Battery charge state estimation method and device of energy storage transformer and medium

Abstract

The invention discloses a battery state-of-charge estimation method and device for an energy storage transformer and a medium. The problem that it is difficult to estimate the state of charge of a battery of an energy storage type transformer is solved. According to the method, the collected operation parameters of the transformer and the energy storage equipment are converted into the input spectrogram of the time-frequency domain, the time sequence and frequency information characteristics are achieved, the analysis result is more accurate, and the generalization ability for different battery states is improved through the converted time-frequency domain data. The combination of the residual block and the uniform spatial pyramid pooling layer optimizes the feature extraction process, improves the performance of the model, and more accurately estimates the state of charge of the energy storage type transformer battery.

Description

Technical Field

[0001] The present application relates to the field of transformers, and in particular to a method, device and medium for estimating the battery state of charge of an energy storage transformer. Background Art

[0002] A transformer is a device that uses the principle of electromagnetic induction to change AC voltage. Energy storage transformers not only provide an effective method for large-scale access of renewable energy to the power grid, but also help improve the comprehensive utilization rate of power grid equipment, system peak and frequency regulation, and reactive power support capabilities, thereby improving the stability, economy, and flexibility of power grid operation. As the basis of the battery management system for energy storage transformers, accurate state estimation can provide precise parameters for subsequent battery management and provide important support for the reliable operation of various types of equipment. State of charge (SOC) is a parameter of battery status that is used to describe the current remaining capacity of the battery. Accurate state of charge estimation can make applications run more safely and efficiently by optimizing power distribution and maintaining the safe operating range of the battery.

[0003] However, many of the methods proposed so far require accurate modeling of the battery and its internal state, which cannot be measured directly, and require additional processes for parameter estimation, relying on the extraction of time-domain features of the battery data.

[0004] It can be seen that how to solve the problem of difficult estimation of the battery state of charge of the energy storage transformer is a technical problem that needs to be solved urgently by people in this field. Summary of the invention

[0005] The purpose of this application is to provide a method, device and medium for estimating the battery state of charge of an energy storage transformer.

[0006] In order to solve the above technical problems, the present application provides a method for estimating the state of charge of a battery of an energy storage transformer, comprising:

[0007] Collect the operating parameters of the transformer and the status parameters of the energy storage equipment through the measuring device;

[0008] Obtaining an input spectrogram in the time-frequency domain according to the operating parameters;

[0009] Inputting the input spectrogram in the time-frequency domain into a uniform spatial pyramid pooling layer to obtain a global feature map;

[0010] Inputting the global feature map into a preset residual block for convolution operation to extract a high-level feature map;

[0011] Inputting the high-level feature map into a global average pooling layer for aggregation to obtain a one-dimensional feature parameter;

[0012] A battery charge estimation result is obtained according to the one-dimensional characteristic parameter.

[0013] As an optional solution, in the above energy storage transformer battery charge state estimation method, the operating parameters of the transformer and the state parameters of the energy storage device are collected by a measuring device, including:

[0014] The voltage value on both sides of the transformer, the current value on both sides of the transformer, and the transformer power are collected by the measuring device;

[0015] The remaining capacity of the energy storage device, the temperature of the charging / discharging battery, and the chamber temperature are collected by the measuring device;

[0016] The voltage values ​​at both sides of the transformer, the current values ​​at both sides of the transformer, the transformer power, the remaining capacity, the charge / discharge battery temperature, and the chamber temperature are used to generate multiple discrete time feature sequences according to the collection time points.

[0017] As an optional solution, in the above energy storage transformer battery charge state estimation method, the step of obtaining the input spectrum in the time-frequency domain according to the operating parameters includes:

[0018] Performing short-time Fourier transform on the discrete-time feature sequence to obtain a spectrum diagram;

[0019] Extracting an amplitude spectrum and a phase spectrum according to the frequency spectrum;

[0020] The extracted amplitude spectrum and the phase spectrum are combined into a two-dimensional image as an input spectrum in the time-frequency domain.

[0021] As an optional solution, in the above energy storage transformer battery state of charge estimation method, the input spectrum in the time-frequency domain is input into a uniform spatial pyramid pooling layer to obtain a global feature map, including:

[0022] Use hole convolution kernels of different scales to perform convolution operations on the input spectrogram in the time-frequency domain to obtain the convolved feature map;

[0023] Concatenate the convolved feature maps;

[0024] The concatenated feature maps are integrated through point-by-point convolution to output the global feature map.

[0025] As an optional solution, in the above energy storage transformer battery charge state estimation method, the global feature map is input into a preset residual block for convolution operation to extract a high-level feature map, including:

[0026] Deploy a depth-separable convolution layer in the residual block to perform convolution processing on the global feature map; wherein each residual block includes multiple depth-separable convolution layers;

[0027] The convolution result output by the depthwise separable convolutional layer is normalized and processed with an activation function to output a high-level feature map.

[0028] As an optional solution, in the above energy storage transformer battery state of charge estimation method, the high-level feature map is input into the global average pooling layer for aggregation to obtain a one-dimensional feature parameter including:

[0029] The high-level feature map output by the residual block is aggregated through the global average pooling layer to obtain a one-dimensional feature parameter;

[0030] Correspondingly, obtaining a battery charge estimation result according to the one-dimensional characteristic parameter includes:

[0031] Performing linear regression processing on the one-dimensional feature parameters through a fully connected layer;

[0032] The characteristic parameters output after linear regression are activated through a truncated rectified linear unit to obtain an estimated value of the battery state of charge.

[0033] As an optional solution, in the above energy storage transformer battery charge state estimation method, after generating multiple discrete time feature sequences, it also includes:

[0034] Filter multiple discrete-time feature sequences.

[0035] In order to solve the above technical problems, the present application also provides a device for estimating the state of charge of a battery of an energy storage transformer, comprising:

[0036] A collection module, used to collect operating parameters of the transformer and status parameters of the energy storage device through a measuring device;

[0037] A conversion module, used for obtaining an input spectrogram in the time-frequency domain according to the operating parameters;

[0038] A pooling module, used for inputting the input spectrogram in the time-frequency domain into a uniform spatial pyramid pooling layer to obtain a global feature map;

[0039] An extraction module, used for inputting the global feature map into a preset residual block for convolution operation to extract a high-level feature map;

[0040] An aggregation module, used for inputting the high-level feature map into a global average pooling layer for aggregation to obtain a one-dimensional feature parameter;

[0041] The estimation module is used to obtain a battery charge estimation result according to the one-dimensional characteristic parameter.

[0042] In order to solve the above technical problems, the present application also provides a device for estimating the state of charge of a battery of an energy storage transformer, comprising:

[0043] Memory for storing computer programs;

[0044] A processor is used to implement the steps of the above-mentioned method for estimating the state of charge of the battery of the energy storage transformer when executing the computer program.

[0045] In order to solve the above technical problems, the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned method for estimating the state of charge of the energy storage transformer battery are implemented.

[0046] The method for estimating the state of charge of the battery of the energy storage transformer provided in the present application includes: collecting the operating parameters of the transformer and the state parameters of the energy storage device through a measuring device; obtaining an input spectrum in the time-frequency domain according to the operating parameters; inputting the input spectrum in the time-frequency domain into a uniform space pyramid pooling layer to obtain a global feature map; inputting the global feature map into a preset residual block for convolution operation to extract a high-level feature map; inputting the high-level feature map into a global average pooling layer for aggregation to obtain a one-dimensional feature parameter; obtaining a battery charge estimation result according to the one-dimensional feature parameter. The present application converts the collected operating parameters of the transformer and the energy storage device into an input spectrum in the time-frequency domain, which has the characteristics of time series and frequency information, so that the analysis results are more accurate, and the converted time-frequency domain data improves the generalization ability of different battery states. The combination of the residual block and the uniform space pyramid pooling layer optimizes the feature extraction process, improves the performance of the model, and more accurately estimates the state of charge of the battery of the energy storage transformer.

[0047] In addition, the present application also provides a device and a medium, which correspond to the above-mentioned method for estimating the state of charge of the battery of the energy storage transformer, and have the same effect as above. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0049] Figure 1 A flow chart of a method for estimating the state of charge of an energy storage transformer battery provided in an embodiment of the present application;

[0050] Figure 2 A simulation schematic diagram of a method for estimating the state of charge of an energy storage transformer battery provided in an embodiment of the present application;

[0051] Figure 3 A structural diagram of a device for estimating the state of charge of an energy storage transformer battery provided in an embodiment of the present application;

[0052] Figure 4 A structural diagram of another energy storage transformer battery charge state estimation device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0053] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0054] The core of this application is to provide a method, device and medium for estimating the battery state of charge of an energy storage transformer.

[0055] In order to enable those skilled in the art to better understand the present application, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0056] In order to solve the above technical problems, the present application provides a method for estimating the state of charge of a battery of an energy storage transformer, such as Figure 1 As shown, including:

[0057] S11: collecting operating parameters of the transformer and status parameters of the energy storage device through a measuring device;

[0058] S12: Obtaining an input spectrogram in the time-frequency domain according to the operating parameters;

[0059] S13: Input the input spectrogram in the time-frequency domain into the uniform spatial pyramid pooling layer to obtain a global feature map;

[0060] S14: input the global feature map into a preset residual block for convolution operation to extract a high-level feature map;

[0061] S15: Input the high-level feature map into the global average pooling layer for aggregation to obtain one-dimensional feature parameters;

[0062] S16: Obtaining a battery charge estimation result according to the one-dimensional characteristic parameter.

[0063] In step S11, the operating parameters of the transformer and the state parameters of the energy storage device are collected by the measuring device, which can be specifically the electrical parameters such as the voltage, current, and power of the transformer, as well as the remaining capacity of the energy storage device, the temperature of the charging and discharging battery, the chamber temperature and other state parameters. These parameters are used for subsequent state of charge estimation and provide basic data for battery management; in this step, by real-time monitoring of the key parameters of the transformer and the energy storage device, the accuracy and real-time nature of the data can be ensured, which is crucial for subsequent data processing and state of charge estimation; because the state of charge of the battery needs to be accurately estimated, data collection is performed when the device is in operation.

[0064] In step S12, obtaining the input spectrogram in the time-frequency domain according to the operating parameters means converting the collected time-domain signals (such as voltage and current) into the frequency domain through short-time Fourier transform (STFT) to obtain a spectrum. The input spectrogram in the time-frequency domain is used as the input of the neural network model to extract the time-frequency features; in this step, the time-frequency domain conversion is realized through STFT, which can dig out the deep-level features of the battery state change; because the frequency domain features can provide more information about the battery state change, feature extraction is selected in the frequency domain.

[0065] In step S13, the input spectrogram in the time-frequency domain is input into the uniform spatial pyramid pooling layer, and obtaining the global feature map refers to using the uniform spatial pyramid pooling (ASPP) structure in deep learning to process the input spectrogram to capture feature information of different scales. The global feature map is used to extract global and local features in the spectrogram; in this step, through the processing of the ASPP layer, the multi-scale features of the spectrogram can be captured, and the feature extraction capability of the model can be enhanced; because the features of different scales are complementary to the description of the battery state, the ASPP layer is selected to extract the global features.

[0066] In step S14, the global feature map is input into the preset residual block for convolution operation. Extracting high-level feature maps refers to using the residual block in the residual network (ResNet) to further convolve the feature map output by the ASPP layer to extract deeper features. High-level feature maps are used to capture more complex patterns and relationships in the spectrum graph. In this step, the deep convolution of the residual block can improve the model's ability to extract high-level features and enhance the model's expressiveness. Because deep networks are prone to the vanishing gradient problem, residual connections are chosen to alleviate this problem.

[0067] In step S15, the high-level feature map is input into the global average pooling layer for aggregation to obtain one-dimensional feature parameters. Specifically, the high-level feature map output by the residual block is spatially aggregated using the global average pooling (GAP) layer to obtain a one-dimensional feature vector. The one-dimensional feature parameters are used for the final state of charge estimation. In this step, the spatial features can be compressed into one-dimensional features through the aggregation of the GAP layer to simplify the model complexity. Because global average pooling can reduce overfitting and improve the generalization ability of the model, the GAP layer is selected for feature aggregation.

[0068] In step S16, obtaining the battery charge estimation result according to the one-dimensional feature parameter means using one or more fully connected layers to process the one-dimensional feature parameter and finally output the battery state of charge estimation value. The battery charge estimation result is used to guide the battery charge and discharge management; in this step, the accurate estimation of the battery state of charge can be obtained through the linear or nonlinear transformation of the fully connected layer; because the battery state of charge is crucial to battery management, the state of charge estimation result is output in the last step of the model.

[0069] The method for estimating the state of charge of the battery of the energy storage transformer provided by the embodiment of the present application includes: collecting the operating parameters of the transformer and the state parameters of the energy storage device through a measuring device; obtaining an input spectrum in the time-frequency domain according to the operating parameters; inputting the input spectrum in the time-frequency domain into a uniform space pyramid pooling layer to obtain a global feature map; inputting the global feature map into a preset residual block for convolution operation to extract a high-level feature map; inputting the high-level feature map into a global average pooling layer for aggregation to obtain a one-dimensional feature parameter; obtaining a battery charge estimation result according to the one-dimensional feature parameter. The collected operating parameters of the transformer and the energy storage device are converted into an input spectrum in the time-frequency domain, which has the characteristics of time series and frequency information, so that the analysis results are more accurate, and the converted time-frequency domain data improves the generalization ability of different battery states. The combination of the residual block and the uniform space pyramid pooling layer optimizes the feature extraction process, improves the performance of the model, and more accurately estimates the state of charge of the battery of the energy storage transformer.

[0070] According to the above embodiment, in a specific embodiment, a method for estimating the state of charge of a battery of an energy storage transformer collects operating parameters of the transformer and state parameters of the energy storage device through a measuring device, including:

[0071] The voltage value on both sides of the transformer, the current value on both sides of the transformer, and the transformer power are collected by the measuring device;

[0072] The remaining capacity of the energy storage device, the charging / discharging battery temperature, and the chamber temperature are collected by the measuring device;

[0073] The voltage values ​​on both sides of the transformer, the current values ​​on both sides of the transformer, the transformer power, the remaining capacity, the charge / discharge battery temperature, and the chamber temperature are used to generate multiple discrete time feature sequences according to the acquisition time points.

[0074] The measuring device is used to accurately collect the voltage and current values ​​on both sides of the transformer, as well as the real-time power of the transformer. The measuring device is also used to collect the remaining capacity of the energy storage device, the battery temperature during charging and discharging, and the chamber temperature. These parameters help to understand the current state and health of the battery. This technology does not limit the specific structure of the measuring device, and it can set up appropriate collection equipment according to actual needs.

[0075] Arrange the collected parameter values ​​in time order to generate multiple discrete time feature sequences. These sequences include not only the changes in a single parameter, but also the relationship between parameters, such as the relationship between power and current.

[0076] The generated time series data can further be subjected to outlier detection and processing to ensure the accuracy and reliability of the data.

[0077] Furthermore, data of different dimensions and magnitudes can be normalized so that the data are on the same scale, which is convenient for subsequent analysis and processing.

[0078] According to the above embodiment, in a specific embodiment, obtaining an input spectrogram in the time-frequency domain according to the operating parameters includes:

[0079] Perform short-time Fourier transform on the discrete-time feature sequence to obtain a spectrum diagram;

[0080] According to the spectrum diagram, an amplitude spectrum diagram and a phase spectrum diagram are extracted;

[0081] The extracted amplitude spectrum and phase spectrum are combined into a two-dimensional image as the input spectrum in the time-frequency domain.

[0082] The collected discrete-time feature sequence (voltage value, current value, power on both sides of the transformer, remaining capacity of the energy storage device, charging and discharging battery temperature, and chamber temperature) is subjected to short-time Fourier transform to obtain a spectrum diagram.

[0083] Short-time Fourier transform (STFT) is applied to each discrete time feature sequence to convert the time domain signal into a time-frequency domain signal and obtain a spectrum diagram, which reveals the frequency components and time variation characteristics of the signal.

[0084] Amplitude spectrum and phase spectrum extraction: The amplitude spectrum and phase spectrum are extracted from the complex-valued coefficients obtained by STFT. These two spectra reflect the intensity and phase information of the signal respectively. The amplitude spectrum represents the intensity of the signal at each frequency, while the phase spectrum represents the phase information of the signal at each frequency. These spectra are extracted to capture the energy distribution and phase changes of the signal respectively.

[0085] For example, the transformer voltage signal v(k) is a discrete time series Two types of frequency components can be extracted:

[0086] ;

[0087] ;

[0088] in, represents the amplitude of the spectrum, represents the phase of the spectrum, and represent the real and imaginary parts of the complex coefficients respectively.

[0089] The amplitude spectrogram is used as the input of the proposed neural network, which represents the energy distribution of the signal in the frequency domain and can be easily processed by general neural network operations. As the amplitude spectrum of frame m, where F represents the number of frequency components. Therefore, the amplitude spectrum at X(m) can be expressed as: .

[0090] Above, represents the measurement vector of voltage v, current i and temperature t at frame m. Represent the measured values ​​of voltage v, current i and temperature t at frame m respectively.

[0091] Function mapping between input and output It can be expressed as:

[0092] ;

[0093] Where T is the number of spectrogram frames, f Φ :R 3×F×T →R 1 The sequence of magnitude spectrogram frames forms a two-dimensional tensor, which can be visualized as an image, the spectrogram. In addition, the channel-concatenated spectrogram constructs a three-dimensional tensor, similar to an RGB (a color representation method, red, green, blue) image.

[0094] The spectrograms were examined by mapping them into RGB images for different driving cycles and conditions to verify that the concatenated spectrograms were suitable network inputs.

[0095] A model based on a 2D convolutional neural network (CNN) is built using the spectrogram input as the optimal descriptor of f. The spectrogram as network input has the following advantages. First, the Fourier transform provides frequency information that is difficult to capture with a one-dimensional signal. Second, the input data can be enriched using the SpecAugment method, a data augmentation method for speech recognition and audio classification tasks, to avoid overfitting. However, the time and frequency resolution STFT is a trade-off. Therefore, the STFT kernels should be carefully selected as they are related to the computational load and accuracy of the model. Time series input of appropriate length is also required to properly capture the dynamics from the measurements.

[0096] The extracted amplitude spectrum and phase spectrum are merged into a two-dimensional image as the input spectrum in the time-frequency domain. Merging refers to integrating the information of the two spectra into one two-dimensional image. Specifically, the amplitude spectrum and the phase spectrum are used as the two channels of the image. The merged two-dimensional image is used as the input of the deep learning model to extract time-frequency features.

[0097] According to the above embodiment, in a specific embodiment, the input spectrogram in the time-frequency domain is input into the uniform spatial pyramid pooling layer to obtain a global feature map, including:

[0098] Use hole convolution kernels of different scales to perform convolution operations on the input spectrogram in the time-frequency domain to obtain the convolved feature map;

[0099] Concatenate the convolved feature maps;

[0100] The concatenated feature maps are integrated through point-by-point convolution to output the global feature map.

[0101] The uniform spatial pyramid pooling layer (ASPP) uses multiple dilated convolutions to expand its receptive field by inserting zeros between kernel elements. Through feature concatenation, point-by-point convolution is performed, and dilated convolution is used to aggregate multi-scale features.

[0102] A dilated convolution kernel is a convolution kernel with gaps (holes) between its elements. It can be specifically a convolution kernel with different dilation rates, which is used to capture features of different scales. The dilated convolution kernel is used to capture features in the input spectrogram at different scales. By applying dilated convolution kernels of different scales, local details and global context information can be captured simultaneously, enhancing the model's ability to perceive features of different scales. The choice of using dilated convolution kernels of different scales is because the signal in the time-frequency domain may contain useful information at different scales, and a single-scale convolution may not be able to fully capture this information.

[0103] ASPP is used as a replacement for the input convolutional layer in the traditional 2D convolutional neural network model to capture the long-term and short-term dynamics encoded in the spectrogram. The following modifications are made. First, considering the feature map shape of the spectrogram, an asymmetric attribute rate {(1,2),(1,4)} is adopted. Second, a GAP layer is added in the parallel branch to directly capture the global information. Finally, a skip connection is introduced from the time series input as an additional branch of ASPP. Considering that the input to the amplitude spectrogram is not a 1:1 mapping, it leads to information loss. However, direct skip connection is not possible because the spectrogram and time series input shapes can vary. Therefore, one-dimensional average pooling and expansion are used to match the width and height of the spectrogram, respectively. The kernel size and stride of the one-dimensional pooling are the same as the window and jump length of the STFT kernel.

[0104] The input of the ASPP block is the concatenated spectrograms of the time series input. First, the convolution layer generates a feature map from the input spectrogram. The feature map is then fed back to a parallel branch consisting of three convolutions and a GAP layer. Each branch generates a feature map, and all feature maps are concatenated to obtain a global feature map. Finally, the output feature map is generated using point-by-point convolution.

[0105] Multiple feature maps obtained after processing with different-scale dilated convolution kernels are concatenated. Multiple feature maps are merged along the depth direction (i.e., channel direction) to form a deeper feature map. Features extracted by convolution kernels of different scales are integrated so that subsequent processing can consider multi-scale information at the same time.

[0106] According to the above embodiment, in a specific embodiment, the global feature map is input into a preset residual block for convolution operation to extract a high-level feature map, including:

[0107] Deploy a depth-wise separable convolutional layer in the residual block to perform convolution processing on the global feature map; each residual block contains multiple depth-wise separable convolutional layers;

[0108] The convolution result output by the depthwise separable convolutional layer is normalized and processed with an activation function to output a high-level feature map.

[0109] The input is sent to the seven pre-deployed residual blocks (ResBlock) to extract high-level features. All convolution operations in the residual blocks are replaced with depthwise separable convolution (DWS CNN). Deploy the depthwise separable convolution layer in the residual block to perform convolution on the global feature map.

[0110] The residual connection (or skip connection) introduced in ResNet is widely used to improve the gradient flow and the convexity of the loss function. The residual block consists of two convolutional layers with a residual path. Normalization is performed after each convolutional layer to alleviate the internal covariate shift problem. At the same time, instead of the standard ReLU, mish (a deep learning activation function) is selected as the activation function. Each residual block contains multiple depth-wise separable convolutional layers, which process the global feature map in turn. By stacking multiple depth-wise separable convolutional layers in the residual block, more abstract features can be gradually extracted while maintaining the integrity of the features through the residual connection.

[0111] Depthwise separable convolution is a well-known modification of the standard convolution operation. By reducing the number of parameters, features can be extracted efficiently by separating the image into depthwise convolution and pointwise convolution. Since the input of the model consists of images concatenated from three domains, it can be processed efficiently with separable convolution. Therefore, all convolutional layers in the proposed model are replaced to reduce the number of parameters.

[0112] Through normalization, the problem of internal covariate shift can be reduced and the convergence of the model can be accelerated; through the activation function, nonlinearity can be introduced, enabling the model to learn more complex feature representations.

[0113] According to the above embodiment, in a specific embodiment, the high-level feature map is input into the global average pooling layer for aggregation, and the one-dimensional feature parameters obtained include:

[0114] The high-level feature map output by the residual block is aggregated through the global average pooling layer to obtain a one-dimensional feature parameter;

[0115] Correspondingly, the battery charge estimation result is obtained according to the one-dimensional characteristic parameters, including:

[0116] The one-dimensional feature parameters are processed by linear regression through a fully connected layer;

[0117] The characteristic parameters output after linear regression are activated through a truncated rectified linear unit to obtain an estimated value of the battery state of charge.

[0118] The high-level feature map output by the residual block is aggregated through the Global Average Pooling (GAP) layer to obtain a one-dimensional feature parameter. The global average pooling layer is a pooling technique that calculates the average value of each channel of the entire feature map and converts the high-dimensional feature map into a one-dimensional feature vector. Extracting global features avoids feature loss that may be caused by traditional pooling methods, while reducing the number of parameters and computational complexity of the model. Through the GAP layer, the spatial information (height and width) of each channel is integrated into a single value to obtain a one-dimensional feature parameter containing global information.

[0119] The extracted features are input into the global average pooling (GAP) layer to aggregate the remaining spatial features to obtain one-dimensional features. A single fully connected (FC) layer is used for linear regression of battery SOC estimation. The FC layer is followed by a truncated rectified linear unit (ReLU) activation with an upper limit of 1.0. For each block and layer, the (C×H×W) in the left column represents the channel, height, and width of the output feature map, respectively.

[0120] The fully connected layer is used to map the one-dimensional feature parameters to the prediction space of the battery state of charge, realizing the linear transformation from feature to prediction. The one-dimensional feature parameters are linearly combined to generate the preliminary prediction value of the battery state of charge.

[0121] The activation function is used to introduce nonlinearity and limit the output to a reasonable range, making the output more consistent with the actual value range of the battery state of charge. By truncating the ReLU, the output of the fully connected layer is limited to the range of [0, 1] to obtain the final battery state of charge estimate, which represents the current charge percentage of the battery.

[0122] Figure 2 A simulation schematic diagram of a method for estimating the state of charge of an energy storage transformer battery provided in an embodiment of the present application is shown in FIG. Figure 2 As shown, according to the data in the measurement device, a convolutional layer is used to generate a feature map from the input spectrogram; then, the uniform spatial pyramid pooling layer (ASPP) captures long-term and short-term dynamics by introducing multi-field convolution; then, seven residual blocks are deployed to extract high-level features; finally, the extracted features are input into the global average pooling layer, the remaining spatial features are aggregated to obtain one-dimensional features, and the energy storage transformer is controlled.

[0123] According to the above embodiment, in a specific embodiment, after generating a plurality of discrete time feature sequences, the method further includes:

[0124] Filter multiple discrete-time feature sequences.

[0125] The multiple discrete time characteristic sequences of transformers and energy storage devices collected are filtered to improve data quality and remove noise and interference that may affect subsequent analysis and model performance.

[0126] It is necessary to select a suitable filtering algorithm to process the discrete time feature sequence. The filtering algorithm includes but is not limited to Butterworth filter, Chebyshev filter, elliptic filter, etc. According to the characteristics of the signal and the frequency range of the noise, a suitable filter is selected and designed to filter each discrete time feature sequence.

[0127] The quality of filtering directly affects the effect of STFT and the accuracy of subsequent feature extraction, which in turn affects the performance of battery state of charge estimation. Ensure that the filtered signal can accurately reflect the actual operating status of the transformer and energy storage equipment, and provide high-quality data for subsequent time-frequency analysis.

[0128] In the above embodiments, the method for estimating the state of charge of the battery of the energy storage transformer is described in detail, and the present application also provides an embodiment corresponding to the device for estimating the state of charge of the battery of the energy storage transformer. It should be noted that the present application describes the embodiments of the device part from two perspectives, one is based on the perspective of the functional module, and the other is based on the perspective of the hardware.

[0129] From the perspective of functional modules, Figure 3 A structural diagram of a device for estimating the state of charge of a battery of an energy storage transformer provided in an embodiment of the present application, such as Figure 3 As shown, including:

[0130] The acquisition module 21 is used to acquire the operating parameters of the transformer and the state parameters of the energy storage device through a measuring device;

[0131] A conversion module 22, used to obtain an input spectrogram in the time-frequency domain according to the operating parameters;

[0132] A pooling module 23 is used to input the input spectrogram in the time-frequency domain into a uniform spatial pyramid pooling layer to obtain a global feature map;

[0133] An extraction module 24 is used to input the global feature map into a preset residual block for convolution operation to extract a high-level feature map;

[0134] Aggregation module 25, used for inputting high-level feature maps into a global average pooling layer for aggregation to obtain one-dimensional feature parameters;

[0135] The estimation module 26 is used to obtain a battery charge estimation result according to the one-dimensional characteristic parameter.

[0136] Since the embodiments of the apparatus part correspond to the embodiments of the method part, please refer to the description of the embodiments of the method part for the embodiments of the apparatus part, which will not be repeated here.

[0137] Figure 4 A structural diagram of another energy storage transformer battery charge state estimation device provided in an embodiment of the present application, such as Figure 4 As shown, the energy storage transformer battery charge state estimation device includes: a memory 30 for storing a computer program;

[0138] The processor 31 is used to implement the steps of the method for obtaining user operation habit information in the above embodiment (energy storage transformer battery charge state estimation method) when executing the computer program.

[0139] The energy storage transformer battery charge state estimation device provided in this embodiment may include but is not limited to a smart phone, a tablet computer, a laptop computer or a desktop computer.

[0140] Among them, the processor 31 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 31 may be implemented in at least one hardware form of a digital signal processor (DSP), a field-programmable gate array (FPGA), and a programmable logic array (PLA). The processor 31 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a central processing unit (CPU); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 31 may be integrated with a graphics processing unit (GPU), and the GPU is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 31 may also include an artificial intelligence (AI) processor, which is used to process computing operations related to machine learning.

[0141] The memory 30 may include one or more computer-readable storage media, which may be non-transitory. The memory 30 may also include a high-speed random access memory, and a non-volatile memory, such as one or more disk storage devices, flash memory storage devices. In this embodiment, the memory 30 is at least used to store the following computer program 301, wherein, after the computer program is loaded and executed by the processor 31, it can implement the relevant steps of the energy storage transformer battery state of charge estimation method disclosed in any of the aforementioned embodiments. In addition, the resources stored in the memory 30 may also include an operating system 302 and data 303, etc., and the storage method may be temporary storage or permanent storage. Among them, the operating system 302 may include Windows, Unix, Linux, etc. Data 303 may include but is not limited to data involved in implementing the energy storage transformer battery state of charge estimation method, etc.

[0142] In some embodiments, the energy storage transformer battery charge state estimation device may further include a display screen 32 , an input and output interface 33 , a communication interface 34 , a power supply 35 , and a communication bus 36 .

[0143] Those skilled in the art will understand that Figure 4 The structure shown in the figure does not constitute a limitation on the device for estimating the state of charge of a battery of an energy storage transformer, and may include more or fewer components than those shown in the figure.

[0144] The energy storage transformer battery state of charge estimation device provided in the embodiment of the present application includes a memory and a processor. When the processor executes the program stored in the memory, it can implement the following method: energy storage transformer battery state of charge estimation method.

[0145] Finally, the present application also provides an embodiment corresponding to a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps described in the embodiment of the method for estimating the state of charge of a battery of an energy storage transformer are implemented.

[0146] It is understandable that if the method in the above embodiment 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 this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0147] The computer-readable storage medium provided in this embodiment stores a computer program, and when a processor executes the program, the following method can be implemented: a method for estimating the state of charge of a battery of an energy storage transformer.

[0148] The above is a detailed introduction to the battery state of charge estimation method, device and medium of the energy storage transformer provided by the present application. The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same and similar parts between the embodiments can be referenced to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of the present application, several improvements and modifications can be made to the present application, and these improvements and modifications also fall within the scope of protection of the claims of the present application.

[0149] It should also be noted that, in this specification, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the statement "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device including the element.

Claims

1. A method for estimating the state of charge of a battery of an energy storage transformer, characterized in that: include: Collect the operating parameters of the transformer and the status parameters of the energy storage equipment through the measuring device; Obtaining an input spectrogram in the time-frequency domain according to the operating parameters; Inputting the input spectrogram in the time-frequency domain into a uniform spatial pyramid pooling layer to obtain a global feature map; Inputting the global feature map into a preset residual block for convolution operation to extract a high-level feature map; Inputting the high-level feature map into a global average pooling layer for aggregation to obtain a one-dimensional feature parameter; A battery charge estimation result is obtained according to the one-dimensional characteristic parameter.

2. The method for estimating the state of charge of a battery of an energy storage transformer according to claim 1, characterized in that: The measuring device collects the operating parameters of the transformer and the status parameters of the energy storage device, including: The voltage value on both sides of the transformer, the current value on both sides of the transformer, and the transformer power are collected by the measuring device; The remaining capacity of the energy storage device, the charging / discharging battery temperature, and the chamber temperature are collected by the measuring device; The voltage values ​​at both sides of the transformer, the current values ​​at both sides of the transformer, the transformer power, the remaining capacity, the charge / discharge battery temperature, and the chamber temperature are used to generate multiple discrete time feature sequences according to the collection time points.

3. The method for estimating the state of charge of the energy storage transformer battery according to claim 2, characterized in that: The step of obtaining an input spectrogram in the time-frequency domain according to the operating parameters comprises: Performing short-time Fourier transform on the discrete-time feature sequence to obtain a spectrum diagram; Extracting an amplitude spectrum and a phase spectrum according to the frequency spectrum; The extracted amplitude spectrum and the phase spectrum are combined into a two-dimensional image as an input spectrum in the time-frequency domain.

4. The method for estimating the state of charge of the energy storage transformer battery according to claim 3, characterized in that: The input spectrogram in the time-frequency domain is input into the uniform spatial pyramid pooling layer to obtain a global feature map, including: Use hole convolution kernels of different scales to perform convolution operations on the input spectrogram in the time-frequency domain to obtain the convolved feature map; Concatenate the convolutional feature maps; The concatenated feature maps are integrated through point-by-point convolution to output the global feature map.

5. The method for estimating the state of charge of the energy storage transformer battery according to claim 4, characterized in that: The global feature map is input into a preset residual block for convolution operation to extract a high-level feature map, including: Deploy a depth-separable convolution layer in the residual block to perform convolution processing on the global feature map; wherein each residual block includes multiple depth-separable convolution layers; The convolution result output by the depthwise separable convolutional layer is normalized and processed with an activation function to output a high-level feature map.

6. The method for estimating the state of charge of the energy storage transformer battery according to claim 5, characterized in that: The step of inputting the high-level feature map into the global average pooling layer for aggregation to obtain one-dimensional feature parameters includes: The high-level feature map output by the residual block is aggregated through the global average pooling layer to obtain a one-dimensional feature parameter; Correspondingly, obtaining a battery charge estimation result according to the one-dimensional characteristic parameter includes: Performing linear regression processing on the one-dimensional feature parameters through a fully connected layer; The characteristic parameters output after linear regression are activated through a truncated rectified linear unit to obtain an estimated value of the battery state of charge.

7. The method for estimating the state of charge of the energy storage transformer battery according to claim 6, characterized in that: After generating multiple discrete time feature sequences, it also includes: Filter multiple discrete-time feature sequences.

8. A device for estimating the state of charge of a battery of an energy storage transformer, characterized in that: include: A collection module, used to collect operating parameters of the transformer and status parameters of the energy storage device through a measuring device; A conversion module, used for obtaining an input spectrogram in the time-frequency domain according to the operating parameters; A pooling module, used for inputting the input spectrogram in the time-frequency domain into a uniform spatial pyramid pooling layer to obtain a global feature map; An extraction module, used for inputting the global feature map into a preset residual block for convolution operation to extract a high-level feature map; An aggregation module, used for inputting the high-level feature map into a global average pooling layer for aggregation to obtain a one-dimensional feature parameter; The estimation module is used to obtain a battery charge estimation result according to the one-dimensional characteristic parameter.

9. A device for estimating the state of charge of a battery of an energy storage transformer, characterized in that: include: Memory for storing computer programs; A processor is used to implement the steps of the method for estimating the state of charge of the energy storage transformer battery as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for estimating the state of charge of a battery of an energy storage transformer are implemented as described in any one of claims 1 to 7.

Images