine the specific type of photovoltaic fault, the application provides a photovoltaic system fault diagnosis method based on an improved adaptive variational mode decomposition algorithm and multi-scale discrete entropy, which can perform online diagnosis based on the current signal output of the photovoltaic power station, and realize rapid diagnosis of the type of photovoltaic fault in the case of low cost and easy operation.

[0005] To achieve the above purpose, the application provides the following technical scheme: a photovoltaic array fault diagnosis method based on current signal output of a photovoltaic power station, comprising the following steps:

[0006] Step 1. Connect the operation and management system of the photovoltaic power station to the current signal collection of the inverter of the power grid, collect the current signal of the shadow-shielded photovoltaic component and the current signal of the normal weather, use Kalman filtering to denoise the input current signal, and use an improved AVMD algorithm to decompose the denoised current signal into multiple intrinsic mode functions (IMFs); the improved AVMD algorithm introduces an adaptive learning rate adjustment, and combines an improved Adam algorithm to optimize and update the bandwidth parameter and the frequency parameter;

[0007] Step 2. Apply the Lasso regression model to use cross-validation to select the optimal regularization parameter to balance the bias and variance of the model, use the training data to fit the Lasso regression model, calculate the regression coefficients of the IMFs, evaluate the importance of the IMFs in fault diagnosis, and then determine the IMFs with higher importance as information carriers for diagnosing photovoltaic system faults;

[0008] Step 3. Perform wavelet transform on the selected IMFs to decompose the original time series of the output current of the photovoltaic power station over time into low-frequency components and high-frequency components in different frequency bands;

[0009] Step 4. Perform coarse-grained processing on the low-frequency components, and then perform multi-scale discrete entropy (MDE) calculation on the coarse-grained sub-sequences of different time scales decomposed;

[0010] Step 5. Compare the normalized discrete entropy value curves of the fault-free sunny day, fault-free overcast day, and fault sunny day, observe the differences in the discrete entropy values at different time scales, find the discrete entropy values that are significantly different between the shadow-shielded fault and the fault-free photovoltaic system, and set them as threshold values to determine whether the photovoltaic system is in a normal state, a shadow-shielded fault state, or a shadow-shielded fault occurrence or end state, and realize fault diagnosis of the photovoltaic system.

[0011] Further, the specific method of decomposing the current signal output by the inverter of the photovoltaic power station into multiple intrinsic mode functions in step 1 is as follows:

[0012] For the current signal denoised by Kalman filtering, the goal of the improved AVMD is to decompose the current signal into K modal functions u k (t) and corresponding center frequencies ω k The goal of the decomposition is to minimize the following variational formula with bandwidth constraints:

[0013]

[0014] where, denotes the derivative with respect to time, which is used to represent the change in bandwidth, u k (t) is the kth modal function of the signal, and ω kIt is the center frequency of each modality function, which is dynamically updated with adaptive learning rate and improved Adam optimization;

[0015] The improved AVMD introduces the Lagrange multiplier λ(t), which allows the sum of the modal components to reconstruct the original signal, making the reconstructed signal as close as possible to the original signal and ensuring signal integrity. The defined Lagrange function is:

[0016]

[0017] Where α is the adaptive learning rate, used to dynamically control the optimization step size of the bandwidth, λ(t) is the Lagrange multiplier, which ensures the reconstruction accuracy of the signal, f(t) is the input original current signal of the photovoltaic power station, and K is the number of modal components, representing the number of intrinsic mode functions obtained when decomposing the signal;

[0018] An improved Adam optimization algorithm is introduced to optimize u. k (t) and ω k The dynamic update first involves initializing the parameters of the Adam optimization algorithm, setting the initial learning rate and two exponential decay rate parameters β1 and β2, initializing the first moment m0 ​​= 0 and the second moment v0 = 0, and setting a small value ∈ to avoid division by zero error;

[0019] The computational steps of the improved Adam optimization algorithm are as follows:

[0020] a. Calculate the gradient

[0021]

[0022] b. Update bandwidth and frequency parameters

[0023] For the bandwidth parameter μ k :

[0024]

[0025] For the frequency parameter ω k :

[0026]

[0027] Where, m t and v t These are adaptive estimates of the first and second moments, respectively, where α is the learning rate for bandwidth updates. ω The learning rate is updated at a specific frequency, and all parameters are adaptively adjusted.

[0028] An adaptive learning rate adjustment is added to the improved Adam optimization algorithm, allowing the learning rate to automatically adjust according to gradient changes. The formula for calculating the adaptive learning rate adjustment is as follows:

[0029]

[0030] Where δ is the adjustment factor, which is 10. -6 ,|g t | represents the absolute value of the current gradient. After introducing adaptive learning rate adjustment, the learning rate is appropriately reduced when the gradient is large to avoid drastic updates, and appropriately increased when the gradient is small to speed up convergence.

[0031] Using an improved Adam optimization algorithm and adaptive learning rate adjustment, for u k (t) and ω k The process involves iterative updates to minimize the objective function. In each update round, the bandwidth change is checked to ensure it meets the preset convergence condition. If it does, the iteration ends. After multiple iterations, K convergent eigenmode functions are obtained, denoted as:

[0032]

[0033] Furthermore, the specific method for using the Lasso regression model in step 2 to screen out IMFs with higher importance as information carriers is as follows:

[0034] The decomposed IMFs are used to form a feature matrix X, where each column represents an IMF and each row represents a time node. The corresponding target variable y is set, which is the label of the fault state.

[0035] The goal of Lasso regression is to minimize the following loss function:

[0036]

[0037] Where n is the number of samples, y i The target variable is X. ij These are elements of the characteristic matrix, where β0 is the intercept and β... j λ is the regression coefficient of the target variable by the IMF, and λ is the regularization parameter, which controls the strength of the constraint on the regression coefficient. The larger the value, the stronger the constraint.

[0038] By using a regularization parameter λ to balance the model's bias and variance, and fitting the Lasso regression model with the training data, the regression coefficients β of the IMF are obtained. j The training data consists of the IMF feature matrix X and the target variable y. The number of IMFs selected is controlled by adjusting the regularization parameter. The regression coefficient β of each IMF is compared. j The size of the IMF is used to assess its importance in fault diagnosis. A threshold is set to select IMFs with large non-zero coefficients as information carriers for fault diagnosis of photovoltaic systems.

[0039] Furthermore, in step 3, the selected IMFs are decomposed into low-frequency and high-frequency components in different frequency bands using wavelet transform. The specific method is as follows:

[0040] The selected IMF components are decomposed using Discrete Wavelet Transform (DWT). Each wavelet decomposition level includes two filtering steps: low-pass filtering and high-pass filtering. In the j-th level decomposition, the low-frequency coefficients are represented as:

[0041]

[0042] High-frequency coefficients are represented as:

[0043]

[0044] Among them, A j D represents the low-frequency component of the j-th layer. j Let h and g represent the high-frequency components of the j-th layer, respectively, and let h and g represent the low-pass and high-pass filter coefficients of the wavelet. After J-level wavelet decomposition, the multi-scale subsequence A is obtained. J And D1, D2, ..., D J Among them, A J This is the lowest frequency approximation signal, representing the main trend of the signal, and is suitable for further coarse-grained processing; D J Let be the detail signal of the j-th layer, representing the high-frequency components of that layer. It reflects local fluctuations and abrupt changes in the current signal, and can identify short-term variations and noise. When decomposed to the L-th layer, the original signal is decomposed into low-frequency coefficients of the L-th layer and high-frequency coefficients at multiple scales, expressed as:

[0045]

[0046] Among them, A L To decompose to the low-frequency components at layer L, D l This refers to the high-frequency components of the l-th layer when decomposed to the L-th layer.

[0047] Furthermore, the low-frequency components obtained after wavelet transform are coarsened, and then the discrete entropy values ​​are calculated for the coarsened subsequences at different time scales. The specific method is as follows:

[0048] First, considering the different time scales of the various phenomena involved in the output current signal, the time scale factor is determined, and the low-frequency components obtained after wavelet transform are coarsened according to the scale factor ratio. Next, the low-frequency components obtained after wavelet transform are grouped according to the time scale factor ratio, and the data within each group are averaged. Finally, a new coarse-grained time series is obtained, with each series having a length of N / τ. cs Establish a new time series based on the scale:

[0049]

[0050] Where, τ sc It is the time scale factor that determines how many subsequences the low-frequency components are coarsened into. For each time scale factor τ sc The data in the low-frequency components are grouped, and the average value of each group is calculated to obtain new data points y. j (τsc) It represents the original time series at τ sc The coarsening results on the time scale show that the length of the coarse-grained time series decreases as the time scale ratio increases.

[0051] Furthermore, the coarse-grained subsequences at different time scales obtained from the decomposition are subjected to multi-scale discrete entropy (MDE) calculation. The discrete entropy (MDE) is defined as follows:

[0052]

[0053] Where, p i Let represent the probability of the system in unit i, and n represent the characteristic dimension of the signal.

[0054] Furthermore, in step 5, the trend changes of discrete entropy curves at different time scales are observed, and the health status of the photovoltaic system is judged based on the shape of the curves and thresholds. The specific method for diagnosing whether the photovoltaic system has a shading fault is as follows:

[0055] First, observe the curve shape to make a preliminary judgment on whether there is a fault in the photovoltaic system. When the photovoltaic system is in sunny weather, the MDE curve usually presents a bell shape; when the photovoltaic system is in cloudy weather, the MDE curve usually presents a flat shape; when the photovoltaic system has a fault, the MDE curve will shift from the bell-shaped curve in the normal state, moving upward or downward. When a fault is observed in the photovoltaic system, compare the discrete entropy curves of a fault-free sunny day, a fault-free cloudy day, and a faulty sunny day to obtain the MDE value that differs significantly from the fault-free sunny day and fault-free cloudy day, and set it as a threshold. The threshold is used to determine whether there is a shading fault in the photovoltaic system, and whether the shading fault has started or ended, thereby achieving the effect of fault diagnosis.

[0056] Beneficial effects:

[0057] (1) This invention overcomes the limitations of the adaptive variational mode decomposition algorithm by improving the adaptive variational mode decomposition algorithm. The improved AVMD overcomes the limitation of setting fixed initial values ​​for learning rate, bandwidth, and frequency parameters in AVMD by introducing adaptive learning rate adjustment. Combined with the improved Adam algorithm, the bandwidth and frequency parameters are optimized and updated, making the decomposition process more flexible and better adaptable to signal characteristics. It has higher computational efficiency and convergence speed when processing non-stationary and nonlinear signals, and can more accurately decompose current signals into intrinsic mode functions. Furthermore, leveraging the advantages of entropy, multi-scale entropy is applied to the diagnostic field. Multi-scale methods can evaluate the entropy of different phenomena at different time scales. Using the above techniques to detect faults on photovoltaic equipment, the AVMD algorithm with Adam optimization and adaptive learning rate is used to extract current signals carrying photovoltaic system state and fault characteristic information. Then, multi-scale discrete entropy is used to diagnose photovoltaic system faults at different time scales. This method has good robustness and is easy to implement.

[0058] (2) This invention utilizes the current data collected by the inverter and adopts an adaptive variational mode decomposition algorithm with Adam optimization algorithm and adaptive learning rate to decompose the current signal into intrinsic mode functions (IMFs). It has high accuracy, does not depend on weather changes, and has good robustness.

[0059] (3) This invention coarse-grained the low-frequency components obtained after wavelet transform. Coarse-graining is a signal processing technique that creates new time series, ensures that the length of each time series meets the conditions for calculating discrete entropy, improves the reliability of entropy values, and also improves the efficiency of calculation.

[0060] (4) MDE can analyze the complexity of signals at different time scales and more comprehensively assess the health status of photovoltaic systems. It can reduce the impact of noise by coarsening the signals. Compared with other entropy calculation methods, discrete entropy has higher calculation efficiency and is suitable for real-time monitoring.

[0061] (5) Combining the improved adaptive variational mode decomposition algorithm and the multi-scale discrete entropy algorithm to diagnose photovoltaic system faults can distinguish the characteristics of different types of faults at different time scales, thereby improving the accuracy and efficiency of fault detection. Attached Figure Description

[0062] Figure 1 This is a flowchart illustrating the process of combining the improved adaptive variational mode decomposition algorithm with multi-scale discrete entropy in this invention.

[0063] Figure 2These are the time scale factor and normalized discrete entropy curves of the current signal output by the photovoltaic array under normal and fault conditions in this invention after processing with improved AVMD-MDE.

[0064] Figure 3 This is a flowchart illustrating the adaptive variational mode decomposition based on improved current signals in this invention.

[0065] Figure 4 This is a flowchart illustrating the process of calculating discrete entropy in this invention. Detailed Implementation

[0066] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0067] The photovoltaic array fault diagnosis method based on the current signal output of a photovoltaic power station disclosed in this invention includes the following steps:

[0068] Step 1: Collect the current signal transmitted from the inverter on the photovoltaic power station. The inverter converts the DC power generated by the photovoltaic array into AC power for grid connection. The photovoltaic power station consists of two groups of 12 modules, each composed of 72 solar cells, with 18 cells connected in series in each group, and equipped with a bypass diode. The solar panels are tilted at 26° to the ground, facing north. Two sets of lines are tested: line one is used to collect current data during shading faults, and line two is used to collect data during normal weather. By comparing the differences between the two sets of data curves, a fault curve graph is obtained. The calculated power of line one is 2.04 kWp, and the calculated power of line two is 2.09 kWp. The characteristics of the photovoltaic modules are shown in Table 1. To implement data transmission conversion and security, a data pipeline consisting of a gateway, network services, a database for storing data, and an application server is implemented. Electrical data measurements are acquired by the operation and management system of the photovoltaic power station connected to the grid and coupled to the inverter. The frequency of data transmission is selected by telemetry control in the software. All radiation and meteorological sensors are connected to the data logger for data collection, and the data is then transmitted to the server via TCP / IP. To enable real-time data utilization, the data is stored at a sampling frequency of 1 Hz. To ensure real-time data availability, the software is written in C, monitors data files from various sources, and injects the data into the database in time-series format. Therefore, an end-to-end data pipeline relying on open-source technologies was developed, where data from sensors and inverters is periodically sent, collected, and stored in a centralized database system.

[0069] Table 1

[0070]

[0071] Step 2: Denoise the inverter output current signal using Kalman filtering, employing an improved adaptive variational mode decomposition algorithm (improved AVMD algorithm). This involves replacing the traditional optimization algorithm with an improved Adam optimization algorithm to decompose the denoised current signal into multiple intrinsic mode functions (IMFs), and iteratively updating the bandwidth parameter during the decomposition process. Specific steps include:

[0072] Inverters connected to the grid through the photovoltaic power plant's operation and management system collect current data, including current signals from shaded photovoltaic modules. Kalman denoising is applied to the inverter's output current data to generate a denoised current signal. An improved AVMD algorithm is then used to decompose the denoised current signal. The improved AVMD algorithm is an adaptive variational mode decomposition algorithm for signal processing that can decompose complex signals into several eigenmode functions. The improved AVMD overcomes the limitations of AVMD in setting fixed initial values ​​for learning rate, bandwidth, and frequency parameters by introducing adaptive learning rate adjustment and combining it with an improved Adam algorithm to optimize and update the bandwidth and frequency parameters, making the decomposition process more flexible.

[0073] Kalman filtering requires setting initial states and initial values ​​for the noise covariance matrix, defining process noise covariance and measurement noise covariance to control the filter strength and signal smoothness. A state-space model of the Kalman filter is constructed. Based on the current state and the noise model, the state and error covariance at the next time step are predicted. The predicted state is then updated using the current measurement value to reduce prediction error. In the denoising stage, the current signal is continuously predicted and updated, iterating through Kalman filtering to generate a denoised current signal sequence.

[0074] For the current signal after Kalman filtering and denoising, the goal of improving AVMD is to decompose it into K mode functions u k (t) and the corresponding center frequency ω k The objective of the decomposition is to minimize the variational formula for the following bandwidth constraint:

[0075]

[0076] in, The time derivative, u, is used to represent the change in bandwidth. k (t) is the Kth mode function of the signal, ω k It is the center frequency of each modality function, which is dynamically updated with adaptive learning rate and improved Adam optimization.

[0077] The improved AVMD introduces the Lagrange multiplier λ(t), which allows the sum of the modal components to reconstruct the original signal, making the reconstructed signal as close as possible to the original signal and ensuring signal integrity. The defined Lagrange function is:

[0078]

[0079] Where α is the adaptive learning rate, used to dynamically control the optimization step size of the bandwidth, λ(t) is the Lagrange multiplier, which ensures the reconstruction accuracy of the signal, f(t) is the input original current signal of the photovoltaic power station, and K is the number of modal components, representing the number of intrinsic mode functions obtained when decomposing the signal.

[0080] To solve the above optimization problem, an improved Adam optimization algorithm is introduced to optimize u. k (t) and ω k Dynamic updates. First, the Adam optimization algorithm needs to initialize parameters, setting the initial learning rate and two exponential decay rate parameters β1 and β2, initializing the first moment (momentum) m0 = 0 and the second moment (weighted non-central variance) v0 = 0, and setting a small value ∈ to avoid division by zero error.

[0081] The core computational steps of the improved Adam optimization algorithm are as follows:

[0082] a. Calculate the gradient

[0083]

[0084] b. Update bandwidth and frequency parameters

[0085] For the bandwidth parameter μ k :

[0086]

[0087] For the frequency parameter ω k :

[0088]

[0089] Where, m t and v t These are adaptive estimates of the first and second moments, respectively, where α is the learning rate for bandwidth updates. ω The learning rate is updated at a specific frequency, and all parameters are adaptively adjusted.

[0090] An adaptive learning rate adjustment is incorporated into the improved Adam optimization algorithm, allowing the learning rate to automatically adjust based on gradient changes. The formula for calculating the adaptive learning rate adjustment is:

[0091]

[0092] Where δ is the adjustment factor, which is usually taken as 10. -6 ,|g t | represents the absolute value of the current gradient. Introducing adaptive learning rate adjustment allows the learning rate to be appropriately reduced when the gradient is large to avoid drastic updates, and also appropriately increased when the gradient is small to accelerate convergence.

[0093] Using an improved Adam optimization algorithm and adaptive learning rate adjustment, for u k (t) and ω k The process involves iterative updates to minimize the objective function. In each update round, the bandwidth change is checked to ensure it meets the preset convergence condition. If it does, the iteration ends. After multiple iterations, K convergent eigenmode functions are obtained, denoted as:

[0094]

[0095] The optimization process of the improved adaptive variational mode decomposition mainly involves the following steps:

[0096] Step a: Initialize the mode function for each mode at iteration number 1. and center frequency ω k and Lagrange multipliers The spectrum and center frequency of each K mode.

[0097] Step b: In the improved Adam optimization algorithm, momentum is used to update the mode function:

[0098]

[0099] Step c: Update the Lagrange multipliers:

[0100]

[0101] Step d: Update the center frequency

[0102]

[0103] Step e: Repeat steps b to e to determine whether the change in bandwidth meets the convergence condition. If the convergence condition is met, stop the iteration.

[0104] The improved AVMD algorithm no longer directly relies on the alternating direction multiplier method to update frequency and bandwidth parameters. Instead, it utilizes an improved Adam optimization algorithm and an adaptive learning rate to gradually adjust u. k (t) and center frequency ω kThe algorithm first initializes the center frequency, mode function, and Lagrange multipliers for each mode. Then, in each iteration, it updates the mode function, Lagrange multipliers, and center frequency sequentially. An improved Adam optimization algorithm is used to update the mode function, the Lagrange multipliers are updated to maintain bandwidth constraints, and the center frequency is updated to ensure it accurately reflects the main components of the signal. Finally, it checks if the bandwidth change satisfies the convergence condition. If convergence is achieved, the iteration stops; otherwise, it returns to the previous iteration step. Compared to the original adaptive variational mode decomposition algorithm, the improved Adam optimization algorithm and adaptive learning rate adjustment method allow for faster and more stable adjustment of mode and frequency parameters in each iteration, decomposing the signal into several eigenmode functions with specific bandwidths and center frequencies.

[0105] The improved AVMD algorithm requires initialization of some parameters at the beginning, setting the initial mode function to 0 and the initial center frequency to 0.06. During the iteration process, the improved Adam optimization algorithm and adaptive learning rate are used to gradually adjust u. k (t) and center frequency ω k The value of is determined in each iteration. The mode function, Lagrange multipliers, and center frequency are updated sequentially. The improved Adam optimization algorithm is used to update the mode function, the Lagrange multipliers to maintain bandwidth constraints, and the center frequency to ensure it accurately reflects the main components of the signal. Finally, it is determined whether the bandwidth change satisfies the convergence condition. If convergence is achieved, the iteration stops; otherwise, the iteration returns to the previous step. The AVMD algorithm, which incorporates the improved Adam optimization algorithm and adaptive learning rate, is more efficient than the original AVMD algorithm when processing non-stationary, multimodal signals, and has greater advantages in practical applications.

[0106] Step 3: Decompose the current signal into multiple intrinsic mode functions (IMFs), and select the more important IMFs as information carriers based on the Lasso regression model. Specific steps include:

[0107] The improved AVMD algorithm decomposes the IMFs into a feature matrix X, where each column represents an IMF and each row represents a time node. The corresponding target variable y is set, which is the label of the fault state.

[0108] Lasso regression is a linear regression model that, with the introduction of L1 regularization, can effectively perform feature selection. The goal of Lasso regression is to minimize the following loss function:

[0109]

[0110] Where n is the number of samples, yi The target variable is X. ij These are elements of the characteristic matrix, where β0 is the intercept and β... j λ is the regression coefficient of the target variable by the IMF, and λ is the regularization parameter, which controls the strength of the constraint on the regression coefficient. The larger the value, the stronger the constraint.

[0111] By using a regularization parameter λ to balance the model's bias and variance, and fitting a Lasso regression model with the training data (IMF feature matrix and target variable), the regression coefficients β of the IMF are obtained. j The number of IMFs selected can be controlled by adjusting the regularization parameter. The regression coefficient β of each IMF is compared. j The magnitude of the coefficient is used to assess the importance of the IMF in fault diagnosis; generally, a larger absolute value indicates a greater impact of the IMF on the target variable. IMFs containing a large amount of information are selected as information carriers for diagnosis. A threshold is set to select IMFs with large non-zero coefficients as information carriers for fault diagnosis of photovoltaic systems.

[0112] Step 4: Perform wavelet transform on the selected IMFs to reduce the complexity of subsequent discrete entropy calculation. Wavelet transform decomposes the original time series of photovoltaic power plant output current change over time into low-frequency and high-frequency components in different frequency bands.

[0113] Wavelet transform is a time-frequency localization analysis method that decomposes a signal into different time and frequency components. The selected IMF components are decomposed using Discrete Wavelet Transform (DWT), which is obtained through cascaded filtering operations. Each level of wavelet decomposition includes two filtering steps: low-pass filtering and high-pass filtering. In the j-th level decomposition, the low-frequency coefficients can be expressed as:

[0114]

[0115] High-frequency coefficients can be expressed as:

[0116]

[0117] Among them, A j D represents the low-frequency component of the j-th layer. j Let represent the high-frequency components of the j-th level, and h and g represent the low-pass and high-pass filter coefficients of the wavelet, respectively. After J-level wavelet decomposition, the multi-scale subsequence A is obtained. J And D1, D2, ..., D J Among them, A J This is the lowest frequency approximation signal, representing the main trend of the signal, and is suitable for further coarse-grained processing. D JThe j-th layer represents the high-frequency components of that layer, reflecting local fluctuations and abrupt changes in the current signal, and can identify short-term variations and noise. These subsequences provide multi-scale information from high to low frequencies, helping to capture the signal's changing characteristics at different time scales. When decomposed to the L-th layer, the original signal is decomposed into low-frequency coefficients of layer L and high-frequency coefficients at multiple scales, expressed as:

[0118]

[0119] Step 5: To further capture the variation characteristics of each subsequence at different time scales, the low-frequency components obtained after wavelet transform are coarsened. Then, multi-scale discrete entropy (MDE) calculation is performed on the coarsened subsequences at different time scales. Considering the different time scales involved in the current signal, discrete entropy is calculated for the coarsened subsequences at different time scales to obtain discrete entropy values. The calculated discrete entropy values ​​are then normalized. The parameter selection and results are shown below:

[0120] Multi-scale methods are a common signal analysis technique that allows for the observation and analysis of signal characteristics at different time scales. In this patent, the multi-scale method considers the different time scales of various phenomena involved in the output current signal, and coarsely processes the low-frequency components obtained after wavelet transform in terms of time scale, assigning each component a length of N / τ. cs A new time series is established at the scale.

[0121]

[0122] Where, τ sc It is the time scale factor, which determines how many subsequences the low-frequency components are coarsened into. For each time scale factor τ sc The data in the low-frequency components are grouped, and the average value of each group is calculated to obtain new data points y. j (τsc) It represents the original time series at τ sc Coarse-grained results over time.

[0123] Multiscale discrete entropy builds upon discrete entropy by incorporating multiscale methods, considering different time scales of the phenomena involved in the output signal. Entropy is a statistical algorithm that quantifies the probability of adjacent points in a time series falling within a predetermined range, thus measuring the complexity and richness of information in a time series. The definition of entropy is as follows:

[0124]

[0125] Where, p iis the probability that the system is located in unit i, and n represents the characteristic dimension of the signal.

[0126] Discrete entropy algorithm is a newly developed algorithm, and its flowchart is as follows: Figure 4 As shown, the specific steps of the algorithm are as follows:

[0127] Step 1: Category Mapping In this work, the original signal is mapped to a normal cumulative distribution function (NCDF): x = {x1, x2, ..., x...} N}→y={y1,y2,...,y N The value of y ranges from 0 to 1. To assign each value of yi an integer value from 1 to c, an algorithm is used, where c is the number of categories.

[0128]

[0129] Step 2: Create the embedding vector and scatter pattern. The created embedding vector has a length of m and takes into account the time delay d.

[0130]

[0131] Mapping to dispersion mode in The number of possible patterns is equal to c m .

[0132] Step 3: Calculate the relative frequency. The relative frequency of each pattern is defined as the number of times each pattern occurs divided by the total number of embedding vectors.

[0133]

[0134] Step 4: Entropy Calculation Based on the Definition of Entropy

[0135]

[0136] In this invention, to facilitate numerical comparison, the discrete entropy values ​​are normalized using the following formula:

[0137]

[0138] The selection of DE parameters needs to be determined, and the following suggestions must be considered:

[0139] (1) d should be set to 1 to avoid aliasing;

[0140] (2)c m The length must be less than the length of the time series;

[0141] (3) c must be greater than 1 to have different modes. If c is too small, very different values ​​may be assigned to the same mode; if c is too large, DE will be sensitive to noise.

[0142] (4) c can be chosen from 4 to 8;

[0143] (5) m must be carefully selected; when the value of m is too small, dynamic changes cannot be detected; when the value of m is too large, small dynamic changes cannot be detected.

[0144] (6) The larger the values ​​of m and c, the longer the calculation time;

[0145] Furthermore, considering a multi-scale approach, the original time series of the photovoltaic power plant's output current variation over time is coarsened, resulting in a single time series of N / τ. cs A new time series is established at the scale.

[0146] Then, the discrete entropy of each coarse-grained time series is calculated. The length of the coarse-grained time series decreases as the time scale ratio increases. According to relevant literature, the time series length must be at least 10. m By 20 m Only through this process can a reliable entropy value be obtained.

[0147] The time delay d, the number of classes c, and the length of the embedding vector are selected according to previous recommendations. Maximum scaling factor τ s The definitions of c and the time series length N are the result of the following considerations:

[0148] 1. The time series size should be short enough to minimize the necessary data volume and computation time, and maximize monitoring efficiency. The window width should not exceed 30 minutes.

[0149] 2. The photovoltaic power station data acquisition system studied had a very low frequency and no external measurement equipment was added. Approximately 300 data points were collected within 30 minutes.

[0150] 3. Dispersion entropy requires a minimum number of points to ensure the entropy value is valid. Considering the length of the coarse-grained time series, N / τ s c must be significantly higher than c m =6 2 =36.

[0151] Table 4 lists the parameters of all multiscale dispersion entropy.

[0152]

[0153] Table 4

[0154] Step 6: Calculate the multi-scale discrete entropy for each IMF based on the data in Table 4, select the IMF map with the most information, and analyze it:

[0155] Figure 2 'a' represents the experimental dataset. Data from sunny days without faults is represented by a black dashed line, data from cloudy days without faults is represented by a black solid line, and data from faulty days is represented by a black dotted line. Curves with triangles represent fault occurrence points; downward curves indicate the fault's initiation point, and upward curves indicate the fault's endpoint. For example... Figure 2 As shown in b, the IMF5 MDE curve exhibits three distinct patterns depending on the experimental conditions. Sunny-day fault data and fault-free data produce bell-shaped curves, while fault data produces a shifted curve resembling the sunny-day data. In fact, Figure 2 The data shows that the output current varies significantly around 10:00 AM, which explains the results of the improved AVMD-MDE. To evaluate the performance of the improved AVMD-MDE in terms of shading faults, two types of data were collected: one for cloudy days and one for sunny days, and the data were analyzed using the improved AVMD-MDE algorithm.

[0156] Figure 2 c shows two time periods: cloudy and sunny. The period from 8:00 to 10:00 is sunny, and the later period is cloudy. Figure 2 Figure d shows the MDE analysis results, presenting two different MDE curve patterns: a straight line corresponding to cloudy days and a bell-shaped curve corresponding to sunny days. One of the curves is a misclustered cluster, corresponding to the tenth time period. In fact, this time period corresponds to sustained low irradiance, which led to low current generation and low variability, thus explaining the trend of the curve. Figure 2 e describes a sunny day, a day when a malfunction occurs, and a cloudy day after a malfunction. Figure 2 f shows that the sunny day curve and the cloudy day curve have different trends. In the event of a fault, the MDE curve can be observed to transform from the healthy sunny day curve, such as... Figure 2 As shown in b. However, compared to the cloudy day curve, the fault start and end curves have the same trend and values.

[0157] As shown in Table 5, a total of 36 datasets were collected. These preliminary results demonstrate the relationship between MDE values ​​and system health. Based on the previously shown curves, two thresholds can be retained to precisely define the health of the photovoltaic system. Figure 2As can be seen from f, at a scale factor of 1, a DE value below the critical value of 0.6 indicates overcast skies or the start or end of a fault. Although the DE values ​​for sunny days are very similar at a scale factor of 1, regardless of whether the data is faulty or fault-free, the DE values ​​differ significantly at scale factors of 4 and 5. In fact, at a scale factor of 4, a DE value above 0.9 indicates a faulty state. In this work, scale factors of 1 and 4 were chosen with thresholds of 0.6 and 0.9, respectively, to determine the health status of photovoltaic systems. The overall accuracy of the proposed method was evaluated to be 94.30%. In this invention, health status is determined through a simple rule-based diagnosis, and the selected scale and relevant thresholds are set based on observations.

[0158] Table 5

[0159]

[0160]

[0161] In summary, this invention develops an innovative online fault detection and diagnosis method. This method relies on an improved adaptive variational mode decomposition algorithm (improved AVMD) and multi-scale discrete entropy (MDE). Selecting an entropy value at a specific scale serves as a quantitative indicator for determining the system's health status. Diagnosis based on the improved AVMD-MDE requires no additional equipment and can be easily implemented online. Combining improved adaptive variational mode decomposition and multi-scale discrete entropy shows great promise and can be applied to more complex fault diagnoses in the future. A flowchart based on improved adaptive variational mode decomposition and multi-scale discrete entropy is shown below. Figure 1 As shown.

[0162] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for fault diagnosis of photovoltaic arrays based on the current signal output of a photovoltaic power station, characterized in that, Includes the following steps: Step 1. Connect the photovoltaic power station's operation and management system to the inverter connected to the grid to collect current signals. Collect current signals from the photovoltaic modules under shade and from normal weather conditions. Use Kalman filtering to denoise the input current signals. Use an improved AVMD algorithm to decompose the denoised current signals into multiple intrinsic mode functions (IMFs). The improved AVMD algorithm introduces adaptive learning rate adjustment and combines it with an improved Adam algorithm to optimize and update bandwidth and frequency parameters. An improved Adam optimization algorithm is introduced for optimization. and The dynamic update first involves initializing the Adam optimization algorithm parameters, setting the initial learning rate and two exponential decay rate parameters. and Initialize first-order moments =0 and second moment =0, sets a small value. To avoid division by zero errors; The computational steps of the improved Adam optimization algorithm are as follows: a. Calculate the gradient: （3）； b. Update bandwidth and frequency parameters: For bandwidth parameters : （4）； For frequency parameters : （5）； in, and Adaptive estimates of the first and second moments, respectively. The learning rate is updated for bandwidth. The learning rate is updated at a specific frequency, and all parameters are adaptively adjusted. An adaptive learning rate adjustment is added to the improved Adam optimization algorithm, allowing the learning rate to automatically adjust according to gradient changes. The formula for calculating the adaptive learning rate adjustment is as follows: （6）； in, It is a regulating factor, set to 10. -6 , The absolute value of the current gradient is used. After introducing adaptive learning rate adjustment, the learning rate is appropriately reduced when the gradient is large to avoid drastic updates, and appropriately increased when the gradient is small to speed up the convergence speed. Using an improved Adam optimization algorithm and adaptive learning rate adjustment, for and The process involves iterative updates to minimize the objective function. In each update round, the bandwidth change is checked to ensure it meets the preset convergence condition. If it does, the iteration ends. After multiple iterations, K convergent eigenmode functions are obtained, denoted as: （7）； Step 2. Apply the Lasso regression model and use cross-validation to select the optimal regularization parameter to balance the bias and variance of the model. Fit the Lasso regression model using the training data, calculate the regression coefficient of the IMF, evaluate the importance of the IMF in fault diagnosis, and then identify the more important IMFs as information carriers for diagnosing photovoltaic system faults. Step 3. Perform wavelet transform on the selected IMF to decompose the original time series of the photovoltaic power station output current change over time into low-frequency and high-frequency components in different frequency bands. Step 4. Coarse-grained processing is performed on the low-frequency components, and then multi-scale discrete entropy (MDE) calculation is performed on the coarse-grained subsequences of different time scales decomposed. The calculated discrete entropy values ​​are then normalized. Step 5. Compare the normalized discrete entropy curves of a fault-free sunny day, a fault-free cloudy day, and a faulty sunny day. Observe the differences in discrete entropy values ​​at different time scales, find the discrete entropy values ​​that differ significantly between the shading fault and fault-free photovoltaic systems, and set them as thresholds to determine whether the photovoltaic system is in a normal state, a shading fault state, or a state where the shading fault has occurred or ended, thus realizing fault diagnosis of the photovoltaic system.

2. The photovoltaic array fault diagnosis method based on photovoltaic power station current signal output according to claim 1, characterized in that: The specific method for decomposing the current signal output by the photovoltaic power station inverter into multiple intrinsic mode functions using the improved AVMD in step 1 is as follows: For the current signal after Kalman filtering and denoising, the goal of improving AVMD is to decompose the current signal into K mode functions. and the corresponding center frequency The objective of the decomposition is to minimize the variational formula for the following bandwidth constraint: （1）； in, The time derivative is used to represent the change in bandwidth. It is the k-th mode function of the signal. It is the center frequency of each modality function, which is dynamically updated with adaptive learning rate and improved Adam optimization; The improved AVMD introduced Lagrange multipliers. This ensures that the sum of the modal components can reconstruct the original signal, making the reconstructed signal as close as possible to the original signal and ensuring signal integrity; the Lagrange function is defined as: （2）； in, It is an adaptive learning rate, used to dynamically control the optimization step size of bandwidth. These are Lagrange multipliers, ensuring the accuracy of signal reconstruction. It is the input original current signal of the photovoltaic power station, and K is the number of modal components, representing the number of intrinsic mode functions obtained when decomposing the signal.

3. The photovoltaic array fault diagnosis method based on photovoltaic power station current signal output according to claim 1, characterized in that: The specific method for using the Lasso regression model in step 2 to screen out the more important IMFs as information carriers is as follows: The decomposed IMFs are used to form a feature matrix X, where each column represents an IMF and each row represents a time node. The corresponding target variable y is set, which is the label of the fault state. The goal of Lasso regression is to minimize the following loss function: （8）； Where n is the number of samples, y i The target variable is X. ij These are elements of the characteristic matrix. It is the intercept. These are the regression coefficients of the IMF on the target variable. It is a regularization parameter that controls the strength of the constraint on the regression coefficients; the larger the value, the stronger the constraint. By regularization parameters To balance the model's bias and variance, the training data is used to fit the Lasso regression model to obtain the regression coefficients of the IMF. The training data consists of the IMF feature matrix X and the target variable y. The number of IMFs selected is controlled by adjusting the regularization parameter. The regression coefficients of each IMF are compared. The size of the IMF is used to assess its importance in fault diagnosis. A threshold is set to select IMFs with large non-zero coefficients as information carriers for fault diagnosis of photovoltaic systems.

4. The photovoltaic array fault diagnosis method based on photovoltaic power station current signal output according to claim 1, characterized in that: In step 3, the selected IMFs are decomposed into low-frequency and high-frequency components in different frequency bands using wavelet transform. The specific method is as follows: The selected IMF components are decomposed using Discrete Wavelet Transform (DWT). Each wavelet decomposition level includes two filtering steps: low-pass filtering and high-pass filtering. In the j-th level decomposition, the low-frequency coefficients are represented as: （9）； High-frequency coefficients are represented as: （10）； Among them, A j D represents the low-frequency component of the j-th layer. j Let h and g represent the high-frequency components of the j-th layer, respectively, and let h and g represent the low-pass and high-pass filter coefficients of the wavelet. After J-level wavelet decomposition, the multi-scale subsequence A is obtained. J And D1, D2, ..., D J Among them, A J This is the lowest frequency approximation signal, representing the main trend of the signal, and is suitable for further coarse-grained processing; D J Let be the detail signal of the j-th layer, representing the high-frequency components of that layer. It reflects local fluctuations and abrupt changes in the current signal, and can identify short-term variations and noise. When decomposed to the L-th layer, the original signal is decomposed into low-frequency coefficients of the L-th layer and high-frequency coefficients at multiple scales, expressed as: （11）； in, This refers to the low-frequency components when decomposed down to the Lth layer. This refers to the high-frequency components of the l-th layer when decomposed to the L-th layer.

5. The photovoltaic array fault diagnosis method based on photovoltaic power station current signal output according to claim 1, characterized in that: The low-frequency components obtained after wavelet transform are coarsened, and then the discrete entropy values ​​of the coarsened subsequences at different time scales are calculated. The specific method is as follows: First, considering the different time scales of the various phenomena involved in the output current signal, the time scale factor is determined, and the low-frequency components obtained after wavelet transform are coarsened according to the scale factor ratio. Next, the low-frequency components obtained after wavelet transform are grouped according to the time scale factor ratio, and the data within each group are averaged. Finally, a new coarse-grained time series is obtained, with each series having a length of N / τ. cs Establish a new time series based on the scale: （12）； Where, τ sc It is the time scale factor that determines how many subsequences the low-frequency components are coarsened into. For each time scale factor τ sc The data in the low-frequency components are grouped, and the average value of each group is calculated to obtain new data points y. j (τsc) It represents the original time series at τ sc The coarsening results on the time scale show that the length of the coarse-grained time series decreases as the time scale ratio increases.

6. The photovoltaic array fault diagnosis method based on photovoltaic power station current signal output according to claim 5, characterized in that: The coarse-grained subsequences at different time scales obtained from the decomposition are subjected to multi-scale discrete entropy (MDE) calculation. The discrete entropy (MDE) is defined as follows: （13）； in, Let represent the probability of the system in unit i, and n represent the characteristic dimension of the signal.

7. The photovoltaic array fault diagnosis method based on photovoltaic power station current signal output according to claim 1, characterized in that: In step 5, the trend changes of discrete entropy curves at different time scales are observed, and the health status of the photovoltaic system is judged based on the shape of the curves and the threshold. The specific method for diagnosing whether the photovoltaic system has a shading fault is as follows: First, observe the curve shape to make a preliminary judgment on whether there is a fault in the photovoltaic system. When the photovoltaic system is in sunny weather, the MDE curve usually presents a bell shape; when the photovoltaic system is in cloudy weather, the MDE curve usually presents a flat shape; when the photovoltaic system has a fault, the MDE curve will shift from the bell-shaped curve in the normal state, moving upward or downward. When a fault is observed in the photovoltaic system, compare the discrete entropy curves of a fault-free sunny day, a fault-free cloudy day, and a faulty sunny day to obtain the MDE value that differs significantly from the fault-free sunny day and fault-free cloudy day, and set it as a threshold. The threshold is used to determine whether there is a shading fault in the photovoltaic system, and whether the shading fault has started or ended, thereby achieving the effect of fault diagnosis.

Images