An error change prediction method based on supercapacitor energy storage

Abstract

The present invention discloses an error change prediction method based on supercapacitor energy storage, which relates to the technical field of supercapacitor energy storage. The method includes collecting data information, performing intelligent fault detection using a deep learning model, monitoring the operating state of the supercapacitor energy storage system, identifying and marking abnormal data, and correcting and eliminating the marked abnormal data; performing multi-scale analysis on the corrected data using the wavelet transform method to extract multi-level features of the error change, and further extracting non-linear features in the error change through non-linear feature extraction technology; constructing a hybrid prediction model for error change prediction, deploying a lightweight prediction model on edge devices for real-time data processing and prediction, and uploading the results to the cloud. The method of the present invention can not only effectively reduce the limitations and instabilities of a single model, but also optimize the prediction results through an integration strategy, thereby improving the robustness and accuracy of the model.

Description

Technical Field

[0001] The present invention relates to the technical field of supercapacitor energy storage, and in particular, to an error change prediction method based on supercapacitor energy storage. Background Art

[0002] In the field of modern data analysis and prediction, with the continuous increase in data scale and complexity, a single prediction model often struggles to handle various complex practical application scenarios. Traditional prediction models, such as random forest, support vector machine (SVM), and long short-term memory network (LSTM), each have their own unique advantages and application scenarios. However, when dealing with complex and multi-dimensional data, the performance of a single model may be limited. For example, random forest performs well in handling high-dimensional data and non-linear problems, SVM has advantages in dealing with small samples and high-dimensional data, and LSTM has a unique ability to capture long-term dependencies in time series. Therefore, how to effectively combine the advantages of multiple models to construct an efficient and accurate hybrid prediction model has become an important research direction. Summary of the Invention

[0003] In view of the above existing problems, the present invention is proposed.

[0004] Therefore, the present invention provides an error change prediction method based on supercapacitor energy storage. By combining random forest, support vector machine, and LSTM neural network, and using a weighted voting or stacking ensemble strategy, high-precision prediction of error change is achieved. In practical applications, a single model may result in unsatisfactory prediction results due to the characteristics of the data or the limitations of the model itself. To solve this problem, the present invention conducts parallel training with multiple algorithms and assigns weights based on the prediction accuracy of each model, integrating the advantages of multiple models to improve the overall prediction performance. This method can not only effectively reduce the limitations and instability of a single model but also optimize the prediction results through the ensemble strategy, thereby improving the robustness and accuracy of the model.

[0005] To solve the above technical problems, the present invention provides the following technical solution. An error change prediction method based on supercapacitor energy storage includes:

[0006] Collect data information, use a deep learning model for intelligent fault detection, monitor the operating status of the supercapacitor energy storage system, identify and mark abnormal data, and correct and eliminate the marked abnormal data; use the wavelet transform method to perform multi-scale analysis on the corrected data, extract multi-level features of error changes, and further extract non-linear features in error changes through non-linear feature extraction technology; construct a hybrid prediction model, combine random forest, support vector machine, and LSTM neural network for error change prediction, deploy a lightweight prediction model on edge devices for real-time data processing and prediction, and upload the results to the cloud.

[0007] As a preferred solution of the error change prediction method based on supercapacitor energy storage according to the present invention, wherein: the collection of data information includes collecting voltage data, current data, temperature data, power data, status data, environmental data, historical operation data, frequency data, capacity data, vibration and noise data; the status data includes the charging status and discharging status of the supercapacitor energy storage unit; the environmental data includes external environmental data of the operating environment of the supercapacitor energy storage system; the capacity data includes the remaining capacity and charge-discharge cycle times of the supercapacitor energy storage unit.

[0008] As a preferred solution of the error change prediction method based on supercapacitor energy storage according to the present invention, wherein: the use of a deep learning model for intelligent fault detection includes constructing a fault detection result function, representing the fault detection result at time t:

[0009] ;

[0010] Wherein, represents the original input data, representing the th convolutional kernel at the th layer of the th input signal; represents the weight matrix, representing the weight parameters of the th convolutional kernel at the th layer, obtained through training data; represents the convolutional kernel parameter, representing the th convolutional kernel at the th layer of the th parameter; represents the bias term, representing the bias parameter of the th convolutional kernel at the th layer, obtained through training data; represents the activation function, selects the ReLU activation function, defined as ; represents the pooling result, representing the th convolutional kernel at the The pooling result of the layer is obtained through the max pooling or average pooling function; Q represents the total number of convolution kernels; L represents the total number of layers; R represents the total number of input signals. represents the convolution kernel index. is the layer index, and r is the input signal index.

[0011] As a preferred solution of the error change prediction method based on supercapacitor energy storage according to the present invention, wherein: the correction and elimination of the marked abnormal data include identifying and marking abnormal data points through the intelligent fault detection of the convolutional neural network, and the abnormal data points refer to data deviating from the preset normal range caused by sensor failures, communication errors or unexpected situations.

[0012] Statistical analysis is performed on the collected historical data, and the statistical characteristics of each parameter are calculated, including the mean value, standard deviation, median, and quartiles.

[0013] Data correction is performed on the abnormal data points. For a single abnormal data point, the mean value of the historical data is used for replacement:

[0014] ;

[0015] For multiple consecutive abnormal data points, the linear interpolation method is used to perform linear interpolation according to adjacent normal data points:

[0016] ;

[0017] Wherein, is the interpolated data, and are the normal data points before and after the abnormal data segment respectively, is the total number of abnormal data points; is the position of the current abnormal data point in the abnormal data segment, from 1 to ; The missing data is filled using a sliding window, and the window size is The average value of the data points within the window is used as the filling value:

[0018] ;

[0019] Wherein, represents the value of the hth data point within the sliding window, represents the value of the data point after correction and filling;

[0020] Polynomial fitting is performed on the historical data, and the fitting function is used to fill the missing data.

[0021] ,

[0022] Wherein, is the polynomial coefficient, is the time; is the highest order of the polynomial, is the order of the polynomial.

[0023] As a preferred solution of the error change prediction method based on supercapacitor energy storage according to the present invention, wherein: the wavelet transform method includes decomposing the corrected data into multi-scale components through wavelet transform, and each component represents signal components of different frequencies. The wavelet transform includes discrete wavelet transform and continuous wavelet transform;

[0024] Arrange all the corrected data into a in the time series, represents the corrected original data time series, including voltage, current, temperature, and power; represents the wavelet basis function, is the mother wavelet, which is obtained by dilation and translation;

[0025] ;

[0026] wherein, represents the scale index, which is used to control the dilation of the wavelet; represents the translation index, which is used to control the position of the wavelet; represents the approximation coefficient of the scale component, is the scaling function, which is defined as:

[0027] ;

[0028] Process the data using discrete wavelet transform DWT and continuous wavelet transform CWT respectively;

[0029] ;

[0030] wherein, a is the dilation factor and b is the translation factor; the scale index and the translation index are determined through the wavelet transform process;

[0031] Wavelet basis function Select the mother wavelet and choose the optimal mother wavelet according to the data characteristics; Select the mother wavelet according to the data characteristics and obtain through dilation and translation;

[0032] Approximation coefficient is obtained by calculating the cumulative result of each scale component, and the formula is:

[0033] ;

[0034] For each scale and each translation perform wavelet transform to obtain multi-scale components , The smaller it is, the smaller the signal component on the corresponding scale; is time, is the scale index, is the translation index, is the maximum scale, is the corrected original data time series, is the scaling function, is the continuous wavelet transform result.

[0035] As a preferred solution of the error change prediction method based on supercapacitor energy storage described in the present invention, wherein: the multi-scale analysis includes analyzing the multi-scale components, extracting the features affecting the error change, and using them as the input of the prediction model. The features affecting the error change include frequency components, amplitudes, and phase information;

[0036] Select the mother wavelet according to the data characteristics , and obtain the multi-scale components through stretching and translation. For each multi-scale component calculate the amplitude and the phase ; calculate the standard deviation at each scale and translation position using historical data, and standardize the frequency components; combine the standardized amplitude and phase information to calculate the comprehensive feature function :

[0037] ;

[0038] ;

[0039] wherein, is the frequency, is the total number of translation positions, and i1 is the imaginary unit; The multi-scale component at the th scale and the th translation position obtained through wavelet transform is used as a function of the frequency ; represents the standard deviation at the th scale and the th translation position.

[0040] As a preferred embodiment of the error change prediction method based on supercapacitor energy storage according to the present invention, wherein: extracting the non-linear features in the error change includes processing the input data by a kernel method or an autoencoder to extract non-linear features in a high-dimensional space, and the kernel method includes a radial basis function kernel and a polynomial kernel;

[0041] Normalize the input data Select a preset kernel width parameter and calculate the radial basis function kernel ; Select a preset constant and the degree of the polynomial to calculate the polynomial kernel Train the autoencoder through an unsupervised learning method to obtain the weight matrix and bias term;

[0042] Combine the outputs of the radial basis function kernel, the polynomial kernel and the autoencoder to calculate the non-linear feature mapping in the high-dimensional space Specifically as follows:

[0043] ,

[0044] wherein, is the training sample index, is the total number of training samples, is the kernel width parameter; is the degree of the polynomial, representing the inner product of the input data x and the training sample ; is the autoencoder neuron index, is the total number of autoencoder neurons, is the input data feature index, is the total number of input data features, is the weight of the radial basis function kernel, is the weight of the autoencoder, is the weight matrix of the autoencoder, is the bias term of the autoencoder, is the activation function.

[0045] As a preferred embodiment of the error change prediction method based on supercapacitor energy storage according to the present invention, wherein: constructing the hybrid prediction model includes combining a random forest, a support vector machine and an LSTM neural network through an ensemble learning method to form a hybrid prediction model, and the ensemble learning method includes a weighted voting and a stacking ensemble strategy;

[0046] Perform single-model predictions of the random forest, the support vector machine and the LSTM neural network, introduce a weighted voting strategy, assign weights to the prediction results of each model, and calculate the mean absolute error MAE of each model based on the predictions of each model:

[0047] ;

[0048] Assign weights according to MAE, and the weight is the reciprocal of MAE:

[0049] ;

[0050] Use the weighted voting method to weight the prediction results of each model:

[0051] ;

[0052] Use a secondary learner to combine the prediction results of single models and calculate the root mean square error RMSE of the secondary learner:

[0053] ;

[0054] Assign weights according to RMSE, and the weight is the reciprocal of RMSE:

[0055] ;

[0056] The final prediction result is obtained by combining the weighted voting result and the result of the secondary learner:

[0057] ,

[0058] where represents the prediction result of the random forest model for the input data ; represents the prediction result of the support vector machine model for the input data ; represents the prediction result of the long short-term memory network model for the input data ; is the final prediction result; is the mean absolute error of model f, w f is the weight of model f, is the prediction result of the secondary learner for the input data X, is the root mean square error of the secondary learner s, is the weight of the secondary learner s, F is the total number of models participating in weighted voting, D is the total number of secondary learners participating in the fusion of the secondary learner, s represents the secondary learner index, B is the total number of samples for calculating the error, represents the th sample's true value, used to calculate the prediction error; represents the prediction value of the fth model for the th sample.

[0059] A computer device includes a memory and a processor. The memory stores a computer program. It is characterized in that when the processor executes the computer program, the steps of an error change prediction method based on supercapacitor energy storage are realized.

[0060] A computer-readable storage medium stores a computer program thereon. It is characterized in that when the computer program is executed by a processor, the steps of an error change prediction method based on supercapacitor energy storage are realized.

[0061] Advantages of the present invention: The hybrid prediction model of the present invention effectively combines the advantages of random forest, support vector machine, and LSTM neural network through multi-algorithm parallel training and weighted voting or stacking ensemble strategy, thereby achieving high-precision error change prediction. Compared with traditional single models, the present invention has significantly improved prediction accuracy and model robustness. Specifically, through weight allocation based on the prediction error of the model, models with smaller errors have a greater impact on the final result, thus improving the prediction accuracy. At the same time, by integrating the prediction results of multiple models, the influence of outliers or abnormal performances of a single model on the final result can be effectively reduced, enhancing the stability and robustness of the model. In addition, the method of the present invention has good scalability and adaptability, can be applied to various complex data prediction scenarios, provides reliable prediction results, and has broad application prospects. Description of the Drawings

[0062] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for description in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0063] Figure 1 It is a schematic flowchart of an error change prediction method based on supercapacitor energy storage provided by an embodiment of the present invention. Detailed Embodiments

[0064] 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, rather than all, embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0065] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention. However, the present invention may be practiced in other ways different from those described herein. Persons skilled in the art can make similar extensions without departing from the spirit of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0066] Secondly, the so-called "one embodiment" or "embodiment" herein refers to 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 alternative embodiments that exclude each other with other embodiments.

[0067] The present invention is described in detail in conjunction with schematic diagrams. When detailing the embodiments of the present invention, for the convenience of explanation, the cross-sectional views showing the device structure will be enlarged locally out of the general proportion, and the schematic diagrams are only examples and should not limit the scope of protection of the present invention herein. In addition, in actual production, three-dimensional spatial dimensions including length, width and depth should be included.

[0068] Meanwhile, in the description of the present invention, it should be noted that the orientation or positional relationship indicated by terms such as "upper, lower, inner and outer" is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation to the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0069] Unless otherwise clearly defined and limited in the present invention, the terms "mounted, connected and coupled" shall be understood in a broad sense. For example, it may be a fixed connection, a detachable connection or an integral connection; it may also be a mechanical connection, an electrical connection or a direct connection, or may be indirectly connected through an intermediate medium, or may be the communication inside two elements. For persons of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0070] Example 1, referring to Figure 1 , which is the first embodiment of the present invention. This embodiment provides an error change prediction method based on supercapacitor energy storage, including:

[0071] Collect data information, perform intelligent fault detection using a deep learning model, monitor the operating status of the supercapacitor energy storage system, identify and mark abnormal data, and correct and eliminate the marked abnormal data; use the wavelet transform method to perform multi-scale analysis on the corrected data, extract multi-level features of the error change, and further extract non-linear features in the error change through non-linear feature extraction technology; construct a hybrid prediction model, combine random forest, support vector machine and LSTM neural network to predict the error change, deploy a lightweight prediction model on edge devices for real-time data processing and prediction, and upload the results to the cloud.

[0072] The data information collection includes collecting voltage data, current data, temperature data, power data, status data, environmental data, historical operation data, frequency data, capacity data, vibration and noise data; the status data includes the charging status and discharging status of the supercapacitor energy storage unit; the environmental data includes the external environmental data of the operating environment of the supercapacitor energy storage system; the capacity data includes the remaining capacity and charge-discharge cycle times of the supercapacitor energy storage unit.

[0073] The intelligent fault detection using the deep learning model includes constructing a fault detection result function, representing the fault detection result at time t:

[0074] ;

[0075] where, represents the original input data, representing the th convolutional kernel at the th layer for the th input signal; represents the weight matrix, representing the weight parameters of the th convolutional kernel at the th layer, obtained through training data; represents the convolutional kernel parameter, representing the th convolutional kernel at the th layer for the th parameter; represents the bias term, representing the bias parameter of the th convolutional kernel at the th layer, obtained through training data; represents the activation function, selects the ReLU activation function, defined as ; represents the pooling result, representing the pooling result of the th convolutional kernel at the th layer, obtained through the max pooling or average pooling function; Q represents the total number of convolutional kernels; L represents the total number of layers; R represents the total number of input signals, represents the convolutional kernel index, is the layer index, and r is the input signal index.

[0076] The correction and elimination of the marked abnormal data include identifying and marking abnormal data points through the intelligent fault detection of the convolutional neural network. The abnormal data points refer to data deviating from the preset normal range due to sensor failures, communication errors, or unexpected situations.

[0077] Statistical analysis is performed on the collected historical data to calculate the statistical characteristics of each parameter, including the mean, standard deviation, median, and quartiles.

[0078] Data correction is performed on the abnormal data points. For a single abnormal data point, the mean of the historical data is used for replacement:

[0079] ;

[0080] For multiple consecutive abnormal data points, linear interpolation is used to perform linear interpolation based on adjacent normal data points:

[0081] ,

[0082] where is the interpolated data, and are the normal data points before and after the abnormal data segment respectively, is the total number of abnormal data points; is the position of the current abnormal data point in the abnormal data segment, ranging from 1 to .

[0083] The missing data is filled using a sliding window with a window size of . The average value of the data points within the window is used as the filling value:

[0084] ,

[0085] where represents the value of the h-th data point within the sliding window, represents the value of the data point after correction and filling.

[0086] The historical data is polynomially fitted, and the fitting function is used to fill the missing data.

[0087] ,

[0088] where are the polynomial coefficients, is the time; is the highest order of the polynomial, is the order of the polynomial.

[0089] Specifically, in this embodiment, assume that we need to fill in the marked abnormal data to ensure the quality of the input data:

[0090] Suppose there is a set of voltage data as follows, where some data points are marked as abnormal:

[0091] ,

[0092] Among them, 300 and 305 are the data points marked as abnormal.

[0093] Select a sliding window size of 3 (i.e., calculate the average of three data points each time):

[0094] The first window (including the 1st, 2nd, and 3rd data points): , specifically calculated as:

[0095] ;

[0096] The second window (including the 2nd, 3rd, and 4th data points): , specifically calculated as:

[0097] ;

[0098] (Note that 300 here is an abnormal value, but due to the average calculation of the sliding window, the calculation result will be unreliable).

[0099] The third window (including the 3rd, 4th, and 5th data points): , specifically calculated as:

[0100] ;

[0101] The fourth window (including the 4th, 5th, and 6th data points): , specifically calculated as:

[0102] ;

[0103] The fifth window (including the 5th, 6th, and 7th data points): , specifically calculated as:

[0104] .

[0105] Suppose a quadratic polynomial (i.e., a second-order polynomial) is selected for fitting. The form of the quadratic polynomial is:

[0106] ;

[0107] Determine the fitting points and perform fitting using the normal data points before and after the abnormal data points and the average value calculated by the sliding window method: ;

[0108] Perform fitting on these points, assuming the corresponding time points are .

[0109] Construct a fitting equation system based on the selected data points:

[0110] ;

[0111] Solve the equation system to obtain the fitting coefficients :

[0112] ;

[0113] Calculate the correction value of the abnormal data point using the fitted polynomial. Assume the time points corresponding to the abnormal data points are and :

[0114]

[0115] The corrected data sequence using the combination of polynomial fitting method and sliding window method is:

[0116] ; where is the corrected data point value calculated by the sliding window method, is the fitting polynomial function, t is the time point, a0 is the constant term of the fitting polynomial, a1 is the first-order coefficient of the fitting polynomial, a2 is the second-order coefficient of the fitting polynomial, X' is the corrected data point value calculated by the sliding window method, w is the size of the sliding window, 220, 221, 225, 223, 222 are the original data points, 220.7, 221 are the corrected data point values calculated by the fitting polynomial, 248.67, 276.67, 276, 250 are the intermediate results calculated by the sliding window method.

[0117] The wavelet transform method includes decomposing the corrected data into multi-scale components through wavelet transform, and each component represents signal components of different frequencies. The wavelet transform includes discrete wavelet transform and continuous wavelet transform;

[0118] Arrange all the corrected data in time series into , represents the corrected original data time series, including voltage, current, temperature, power; represents the wavelet basis function, is the mother wavelet, obtained by dilation and translation;

[0119] ,

[0120] Among them, represents the scale index, which is used to control the dilation of the wavelet; represents the translation index, which is used to control the position of the wavelet; represents the approximation coefficient of the scale component.

[0121] is the scaling function, defined as:

[0122] ;

[0123] The discrete wavelet transform DWT and the continuous wavelet transform CWT are used to process the data respectively:

[0124] ,

[0125] where a is the dilation factor and b is the translation factor; the scale index and the translation index are determined through the wavelet transform process.

[0126] Wavelet basis function Select the mother wavelet, and select the optimal mother wavelet according to the data characteristics; select the mother wavelet according to the data characteristics , and obtain through dilation and translation;

[0127] Approximation coefficient is obtained by calculating the cumulative result of each scale component, and the formula is:

[0128] For each scale and each translation perform wavelet transform to obtain the multi-scale component , The smaller it is, the smaller the signal component on the corresponding scale; is time, is the scale index, is the translation index, is the maximum scale, is the corrected original data time series, is the scaling function, is the result of the continuous wavelet transform.

[0129] The multi-scale analysis includes analyzing the multi-scale components, extracting the features that affect the error change, and using them as the input of the prediction model. The features that affect the error change include frequency components, amplitudes, and phase information;

[0130] Select the mother wavelet according to the data characteristics , and multi-scale components are obtained through stretching and translation , for each multi-scale component calculate the amplitude and the phase ; calculate the standard deviation of each scale and translation position using historical data , standardize the frequency components; combine the standardized amplitude and phase information to calculate the comprehensive feature function :

[0131] ;

[0132] ;

[0133] where, is the frequency, is the total number of translation positions, and i1 is the imaginary unit; the multi-scale component at the scale and the th translation position obtained by wavelet transform, as a function of the frequency ; represents the standard deviation at the scale and the translation position.

[0134] The extraction of non-linear features in the error change includes processing the input data through a kernel method or an autoencoder to extract non-linear features in a high-dimensional space. The kernel method includes a radial basis function kernel and a polynomial kernel;

[0135] Normalize the input data , select a preset kernel width parameter, and calculate the radial basis function kernel ; select a preset constant and the degree of the polynomial , calculate the polynomial kernel , train the autoencoder through an unsupervised learning method to obtain the weight matrix and bias term;

[0136] Combine the outputs of the radial basis function kernel, polynomial kernel, and autoencoder to calculate the non-linear feature mapping in the high-dimensional space Specifically as follows:

[0137] ,

[0138] where, is the training sample index, is the total number of training samples, is the kernel width parameter; is the degree of the polynomial, indicating the input data x and the training sample Inner product; Is the index of the autoencoder neuron, Is the total number of autoencoder neurons, Is the index of the input data feature, Is the total number of input data features, Is the weight of the radial basis function kernel, Is the weight of the autoencoder, Is the weight matrix of the autoencoder, Is the bias term of the autoencoder, Is the activation function.

[0139] The construction of the hybrid prediction model includes combining random forest, support vector machine and LSTM neural network through the method of ensemble learning to form a hybrid prediction model. The method of ensemble learning includes weighted voting and stacking ensemble strategies;

[0140] Perform single-model predictions for random forest, support vector machine and LSTM neural network, introduce the weighted voting strategy, assign weights to the prediction results of each model, and calculate the mean absolute error MAE of each model based on the predictions of each model:

[0141] ;

[0142] Assign weights according to MAE, and the weight is the reciprocal of MAE:

[0143] ;

[0144] Use the weighted voting method to weight the prediction results of each model:

[0145] ;

[0146] Use the secondary learner Combine the single-model prediction results and calculate the root mean square error RMSE of the secondary learner:

[0147] ;

[0148] Assign weights according to RMSE, and the weight is the reciprocal of RMSE:

[0149] ;

[0150] The final prediction result is obtained by combining the weighted voting result and the secondary learner result:

[0151] ;

[0152] Wherein, Indicates the random forest model for the input data The prediction result; Indicates the prediction result of the support vector machine model for the input data The prediction result; Indicates the prediction result of the long short-term memory network model for the input data The prediction result; Is the final prediction result; Is the mean absolute error of model f, w f Is the weight of model f, Is the prediction result of the secondary learner for the input data X, Is the root mean square error of the secondary learner s, Is the weight of the secondary learner s, F is the total number of models participating in weighted voting, D is the total number of secondary learners participating in the fusion of secondary learners, s represents the secondary learner index, B represents the total number of samples for calculating the error, Indicates the th sample's true value, used to calculate the prediction error; Indicates the prediction value of the fth model for the th sample.

[0153] Example 2, the second example of the present invention, which is different from the previous example in that:

[0154] If the described function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or a part of this technical solution can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. And the aforementioned storage medium includes: USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs, etc., various media that can store program codes.

[0155] This application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, as well as the combination of flows and / or blocks in the flowchart and / or block diagram. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or a device for implementing the functions specified in multiple blocks.

[0156] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device that implements the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or a device for implementing the functions specified in multiple blocks.

[0157] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or a device for implementing the functions specified in multiple blocks.

[0158] Although the preferred embodiments of the present application have been described, those skilled in the art can make additional changes and modifications once they know the basic creative concepts. Therefore, the appended claims are intended to be construed to include the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0159] Obviously, those skilled in the art can make various changes and variations to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application is also intended to include these modifications and variations.

Claims

1. A method for predicting error variation based on supercapacitor energy storage, characterized in that: Including, Collecting data information, performing intelligent fault detection using a deep learning model, monitoring the operating status of the supercapacitor energy storage system, identifying and marking abnormal data, and correcting and eliminating the marked abnormal data; Performing multi-scale analysis on the corrected data using the wavelet transform method, extracting multi-level features of the error change, and further extracting non-linear features in the error change through non-linear feature extraction technology; Constructing a hybrid prediction model, combining random forest, support vector machine and LSTM neural network for error change prediction, deploying a lightweight prediction model on edge devices for real-time data processing and prediction, and uploading the results to the cloud; The collecting data information includes collecting voltage data, current data, temperature data, power data, status data, environmental data, historical operation data, frequency data, capacity data, vibration and noise data; the status data includes the charging status and discharging status of the supercapacitor energy storage unit; the environmental data includes external environmental data of the operating environment of the supercapacitor energy storage system; the capacity data includes the remaining capacity and charge-discharge cycle times of the supercapacitor energy storage unit.

2. The error change prediction method based on supercapacitor energy storage according to claim 1, wherein: The performing intelligent fault detection using a deep learning model includes constructing a fault detection result function representing the fault detection result at time t: Among them, X qlr represents the original input data, which represents the r-th input signal of the q-th convolutional kernel in the l-th layer; W ql represents the weight matrix, which represents the weight parameter of the q-th convolutional kernel in the l-th layer and is obtained through training data; K qlr represents the convolutional kernel parameter, which represents the r-th parameter of the q-th convolutional kernel in the l-th layer; b ql represents the bias term, which represents the bias parameter of the q-th convolutional kernel in the l-th layer and is obtained through training data; σ represents the activation function, and the ReLU activation function is selected, defined as σ(x) = max(0, x); P ql represents the pooling result, which represents the pooling result of the q-th convolutional kernel in the l-th layer and is obtained through the max pooling or average pooling function; Q represents the total number of convolutional kernels; L represents the total number of layers; R represents the total number of input signals, q represents the convolutional kernel index, l is the layer index, and r is the input signal index.

3. The error change prediction method based on supercapacitor energy storage according to claim 2, characterized in that: The correcting and eliminating the marked abnormal data includes identifying and marking abnormal data points through intelligent fault detection of the convolutional neural network, where the abnormal data points refer to data deviating from the preset normal range due to sensor faults, communication errors or unexpected situations; Performing statistical analysis on the collected historical data, calculating the statistical features of each parameter, including mean, standard deviation, median, quartiles; Correcting the abnormal data points, where for a single abnormal data point, the mean μ of the historical data is used for replacement: X′ = μ; For multiple consecutive abnormal data points, linear interpolation is used to perform linear interpolation according to adjacent normal data points; where X′ is the interpolated data, X prev and X next are the normal data points before and after the abnormal data segment respectively, N is the total number of abnormal data points; k1 is the position of the current abnormal data point in the abnormal data segment, ranging from 1 to N; Using a sliding window to fill in missing data, with the window size of w, and the average value of the data points within the window as the filling value; where X h represents the value of the h-th data point within the sliding window, and X' represents the value of the data point after correction and filling; Performing polynomial fitting on the historical data and using the fitting function to fill in the missing data; where a n' is the polynomial coefficient, t is the time; N' is the highest order of the polynomial, and n' is the order of the polynomial.

4. A method for predicting error variation based on supercapacitor energy storage as claimed in claim 3, wherein: The wavelet transform method includes decomposing the corrected data into multi-scale components through wavelet transform, where each component represents signal components of different frequencies, and the wavelet transform includes discrete wavelet transform and continuous wavelet transform; Arrange all the corrected data into X(t) according to the time series. X(t) represents the time series of the corrected original data, including voltage, current, temperature, and power; ψ j,k (t) represents the wavelet basis function. ψ(t) is the mother wavelet, which is obtained by dilation and translation; ψ j,k (t) = 2 j / 2 ψ(2 j t - k); where j represents the scale index for controlling the dilation of the wavelet, k represents the translation index for controlling the position of the wavelet, and C s denotes the approximation coefficient of the scale component, and φ(t) is the scaling function, defined as: Processing the data using discrete wavelet transform DWT and continuous wavelet transform CWT respectively; where, a is the scaling factor and b is the translation factor; the scale index j and the translation index k are determined through the wavelet transform process; Wavelet basis function ψ j,k (t) selects the mother wavelet and chooses the optimal mother wavelet according to the data characteristics; selects the mother wavelet ψ(t) according to the data characteristics and obtains ψ j,k (t); Approximation coefficient C s Obtained by calculating the cumulative result of each scale component, and the formula is: Perform wavelet transform for each scale j and each translation k to obtain the multi-scale component D s (t). The smaller D s (t) is, the smaller the signal component is at the corresponding scale; t is time, j is the scale index, k is the translation index, J is the maximum scale, X(t) is the corrected original data time series, φ(t) is the scaling function, and W ψ (a, b) is the result of continuous wavelet transform.

5. The error change prediction method based on supercapacitor energy storage according to claim 4, characterized in that: The multi-scale analysis includes analyzing the multi-scale components, extracting the features affecting the error change as the input of the prediction model, and the features affecting the error change include frequency components, amplitude and phase information; Select the mother wavelet ψ(t) according to the data characteristics, and obtain the multi-scale component X j,k (ω) by dilation and translation. Calculate the amplitude |X j,k (ω)| and phase θ j,k (ω) for each multi-scale component X j,k (ω); calculate the standard deviation σ j,k at each scale and translation position using historical data, and standardize the frequency components; combine the standardized amplitude and phase information to calculate the comprehensive characteristic function F(ω): θ j,k (ω) = arg(X j,k (ω)); where ω is the frequency, K is the total number of translation positions, and i1 is the imaginary unit; X j,k (ω) is the multi-scale component at the j-th scale and the k-th translation position obtained by wavelet transform and is a function of the frequency ω; σ j,k represents the standard deviation at the j-th scale and the k-th translation position.

6. The error change prediction method based on supercapacitor energy storage according to claim 5, wherein: The extracting non-linear features in the error change includes processing the input data through the kernel method or autoencoder to extract non-linear features in the high-dimensional space, and the kernel method includes radial basis function kernel and polynomial kernel; Normalize the input data x, select the preset kernel width parameter, and calculate the radial basis function kernel Select the preset constant c and the degree d of the polynomial, and calculate the polynomial kernel (x·x u +c) d , train the autoencoder through an unsupervised learning method to obtain the weight matrix and the bias term; Combine the outputs of the radial basis function kernel, polynomial kernel, and autoencoder to calculate the non-linear feature mapping Φ(x) in the high-dimensional space as follows: where \(u\) is the training sample index, \(U\) is the total number of training samples, and \(\sigma_1\) is the kernel width parameter; \(d\) is the degree of the polynomial, representing the inner product of the input data \(x\) and the training sample \(x\) u ; \(m\) is the autoencoder neuron index, \(M\) is the total number of autoencoder neurons, \(v\) is the input data feature index, \(p\) is the total number of input data features, \(\alpha\) u is the weight of the radial basis function kernel, \(\beta\) m is the weight of the autoencoder, \(W\) mv is the weight matrix of the autoencoder, \(b\) m is the bias term of the autoencoder, and \(\sigma(z)\) is the activation function.

7. The error change prediction method based on supercapacitor energy storage according to claim 6, wherein: The construction of the hybrid prediction model includes combining a random forest, a support vector machine, and an LSTM neural network through an ensemble learning method to form a hybrid prediction model. The ensemble learning method includes weighted voting and stacking ensemble strategies; Perform single-model predictions for the random forest, support vector machine, and LSTM neural network, introduce a weighted voting strategy, assign weights to the prediction results of each model, and calculate the mean absolute error MAE of each model based on the predictions of each model: Assign weights according to MAE, and the weight is the reciprocal of MAE: Use the weighted voting method to weight the prediction results of each model: Use the secondary learner g to combine the single-model prediction results and calculate the root mean square error RMSE of the secondary learner: Assign weights according to RMSE, and the weight is the reciprocal of RMSE: The final prediction result is obtained by combining the weighted voting result and the secondary learner result: where \(f\in\{RF, SVM, LSTM\}\), \(P\) RF represents the prediction result of the random forest model for the input data \(X\); \(P\) SVM represents the prediction result of the support vector machine model for the input data \(X\); \(P\) LSTM represents the prediction result of the long short - term memory network model for the input data \(X\); \(P\) final is the final prediction result; \(MAE\) f is the mean absolute error of the model \(f\), \(w\) f is the weight of the model \(f\), \(g\) s (\(X\)) is the prediction result of the secondary learner for the input data \(X\), \(RMSE\) s is the root mean square error of the secondary learner \(s\), \(w\) s is the weight of the secondary learner \(s\), \(F\) is the total number of models participating in the weighted voting, \(D\) is the total number of secondary learners participating in the secondary learner fusion, \(s\) represents the secondary learner index, \(B\) is the total number of samples for calculating the error, \(y\) b' represents the true value of the \(b'\) - th sample, used to calculate the prediction error; \(P\) f,b' represents the predicted value of the \(b'\) - th sample by the \(f\) - th model.

8. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of a method for predicting error changes based on supercapacitor energy storage according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of a method for predicting error changes based on supercapacitor energy storage according to any one of claims 1 to 7.

Images