Compression parameter optimization method and system for distributed photovoltaic abnormal transient waveform data

Abstract

The present invention provides a compression parameter optimization method and system for distributed photovoltaic abnormal transient waveform data, the method comprising: obtaining the original abnormal transient waveform data of distributed photovoltaic according to a preset time sequence; selecting a detail coefficient matrix; calculating the initial threshold of each scale; constructing a fitness function; selecting a threshold; obtaining an optimized threshold vector; using the optimal threshold vector for compression processing to form an applicable detail coefficient matrix. The present invention solves the technical problems of fixed parameter settings, inconsistent compression accuracy, poor adaptability, inability to process large-scale data, high energy consumption, and poor real-time performance.

Description

Technical Field

[0001] The present invention relates to the field of distributed photovoltaic data measurement, and in particular to a compression parameter optimization method and system for distributed photovoltaic abnormal transient waveform data. Background Art

[0002] With the popularization and widespread application of distributed photovoltaic power generation systems in power grids, it is necessary to effectively monitor and manage the operating status of distributed photovoltaic power generation systems. Due to the influence of weather and light factors, it is easy to cause abnormal conditions in distributed photovoltaic power generation systems, such as voltage fluctuations and frequency changes. The aforementioned abnormalities may affect the stability and security of the power grid, so timely and efficient monitoring and response are required.

[0003] The monitoring of distributed photovoltaic systems requires the collection of a large amount of waveform data to record in detail the operating information of distributed photovoltaic power generation systems in different periods of time. However, due to the large amount of data and high transmission costs, it is not practical to directly transmit and store this data. Therefore, the waveform data needs to be compressed to reduce the amount of data while retaining key information for subsequent analysis and processing.

[0004] Distributed photovoltaic abnormal transient waveform data is different from other power grid data. For example, it has obvious time-varying characteristics and changes with the working state of the photovoltaic power generation system; it is greatly affected by external factors, resulting in uncertainty in the data, including the uncertainty of the amplitude and frequency of the waveform; the waveform data may contain high-frequency noise, which mainly comes from system operation and power grid interference; the waveform includes different frequency components and multiple waveform cycles. This puts forward more stringent requirements for the optimization operation of distributed photovoltaic abnormal transient waveform data compression parameters. At the same time, due to the characteristics of distributed photovoltaics with decentralized layout, the data scale increases, and a large amount of data needs to be transmitted, stored and processed, which increases the complexity and cost of data processing. Therefore, for the optimization of distributed photovoltaic abnormal transient waveform data compression parameters, a method with strong real-time performance, small compression error, high compression efficiency, high adaptability and fast operation speed is required.

[0005] The traditional distributed photovoltaic abnormal transient waveform data compression parameter optimization method has the following disadvantages:

[0006] (1) Relatively fixed parameter settings: Traditional methods use fixed parameter settings, which may be determined based on experience or fixed rules. This fixed parameter setting cannot adapt to changes in the system's operating state, resulting in unstable compression effects under different working environments.

[0007] (2) Inconsistent compression accuracy: The compression effect of traditional methods is unstable under different operating conditions, resulting in some data losing important information after compression, which reduces the accuracy of subsequent analysis and diagnosis.

[0008] (3) Poor adaptability: Traditional methods lack adaptability to changes in system status. Once the parameter settings are determined, it is difficult to respond to system changes in a timely manner. This may lead to performance degradation or failure during system operation.

[0009] (4) Unable to process large-scale data: As the scale of distributed photovoltaic systems continues to expand, the amount of data generated by the system is also increasing. Traditional compression parameter optimization methods may not be able to effectively process large-scale data, resulting in poor optimization results.

[0010] (5) High energy consumption: Since the compression algorithm of traditional methods has a large amount of computation, it consumes a lot of computing resources when processing data. In large-scale systems, this high energy consumption characteristic will affect the overall efficiency of the system.

[0011] (6) Poor real-time performance: When abnormal transient waveforms appear in the system, traditional methods require a long time to adjust parameters, resulting in large processing delays and inability to detect and handle abnormal situations in a timely manner.

[0012] In summary, the existing technology has technical problems such as fixed parameter settings, inconsistent compression accuracy, poor adaptability, inability to process large-scale data, high energy consumption and poor real-time performance. Summary of the invention

[0013] The technical problem to be solved by the present invention is: how to solve the technical problems in the prior art of fixed parameter settings, inconsistent compression accuracy, poor adaptability, inability to process large-scale data, high energy consumption and poor real-time performance.

[0014] The present invention adopts the following technical solutions to solve the above technical problems: A compression parameter optimization method for abnormal transient waveform data of distributed photovoltaics includes:

[0015] S1. Obtaining the original abnormal transient waveform data of distributed photovoltaic according to the preset time sequence;

[0016] S2. Performing empirical mode decomposition on the original abnormal transient waveform data to obtain a neural network input vector, using a preset neural network model to determine an applicable wavelet function and an applicable wavelet decomposition scale, performing wavelet decomposition on the original abnormal transient waveform data to obtain a detail coefficient matrix;

[0017] S3. Calculate the initial threshold for each applicable wavelet decomposition scale T i , obtain the initial population individuals;

[0018] S4. Based on the original abnormal transient waveform data S[n] and initial threshold to construct fitness function;

[0019] S5. Set the adaptive mutation operator and the coefficient of variation, perform mutation operations on the individuals in the initial population, and generate a mutation vector V i , perform a crossover operation to generate a test vector U i , according to the mutation vector V i , test vector U i Select an optimized threshold vector;

[0020] S6, compress and optimize the threshold vector, calculate the coding length of the detail coefficient matrix and the compression error description length, perform a minimum description length (MDL) operation, and obtain the optimal threshold vector;

[0021] S7. Use the optimal threshold vector to perform compression processing to form a suitable detail coefficient matrix.

[0022] The present invention has the advantages of small compression error, high compression rate and adaptive threshold selection. The signal is effectively decomposed into several intrinsic mode functions through empirical mode decomposition, and only the features related to the special-shaped waveform are retained, which speeds up the data compression. The appropriate wavelet function and decomposition scale are selected through the neural network model to reduce the error of data compression. The minimum description length MDL criterion reduces the storage space of distributed photovoltaic abnormal transient waveform data, retains as many features of distributed photovoltaic abnormal transient waveform data as possible, can improve the stability of distributed photovoltaic systems, reduce the impact of data compression on fault data application, and provide guarantee for the smooth operation of distributed photovoltaics.

[0023] In a more specific technical solution, in S2, the hidden layers and branches in the neural network model are used to process the neural network input vector, and the wavelet function selection branches and decomposition scale layers are obtained by processing, and the neural network model is used to select the applicable wavelet function and the applicable wavelet decomposition scale. S2 includes:

[0024] S21, using the following logic, the original abnormal transient waveform data S[n] Perform empirical mode decomposition:

[0025] (1)

[0026] In formula (1), is the residual term, which is the low-frequency part of the signal, n represents the number of original abnormal transient waveform data, and K is the neural network input vector obtained by decomposition. IMF The number of

[0027] S22, k Neural network input vector IMF k [n] Expressed as IMF k =[x 1 ,x 2 ,...,x n ] , as the input of the neural network model, where x represents the components of the neural network input vector, n Indicates the number of components;

[0028] S23. Using the following logic, apply a fully connected layer in the hidden layer, using ReLU As an activation function:

[0029] (2)

[0030] In formula (2), is the weight matrix of the hidden layer, is the bias matrix of the hidden layer, H is the output of the hidden layer, where the fully connected layer includes: h neurons;

[0031] S24, using the following logic, the output of the hidden layer H Pass it to no less than 2 branches to obtain the wavelet function selection branch and the decomposition scale layer number respectively:

[0032] (3)

[0033] In formula (3), is the weight matrix of the wavelet function selection branch, is the bias matrix of the wavelet function selection branch, is the output vector of the wavelet function selection branch, is the normalized exponential function;

[0034] S25. According to the following logic, using the neural network model, select the applicable wavelet function and the applicable wavelet decomposition scale:

[0035] (5)

[0036] In formula (5), represents the applicable wavelet function, L To apply the wavelet decomposition scale, NN Represents a neural network model;

[0037] S26, original abnormal transient waveform data S[n] , perform wavelet decomposition according to the applicable wavelet function and the applicable wavelet decomposition scale, and obtain the decomposition scalei The detail coefficient vector under :

[0038]

[0039] Processed I detail coefficient vectors, forming a detail coefficient matrix :

[0040]

[0041] in, Represents the decomposition scale i No. j The detail coefficient, I is the maximum wavelet decomposition scale.

[0042] In a more specific technical solution, in S23, the activation function is determined using the following logic: ReLU :

[0043] (4)

[0044] In formula (4), is the branch weight matrix of the decomposition scale layer, is the decomposition scale layer branch bias scalar, is the number of decomposition scales of the output.

[0045] In a more specific technical solution, in S3, the following logic is used to obtain the initial threshold value: T i :

[0046] (6)

[0047] In formula (6), is the decomposition scale i The detail coefficient vector under The standard deviation of is the threshold adjustment constant;

[0048] The initial threshold vector of all applicable wavelet decomposition scales , as the initial population individuals of the differential evolution algorithm, where Represents the maximum wavelet decomposition scale I The components of the corresponding initial threshold vector.

[0049] In a more specific technical solution, in S4, the following logic is used to construct the fitness function: :

[0050] (7)

[0051] In formula (7), It is j The original data, It is j Use the current threshold vector The restored data after compression.

[0052] In a more specific technical solution, in S5, for each individual in the initial population T i , perform mutation operation and generate mutation vector V i , combined with the adaptive crossover probability to adjust the crossover strategy, perform the crossover operation, and generate the test vector U i , according to the mutation vector V i , test vector U i Select the optimized threshold vector, S5 includes:

[0053] S51. Use the following logic to set an adaptive mutation operator:

[0054] (8)

[0055] In formula (8), represents the maximum number of iterations, Indicates the current iteration number, represents the adaptive constant;

[0056] S52, using the following logic, set the coefficient of variation F :

[0057] (9)

[0058] In formula (9), is the initial coefficient of variation;

[0059] S53, using the following logic, for each individual T i Perform mutation operation to generate mutation vector V i :

[0060] (10)

[0061] In formula (10), , , represents the first individual, the second individual, and the third individual randomly selected from the population;

[0062] S54, according to the following logic, combined with the adaptive crossover probability, the crossover strategy is adjusted, and the crossover operation is performed to obtain the adaptive crossover probability. , calculate the trial vector U i :

[0063] (11)

[0064] In formula (11), is the fitness of the current population, is the maximum value of fitness, is the minimum crossover probability, is the maximum value of the crossover probability;

[0065] S55. Calculate the test vector U i The fitness value of f(U i ) , if it satisfies: f(U i ) < f(T i ) , then use the test vector U i Replace individual T i , f(T i） Represents an individual T i Corresponding fitness function; repeat the crossover operation and selection operation until the optimized threshold vector is obtained.

[0066] In a more specific technical solution, in S54, the test vector is determined using the following logic: U i :

[0067] (12)

[0068] In formula (12), is the adaptive crossover probability, is a randomly selected index, ensuring that the trial vector U i Inherits at least one variant element, Indicates j mutation vector, Indicates selecting a random number in the interval (0, 1).

[0069] Aiming at the shortcomings of fixed parameter settings and poor adaptability of the compression parameter optimization method for distributed photovoltaic abnormal transient waveform data, the present invention adopts a genetic algorithm combined with an adaptive crossover probability to adjust the crossover strategy for adaptive optimization, so that the compression average error reaches the expected target. There is no need to manually adjust the compression parameters according to experience, environment, and fault type, thereby improving the system operation efficiency.

[0070] In a more specific technical solution, in S6, the compression optimization threshold vector is compressed to obtain a compression threshold, the coding length of the detail coefficient matrix and the compression error description length are calculated, and the MDL criterion and the compression threshold are combined to perform a minimized MDL operation to obtain the optimal threshold vector. S6 includes:

[0071] S61, optimizing the threshold vector To perform compression, Represents the optimization operation, forming a compressed detail coefficient matrix :

[0072]

[0073] (13)

[0074] in, Represents the maximum wavelet decomposition scale I The corresponding threshold vector components, Indicates the maximum wavelet decomposition scale after compression I The corresponding detail coefficient is Indicates the decomposition scale after compression i The next j The detail coefficient, Represents the decomposition scale i The next j The detail coefficient, Indicates i The threshold vector component corresponding to the wavelet decomposition scale;

[0075] S62, calculate the compressed detail coefficient matrix The encoding length :

[0076] (14)

[0077] In formula (14), For the i Layer compression detail coefficient probability distribution;

[0078] S63. Calculate the compression error description length L err :

[0079] (15)

[0080] In formula (15), To optimize the threshold vector The signal data obtained by compression and restoration, Represents the original data, P represents a probability distribution;

[0081] S64, combine the MDL criterion and the compression threshold, minimize the minimum description length MDL, and calculate the optimal threshold vector :

[0082] (16)

[0083] in, represents the threshold minimization function.

[0084] The present invention performs wavelet decomposition on the distributed photovoltaic abnormal transient waveform data, extracts data features, uses a differential evolution algorithm to obtain a threshold vector, and combines the MDL criterion to obtain an optimal threshold vector for compression, which can better retain the features of the original distributed photovoltaic abnormal transient waveform data. The present invention also considers the complexity of the model and the error after compression, takes into account compression and data fidelity, and reduces the impact of data compression on the application of distributed photovoltaic abnormal transient waveforms.

[0085] In a more specific technical solution, in S7, the following logic is used to obtain the applicable detail coefficient matrix: :

[0086]

[0087] (17)

[0088] In formula (17), is the optimal threshold vector, Represents the maximum wavelet decomposition scale I The corresponding applicable detail coefficient, Indicates i The applicable detail coefficient corresponding to the decomposition scale is, Indicates j The applicable detail coefficient matrix, Indicates the decomposition scale after compression i The next j A detail factor.

[0089] In a more specific technical solution, the compression parameter optimization system of distributed photovoltaic abnormal transient waveform data includes:

[0090] A transient waveform acquisition module is used to acquire the original abnormal transient waveform data of distributed photovoltaics according to a preset time sequence;

[0091] A waveform data decomposition module is used to perform empirical mode decomposition on the original abnormal transient waveform data, obtain a neural network input vector, use a preset neural network model to determine an applicable wavelet function and an applicable wavelet decomposition scale, perform wavelet decomposition on the original abnormal transient waveform data, and obtain a detail coefficient matrix. The waveform data decomposition module is connected to the transient waveform acquisition module;

[0092] Population individual acquisition module, used to calculate each applicable wavelet decomposition scale D i The initial threshold T i , obtain the initial population individuals, and the population individual acquisition module is connected with the waveform data decomposition module;

[0093] Fitness function building module, used to calculate the original abnormal transient waveform data S[n] and initial threshold to construct a fitness function, the fitness function construction module is connected with the population individual acquisition module and the transient waveform acquisition module;

[0094] The threshold vector optimization module is used to set the adaptive mutation operator and coefficient of variation, perform mutation operations on the initial population individuals, and generate mutation vectors V i , perform a crossover operation and generate a test vector U i , according to the mutation vector V i , test vector U i Select an optimized threshold vector, and connect the optimized threshold vector module with the population individual acquisition module;

[0095] An optimal threshold vector obtaining module is used to compress and optimize the threshold vector, calculate the coding length of the detail coefficient matrix and the compression error description length, perform the minimization MDL operation, and obtain the optimal threshold vector. The optimal threshold vector obtaining module is connected to the optimization threshold vector module;

[0096] The detail coefficient matrix acquisition module is used to use the optimal threshold vector for compression processing to form a suitable detail coefficient matrix. The detail coefficient matrix acquisition module is connected to the optimal threshold vector obtaining module.

[0097] Compared with the prior art, the present invention has the following advantages:

[0098] The invention has the advantages of small compression error, high compression rate and adaptive threshold selection. The signal is effectively decomposed into several intrinsic mode functions through empirical mode decomposition, and only the features related to the special-shaped waveform are retained, which speeds up the data compression. The appropriate wavelet function and decomposition scale are selected through the neural network model to reduce the error of data compression. The minimum description length MDL (Minimun Description Length) criterion reduces the storage space of distributed photovoltaic abnormal transient waveform data, retains as many features of distributed photovoltaic abnormal transient waveform data as possible, can improve the stability of distributed photovoltaic systems, reduce the impact of data compression on fault data application, and provide guarantee for the smooth operation of distributed photovoltaics.

[0099] Aiming at the shortcomings of fixed parameter settings and poor adaptability of the compression parameter optimization method for distributed photovoltaic abnormal transient waveform data, the present invention adopts a genetic algorithm combined with an adaptive crossover probability to adjust the crossover strategy for adaptive optimization, so that the compression average error reaches the expected target. There is no need to manually adjust the compression parameters according to experience, environment, and fault type, thereby improving the system operation efficiency.

[0100] The present invention performs wavelet decomposition on the distributed photovoltaic abnormal transient waveform data, extracts data features, uses a differential evolution algorithm to obtain a threshold vector, and combines the MDL criterion to obtain an optimal threshold vector for compression, which can better retain the features of the original distributed photovoltaic abnormal transient waveform data. The present invention also considers the complexity of the model and the error after compression, takes into account compression and data fidelity, and reduces the impact of data compression on the application of distributed photovoltaic abnormal transient waveforms.

[0101] The present invention solves the technical problems existing in the prior art of fixed parameter settings, inconsistent compression accuracy, poor adaptability, inability to process large-scale data, high energy consumption and poor real-time performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0102] Figure 1 This is a schematic diagram of the basic steps of the compression parameter optimization method for distributed photovoltaic abnormal transient waveform data according to Embodiment 1 of the present invention;

[0103] Figure 2 Schematic diagram of specific steps of selecting a detail coefficient matrix according to Embodiment 1 of the present invention;

[0104] Figure 3 This is a schematic diagram of the neural network structure of Example 1 of the present invention;

[0105] Figure 4 Schematic diagram of the wavelet decomposition process of transient waveform data according to Embodiment 1 of the present invention;

[0106] Figure 5A schematic diagram of specific steps for selecting a threshold value in Example 1 of the present invention;

[0107] Figure 6 This is a schematic diagram of specific steps for optimizing a threshold vector in Embodiment 1 of the present invention;

[0108] Figure 7 This is a graph showing the algorithm test results of Example 1 of the present invention;

[0109] Figure 8 This is a schematic diagram of specific implementation steps of the compression parameter optimization method for distributed photovoltaic abnormal transient waveform data according to Example 2 of the present invention. DETAILED DESCRIPTION

[0110] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described in combination with the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0111] Example 1

[0112] like Figure 1 As shown, the compression parameter optimization method of distributed photovoltaic abnormal transient waveform data provided by the present invention includes the following basic steps:

[0113] S1. Obtaining the original abnormal transient waveform data of distributed photovoltaic according to the preset time sequence;

[0114] S2. Performing empirical mode decomposition on the original abnormal transient waveform data to obtain a neural network input vector, using a preset neural network model to determine an applicable wavelet function and an applicable wavelet decomposition scale, performing wavelet decomposition on the original abnormal transient waveform data to obtain a detail coefficient matrix;

[0115] like Figure 2 As shown, in this embodiment, the step S2 of selecting the detail coefficient matrix also includes the following specific steps:

[0116] S21, using the following logic, the original abnormal transient waveform data S[n] Perform empirical mode decomposition:

[0117] (1)

[0118] S22, k Neural network input vector IMF k [n] Expressed as IMF k=[x 1 ,x 2 ,...,x n ] , as the input of the neural network model, where x represents the components of the neural network input vector, n Indicates the number of components;

[0119] S23. Using the following logic, apply a fully connected layer in the hidden layer, using ReLU As an activation function:

[0120] (2)

[0121] S24, using the following logic, the output of the hidden layer H Pass it to no less than 2 branches to obtain the wavelet function selection branch and the decomposition scale layer number respectively:

[0122] (3)

[0123] It is the output vector of the wavelet function selection branch, and each value represents the probability of selecting a certain wavelet function.

[0124] Use the following logic to determine the activation function ReLU :

[0125] (4)

[0126] In formula (4), is the branch weight matrix of the decomposition scale layer, is the decomposition scale layer branch bias scalar, L is the number of decomposition scales of the output.

[0127] S25. According to the following logic, using the neural network model, select the applicable wavelet function and the applicable wavelet decomposition scale:

[0128] (5)

[0129] S26, original abnormal transient waveform data S[n] , perform wavelet decomposition according to the applicable wavelet function and the applicable wavelet decomposition scale, and obtain the decomposition scale i The detail coefficient vector under :

[0130]

[0131] Processed I detail coefficient vectors, forming a detail coefficient matrix :

[0132]

[0133] in, Represents the decomposition scale i No. j The detail coefficient, I is the maximum wavelet decomposition scale.

[0134] The original abnormal transient waveform data S[n] , perform wavelet decomposition by executing the applicable wavelet function and applicable wavelet decomposition scale obtained by executing S25. For details, see Figure 4 ;

[0135] like Figure 3 , Figure 4 As shown, in this embodiment, Figure 4 middle represents the low-pass filter at the first decomposition scale, represents the high-pass filter of the first decomposition scale 1, then Decomposition scale i The low-pass filter below Decomposition scale i The high-pass filter below and The convolution operation can obtain the decomposition scale i The detail coefficient vector under I The detail coefficient matrix is ​​composed of detail coefficient vectors. In this embodiment, the number of sampling points after decomposition is twice that before decomposition. Two downsamplings are required to ensure that the number of sampling points is equal. For details, see Figure 3 The reconstruction process is Figure 3 The reverse process of decomposition.

[0136] S3. Calculate the initial threshold for each applicable wavelet decomposition scale T i , get the initial population individuals:

[0137] (6)

[0138] In this embodiment, the initial threshold vectors of all scales are As the initial population individuals of the differential evolution algorithm;

[0139] S4. Constructing fitness function :

[0140] (7)

[0141] S5. Set the adaptive mutation operator and the coefficient of variation, perform mutation operations on the individuals in the initial population, and generate a mutation vector Vi , perform a crossover operation and generate a test vector U i , according to the mutation vector V i , test vector U i Select an optimized threshold vector;

[0142] In this embodiment, for each individual in the initial population T i , perform mutation operation and generate mutation vector V i , combined with the adaptive crossover probability to adjust the crossover strategy, perform the crossover operation, and generate the test vector U i , according to the mutation vector V i , test vector U i Select an optimized threshold vector;

[0143] like Figure 5 As shown, in this embodiment, the step S5 of threshold selection also includes the following specific steps:

[0144] S51. Use the following logic to set an adaptive mutation operator:

[0145] (8)

[0146] S52, set the coefficient of variation F ;

[0147] In this embodiment, the coefficient of variation F Set to:

[0148] (9)

[0149] S53. For each individual T i Perform mutation operation to generate mutation vector V i ;

[0150] (10)

[0151] S54, combining the adaptive crossover probability to adjust the crossover strategy, perform the crossover operation, and obtain the adaptive crossover probability , calculate the trial vector U i :

[0152] (11)

[0153] Specifically, the test vector is determined using the following logic: U i :

[0154] (12)

[0155] In formula (12), is the adaptive crossover probability, is a randomly selected index, ensuring Inherits at least one variant element.

[0156] S55. Calculate the test vector U i The fitness value of f(U i ) , if it satisfies: f(U i ) < f(T i ) , then use the test vector U i Replace individual T i , f(T i） Represents an individual T i Corresponding fitness function; repeat the crossover operation and selection operation until the optimized threshold vector is obtained.

[0157] S6, compressing and optimizing the threshold vector, calculating the coding length and compression error description length of the detail coefficient matrix, performing a minimization MDL operation, and obtaining the optimal threshold vector;

[0158] like Figure 6 As shown, in this embodiment, step S6 of obtaining the optimized threshold vector further includes the following specific steps:

[0159] S61, optimizing the threshold vector To perform compression, Represents the optimization operation, forming a compressed detail coefficient matrix :

[0160]

[0161] (13)

[0162] S62, calculate the compressed detail coefficient matrix The encoding length :

[0163] (14)

[0164] S63. Calculate the compression error description length L err :

[0165] (15)

[0166] S64, combine the MDL criterion and the compression threshold, minimize the minimum description length MDL, and calculate the optimal threshold vector :

[0167] (16)

[0168] S7, use the optimal threshold vector to perform compression processing to form an applicable detail coefficient matrix :

[0169]

[0170] (17)

[0171] like Figure 7 As shown, the data obtained by compressing and reconstructing the original abnormal transient waveform data according to the present invention is compared with the original abnormal transient waveform data, wherein the red curve is the original abnormal transient waveform data, and the blue curve is the abnormal transient waveform data after compression and reconstruction. Figure 7 As can be seen in the figure, the two curves completely overlap, indicating that this method has the characteristics of high compression accuracy.

[0172] Example 2

[0173] like Figure 1 As shown, in this embodiment, the compression parameter optimization method of distributed photovoltaic abnormal transient waveform data also includes the following specific implementation steps:

[0174] S1', obtaining original abnormal transient waveform data;

[0175] S2', performing empirical mode decomposition on the original abnormal transient waveform data and inputting the data into the neural network model to obtain abnormal transient model data;

[0176] S3', perform wavelet decomposition on abnormal transient model data;

[0177] S4', judging whether the conditions are met;

[0178] S5', if yes, then calculate the optimal threshold vector in combination with the MDL criterion;

[0179] S6', perform hard threshold compression;

[0180] S7', if not, define the fitness function;

[0181] S8', set the adaptive mutation operator;

[0182] S9', perform mutation and crossover operations;

[0183] S10', perform boundary condition processing;

[0184] S11', calculate the objective function;

[0185] S12', perform a threshold selection operation.

[0186] In summary, the present invention has the advantages of small compression error, high compression rate, and adaptive threshold selection. The signal is effectively decomposed into several intrinsic mode functions through empirical mode decomposition, and only the features related to the special-shaped waveform are retained, which speeds up the data compression. The appropriate wavelet function and decomposition scale are selected through the neural network model to reduce the error of data compression. The minimum description length MDL (Minimun Description Length) criterion reduces the storage space of distributed photovoltaic abnormal transient waveform data, retains as many features of distributed photovoltaic abnormal transient waveform data as possible, can improve the stability of distributed photovoltaic systems, reduce the impact of data compression on fault data applications, and provide guarantee for the smooth operation of distributed photovoltaics.

[0187] Aiming at the shortcomings of fixed parameter settings and poor adaptability of the compression parameter optimization method for distributed photovoltaic abnormal transient waveform data, the present invention adopts a genetic algorithm combined with an adaptive crossover probability to adjust the crossover strategy for adaptive optimization, so that the compression average error reaches the expected target. There is no need to manually adjust the compression parameters according to experience, environment, and fault type, thereby improving the system operation efficiency.

[0188] The present invention performs wavelet decomposition on the distributed photovoltaic abnormal transient waveform data, extracts data features, uses a differential evolution algorithm to obtain a threshold vector, and combines the MDL criterion to obtain an optimal threshold vector for compression, which can better retain the features of the original distributed photovoltaic abnormal transient waveform data. The present invention also considers the complexity of the model and the error after compression, takes into account compression and data fidelity, and reduces the impact of data compression on the application of distributed photovoltaic abnormal transient waveforms.

[0189] The present invention solves the technical problems existing in the prior art of fixed parameter settings, inconsistent compression accuracy, poor adaptability, inability to process large-scale data, high energy consumption and poor real-time performance.

[0190] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A compression parameter optimization method for distributed photovoltaic abnormal transient waveform data, characterized in that: The method comprises: S1. Obtaining original abnormal transient waveform data of distributed photovoltaics according to a preset time sequence; S2. Performing empirical mode decomposition on the original abnormal transient waveform data to obtain a neural network input vector, using a preset neural network model to determine an applicable wavelet function and an applicable wavelet decomposition scale, performing wavelet decomposition on the original abnormal transient waveform data to obtain a detail coefficient matrix; S3. Calculate the initial threshold for each applicable wavelet decomposition scale T i , obtain the initial population individuals; S4. Based on the original abnormal transient waveform data S[n] and initial threshold to construct fitness function; S5. Set the adaptive mutation operator and the coefficient of variation, perform mutation operations on the individuals in the initial population, and generate mutation vectors V i , perform a crossover operation to generate a test vector U i , according to the mutation vector V i , test vector U i Select an optimized threshold vector; S6, compressing and optimizing the threshold vector, calculating the coding length and compression error description length of the detail coefficient matrix, performing a minimization MDL operation, and obtaining the optimal threshold vector; Among them, the calculation detail coefficient matrix The encoding length : (14) In formula (14), For the i Layer compression detail coefficient probability distribution, I is the maximum wavelet decomposition scale; Calculate the compression error description length L err : (15) In formula (15), To optimize the threshold vector The signal data obtained by compression and restoration, Represents the original data, P represents a probability distribution; Calculate the optimal threshold vector : (16) In the formula, represents the threshold minimization function; S7. Use the optimal threshold vector to perform compression processing to form a suitable detail coefficient matrix.

2. The compression parameter optimization method for distributed photovoltaic abnormal transient waveform data according to claim 1 is characterized in that: In S2, the neural network input vector is processed by using the hidden layer and branch in the neural network model to obtain the wavelet function selection branch and the number of decomposition scale layers, and the applicable wavelet function and the applicable wavelet decomposition scale are selected by using the neural network model. S2 includes: S21, using the following logic, the original abnormal transient waveform data S[n] Perform the empirical mode decomposition: (1) In the formula (1), is the residual term, which is the low-frequency part of the signal, n represents the number of the original abnormal transient waveform data, and K is the neural network input vector obtained by decomposition. IMF The number of S22, k The neural network input vector IMF k [n] Expressed as IMF k =[x 1 ,x 2 ,...,x n ] , as the input of the neural network model, where x represents the components of the neural network input vector, n Indicates the number of components; S23, using the following logic, apply a fully connected layer in the hidden layer, using ReLU As an activation function: (2) In the formula (2), is the weight matrix of the hidden layer, is the bias matrix of the hidden layer, H is the output of the hidden layer, wherein the fully connected layer includes: h neurons; S24, using the following logic, the output of the hidden layer H Pass it to no less than 2 branches to obtain the wavelet function selection branch and the decomposition scale layer number respectively: (3) In the formula (3), is the weight matrix of the wavelet function selection branch, is the bias matrix of the wavelet function selection branch, is the output vector of the wavelet function selection branch, is the normalized exponential function; S25, according to the following logic, using the neural network model, select the applicable wavelet function and the applicable wavelet decomposition scale: (5) In the formula (5), represents the applicable wavelet function, L is the applicable wavelet decomposition scale, NN represents the neural network model; S26, the original abnormal transient waveform data S[n] , perform wavelet decomposition according to the applicable wavelet function and the applicable wavelet decomposition scale to obtain a decomposition scale i The detail coefficient vector under : Processed I The detail coefficient vectors constitute a detail coefficient matrix : In the formula, Represents the decomposition scale i No. j A detail factor.

3. The compression parameter optimization method for distributed photovoltaic abnormal transient waveform data according to claim 2 is characterized in that: In S23, the activation function is determined using the following logic: ReLU : (4) In the formula (4), is the branch weight matrix of the decomposition scale layer, is the decomposition scale layer branch bias scalar, is the number of decomposition scales of the output.

4. The compression parameter optimization method for distributed photovoltaic abnormal transient waveform data according to claim 1 is characterized in that: In S3, the initial threshold is obtained using the following logic: T i : (6) In the formula (6), is the decomposition scale i The detail coefficient vector under The standard deviation of is the threshold adjustment constant; The initial threshold vector of all the applicable wavelet decomposition scales , as the initial population individuals of the differential evolution algorithm, where, Represents the maximum wavelet decomposition scale I The components of the initial threshold vector corresponding to .

5. The compression parameter optimization method for distributed photovoltaic abnormal transient waveform data according to claim 1 is characterized in that: In S4, the fitness function is constructed using the following logic: : (7) In the formula (7), It is j The original data, It is j Use the current threshold vector The restored data after compression.

6. The compression parameter optimization method for distributed photovoltaic abnormal transient waveform data according to claim 1 is characterized in that: In S5, for each individual in the initial population, T i , perform the mutation operation to generate the mutation vector V i , combined with the adaptive crossover probability to adjust the crossover strategy, perform the crossover operation, and generate the test vector U i , according to the mutation vector V i , the test vector U i Selecting the optimized threshold vector, S5 includes: S51. Using the following logic, set the adaptive mutation operator: (8) In the formula (8), represents the maximum number of iterations, Indicates the current iteration number, represents the adaptive constant; S52, using the following logic, set the coefficient of variation F : (9) In the formula (9), is the initial coefficient of variation; S53, using the following logic, for each of the individuals T i Perform the mutation operation to generate the mutation vector V i : (10) In the formula (10), , , represents the first individual, the second individual, and the third individual randomly selected from the population; S54, according to the following logic, combining the adaptive crossover probability to adjust the crossover strategy, perform the crossover operation, and obtain the adaptive crossover probability , calculate the test vector U i : (11) In the formula (11), is the fitness of the current population, is the maximum value of fitness, is the minimum crossover probability, is the maximum value of the crossover probability; S55, calculating the test vector U i The fitness value of f(U i ) , if it satisfies: f(U i ) < f(T i ) , then use the test vector U i Replace the individual T i , f(T i） Indicates the individual T i The corresponding fitness function; repeating the crossover operation and the selection operation until the optimized threshold vector is obtained.

7. The compression parameter optimization method for distributed photovoltaic abnormal transient waveform data according to claim 6 is characterized in that: In S54, the test vector is determined using the following logic: U i : (12) In the formula (12), is the adaptive crossover probability, is a randomly selected index, ensuring that the test vector U i Inherits at least one variant element, Indicates j The mutation vector, Indicates selecting a random number in the interval (0, 1).

8. The compression parameter optimization method for distributed photovoltaic abnormal transient waveform data according to claim 1 is characterized in that: In the S6, the optimized threshold vector is compressed to obtain a compression threshold, the coding length and the compression error description length of the detail coefficient matrix are calculated, and the MDL minimization operation is performed in combination with the MDL criterion and the compression threshold to obtain the optimal threshold vector. The S6 includes: S61, optimizing the threshold vector To perform compression, Represents the optimization operation, forming a compressed detail coefficient matrix : (13) In the formula, Represents the maximum wavelet decomposition scale I The corresponding threshold vector components, Represents the maximum wavelet decomposition scale after compression I The corresponding detail coefficient is Indicates the decomposition scale after compression i The next j The detail coefficient, Represents the decomposition scale i The next j The detail coefficient, Indicates i The threshold vector component corresponding to the wavelet decomposition scale; S62, calculating the compressed detail coefficient matrix The encoding length ; S63. Calculate the compression error description length L err ; S64: Combine the MDL criterion and the compression threshold to minimize the minimum description length MDL and calculate the optimal threshold vector .

9. The compression parameter optimization method for distributed photovoltaic abnormal transient waveform data according to claim 1, characterized in that: In S7, the applicable detail coefficient matrix is ​​obtained using the following logic: : (17) In the formula (17), is the optimal threshold vector, Represents the maximum wavelet decomposition scale I The corresponding applicable detail coefficient, Indicates i The applicable detail coefficient corresponding to the decomposition scale is, Indicates j The applicable detail coefficient matrix, Indicates the decomposition scale after compression i The next j A detail factor.

10. A compression parameter optimization system for distributed photovoltaic abnormal transient waveform data, used to execute the compression parameter optimization method for distributed photovoltaic abnormal transient waveform data as claimed in any one of claims 1 to 9, characterized in that: The system comprises: A transient waveform acquisition module is used to acquire the original abnormal transient waveform data of distributed photovoltaics according to a preset time sequence; A waveform data decomposition module, used to perform empirical mode decomposition on the original abnormal transient waveform data, obtain a neural network input vector, use a preset neural network model to determine an applicable wavelet function and an applicable wavelet decomposition scale, perform wavelet decomposition on the original abnormal transient waveform data, and obtain a detail coefficient matrix. The waveform data decomposition module is connected to the transient waveform acquisition module; A population individual acquisition module is used to calculate each applicable wavelet decomposition scale D i The initial threshold T i , obtaining initial population individuals, wherein the population individual obtaining module is connected to the waveform data decomposition module; A fitness function building module is used to build a fitness function according to the original abnormal transient waveform data. S[n] and the initial threshold value constructs a fitness function, the fitness function construction module is connected with the population individual acquisition module and the transient waveform acquisition module; The threshold vector optimization module is used to set the adaptive mutation operator and the coefficient of variation, perform mutation operations on the individuals of the initial population, and generate a mutation vector V i , perform a crossover operation to generate a test vector U i , according to the mutation vector V i , the test vector U i Selecting an optimized threshold vector, wherein the optimized threshold vector module is connected to the population individual acquisition module; An optimal threshold vector obtaining module, used for compressing the optimized threshold vector, calculating the coding length and compression error description length of the detail coefficient matrix, performing a minimization MDL operation, and obtaining the optimal threshold vector, wherein the optimal threshold vector obtaining module is connected to the optimized threshold vector module; The detail coefficient matrix acquisition module is used to use the optimal threshold vector to perform compression processing to form a suitable detail coefficient matrix. The detail coefficient matrix acquisition module is connected to the optimal threshold vector obtaining module.

Images