A Hybrid Energy Storage Frequency Regulation Command Prediction Method and System Based on Error Feedback

Abstract

The present invention discloses a hybrid energy storage frequency modulation command prediction method and system based on error feedback, which relates to the field of hybrid energy storage frequency modulation prediction optimization. The method includes: collecting frequency modulation commands to obtain first data, and performing first screening and second screening; based on the first data and the results of the first screening and the second screening, finding data outliers and repairing them to obtain second data; decomposing the second data to obtain third data; inputting the third data into a convolutional neural network to obtain the prediction result of the frequency modulation command. The present invention significantly improves the accuracy, stability, real-time performance and adaptability of the frequency modulation command prediction through multi-level outlier detection and repair, optimized decomposition strategy, using a convolutional neural network for feature extraction and prediction, and a multi-level error feedback mechanism; it can be adjusted and optimized according to different application scenarios, and has important application value in the fields of power system frequency modulation, hybrid energy storage system control, etc.

Description

Technical Field

[0001] The present invention relates to the field of hybrid energy storage frequency modulation prediction optimization, and particularly to a method and system for predicting hybrid energy storage frequency modulation commands based on error feedback. Background Art

[0002] In a power system, thermal power units, as one of the main power generation methods, their frequency modulation capabilities are crucial for the stable operation of the power grid. Traditional frequency modulation methods mainly rely on the regulation capabilities of thermal power units themselves. However, due to the slow response speed of thermal power units, it is difficult to meet the power grid's demand for rapid frequency modulation. To improve the frequency modulation efficiency, a hybrid energy storage system (such as supercapacitor + lithium battery) is introduced as an auxiliary frequency modulation means. The traditional hybrid energy storage frequency modulation method is to transmit the difference between the frequency modulation command and the actual output of the thermal power unit to the hybrid energy storage system, where the low-frequency part is borne by the lithium battery and the high-frequency part is borne by the supercapacitor.

[0003] There is a certain time required for the frequency modulation command to be transmitted from the thermal power unit to the hybrid energy storage system. At the same time, there are also delays in the responses of the supercapacitor and the lithium battery, resulting in poor overall frequency modulation effects and thus affecting the power plant's revenue; in the data collection stage, due to natural or human factors, the data may be distorted or missing, affecting the accuracy of prediction. Traditional methods lack effective preprocessing of the original data, resulting in unreliable prediction results; when traditional data decomposition methods (such as VMD, CEEMD, etc.) are used to process the frequency modulation command signal, the decomposed subsequences may have mode mixing phenomena, affecting the prediction accuracy; traditional prediction methods usually lack a feedback and correction mechanism for prediction errors, resulting in a deviation between the prediction result and the actual demand and being unable to achieve dynamic optimization. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is: how to improve the response speed and prediction accuracy of hybrid energy storage assisting thermal power units for frequency modulation to enhance the stability of the power grid and the economic benefits of the power plant.

[0006] To solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a method for predicting hybrid energy storage frequency modulation commands based on error feedback, including:

[0008] Collect frequency modulation commands to obtain first data, and perform a first screening and a second screening;

[0009] Based on the first data and the results of the first screening and the second screening, find data outliers and repair them to obtain second data;

[0010] Decompose the second data to obtain third data;

[0011] The decomposition of the second data to obtain the third data includes:

[0012] Set the initial value of the iteration round p to 0, set the initial decomposition layer number K, and perform the first decomposition:

[0013] Perform five random decompositions. Each time, generate K subsequences. After the five random decompositions are completed, perform the first operation. The first operation includes: calculating the error coefficient of each decomposition, expressed as [W1, W2, W3, W4, W5]; finding the decomposition corresponding to the minimum error coefficient;

[0014] Take the decomposition corresponding to the minimum error coefficient as the optimal decomposition of this round, and increment the value of p by one;

[0015] Repeat the first decomposition. When the iteration round p reaches the set value, if K has not reached the set layer number, increment the value of K by one and continue to repeat the first decomposition;

[0016] The five random decomposition directions of the next round of decomposition are based on the direction of the optimal decomposition of the previous round;

[0017] When the iteration round p reaches the set value and K reaches the set layer number, find the decomposition with the lowest decomposition error in the corresponding round, take the subsequences of this decomposition as the final decomposition result, and take the final decomposition result as the third data;

[0018] Input the third data into a convolutional neural network to obtain the prediction result of the frequency modulation command;

[0019] The inputting the third data into a convolutional neural network to obtain the prediction result of the frequency modulation command includes:

[0020] Input the values with a length of 0.9N at the front of each subsequence in the third data into the convolutional neural network, output the corresponding values of the unknown group, and then add up the values of the unknown group predicted for each subsequence to obtain the prediction result, where N is the total length of the subsequence.

[0021] As an optimal scheme of the frequency modulation command prediction method for hybrid energy storage based on error feedback, where:

[0022] The finding of data outliers and their repair based on the first data and the results of the first screening and the second screening to obtain the second data includes:

[0023] If a certain point is both a suspicious point in the first screening and a suspicious point in the second screening, identify this point as an outlier; after repairing the outliers in the first data, obtain the second data.

[0024] As an optimal scheme of the frequency modulation command prediction method for hybrid energy storage based on error feedback, where:

[0025] The first screening is for abnormal size:

[0026] Let the frequency modulation command be P t =[X1, X2, X3,..., X i ,..., X N , and the mean value is J = avg(X1, X2, X3,..., X i ,..., X N ), where X1, X2, X3,..., X i ,..., X N are all frequency modulation values;

[0027] If a certain value of the frequency modulation command is greater than 3 times J or less than 1 / 3 of J, it is listed as a suspicious point.

[0028] As a preferred solution of the frequency modulation command prediction method for hybrid energy storage based on error feedback, where:

[0029] The second screening is for abnormal changes before and after:

[0030] Let three consecutive values in the frequency modulation command be X i-1 、X i 、X i+1 ;

[0031] If it satisfies that sin(X i-1 -X i ) and sin(X i -X i+1 ) have different signs, and:

[0032] , then X i is listed as a suspicious point.

[0033] In a second aspect, an embodiment of the present invention provides a frequency modulation command prediction system for hybrid energy storage based on error feedback, including:

[0034] A screening module, configured to collect frequency modulation commands, obtain first data, and perform the first screening and the second screening;

[0035] A repair module, configured to find data outliers and repair them based on the first data and the results of the first screening and the second screening to obtain second data;

[0036] A decomposition module, configured to decompose the second data to obtain third data;

[0037] The decomposition of the second data to obtain third data includes:

[0038] Set the initial value of the iteration round number p to 0, set the initial decomposition layer number K, and perform the first decomposition:

[0039] Perform five random decompositions, generating K subsequences each time. After the five random decompositions are completed, perform a first operation, which includes: calculating the error coefficient of each decomposition, expressed as [W1, W2, W3, W4, W5]; finding the decomposition corresponding to the minimum error coefficient among them;

[0040] Take the decomposition corresponding to the minimum error coefficient as the optimal decomposition of this round, and increment the value of p by one;

[0041] Repeat the first decomposition. When the iteration round number p reaches the set value, if K has not reached the set value of the number of layers, increment the value of K by one and continue to repeat the first decomposition;

[0042] The directions of the five random decompositions for the next round of decomposition are based on the direction of the optimal decomposition in the previous round;

[0043] When the iteration round number p reaches the set value and K reaches the set value of the number of layers, find the decomposition with the lowest decomposition error in the corresponding round, take the subsequences of this decomposition as the final decomposition result, and take the final decomposition result as the third data;

[0044] A prediction module, configured to input the third data into a convolutional neural network to obtain a prediction result of the frequency modulation instruction;

[0045] The inputting the third data into the convolutional neural network to obtain the prediction result of the frequency modulation instruction includes:

[0046] Input the values with a length of 0.9N at the front of each subsequence in the third data into the convolutional neural network, output the values of the corresponding unknown groups, and then add up the values of the unknown groups predicted for each subsequence to obtain the prediction result, where N is the total length of the subsequences.

[0047] In a third aspect, an embodiment of the present invention provides a computing device, including:

[0048] A memory and a processor;

[0049] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the one or more programs are executed by the one or more processors, the one or more processors implement the error feedback-based hybrid energy storage frequency modulation instruction prediction method as described in any embodiment of the present invention.

[0050] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, which stores computer-executable instructions, and when the computer-executable instructions are executed by a processor, the error feedback-based hybrid energy storage frequency modulation instruction prediction method as described above is implemented.

[0051] Advantages of the present invention: By calculating the mean of the frequency modulation command sequence, identifying outliers, and effectively filtering extreme values, the present invention avoids their negative impacts on subsequent analysis and prediction; after removing or repairing the outliers from the data, the error sources of the prediction model are significantly reduced, and the reliability of the prediction results is improved; by using the stochastic decomposition method, the complex time series signal is decomposed into multiple subsequences, and each subsequence often represents different frequency components or change patterns, which is convenient for subsequent feature extraction and model prediction. This decomposition method can better capture the dynamic characteristics of the signal and improve the fineness of signal processing; by using a convolutional neural network to predict the decomposed subsequences, it can more efficiently process complex frequency modulation command sequences, significantly improving the accuracy and robustness of the prediction. At the same time, the end-to-end training method of the convolutional neural network simplifies the model design and optimization process; through a multi-level error feedback mechanism, the decomposition and prediction processes are gradually optimized, and the data processing and model prediction processes can be dynamically adjusted and optimized, significantly improving the prediction accuracy, especially in complex or noisy frequency modulation command data, showing higher stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings without creative efforts based on these drawings.

[0053] Figure 1 is the overall flowchart of the hybrid energy storage frequency modulation command prediction method based on error feedback according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0054] In order to make the above objects, features, and advantages of the present invention more obvious and understandable, the specific embodiments of the present invention will be described in detail below with reference to the drawings of the specification. Obviously, the described embodiments are some embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments of the present invention shall fall within the protection scope of the present invention.

[0055] Many specific details are set forth in the following description in order to provide a thorough understanding of the present invention, but the present invention may be practiced in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the spirit of the present invention, so the present invention is not limited by the specific embodiments disclosed below.

[0056] Second, the "one embodiment" or "embodiment" referred to herein means a specific feature, structure, or characteristic that may be included in at least one implementation manner of the present invention. The appearances of "in one embodiment" in different places in this specification do not all refer to the same embodiment, nor are they separate or selectively exclusive embodiments from other embodiments.

[0057] Embodiment 1, referring to Figure 1 , which is the first embodiment of the present invention. This embodiment provides a method for predicting frequency modulation commands of a hybrid energy storage based on error feedback, including:

[0058] S1: Collect frequency modulation commands to obtain first data, and perform first screening and second screening;

[0059] S2: Based on the first data and the results of the first screening and second screening, find data outliers and repair them to obtain second data;

[0060] S3: Decompose the second data to obtain third data;

[0061] S4: Input the third data into a convolutional neural network to obtain the prediction result of the frequency modulation command.

[0062] It should be noted that through the above steps, the method for predicting frequency modulation commands of a hybrid energy storage based on error feedback is realized, the data quality is optimized, and the reliability of the input data is improved; signal feature extraction and optimized decomposition are carried out to enhance the effectiveness of signal features; efficient feature learning and prediction are carried out using a CNN, significantly improving the prediction accuracy; through a multi-level error feedback mechanism, the stability and prediction performance of the model are improved; fast response and real-time prediction are realized to meet the real-time control requirements of the power system; a flexible system design is provided to adapt to various application scenarios.

[0063] Embodiment 2, referring to Figure 1 , which is an embodiment of the present invention. Based on the previous embodiment, a method for predicting frequency modulation commands of a hybrid energy storage based on error feedback is provided, including:

[0064] In the embodiment of the present application, the collection of frequency modulation commands in the above step S1 to obtain first data and perform first screening and second screening includes:

[0065] It should be noted that the collected dispatching commands are frequency modulation values at intervals of 1 second and are processed subsequently in the form of a data sequence, such as [1.1, 2.0, 3]. The first data is the screened data sequence.

[0066] Specifically, the screening includes the first screening: screening for abnormal sizes:

[0067] Let the frequency modulation command be P t = [X1, X2, X3,..., Xi ,..., X N , and the mean is J = avg(X1, X2, X3,..., X i ,..., X N ), where X1, X2, X3,..., X i ,..., X N are all frequency modulation values;

[0068] If a certain value of the frequency modulation command is greater than 3 times J or less than 1 / 3 of J, it is listed as a suspicious point.

[0069] The screening also includes a second screening: screening for abnormal changes before and after:

[0070] Let three consecutive values in the frequency modulation command be X i-1 , X i , X i+1 ;

[0071] If it satisfies that sin(X i-1 - X i ) and sin(X i - X i+1 ) have different signs, and:

[0072] , then X i is listed as a suspicious point.

[0073] In another possible implementation, the first screening and the second screening can also be Z - score detection: calculate the Z - score of each data point and identify the points that deviate from the mean by more than a certain standard deviation as outliers.

[0074] Rolling average filtering: Calculate the average value using a rolling window and mark the points that deviate from this average value as abnormal.

[0075] Window size: Select an appropriate window size according to the data characteristics.

[0076] Box - plot method: Use quartiles to identify outliers, for example, values outside Q1 - 1.5IQR and Q3 + 1.5IQR.

[0077] In the embodiment of the present application, based on the first data and the results of the first screening and the second screening in the above step S2, finding data outliers and repairing them to obtain the second data includes:

[0078] If a certain point is both a suspicious point in the first screening and a suspicious point in the second screening, then this point is identified as an outlier.

[0079] Let X i be an outlier, and the correction of X i is expressed as:

[0080] 。

[0081] After repairing the outliers in the first data, the second data is obtained.

[0082] In another possible implementation, polynomial interpolation can also be used to estimate outliers. Machine learning models such as K-nearest neighbors and decision trees can also be used to predict outliers based on surrounding data points.

[0083] In the embodiment of the present application, the decomposition of the second data in step S3 to obtain the third data includes:

[0084] Let the sequence obtained after data processing be P t ' = [X1', X2', X3',..., X i ',..., X N '];

[0085] Set the initial value of the iteration round p to 0, set the initial decomposition layer number K, and perform the first decomposition:

[0086] Perform five random decompositions. Each time, K subsequences are generated. After the five random decompositions are completed, perform the first operation to obtain the optimal decomposition of this round, and increment the value of p by one.

[0087] When the iteration round p reaches the set value, if K has not reached the set layer number value, increment the value of K by one and continue to repeat the first decomposition;

[0088] Each time when performing the next round of five random decompositions, the direction of the five random decompositions in the next round is based on the direction of the optimal decomposition in the previous round.

[0089] Specifically, set the initial value of the iteration round p to 0, set the initial decomposition layer number K. In this embodiment, the preferred value of the iteration round p is 5, K is preferably 6, and the set layer number value is preferably 12; it should be noted that the constraint condition for random decomposition is that the sum of all subsequences in each decomposition is P t '.

[0090] Perform the first decomposition:

[0091] Perform five random decompositions. Each time, P t ' is divided into K subsequences, which are respectively represented as: IMF1, IMF i ,..., IMF5, where i is the order of decomposition; that is, there are K subsequences in IMF1, and there are K subsequences in each of IMF i ,..., IMF5.

[0092] Perform the first operation on the subsequences after five random decompositions: calculate the error coefficient for each decomposition, denoted as [W1, W2, W3, W4, W5]; find the decomposition corresponding to the minimum error coefficient; let W j be the minimum; record the subsequence IMF j generated by this decomposition and the error coefficient W j ;

[0093] It should be noted that the calculation of the error coefficient mainly uses MAPE, and the calculation formula is:

[0094] Exemplarily, calculate MAPE for each of the K subsequences generated by each decomposition, and summarize the K MAPE values for each decomposition into one error coefficient (which can be the average value, the maximum value, the minimum value, etc.). <![CDATA[

[0095] ]]It should also be noted that the directions of the five random decompositions for the next decomposition are based on the direction of the optimal decomposition in the previous step.

[0096] Exemplarily, assume that the error of W3 is the lowest, and the subsequence IMF3 after this decomposition corresponding to W3 is expressed as: IMF 31 =[X 11 ,X 12 ,...,X 1N ,IMF 32 =[X 21 ,X 22 ,...,X 2N ,...,IMF 3K =[X K1 ,X K2 ,...,X KN , let it be set B, then for the random directions of the five random decompositions generated in the next round, there are regulations. For example, assume that the subsequence generated by any one of the five random decompositions in the next round is IMF t , expressed as: IMF t1 =[X 11 ,X 12 ,...,X 1N ,IMF t2 =[X 21 ,X 22 ,...,X 2N ,...,IMF tK =[X K1 ,X K2 ,...,X KN , let it be set A;

[0097] All corresponding set A and set B need to satisfy:

[0098] , where refers to the Nth element in the Kth subsequence of set A, and refers to the Nth element in the Kth subsequence of set B; further, when the iteration round p reaches the set value and K reaches the layer set value, find the corresponding decomposition with the lowest decomposition error in the corresponding round, and use the subsequence of this decomposition as the final decomposition result, and use the final decomposition result as the third data.

[0099] In the embodiment of the present application, inputting the third data into the convolutional neural network in step S4 to obtain the prediction result of the frequency modulation command includes:

[0100] Input the values with a length of 0.9N before each subsequence in the third data into the convolutional neural network, output the values of the corresponding unknown groups, and then add up the values of the unknown groups predicted for each subsequence to obtain the prediction result.

[0101] It should be noted that the convolutional neural network adopted by the present invention has demonstrated excellent performance in the field of signal processing. Its core advantage is that the CNN can automatically extract features from data without manual intervention. The CNN usually consists of multiple layers, mainly including a convolutional layer, a pooling layer, and a fully connected layer. The convolutional layer, as a key part of the CNN, extracts local features from the data through a moving convolutional kernel; the pooling layer is used to reduce the spatial dimension of the features, thereby simplifying the model; at the end of the network, the fully connected layer maps the features extracted by the previous levels to the final output to achieve the classification effect. During the training process of the CNN, the weights and biases of the network are continuously updated through backpropagation.

[0102] Embodiment 3, the above is a schematic solution of the hybrid energy storage frequency modulation command prediction method based on error feedback in this embodiment. It should be noted that the technical solution of the hybrid energy storage frequency modulation command prediction system based on error feedback belongs to the same concept as the technical solution of the hybrid energy storage frequency modulation command prediction method based on error feedback. For the details not described in detail in the technical solution of the hybrid energy storage frequency modulation command prediction system based on error feedback in this embodiment, reference can be made to the description of the technical solution of the hybrid energy storage frequency modulation command prediction method based on error feedback.

[0103] This embodiment also provides a system for the hybrid energy storage frequency modulation command prediction method based on error feedback, including:

[0104] A screening module for collecting frequency modulation commands to obtain first data and performing first screening and second screening;

[0105] A repair module for finding data outliers and repairing them based on the first data and the results of the first screening and the second screening to obtain second data;

[0106] A decomposition module for decomposing the second data to obtain the third data;

[0107] The decomposing the second data to obtain the third data includes:

[0108] Set the initial value of the iteration round p to 0, set the initial decomposition layer number K, and perform the first decomposition:

[0109] Perform five random decompositions. Each time, generate K subsequences. After the five random decompositions are completed, perform the first operation. The first operation includes: calculating the error coefficient of each decomposition, expressed as [W1, W2, W3, W4, W5]; finding the decomposition corresponding to the minimum error coefficient among them;

[0110] Take the decomposition corresponding to the minimum error coefficient as the optimal decomposition of this round, and increment the value of p by one;

[0111] Repeat the first decomposition. When the iteration round p reaches the set value, if K has not reached the layer set value, increment the value of K by one and continue to repeat the first decomposition;

[0112] The five random decomposition directions of the next round of decomposition are based on the direction of the optimal decomposition of the previous round;

[0113] When the iteration round p reaches the set value and K reaches the layer set value, find the decomposition corresponding to the lowest decomposition error in the corresponding round, take the subsequences of this decomposition as the final decomposition result, and take the final decomposition result as the third data;

[0114] A prediction module for inputting the third data into a convolutional neural network to obtain a prediction result of the frequency modulation command;

[0115] The inputting the third data into a convolutional neural network to obtain a prediction result of the frequency modulation command includes:

[0116] Input the values with a length of 0.9N at the front of each subsequence in the third data into the convolutional neural network, output the values of the corresponding unknown groups, and then add up the values of the unknown groups predicted by each subsequence to obtain the prediction result, where N is the total length of the subsequences.

[0117] This embodiment also provides a computing device applicable to the case of the frequency modulation command prediction method for hybrid energy storage based on error feedback, including:

[0118] A memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the frequency modulation command prediction method for hybrid energy storage based on error feedback as proposed in the above embodiment.

[0119] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the method for predicting the frequency modulation command of hybrid energy storage based on error feedback proposed in the above embodiment.

[0120] The storage medium proposed in this embodiment and the method for predicting the frequency modulation command of hybrid energy storage based on error feedback proposed in the above embodiment belong to the same inventive concept. The technical details not described in detail in this embodiment can be referred to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0121] Example 4, referring to Tables 1-2, is an embodiment of the present invention, which provides a method for predicting the frequency modulation command of hybrid energy storage based on error feedback. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through simulation experiments.

[0122] The frequency modulation sequences are predicted by using the method of the present invention and the prediction method of VMD-CNN respectively, and the results are as follows:

[0123] Table 1 Comparison of prediction results of frequency modulation sequences

[0124] ,

[0125] Table 2 Four evaluation indexes

[0126] ,

[0127] Among them, N represents the sample size, and respectively represent the actual value and the predicted value at time n.

[0128] It can be seen that four evaluation indexes all decrease in the frequency modulation sequence 1, indicating that the method proposed by the present invention can well improve the prediction accuracy. The same problem is illustrated by the frequency modulation sequence 2.

[0129] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A hybrid energy storage frequency modulation command prediction method based on error feedback, characterized in that It includes: Collect frequency modulation instructions to obtain the first data, and perform the first screening and the second screening; The first screening is for screening abnormal sizes: Let the frequency modulation command be P t =[X1,X2,X3,...,X i ,...,X N , and the mean value is J = avg(X1,X2,X3,...,X i ,...,X N ), where X1,X2,X3,...,X i ,...,X N are all frequency modulation values; If a certain value of the frequency modulation instruction is greater than 3 times J or less than 1 / 3 of J, it is listed as a suspicious point; The second screening is for screening abnormal front-back changes: Let three consecutive values in the frequency modulation command be X i-1 , X i , X i+1 ; If sin(X i-1 -X i ) and sin(X i -X i+1 ) have different signs, and: , then X i is listed as a suspicious point; Based on the first data and the results of the first screening and the second screening, find data outliers and repair them to obtain the second data; The finding data outliers and repairing them based on the first data and the results of the first screening and the second screening to obtain the second data includes: If a certain point is both a suspicious point in the first screening and a suspicious point in the second screening, that point is identified as an outlier; after repairing the outliers in the first data, the second data is obtained; Decompose the second data to obtain the third data; The decomposing the second data to obtain the third data includes: Set the initial value of the iteration round p to 0, set the initial decomposition layer number K, and perform the first decomposition: Perform five random decompositions, each time generating K subsequences. After the five random decompositions are completed, perform the first operation. The first operation includes: calculating the error coefficient of each decomposition, expressed as [W1, W2, W3, W4, W5]; finding the decomposition corresponding to the smallest error coefficient; Take the decomposition corresponding to the smallest error coefficient as the optimal decomposition of this round, and increment the value of p by one; Repeat the first decomposition. When the iteration round p reaches the set value, if K has not reached the layer set value, increment the value of K by one and continue to repeat the first decomposition; The directions of the five random decompositions in the next round of decomposition are based on the direction of the optimal decomposition in the previous round; When the iteration round p reaches the set value and K reaches the layer set value, find the corresponding decomposition with the lowest decomposition error in the corresponding round, take the subsequences of this decomposition as the final decomposition result, and take the final decomposition result as the third data; Input the third data into a convolutional neural network to obtain the prediction result of the frequency modulation instruction; The inputting the third data into a convolutional neural network to obtain the prediction result of the frequency modulation instruction includes: Input the values with a length of 0.9N at the front of each subsequence in the third data into the convolutional neural network, output the values of the corresponding unknown groups, and then add up the values of the unknown groups predicted by each subsequence to obtain the prediction result, where N is the total length of the subsequences.

2. A hybrid energy storage frequency modulation command prediction system based on error feedback, which applies the method described in claim 1, is characterized in that It includes: A screening module for collecting frequency modulation instructions to obtain the first data and performing the first screening and the second screening; The first screening is for screening abnormal sizes: Let the frequency modulation command be P t =[X1,X2,X3,...,X i ,...,X N , and the mean value is J = avg(X1,X2,X3,...,X i ,...,X N ), where X1,X2,X3,...,X i ,...,X N are all frequency modulation values; If a certain value of the frequency modulation instruction is greater than 3 times J or less than 1 / 3 of J, it is listed as a suspicious point; The second screening is for screening abnormal front-back changes: Let three consecutive values in the frequency modulation command be X i-1 , X i , X i+1 ; If sin(X i-1 -X i ) and sin(X i -X i+1 ) have different signs, and: , then X i is listed as a suspicious point; A repair module for finding data outliers and repairing them based on the first data and the results of the first screening and the second screening to obtain the second data; The finding data outliers and repairing them based on the first data and the results of the first screening and the second screening to obtain the second data includes: If a certain point is both a suspicious point in the first screening and a suspicious point in the second screening, that point is identified as an outlier; after repairing the outliers in the first data, the second data is obtained; A decomposition module for decomposing the second data to obtain the third data; Decomposing the second data to obtain the third data includes: Set the initial value of the iteration round p to 0, set the initial decomposition layer number K, and perform the first decomposition: Perform five random decompositions. Each time, generate K subsequences. After the five random decompositions are completed, perform the first operation. The first operation includes: calculating the error coefficient of each decomposition, expressed as [W1, W2, W3, W4, W5]; finding the decomposition corresponding to the minimum error coefficient among them; Take the decomposition corresponding to the minimum error coefficient as the optimal decomposition of this round, and increment the value of p by one; Repeat the first decomposition. When the iteration round p reaches the set value, if K has not reached the set layer number value, increment the value of K by one and continue to repeat the first decomposition; The directions of the five random decompositions in the next round of decomposition are based on the direction of the optimal decomposition in the previous round; When the iteration round p reaches the set value and K reaches the set layer number value, find the decomposition corresponding to the lowest decomposition error in the corresponding round, take the subsequences of this decomposition as the final decomposition result, and take the final decomposition result as the third data; A prediction module, configured to input the third data into a convolutional neural network to obtain a prediction result of the frequency modulation instruction; The step of inputting the third data into the convolutional neural network to obtain a prediction result of the frequency modulation instruction includes: Input the values with a length of 0.9N at the front of each subsequence in the third data into the convolutional neural network, output the values of the corresponding unknown groups, and then add up the values of the unknown groups predicted by each subsequence to obtain the prediction result, where N is the total length of the subsequences.

3. A computing device, comprising: A memory and a processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the method described in claim 1 are implemented.

4. A computer-readable storage medium, which stores computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the method described in claim 1 are implemented.

Images