Non-stationary Signal Denoising Method for Distributed Photovoltaic Power Generation System Based on Dual Strategies

Abstract

The present invention belongs to the field of electrical digital data processing, and relates to a non-stationary signal denoising method for a distributed photovoltaic power generation system based on a dual strategy, including: Step 1, based on the acquired original signal, generating a target number of intrinsic mode functions through variational mode decomposition; Step 2, removing the intrinsic mode function with the largest total composite entropy from the target number of intrinsic mode functions to obtain an effective signal; Step 3, identifying abnormal data according to the effective signal through discrete data comparison, removing and replacing the abnormal data to obtain a new effective signal; starting from Step 1 and looping until the iterative cutoff condition is reached to generate a denoised signal. It solves the problems of poor non-stationary signal denoising effect and low signal fidelity.

Description

Technical Field

[0001] The present invention relates to the field of electrical digital data processing, and specifically discloses a method for denoising non-stationary signals of a distributed photovoltaic power generation system based on a dual strategy. Background Art

[0002] With the rapid development of renewable energy technologies, distributed photovoltaic power generation plays an increasingly important role in the global energy structure. However, during the operation of a distributed photovoltaic power generation system, signals such as voltage and current collected are often subject to noise interference, causing the output signal to also be affected by noise. The noise includes broadband noise caused by electromagnetic interference, measurement errors, sensor noise, etc., as well as abnormal data caused by mechanical failures, switching processes of power electronic devices, etc.

[0003] During the operation of a distributed photovoltaic power generation system, existing signal denoising technologies usually only process a certain type of noise, and there are certain limitations in dealing with the noise of non-stationary signals. The main disadvantages are as follows:

[0004] 1. Insufficient processing of abnormal data. Traditional signal denoising methods, such as wavelet transform, empirical mode decomposition, etc., are mainly used to remove broadband noise such as white noise, but have poor removal effects on abnormal data with spikes and mutations caused by faults, switching processes of power electronic devices, etc.

[0005] 2. Single noise removal ability. Statistical methods, such as median filtering, dimensionality reduction analysis, etc., are usually used to detect and remove abnormal data, but have poor removal effects on broadband noise such as randomly distributed white noise.

[0006] 3. Difficulty in balancing signal fidelity. Existing signal denoising methods often affect the intrinsic characteristics of the signal, such as affecting the frequency distribution, resulting in low accuracy in signal analysis and signal control.

[0007] Therefore, how to propose an efficient and robust denoising method for the non-stationary signal characteristics of a distributed photovoltaic power generation system has become an urgent problem to be solved. Summary of the Invention

[0008] The purpose of the present invention is to provide a method for denoising non-stationary signals of a distributed photovoltaic power generation system based on a dual strategy, and solve the problems of poor denoising effect and low signal fidelity of non-stationary signals.

[0009] The specific solution of the present invention is as follows:

[0010] A method for denoising non-stationary signals of a distributed photovoltaic power generation system based on a dual strategy includes:

[0011] Step 1: Based on the acquired original signal, generate a target number of intrinsic mode functions through variational mode decomposition;

[0012] Step 2: Eliminate the intrinsic mode function with the largest total composite entropy from the target intrinsic mode functions to obtain the effective signal;

[0013] Step 3: Identify abnormal data by comparing discrete data based on the effective signal, eliminate and replace the abnormal data to obtain a new effective signal;

[0014] Loop from Step 1 until the iteration cut-off condition is reached to generate the denoised signal.

[0015] In some embodiments, based on the acquired original signal, generating the target intrinsic mode functions through variational mode decomposition includes:

[0016] Set the initial decomposition layer number and the initial penalty factor of the variational mode decomposition;

[0017] Generate fitness function values based on the original signal and the acquired initial intrinsic mode functions;

[0018] Optimize the initial decomposition layer number and the initial penalty factor according to the fitness function values to obtain the target decomposition layer number and the target penalty factor;

[0019] Perform variational mode decomposition on the original signal based on the target decomposition layer number and the target penalty factor to generate the target intrinsic mode functions.

[0020] In some embodiments, generating fitness function values based on the original signal and the acquired initial intrinsic mode functions includes:

[0021] Obtain the initial intrinsic mode functions by performing variational mode decomposition on the original signal;

[0022] Divide each intrinsic mode function of the original signal and the initial intrinsic mode functions into N segments of signals respectively to obtain each segment of the original signal and each segment of each intrinsic mode function;

[0023] Generate the composite entropy of each segment of the original signal and the composite entropy of each segment of each intrinsic mode function based on each segment of the original signal and each segment of each intrinsic mode function respectively;

[0024] Obtain the composite entropy correlation coefficients between the original signal and each intrinsic mode function respectively according to the composite entropy of each segment of the original signal and the composite entropy of each segment of each intrinsic mode function;

[0025] Generate fitness function values based on the composite entropy correlation coefficients between the original signal and each intrinsic mode function.

[0026] In some embodiments, generating the composite entropy of each segment of the original signal and the composite entropy of each segment of each intrinsic mode function based on each segment of the original signal and each segment of each intrinsic mode function respectively includes:

[0027] Based on each segment of the original signal, respectively obtain the envelope entropy and sample entropy of each segment of the original signal;

[0028] Perform weighted summation on the envelope entropy and sample entropy of each segment of the original signal respectively to generate the composite entropy of each segment of the original signal;

[0029] Based on each segment of each intrinsic mode function, respectively obtain the envelope entropy and sample entropy of each segment of each intrinsic mode function;

[0030] Perform weighted summation on the envelope entropy and sample entropy of each segment of each intrinsic mode function respectively to generate the composite entropy of each segment of each intrinsic mode function.

[0031] In some embodiments, generating the fitness function value based on the composite entropy correlation coefficient between the original signal and each intrinsic mode function includes:

[0032] Obtain the minimum value of the composite entropy correlation coefficient between the original signal and the intrinsic mode function from the composite entropy correlation coefficients between the original signal and each intrinsic mode function;

[0033] Generate the fitness function value based on the minimum value of the composite entropy correlation coefficient between the original signal and the intrinsic mode function.

[0034] In some embodiments, optimizing the initial decomposition layer number and the initial penalty factor according to the fitness function value to obtain the target decomposition layer number and the target penalty factor includes:

[0035] Take the initial decomposition layer number and the initial penalty factor as the input variables of the cattle herd optimization algorithm, take the fitness function value as the optimization fitness of the cattle herd optimization algorithm, and iteratively optimize through the cattle herd optimization algorithm to obtain the target decomposition layer number and the target penalty factor.

[0036] In some embodiments, obtaining the effective signal by removing the intrinsic mode function with the largest total composite entropy from the target number of intrinsic mode functions includes:

[0037] Obtain the intrinsic mode function with the largest total composite entropy according to the target number of intrinsic mode functions;

[0038] Remove the intrinsic mode function with the largest total composite entropy from the target number of intrinsic mode functions, and take the sum of all the intrinsic mode functions before the intrinsic mode function with the largest total composite entropy and all the intrinsic mode functions after the intrinsic mode function with the largest total composite entropy as the effective signal.

[0039] In some embodiments, abnormal data is identified by comparing discrete data with the effective signal, and the abnormal data is removed and replaced to obtain a new effective signal, including:

[0040] Set a threshold coefficient, and compare each discrete data in the effective signal one by one to determine whether each discrete data does not belong to the effective data interval, where the effective data interval is [the difference between the average value of the effective signal and the product of the standard deviation of the effective signal multiplied by the threshold coefficient, the sum value of the average value of the effective signal and the product of the standard deviation of the effective signal multiplied by the threshold coefficient];

[0041] If it is determined that a certain discrete data does not belong to the effective data interval, then the certain discrete data is identified as abnormal data; a certain discrete data in the effective signal is removed, and a certain discrete data is replaced to obtain a new effective signal.

[0042] In some embodiments, replacing a certain discrete data includes:

[0043] Using the average value of the previous discrete data and the next discrete data of a certain discrete data to replace the certain discrete data.

[0044] In some embodiments, the iteration termination condition includes that the maximum value of the total composite entropy of the intrinsic mode functions of the new effective signal is less than the first threshold and there is no abnormal data.

[0045] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0046] The present invention obtains the composite entropy by weighted summation of the envelope entropy and the sample entropy, generates the composite entropy correlation coefficient of the original signal and each intrinsic mode function based on the composite entropy of each segment of the original signal and the composite entropy of each segment of the signal of each intrinsic mode function, obtains the fitness function value according to the minimum value of the composite entropy correlation coefficient of the original signal and each intrinsic mode function, performs optimization based on the fitness function value to obtain the target number of intrinsic mode functions, removes the intrinsic mode function with the largest total composite entropy from the target number of intrinsic mode functions to obtain the effective signal, identifies abnormal data by comparing discrete data according to the effective signal, removes and replaces the abnormal data to obtain a new effective signal, and generates the denoised signal until the iteration termination condition is reached, improving the denoising effect and signal fidelity of non-stationary signals; it can not only effectively and robustly remove broadband noise, but also remove abnormal data, while fully retaining the intrinsic characteristics of the original signal, maintaining the integrity of the original signal, improving the accuracy of signal analysis and signal control, and is applicable to the complex operating environment of distributed photovoltaic systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a flowchart of the non-stationary signal denoising method for a distributed photovoltaic power generation system based on a dual strategy in Embodiment 1 of the present invention.

[0048] Figure 2 This is the flowchart for generating the target intrinsic mode functions in Embodiment 1 of the present invention.

[0049] Figure 3 This is the schematic diagram of the original signal in Embodiment 2 of the present invention.

[0050] Figure 4 This is the schematic diagram of the effective signal in Embodiment 2 of the present invention.

[0051] Figure 5 This is the schematic diagram of the new effective signal in Embodiment 2 of the present invention.

[0052] Figure 6 This is the schematic diagram of the denoised signal in Embodiment 2 of the present invention.

[0053] Reference signs: A - current, S - time. Detailed implementation manners

[0054] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Usually, the components of the embodiments of the present invention described and illustrated herein can be arranged and designed in various different configurations.

[0055] Embodiment 1

[0056] A non-stationary signal denoising method for a distributed photovoltaic power generation system based on a dual strategy, as Figure 1 shown, includes the following steps:

[0057] S1. Obtain the original signal;

[0058] Obtain the original signal from the distributed photovoltaic power generation system S ori , the original signal S ori has T discrete data.

[0059] S2. Generate the target number of intrinsic mode functions based on the original signal through variational mode decomposition;

[0060] As Figure 2 shown, generating the target number of intrinsic mode functions specifically includes:

[0061] S21. Initialize the variational mode decomposition parameters;

[0062] The variational mode decomposition parameters include the decomposition layer number and the penalty factor. Set the initial decomposition layer number of the variational mode decomposition to K, the initial penalty factor is α .

[0063] S22. Obtain the initial intrinsic mode functions through variational mode decomposition according to the original signal;

[0064] The initial intrinsic mode functions refer to K intrinsic mode functions, which are respectively: IMF1, IMF2, ……, IMF K .

[0065] S23. Divide the original signal and each intrinsic mode function of the initial intrinsic mode functions into N segments of signals respectively to obtain each segment of signals of the original signal and each segment of signals of each intrinsic mode function;

[0066] N The calculation formula of is:

[0067] ,

[0068] where T is the number of discrete data in the original signal, W is the window length, S is the window sliding step, represents rounding up.

[0069] S24. Generate the composite entropy of each segment of signals of the original signal based on each segment of signals of the original signal;

[0070] Based on each segment of signals of the original signal, obtain the envelope entropy and sample entropy of each segment of signals of the original signal respectively;

[0071] Perform weighted summation on the envelope entropy and sample entropy of each segment of signals of the original signal respectively to generate the composite entropy of each segment of signals of the original signal.

[0072] For the j th segment of signals of the original signal, perform weighted summation on the envelope entropy and sample entropy to generate the composite entropy of the j th segment of signals of the original signal. The formula for the composite entropy of the j th segment of signals of the original signal is:

[0073] ,

[0074] where CE ( S ori, j ) is the composite entropy of the j th segment of signals of the original signal, EE ( S ori, j ) is the envelope entropy of the j th segment of signals of the original signal, SE ( Sori, j ) is the sample entropy of the j th segment of the original signal, w 1 is the weighting coefficient of the envelope entropy of the j th segment of the original signal, w 2 is the weighting coefficient of the sample entropy of the j th segment of the original signal, and w 1 + w 2 = 1, j = 1, 2, ……, N .

[0075] S25. Generate the composite entropy of each segment of the signal of each intrinsic mode function based on each segment of the signal of each intrinsic mode function;

[0076] Based on each segment of the signal of each intrinsic mode function, obtain the envelope entropy and sample entropy of each segment of the signal of each intrinsic mode function respectively;

[0077] Perform weighted summation on the envelope entropy and sample entropy of each segment of the signal of each intrinsic mode function respectively to generate the composite entropy of each segment of the signal of each intrinsic mode function.

[0078] Perform weighted summation on the envelope entropy and sample entropy of the i th segment of the j th intrinsic mode function to generate the composite entropy of the i th segment of the j th intrinsic mode function. The formula for the composite entropy of the i th segment of the j th intrinsic mode function is:

[0079] ,

[0080] where, CE (IMF i, j ) is the composite entropy of the i th segment of the j th intrinsic mode function, EE (IMF i, j ) is the envelope entropy of the i th segment of the j th intrinsic mode function, SE (IMF i, j ) is the sample entropy of the i th segment of the j th intrinsic mode function, w 1 is the weighting coefficient of the envelope entropy of the i th segment of the j th intrinsic mode function, w 2 is the weighting coefficient of the sample entropy of the i th segment of the j th intrinsic mode function, and w 1 +w 2 = 1, i = 1, 2, ……, K , j = 1, 2, ……, N 。

[0081] S26. Obtain the composite entropy correlation coefficient between the original signal and each intrinsic mode function respectively according to the composite entropy of each segment of the original signal and the composite entropy of each segment of each intrinsic mode function;

[0082] The composite entropy correlation coefficient between the original signal and the i th intrinsic mode function is given by the formula:

[0083] ,

[0084] where r i is the composite entropy correlation coefficient between the original signal and the i th intrinsic mode function, CE ( S ori, j ) is the composite entropy of the j - th segment of the original signal, is the average value of the composite entropy of each segment of the original signal, CE (IMF i, j ) is the composite entropy of the i th segment of the j th intrinsic mode function, is the average value of the composite entropy of each segment of the i th intrinsic mode function, i = 1, 2, ……, K , j = 1, 2, ……, N 。

[0085] S27. Generate fitness function values based on the composite entropy correlation coefficients between the original signal and each intrinsic mode function;

[0086] Obtain the minimum value of the composite entropy correlation coefficient between the original signal and the intrinsic mode function from the composite entropy correlation coefficients between the original signal and each intrinsic mode function;

[0087] Generate fitness function values based on the minimum value of the composite entropy correlation coefficient between the original signal and the intrinsic mode function.

[0088] The formula for the fitness function is:

[0089] 。

[0090] where e is the natural constant.

[0091] S28, optimizing the initial number of decomposition layers and the initial penalty factor according to the fitness function value to obtain the target number of decomposition layers and the target penalty factor;

[0092] The initial decomposition level K and the initial penalty factor α As the input variable of the swarm optimization algorithm, the fitness function value is used as the optimization fitness of the swarm optimization algorithm, and the target decomposition layer number is obtained through the iterative optimization of the swarm optimization algorithm. K best and the target penalty factor α best .

[0093] S29. Based on the target decomposition level and the target penalty factor, the original signal is subjected to variational mode decomposition to generate a target number of intrinsic mode functions.

[0094] Using the target decomposition level K best and the target penalty factor α best For the original signal S ori Perform variational mode decomposition to generate the target intrinsic mode functions. The target intrinsic mode function refers to K best IMF 1,best , IMF 2,best ,……, .

[0095] S3, removing the eigenmode function with the largest total value of composite entropy from the target eigenmode functions to obtain a valid signal;

[0096] According to the target intrinsic mode functions, the intrinsic mode function with the largest total value of composite entropy is obtained;

[0097] The eigenmode function with the largest total compound entropy value is eliminated from the target eigenmode functions, and the sum of all the eigenmode functions before the eigenmode function with the largest total compound entropy value and all the eigenmode functions after the eigenmode function with the largest total compound entropy value is taken as the effective signal.

[0098] No. i The formula for the total value of the composite entropy of the eigenmode functions is:

[0099] ,

[0100] in, SCE (IMF i ) is the i The total value of the composite entropy of the eigenmode functions, CE (IMF i, j ) is thei The j section signal composite entropy, i = 1, 2, ……, K best , j = 1, 2, ……, N .

[0101] The intrinsic mode function with the largest total composite entropy is the M th intrinsic mode function, M and its calculation formula is:

[0102] ,

[0103] where, SCE (IMF i ) is the total composite entropy of the i th intrinsic mode function, i = 1, 2, ……, K best .

[0104] The formula for the effective signal is:

[0105] ,

[0106] where, S eff is the effective signal, is the sum of all the intrinsic mode functions before the intrinsic mode function with the largest total composite entropy, is the sum of all the intrinsic mode functions after the intrinsic mode function with the largest total composite entropy, i = 1, 2, ……, K best .

[0107] S4. Identify abnormal data by comparing discrete data based on the effective signal, and eliminate and replace the abnormal data to obtain a new effective signal;

[0108] Set a threshold coefficient, compare all discrete data in the effective signal one by one, and determine whether each discrete data does not belong to the effective data interval. The effective data interval is [the difference between the average value of the effective signal and the product of the standard deviation of the effective signal and the threshold coefficient, the sum of the average value of the effective signal and the product of the standard deviation of the effective signal and the threshold coefficient];

[0109] If it is determined that a certain discrete data does not belong to the effective data interval, then identify the certain discrete data as abnormal data; eliminate a certain discrete data in the effective signal, and replace a certain discrete data to obtain a new effective signal; use the average value of the previous discrete data and the next discrete data of a certain discrete data to replace a certain discrete data.

[0110] Compare each of the discrete data in the valid signal one by one. If it is determined that the S eff nth T discrete data does not belong to the valid data interval, the expression that the t nth t discrete data does not belong to the valid data interval is:

[0111] ,

[0112] where S eff ( t ) is the t nth discrete data in the valid signal, is the average value of the valid signal, is the standard deviation of the valid signal,

[0113] is the threshold coefficient; t If it is determined that the nth t discrete data is an abnormal data, use to replace the

[0114] ,

[0115] where is the t nth S eff ( t -1) is the discrete data immediately preceding the t nth S eff ( t +1) is the discrete data immediately following the t nth

[0116] S5. Determine whether the iteration cut-off condition is reached. If it is determined that the iteration cut-off condition is not reached, repeat steps S2 to S4. If it is determined that the iteration cut-off condition is reached, end the iteration to generate the denoised signal.

[0117] The iteration cut-off conditions include: the upper limit value of the set number of iterations, and the total composite entropy of the intrinsic mode functions of the new valid signal is less than the first threshold and there are no abnormal data.

[0118] The composite entropy is obtained by weighted summation of the envelope entropy and the sample entropy. Based on the composite entropy of each segment of the original signal and the composite entropy of each segment of the signal of each intrinsic mode function, the composite entropy correlation coefficient between the original signal and each intrinsic mode function is generated. The fitness function value is obtained according to the minimum value of the composite entropy correlation coefficient between the original signal and each intrinsic mode function. Optimization is carried out based on the fitness function value to obtain the target number of intrinsic mode functions. The intrinsic mode function with the largest total composite entropy is removed from the target number of intrinsic mode functions to obtain the effective signal. Abnormal data is identified by discrete data comparison based on the effective signal, and the abnormal data is removed and replaced to obtain a new effective signal until the iterative cutoff condition is reached to generate the denoised signal, which improves the denoising effect and signal fidelity of non-stationary signals; it can not only effectively and robustly remove broadband noise, but also remove abnormal data, while fully retaining the intrinsic characteristics of the original signal, maintaining the integrity of the original signal, improving the accuracy of signal analysis and signal control, and is applicable to the complex operating environment of distributed photovoltaic systems.

[0119] Embodiment 2

[0120] A specific embodiment of a non-stationary signal denoising method for a distributed photovoltaic power generation system based on a dual strategy, including:

[0121] Obtain the original current signal, as Figure 3 shown, there are 2500 discrete data in this original current signal.

[0122] Set the initial decomposition layer number of variational mode decomposition to 3, and the initial penalty factor to 600.

[0123] Based on the original current information, the initial decomposition layer number and the initial penalty, the initial number of intrinsic mode functions is obtained through variational mode decomposition. The initial number of intrinsic mode functions is 3 intrinsic mode functions.

[0124] Set the window length W to 500, and the window sliding step S to 50, then the original current signal and the 3 intrinsic mode functions are each divided into (2500 - 500) / 50 = 40 segments.

[0125] According to the 40 segments of the original current signal and the 40 segments of the signal of each intrinsic mode function, the composite entropy correlation coefficient between the original current signal and the first intrinsic mode function is calculated to be -0.21488, the composite entropy correlation coefficient between the original current signal and the second intrinsic mode function is calculated to be -0.01567, and the composite entropy correlation coefficient between the original current signal and the third intrinsic mode function is calculated to be 0.05797. It can be seen that the composite entropy correlation coefficient -0.21488 between the original current signal and the first intrinsic mode function is the minimum value. Then, based on the composite entropy correlation coefficient -0.21488 between the original current signal and the first intrinsic mode function, the moderate function value is generated as e-0.21488 = 0.80664.

[0126] Taking the initial decomposition level 3 and the initial penalty factor 600 as the input variables of the cattle herd optimization algorithm, and taking the fitness function value 0.80664 as the optimization fitness of the cattle herd optimization algorithm, the target decomposition level 5 and the target penalty factor 993 are obtained through the iterative optimization of the cattle herd optimization algorithm.

[0127] Based on the original current signal, the target decomposition level 5 and the target penalty factor 993, the target number of intrinsic mode functions are obtained through variational mode decomposition. The target number of intrinsic mode functions is 5 intrinsic mode functions.

[0128] The intrinsic mode function with the largest total composite entropy value is the 5th intrinsic mode function. The 5th intrinsic mode function is removed from the 5 intrinsic mode functions, and the sum of the remaining 4 intrinsic mode functions is used as the effective signal, as Figure 4 shown.

[0129] The threshold coefficient is set to 1.6. At this time, the average value of the effective signal is -0.020, and the standard deviation of the effective signal is 16.0860. Then the effective data interval of the effective signal is [-25.7577, 25.7177].

[0130] Compare the 2500 discrete data of the effective signal one by one, and find that there is 1 abnormal data. Remove and replace it to obtain a new effective signal, as Figure 5 shown.

[0131] Repeat the above calculation process. Set the upper limit value of the number of iterations for the iterative termination condition to 5 times. When the number of iterations reaches 5 times, a denoised signal is generated, as Figure 6 shown.

[0132] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A non-stationary signal denoising method for distributed photovoltaic power generation system based on dual strategy, characterized in that: include: Step 1: Based on the acquired original signal, generate the target intrinsic mode functions through variational mode decomposition; The method of generating a target number of intrinsic mode functions by variational mode decomposition based on the acquired original signal includes: setting an initial number of decomposition layers and an initial penalty factor of the variational mode decomposition; generating a fitness function value based on the original signal and the acquired initial number of intrinsic mode functions; optimizing the initial number of decomposition layers and the initial penalty factor according to the fitness function value to obtain a target number of decomposition layers and a target penalty factor; performing variational mode decomposition on the original signal based on the target number of decomposition layers and the target penalty factor to generate a target number of intrinsic mode functions; The method of generating a fitness function value based on the original signal and the initial intrinsic mode functions obtained includes: obtaining the initial intrinsic mode functions through variational mode decomposition according to the original signal; dividing the original signal and each intrinsic mode function of the initial intrinsic mode function into N segments of signals, respectively obtaining each segment of the original signal and each segment of the intrinsic mode function; generating the composite entropy of each segment of the original signal and the composite entropy of each segment of the intrinsic mode function based on each segment of the original signal and each segment of the intrinsic mode function; obtaining the composite entropy correlation coefficient between the original signal and each intrinsic mode function based on the composite entropy of each segment of the original signal and the composite entropy of each segment of the intrinsic mode function; generating the fitness function value based on the composite entropy correlation coefficient between the original signal and each intrinsic mode function; The method of generating composite entropies of each signal segment of the original signal and composite entropies of each signal segment of each intrinsic mode function based on each signal segment of the original signal and each signal segment of each intrinsic mode function respectively includes: based on each signal segment of the original signal, respectively obtaining envelope entropies and sample entropies of each signal segment of the original signal; performing weighted summation on the envelope entropies and sample entropies of each signal segment of the original signal to generate composite entropies of each signal segment of the original signal; based on each signal segment of each intrinsic mode function, respectively obtaining envelope entropies and sample entropies of each signal segment of each intrinsic mode function; performing weighted summation on the envelope entropies and sample entropies of each signal segment of each intrinsic mode function to generate composite entropies of each signal segment of each intrinsic mode function; Step 2: Eliminate the eigenmode function with the largest total value of composite entropy from the target eigenmode functions to obtain a valid signal; Step 3: Identify abnormal data by comparing discrete data based on valid signals, remove and replace abnormal data to obtain new valid signals; The denoised signal is generated by looping from step 1 until the iteration cutoff condition is reached.

2. The method for denoising non-stationary signals of distributed photovoltaic power generation systems based on dual strategies according to claim 1 is characterized in that: The generating of the fitness function value based on the composite entropy correlation coefficient between the original signal and each intrinsic mode function includes: The minimum value of the composite entropy correlation coefficient between the original signal and the intrinsic mode function is obtained from the composite entropy correlation coefficient between the original signal and each intrinsic mode function; The fitness function value is generated based on the minimum value of the composite entropy correlation coefficient between the original signal and the intrinsic mode function.

3. The method for denoising non-stationary signals of distributed photovoltaic power generation systems based on dual strategies according to claim 1 is characterized in that: The step of optimizing the initial number of decomposition layers and the initial penalty factor according to the fitness function value to obtain the target number of decomposition layers and the target penalty factor includes: The initial number of decomposition layers and the initial penalty factor are used as input variables of the swarm optimization algorithm, and the fitness function value is used as the optimization fitness of the swarm optimization algorithm. The target number of decomposition layers and the target penalty factor are obtained through iterative optimization of the swarm optimization algorithm.

4. The method for denoising non-stationary signals of distributed photovoltaic power generation systems based on dual strategies according to claim 1 is characterized in that: The step of removing the eigenmode function with the largest total composite entropy value from the target eigenmode functions to obtain a valid signal includes: According to the target intrinsic mode functions, the intrinsic mode function with the largest total value of composite entropy is obtained; The eigenmode function with the largest total compound entropy value is eliminated from the target eigenmode functions, and the sum of all the eigenmode functions before the eigenmode function with the largest total compound entropy value and all the eigenmode functions after the eigenmode function with the largest total compound entropy value is taken as the effective signal.

5. The method for denoising non-stationary signals of distributed photovoltaic power generation systems based on dual strategies according to claim 1 is characterized in that: The method of identifying abnormal data by comparing discrete data according to the effective signal, and removing and replacing the abnormal data to obtain a new effective signal includes: Setting a threshold coefficient, comparing all discrete data in the valid signal one by one, and determining whether each discrete data does not belong to a valid data interval, wherein the valid data interval is [the difference between the mean value of the valid signal and the product of the standard deviation of the valid signal multiplied by the threshold coefficient, and the sum of the mean value of the valid signal and the product of the standard deviation of the valid signal multiplied by the threshold coefficient]; If it is determined that a certain discrete data does not belong to the valid data interval, the certain discrete data is identified as abnormal data; a certain discrete data in the valid signal is eliminated, and a certain discrete data is replaced to obtain a new valid signal.

6. The method for denoising non-stationary signals of distributed photovoltaic power generation systems based on dual strategies according to claim 5 is characterized in that: The replacing of a discrete data comprises: Replace a discrete data with the average of the previous discrete data and the next discrete data.

7. The method for denoising non-stationary signals of a distributed photovoltaic power generation system based on a dual strategy according to claim 1 is characterized in that: The iteration cutoff condition includes that the maximum value of the total value of the composite entropy of the intrinsic mode function of the new effective signal is less than a first threshold and there is no abnormal data.

Images