Energy storage prediction method, system and device

Abstract

The application provides an energy storage prediction method, system and device, and relates to the technical field of electric energy storage.The energy storage prediction method comprises the following steps: obtaining a bus voltage signal in a current time period through a public connection point of a microgrid, and extracting a quantum phase perturbation sequence therefrom; identifying and separating a secondary modulation component of the bus by using an adaptive noise shaping algorithm; performing unsupervised clustering on the secondary modulation component by using a density peak clustering algorithm, obtaining a plurality of aging mode clusters, and extracting statistical features thereof; obtaining an initial residual energy evaluation value by using a variable coefficient regression submodel; performing ampere-hour integration on an instantaneous value sequence, determining an estimated covariance of a statistical mean channel, and obtaining an estimated covariance of a Kalman filtering channel through a Kalman filtering prediction correction cycle; and finally, weighting and fusing the initial residual energy evaluation value according to the estimated covariances of the two channels to obtain a final energy evaluation value.The application improves the adaptability and stability of the prediction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electric energy storage, in particular to an energy storage prediction method, system and device. BACKGROUND

[0002] In the field of microgrid operation control, the residual energy prediction of distributed energy storage generally adopts statistical mean or Kalman filter framework. The original signal is obtained from the monitoring point of the battery side, that is, the current integral and terminal voltage curve, the real-time data is uploaded to the central controller through a special communication link, and the residual energy estimation is completed in the background.

[0003] In the related technology, when the energy storage battery ages or the electrochemical impedance increases, the internal charge redistribution time will modulate the instantaneous power demand of the inverter DC side, and then modulate the microgrid bus twice. The secondary modulation quantity is extremely small and coupled with harmonic non-stationarity, which causes the traditional prediction model based on statistical mean or Kalman filter to completely ignore this weak coupling, thereby causing the error of residual energy estimation to increase sharply in the later stage of battery cycle. SUMMARY

[0004] The problem solved by the present application is how to improve the accuracy of energy storage prediction.

[0005] To solve the above problems, the present application provides an energy storage prediction method, system and device.

[0006] In a first aspect, an energy storage prediction method of the present application comprises:

[0007] Obtaining a bus voltage signal of the microgrid in a current time period through a public connection point of the microgrid;

[0008] Performing phase perturbation extraction according to the bus voltage signal to obtain a quantum phase perturbation sequence;

[0009] Performing identification and separation processing according to the quantum phase perturbation sequence through an adaptive noise shaping algorithm to obtain a secondary modulation component of the bus of the microgrid;

[0010] Performing unsupervised clustering on the secondary modulation component through a density peak clustering algorithm to obtain a plurality of aging mode clusters;

[0011] Performing feature extraction on the secondary modulation component of the aging mode cluster to obtain statistical features of the aging mode cluster;

[0012] Obtaining an initial residual energy evaluation value according to the statistical features of the aging mode cluster through a variable coefficient regression sub-model;

[0013] Integrate the instantaneous value sequence of the bus voltage of the micro-grid in the current time period with a preset time window to determine an estimated covariance of a statistical mean channel, and perform a Kalman wave prediction correction cycle according to the bus voltage to obtain an estimated covariance of a Kalman filter channel;

[0014] According to the estimated covariances corresponding to the Kalman filter channel and the statistical mean channel respectively, the initial residual energy evaluation value is weighted and fused to obtain a final energy evaluation value.

[0015] Optionally, the phase perturbation extraction according to the bus voltage signal to obtain a quantum phase perturbation sequence comprises:

[0016] Performing Hilbert transform on the bus voltage signal to obtain an instantaneous phase sequence;

[0017] According to the fundamental wave phase of the micro-grid, performing difference operation on the instantaneous phase sequence to obtain a phase deviation sequence;

[0018] Removing linear drift in the phase deviation sequence by sliding least squares fitting to obtain a residual phase perturbation sequence;

[0019] Quantum noise screening is performed on the residual phase perturbation sequence to generate the quantum phase perturbation sequence.

[0020] Optionally, the identification and separation processing according to the quantum phase perturbation sequence by the adaptive noise shaping algorithm to obtain the secondary modulation component of the bus of the micro-grid comprises:

[0021] The quantum phase perturbation sequence is divided into a plurality of sub-sequences by a sliding window;

[0022] Wavelet packet decomposition is performed on each of the sub-sequences to obtain a plurality of layers of detail coefficients and approximation coefficients, and adaptive threshold quantization is performed on the plurality of layers of detail coefficients and the approximation coefficients according to energy entropy to obtain a reconstructed perturbation sub-sequence;

[0023] The reconstructed perturbation sub-sequence is input into a noise shaper to perform frequency spectrum shift and shaping on the reconstructed perturbation sub-sequence to obtain a shaped perturbation sub-sequence;

[0024] The shaped perturbation sub-sequence is subjected to Hilbert transform and instantaneous amplitude envelope extraction to obtain the secondary modulation component.

[0025] Optionally, the unsupervised clustering of the secondary modulation component by the density peak value clustering algorithm to obtain a plurality of aging mode clusters comprises:

[0026] Performing short-time Fourier transform on the secondary modulation component to obtain a time-frequency energy spectrum;

[0027] calculate a local density and a relative distance of each preset time point in the current time period based on the time-frequency energy spectrum, and construct a density-distance decision graph according to the local density and the relative distance;

[0028] According to the density-distance decision graph, determine the clustering center, and perform clustering according to the clustering center to obtain a plurality of the aging mode clusters.

[0029] Optionally, the feature extraction is performed on the secondary modulation component of the aging mode cluster to obtain the statistical features of the aging mode cluster, including:

[0030] The mean value, peak factor, kurtosis, entropy value and linear fitting slope of the instantaneous amplitude of each secondary modulation component of the aging mode cluster are calculated respectively, and the mean value, peak factor, kurtosis, entropy value and linear fitting slope are taken as corresponding dimensions to generate the original feature vector of the aging mode cluster;

[0031] The original feature vector is subjected to Z-score standardization, and the original feature vector after standardization is subjected to dimension reduction processing through principal component analysis to obtain the statistical feature vector of the aging mode cluster.

[0032] Optionally, the initial residual energy evaluation value is obtained from the statistical features of the aging mode cluster through the variable coefficient regression sub-model, including:

[0033] The statistical feature vector of the aging mode cluster is taken as the input of the variable coefficient regression sub-model, the features of each dimension of the statistical feature vector are weighted by a radial basis kernel, and the instantaneous nonlinear coefficient corresponding to each dimension is obtained;

[0034] The instantaneous nonlinear coefficient is multiplied by the feature value corresponding to the dimension item by item and accumulated to obtain the initial residual energy evaluation value corresponding to the aging mode cluster, and the variable coefficient regression sub-model is output.

[0035] Optionally, the instantaneous value sequence of the microgrid bus voltage in the current time period is integrated with the preset time window to determine the estimated covariance of the statistical mean channel, and the Kalman wave prediction correction cycle is performed according to the bus voltage to obtain the estimated covariance of the Kalman filtering channel, including:

[0036] The instantaneous value sequence is integrated with the preset time window to obtain an integrated charge quantity sequence, and the sample variance of the integrated charge quantity sequence is determined;

[0037] The sample variance is taken as the estimated covariance of the statistical mean channel;

[0038] obtaining a priori covariance and a priori estimation of residual energy by Kalman filtering prediction through a state space model;

[0039] obtaining posteriori covariance by Kalman filtering correction according to the instantaneous amplitude of the bus voltage, in combination with the a priori covariance;

[0040] taking the posteriori covariance as the estimation covariance of the Kalman filtering channel.

[0041] Optionally, the final energy evaluation value is obtained by weighting and fusing the initial residual energy evaluation value according to the estimation covariance corresponding to the Kalman filtering channel and the statistical mean channel respectively, comprising:

[0042] obtaining a first weight coefficient according to the estimation covariance of the Kalman filtering channel;

[0043] obtaining a second weight coefficient according to the estimation covariance of the statistical mean channel;

[0044] taking the sum of the first weight coefficient and the second weight coefficient as a normalization factor to obtain a weighting coefficient pair;

[0045] the final energy evaluation value is obtained by weighting and summing the initial residual energy evaluation value and the a priori estimation of residual energy through the weighting coefficient pair.

[0046] In a second aspect, an energy storage prediction system is provided, comprising:

[0047] a signal acquisition module for acquiring a bus voltage signal of a microgrid in a current time period through a public connection point of the microgrid;

[0048] a phase perturbation module for extracting quantum phase perturbation sequences according to the bus voltage signal;

[0049] a modulation component separation module for identifying and separating the modulation components of the bus of the microgrid according to the quantum phase perturbation sequences through an adaptive noise shaping algorithm;

[0050] a clustering module for unsupervised clustering of the secondary modulation components through a density peak clustering algorithm to obtain a plurality of aging mode clusters;

[0051] a feature extraction module for feature extraction of the secondary modulation components of the aging mode clusters to obtain statistical features of the aging mode clusters;

[0052] an initial evaluation module for obtaining an initial residual energy evaluation value according to the statistical features of the aging mode clusters through a variable coefficient regression sub-model;

[0053] a covariance estimation module configured to perform ampere-hour integration on the instantaneous value sequence of the bus voltage of the micro-grid in the current time period with a preset time window, determine an estimated covariance of a statistical mean channel, and perform a Kalman wave prediction correction cycle according to the bus voltage to obtain an estimated covariance of a Kalman filtering channel;

[0054] a fusion output module configured to perform weighted fusion on the initial residual energy evaluation value according to the estimated covariances corresponding to the Kalman filtering channel and the statistical mean channel respectively to obtain a final energy evaluation value.

[0055] In a third aspect, an electronic device includes a memory and a processor.

[0056] The memory is configured to store a computer program.

[0057] The processor is configured to implement the energy storage prediction method when executing the computer program.

[0058] The energy storage prediction method, system and device of the present application can accurately identify the quadratic modulation component of the microgrid bus through phase perturbation extraction and adaptive noise shaping algorithm, avoiding the error accumulation caused by ignoring weak coupling in traditional methods. Further, through the intelligent classification of aging patterns by the density peak clustering algorithm, and the weighted fusion of the variable coefficient regression sub-model and the Kalman filter channel, the present application not only improves the adaptability and stability of the prediction, but also enhances the reliability of the microgrid operation. Specifically, the present application obtains the bus voltage signal through the public connection point of the microgrid, and performs phase perturbation extraction to obtain the quantum phase perturbation sequence. The quantum phase perturbation sequence is identified and separated by using the adaptive noise shaping algorithm, so as to accurately extract the quadratic modulation component of the microgrid bus. This process effectively solves the prediction error problem caused by ignoring weak coupling in traditional methods. At the same time, the present application uses the density peak clustering algorithm to perform unsupervised clustering on the quadratic modulation component, obtains multiple aging pattern clusters, and extracts the features of the quadratic modulation component of the aging pattern cluster to obtain the statistical features of the aging pattern cluster. In this way, the present application can intelligently classify energy storage batteries with different aging degrees, so that the prediction model can adopt different prediction strategies for different aging patterns, further improving the adaptability and accuracy of the prediction. Through the variable coefficient regression sub-model, the initial remaining energy evaluation value is obtained according to the statistical features of the aging pattern cluster, and the initial remaining energy evaluation value is weighted and fused by combining the estimation covariance of the preset time window's ampere-hour integral, the Kalman filter channel and the statistical mean channel, to obtain the final energy evaluation value. This method combines a variety of advanced algorithms and technical means, fully utilizes the advantages of each algorithm, improves the stability and reliability of the prediction, and effectively reduces the prediction error. By accurately predicting the remaining energy of the energy storage battery, more accurate basis is provided for the operation control of the microgrid. The present application can better cope with the influence of energy storage battery aging or increasing electrochemical impedance, reduce the problems of microgrid bus voltage fluctuation and power imbalance caused by prediction error, thereby enhancing the stability of the microgrid operation, improving the reliability and economy of the microgrid. BRIEF DESCRIPTION OF DRAWINGS

[0059] Figure 1 The flowchart of the energy storage prediction method of an embodiment of the present application is shown.

[0060] Figure 2 The structural block diagram of another embodiment of the energy storage prediction system of the present application is shown. DETAILED DESCRIPTION

[0061] In order to make the above objectives, characteristics and advantages of the present application more apparent, specific embodiments of the present application will be described in detail below with reference to the accompanying drawings. Although some embodiments of the present application are shown in the drawings, it should be understood that the present application can be implemented in various forms, and should not be interpreted as being limited to the embodiments set forth herein, but rather, these embodiments are provided so as to more completely and thoroughly understand the present application. It should be understood that the drawings and embodiments of the present application are for exemplary purposes only, and are not intended to limit the scope of protection of the present application.

[0062] It should be understood that each of the steps recited in the method embodiments of the present application can be performed in different orders, and / or in parallel. In addition, the method embodiments can include additional steps and / or omit the steps shown. The scope of the present application is not limited in this respect.

[0063] As used herein, the term "comprising" and variations thereof, are open-ended, and mean "including but not limited to"; the term "based on" means "based, at least in part, on"; the term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments"; the term "optional" means "optional in at least some embodiments". Related definitions are given throughout the description. It should be noted that the concepts mentioned in the present application are referred to as "a" "one" or "the" for the purpose of distinguishing different apparatuses, modules or units, and not for limiting the functions of these apparatuses, modules or units.

[0064] It should be noted that the modification of "one" "multiple" mentioned in the present application is illustrative and not restrictive, and those skilled in the art should understand that, unless otherwise explicitly indicated in the context, it should be understood as "one or more".

[0065] In conjunction with Figure 1 As shown, the embodiment of the present application provides an energy storage prediction method, comprising:

[0066] Through the public connection point of the microgrid, the bus voltage signal of the microgrid in the current time period is acquired.

[0067] Specifically, by installing a high-precision voltage sensor at the point of common coupling (PCC) of the microgrid, the bus voltage signal of the microgrid in the current time period is collected in real time. The voltage sensor can record the instantaneous value of the bus voltage at a high sampling rate (e.g. thousands of times per second), and transmit these voltage data to the central controller in real time through a dedicated communication link. The central controller performs preliminary screening and preprocessing on the received voltage data, removes possible noise interference, and obtains a bus voltage signal sequence that can be used for subsequent analysis. For example, if the operating frequency of the microgrid is 50Hz, the sensor can collect voltage data at a sampling rate of 10000 times per second, ensuring that subtle changes in the bus voltage can be captured.

[0068] Phase perturbation extraction is performed according to the bus voltage signal to obtain a quantum phase perturbation sequence.

[0069] Specifically, a phase perturbation extraction method based on Hilbert transform is adopted. First, the collected bus voltage signal is subjected to Hilbert transform to convert it into an analytic signal. The real part of the analytic signal is the original voltage signal, and the imaginary part is the Hilbert transform result of the original signal. By calculating the instantaneous phase of the analytic signal, the phase information of the bus voltage signal can be obtained. Then, the instantaneous phase is differentiated to extract the phase perturbation. Since the amplitude of the quantum phase perturbation sequence is usually small, the present application further adopts wavelet transform to denoise the extracted phase perturbation, so as to improve the signal-to-noise ratio of the phase perturbation sequence. For example, Daubechies wavelet can be selected for multi-scale decomposition, and noise components can be removed through threshold processing, and finally the quantum phase perturbation sequence is reconstructed.

[0070] The secondary modulation component of the bus of the microgrid is obtained by adaptive noise shaping algorithm and identification separation processing according to the quantum phase perturbation sequence.

[0071] Specifically, an adaptive noise shaping algorithm (such as an adaptive filter) is used to process the quantum phase perturbation sequence; this algorithm can effectively separate the weak secondary modulation component coupled with harmonic non-stationarity by dynamically adjusting the parameters of the filter. The coefficient update rule of the adaptive filter is based on the least mean square error (LMS) algorithm, which iteratively optimizes the weights of the filter to minimize the error between the filter output and the target signal (i.e. the secondary modulation component). For example, the weights of the filter can be set to zero initially, and as the algorithm iterates, the weights will gradually adjust to adapt to the characteristics of the input signal. After multiple iterations, the filter can accurately identify and separate the secondary modulation component, thereby providing accurate input data for subsequent aging pattern analysis.

[0072] Multiple aging pattern clusters are obtained by unsupervised clustering of the secondary modulation component through a density peak clustering algorithm.

[0073] Specifically, the separated secondary modulation components are subjected to unsupervised clustering analysis by using a density peak clustering algorithm. First, the local density and relative distance of each data point are calculated. The local density can be determined by counting the number of data points within a certain radius around each data point, while the relative distance represents the distance of the data point to the data point with higher local density. By plotting a decision graph (i.e. a graph of local density vs. relative distance), the cluster centers, i.e. the data points with high local density and large relative distance, can be intuitively identified. Then, according to the distance of each data point to the cluster center, the data points are assigned to different aging mode clusters. For example, if there are two obvious cluster centers in the decision graph, the data points can be divided into two aging mode clusters, corresponding to energy storage batteries with different aging degrees respectively. In this way, the present application can automatically identify the characteristics of different aging modes, providing a basis for subsequent feature extraction and prediction model establishment.

[0074] The secondary modulation components of the aging mode clusters are subjected to feature extraction to obtain statistical features of the aging mode clusters.

[0075] Specifically, the present application extracts features from the secondary modulation components of each aging mode cluster to obtain statistical features that can represent the aging mode. The feature extraction method includes calculating statistical quantities such as mean, variance, skewness, kurtosis of each cluster, as well as extracting frequency features and time domain features of the secondary modulation components. For example, for an aging mode cluster, the mean of its secondary modulation components can be calculated to reflect the central tendency of the cluster, the variance can be calculated to measure the dispersion of the data, and the skewness and kurtosis can be calculated to describe the distribution of the data. In addition, the frequency features of the secondary modulation components can be extracted by fast Fourier transform (FFT) to analyze their frequency spectrum distribution; the time domain features such as peak value and zero-crossing rate of the secondary modulation components can be calculated to further characterize their dynamic characteristics. These statistical features can comprehensively reflect the characteristics of the aging mode cluster, providing accurate input data for the subsequent variable coefficient regression sub-model.

[0076] Through the variable coefficient regression sub-model, an initial remaining energy assessment value is obtained according to the statistical features of the aging mode clusters.

[0077] Specifically, the variable coefficient regression sub-model is a flexible regression model whose coefficients can be dynamically adjusted according to the changes of input features. In the present application, first, the input variables of the regression model are established according to the statistical characteristics (such as mean, variance, frequency characteristics, etc.) of the aging mode cluster. Then, the model is trained by historical data to determine the coefficients of the model. In the model training process, methods such as cross-validation can be used to optimize the parameters of the model and improve the prediction accuracy of the model. For example, if the historical data shows that the mean of the aging mode cluster has a linear relationship with the remaining energy of the energy storage battery, the regression model can take the mean as the main input feature, and the initial remaining energy evaluation value is calculated by the coefficient obtained by training. In this way, the variable coefficient regression sub-model can dynamically adjust the prediction strategy according to different aging modes, thereby improving the accuracy of the initial remaining energy evaluation.

[0078] The instantaneous value sequence of the bus voltage of the microgrid in the current time period is integrated with respect to time within a preset time window to determine an estimated covariance of a statistical mean channel, and a Kalman wave prediction correction cycle is performed according to the bus voltage to obtain an estimated covariance of a Kalman filter channel.

[0079] Specifically, first, a preset time window (for example, 1 minute or 1 hour) is set, and the instantaneous value sequence is integrated with respect to time within the time window to obtain the cumulative power in the time period. Then, the statistical mean of the cumulative power is calculated, and the estimated covariance is calculated according to the statistical mean. The estimated covariance reflects the degree of fluctuation of the data and can be used to evaluate the reliability of the statistical mean channel. At the same time, the present application also uses a Kalman filter algorithm to predict and correct the bus voltage. The Kalman filter algorithm establishes a state space model, combines measurement data and prior knowledge, and estimates and corrects the bus voltage in real time. At each time step, the Kalman filter updates the state estimate and the estimated covariance according to the current measurement data and the state estimate at the previous time. In this way, the Kalman filter channel can dynamically adjust its estimated covariance to reflect the real-time change characteristics of the bus voltage. For example, when the bus voltage fluctuates greatly, the Kalman filter will increase the estimated covariance to represent the increase in uncertainty of the current estimate.

[0080] The initial remaining energy evaluation value is weighted and fused according to the estimated covariances corresponding to the Kalman filter channel and the statistical mean channel, respectively, to obtain a final energy evaluation value.

[0081] Specifically, first, the inverse of the estimated covariance of the two channels is calculated as the weighting coefficient, which reflects the reliability of each channel. The smaller the estimated covariance, the larger the weighting coefficient, indicating that the estimated value of the channel is more reliable. Then, the initial remaining energy evaluation value is weighted and summed according to the weighting coefficient to obtain the final energy evaluation value. For example, if the estimated covariance of the Kalman filtering channel is 0.01 and the estimated covariance of the statistical mean channel is 0.04, the weighting coefficient of the Kalman filtering channel is 100 and the weighting coefficient of the statistical mean channel is 25. Assuming that the initial remaining energy evaluation value is 100 Ah, the estimated value of the Kalman filtering channel is 98 Ah, and the estimated value of the statistical mean channel is 102 Ah, the final energy evaluation value is: final energy evaluation value = (100*98+25*102) / 100+25≈98.4 Ah. In this way, the application can comprehensively consider the reliability of the two channels to obtain a more accurate and reliable final energy evaluation value.

[0082] The energy storage prediction method, system and device of the present application can accurately identify the second modulation component of the microgrid bus through phase perturbation extraction and adaptive noise shaping algorithm, avoiding the error accumulation caused by ignoring weak coupling in traditional methods. Further, through the density peak clustering algorithm, the aging patterns are intelligently classified, and combined with the weighted fusion of the variable coefficient regression sub-model and the Kalman filter channel, the present application not only improves the adaptability and stability of the prediction, but also enhances the reliability of the microgrid operation. Specifically, the present application obtains the bus voltage signal through the point of common coupling of the microgrid, and performs phase perturbation extraction to obtain a quantum phase perturbation sequence. The quantum phase perturbation sequence is identified and separated by using the adaptive noise shaping algorithm, so as to accurately extract the second modulation component of the microgrid bus. This process effectively solves the prediction error problem caused by ignoring weak coupling in traditional methods. At the same time, the present application uses the density peak clustering algorithm to unsupervisedly cluster the second modulation component to obtain multiple aging pattern clusters, and extracts the features of the second modulation component of the aging pattern clusters to obtain the statistical features of the aging pattern clusters. In this way, the present application can intelligently classify energy storage batteries with different aging degrees, so that the prediction model can adopt different prediction strategies for different aging patterns, further improving the adaptability and accuracy of the prediction. Through the variable coefficient regression sub-model, the initial remaining energy evaluation value is obtained according to the statistical features of the aging pattern clusters, and the initial remaining energy evaluation value is weighted and fused by combining the estimation covariance of the preset time window's ampere-hour integral, the Kalman filter channel and the statistical mean channel, to obtain the final energy evaluation value. This method combines a variety of advanced algorithms and technical means, fully utilizes the advantages of each algorithm, improves the stability and reliability of the prediction, and effectively reduces the prediction error. By accurately predicting the remaining energy of the energy storage battery, more accurate basis is provided for the operation control of the microgrid. The present application can better cope with the influence of energy storage battery aging or increasing electrochemical impedance, reduce the problems of microgrid bus voltage fluctuation and power imbalance caused by prediction error, thereby enhancing the stability of the microgrid operation and improving the reliability and economy of the microgrid.

[0083] Optionally, the phase perturbation extraction according to the bus voltage signal to obtain a quantum phase perturbation sequence comprises:

[0084] performing Hilbert transform on the bus voltage signal to obtain an instantaneous phase sequence;

[0085] performing difference operation on the instantaneous phase sequence according to the fundamental phase of the microgrid to obtain a phase deviation sequence;

[0086] removing linear drift in the phase deviation sequence by sliding least squares fitting to obtain a residual phase perturbation sequence;

[0087] quantum noise screening is performed on the residual phase perturbation sequence to generate the quantum phase perturbation sequence.

[0088] Specifically, first, the collected bus voltage signal is subjected to Hilbert transform to obtain its analytic signal . The instantaneous phase of the analytic signal can be obtained by calculation , that is . In this way, the phase information of the bus voltage signal can be extracted from the time domain, providing a basis for subsequent phase perturbation analysis.

[0089] The fundamental phase of the microgrid is usually linearly related to time and can be expressed as , where is the fundamental angular frequency, and is the initial phase. By performing a difference operation on the instantaneous phase sequence and the fundamental phase , that is , the phase deviation sequence is obtained. This phase deviation sequence reflects the dynamic changes of the bus voltage signal relative to the fundamental phase, providing a basis for further extracting perturbation information. In order to remove the linear drift that may exist in the phase deviation sequence , a sliding least squares fitting method is used. Within each sliding window (the window length can be selected according to signal characteristics, such as 10 sampling points), the phase deviation sequence is linearly fitted to obtain the slope and intercept of the fitted straight line. Then, the phase deviation sequence is subtracted by the value of the fitted straight line, that is , to obtain the residual phase perturbation sequence , thus effectively removing the linear trend in the phase deviation, enabling subsequent analysis to focus more accurately on the perturbation component.

[0090] The purpose of quantum noise screening is to extract quantum-level perturbation signals from the residual phase perturbation sequence . By setting a quantum noise threshold (which can be determined according to the theoretical or experimental measured value of quantum noise), the residual phase perturbation sequence is screened. When is less than the quantum noise threshold, the perturbation is considered to belong to the category of quantum noise and is retained and marked as a quantum phase perturbation; otherwise, it is considered to be non-quantum noise and is discarded. The final quantum phase perturbation sequence can more accurately reflect the weak quantum-level changes in the bus voltage signal, providing a high-precision input signal for subsequent energy storage prediction.

[0091] In the embodiments of the present application, the quantum phase perturbation sequence can be accurately extracted from the micro-grid bus voltage signal, which effectively solves the problem of energy storage prediction error caused by the inability to accurately capture weak phase changes in traditional methods. Specifically, the Hilbert transform can extract the phase information of the bus voltage signal from the time domain, the differential operation further highlights the phase deviation, the sliding least squares fitting removes the linear drift, and the quantum noise screening accurately extracts the quantum-level perturbation signal. This series of operations not only improves the accuracy of signal processing, but also provides high-quality input data for the subsequent energy storage prediction model, significantly improving the accuracy and reliability of energy storage prediction, especially when the battery is aging or the electrochemical impedance is increasing, it can still maintain a high prediction accuracy, thereby enhancing the stability and economy of the micro-grid operation.

[0092] Optionally, the identifying and separating processing according to the quantum phase perturbation sequence by the adaptive noise shaping algorithm to obtain the second modulation component of the bus of the micro-grid, comprising:

[0093] dividing the quantum phase perturbation sequence into a plurality of sub-sequences by a sliding window;

[0094] wavelet packet decomposing each of the sub-sequences to obtain a plurality of layers of detail coefficients and approximation coefficients, and adaptively quantizing the plurality of layers of detail coefficients and the approximation coefficients according to energy entropy to obtain a reconstructed perturbation sub-sequence;

[0095] inputting the reconstructed perturbation sub-sequence into a noise shaper to perform frequency spectrum shift and shaping on the reconstructed perturbation sub-sequence to obtain a shaped perturbation sub-sequence;

[0096] performing Hilbert transform on the shaped perturbation sub-sequence and extracting an instantaneous amplitude envelope to obtain the second modulation component.

[0097] Specifically, a sliding window is first defined, whose length can be chosen according to the characteristics of the quantum phase perturbation sequence and the required resolution. For example, if the sampling rate of the quantum phase perturbation sequence is 1000 Hz, the sliding window length can be set to 100 sampling points (i.e. 0.1 seconds). Then, the window is slid along the quantum phase perturbation sequence, moving one or more sampling points at a time, thereby dividing the entire sequence into multiple subsequences. For example, for a quantum phase perturbation sequence with a length of 10000 sampling points, using a sliding window with a length of 100 and moving 10 sampling points at a time, 900 subsequences can be obtained. Each subsequence contains the quantum phase perturbation information within a local time period, providing a basis for subsequent analysis through segmented processing. Wavelet packet decomposition is performed on each subsequence, which is a multi-resolution analysis method that can decompose the signal into detail coefficients and approximation coefficients of different frequency ranges. For example, selecting Daubechies wavelet (such as db4) for 3-layer wavelet packet decomposition can obtain 8 coefficients of different frequency ranges (including detail coefficients and approximation coefficients). Then, the energy entropy of each coefficient is calculated, which reflects the distribution of signal energy in the coefficient. According to the size of the energy entropy, each coefficient is adaptively quantized by threshold. The threshold can be dynamically adjusted according to the statistical characteristics of the energy entropy, for example, setting the threshold to a certain percentage (such as 10%) of the energy entropy. Coefficients below the threshold are considered noise components and are set to zero; coefficients above the threshold are retained. Finally, through the wavelet packet reconstruction algorithm, the quantized coefficients are reconstructed into new subsequences, i.e. reconstructed perturbation subsequences. This process can effectively remove noise interference and retain important features in the signal.

[0098] The reconstructed perturbation subsequences are input into a noise shaper, which is a signal processing module that adjusts the spectral distribution of the signal to transfer noise components in the signal to specific frequency ranges, thereby improving the signal-to-noise ratio of the signal. For example, a band-stop filter can be designed as a noise shaper, with the stop-band frequency range of the filter selected according to the frequency characteristics of the second-order modulation component. When the reconstructed perturbation subsequence passes through the noise shaper, the noise components in its frequency spectrum are transferred outside the stop-band frequency range, while the frequency components of the second-order modulation component are retained. In this way, the shaped perturbation subsequence has a higher signal-to-noise ratio, providing a clearer signal for subsequent extraction of the second-order modulation component.

[0099] The shaped perturbation subsequence is subjected to a Hilbert transform to obtain its analytic signal. The instantaneous amplitude envelope of the analytic signal can be obtained by calculating the modulus of the analytic signal, i.e. where, is the analytic signal. The instantaneous amplitude envelope reflects the amplitude variation of the signal, and can effectively extract the characteristics of the second modulation component. For example, if the shaping perturbation sequence contains a second modulation component, its instantaneous amplitude envelope will show amplitude variation related to the second modulation frequency. By extracting the instantaneous amplitude envelope, the final second modulation component can clearly reflect the weak modulation signal on the microgrid bus, providing accurate input data for subsequent aging pattern analysis and energy storage prediction.

[0100] In the embodiments of the present application, the second modulation component of the microgrid bus can be accurately identified and separated from the quantum phase perturbation sequence. Specifically, the sliding window divides the sub-sequence, which enables localized signal processing, improving the flexibility and accuracy of the processing; wavelet packet decomposition and adaptive threshold quantization can effectively remove noise interference and preserve important features in the signal; the noise shaper further improves the signal-to-noise ratio, making the second modulation component more prominent; Hilbert transform and instantaneous amplitude envelope extraction can clearly reveal the amplitude characteristics of the second modulation component. This series of operations not only improves the accuracy of signal processing, but also provides high-quality input data for subsequent energy storage prediction models, significantly improving the accuracy and reliability of energy storage prediction, especially when the battery ages or the electrochemical impedance increases, it can still maintain a high prediction accuracy, thereby enhancing the stability and economy of the microgrid operation.

[0101] Optionally, the second modulation component is unsupervisedly clustered by a density peak clustering algorithm to obtain a plurality of aging pattern clusters, including:

[0102] Performing short-time Fourier transform on the second modulation component to obtain a time-frequency energy spectrum;

[0103] Based on the time-frequency energy spectrum, calculating the local density and relative distance of each preset time point in the current time period, and constructing a density-distance decision graph according to the local density and the relative distance;

[0104] According to the density-distance decision graph, determining a clustering center, and clustering according to the clustering center to obtain a plurality of the aging pattern clusters.

[0105] Specifically, first, the secondary modulation component signal is subjected to short-time Fourier transform (STFT). STFT is a time-frequency analysis method that can decompose a signal into information in two dimensions of time and frequency. In this embodiment, a suitable window function (such as the Hanning window) and window length (for example, 256 sampling points) are selected, the secondary modulation component is processed in segments, and fast Fourier transform (FFT) is performed on each segment. In this way, the frequency distribution corresponding to each time point can be obtained, thereby generating a time-frequency energy spectrum. The time-frequency energy spectrum reflects the energy distribution of the secondary modulation component at different times and frequencies, providing basic data for subsequent clustering analysis. Based on the time-frequency energy spectrum, the local density and relative distance of each preset time are calculated. The local density refers to the number of points in the time-frequency energy spectrum within a certain range (for example, a time window of 10 sampling points) around a certain time, whose energy value is higher than a certain threshold (for example, 80% of the average energy). The relative distance refers to the distance from the time to the time with higher local density around it. By calculating the local density and relative distance of each time, a density-distance decision diagram can be constructed. In the decision diagram, the horizontal axis represents the local density, and the vertical axis represents the relative distance. The decision diagram can intuitively show which times are potential clustering centers, i.e., points with high local density and large relative distance.

[0106] By analyzing the density-distance decision diagram, the clustering centers are determined, and in the decision diagram, points with high local density and large relative distance are selected as clustering centers. For example, threshold values of local density and relative distance can be set, and points meeting the conditions are selected as clustering centers. Then, according to the distance of each time from the clustering centers, the times are assigned to different aging mode clusters. Specifically, for each time, the Euclidean distance from all clustering centers is calculated, and it is assigned to the cluster corresponding to the nearest clustering center. In this way, multiple aging mode clusters can be obtained, each corresponding to a specific aging mode. For example, if two clustering centers are identified in the decision diagram, the times can be divided into two aging mode clusters, corresponding to energy storage batteries with different aging degrees, respectively.

[0107] In the embodiments of the present application, the secondary modulation component can be effectively unsupervised clustering to obtain multiple aging mode clusters. Specifically, the short-time Fourier transform can clearly show the time-frequency characteristics of the secondary modulation component, providing rich information for clustering analysis; the local density and relative distance are calculated based on the time-frequency energy spectrum, and a density-distance decision diagram is constructed, which can intuitively identify the clustering center; according to the clustering center, the energy storage batteries with different aging degrees can be accurately classified into different aging mode clusters. This series of operations not only improves the accuracy of the aging mode classification, but also provides a clear aging mode division for subsequent feature extraction and energy storage prediction, significantly improving the adaptability and reliability of the energy storage prediction, especially when the battery aging or electrochemical impedance increases, the prediction accuracy can still be maintained at a high level, thereby enhancing the stability and economy of the micro-grid operation.

[0108] Optionally, the feature extraction is performed on the secondary modulation component of the aging mode cluster to obtain statistical features of the aging mode cluster, including:

[0109] The mean value, peak factor, kurtosis, entropy value and linear fitting slope of the instantaneous amplitude of the secondary modulation component of each aging mode cluster are calculated respectively, and the mean value, peak factor, kurtosis, entropy value and linear fitting slope are taken as corresponding dimensions to generate an original feature vector of the aging mode cluster.

[0110] The original feature vector is subjected to Z-score standardization, and the standardized original feature vector is subjected to dimension reduction processing through principal component analysis to obtain a statistical feature vector of the aging mode cluster.

[0111] Specifically, for the secondary modulation component in each aging mode cluster, first, the mean value of the instantaneous amplitude is calculated. The mean value reflects the average intensity of the signal in the aging mode cluster, which can be obtained by simple summation and division by the total number of data points. Second, the peak factor, i.e. the ratio of the maximum instantaneous amplitude to the root mean square value of the signal, is calculated to measure the peak value characteristics of the signal. Then, the kurtosis is calculated, which is a measure of the sharpness of the signal amplitude distribution, calculated by the ratio of the fourth central moment to the square of the variance, which can reflect the frequency of occurrence of abnormal values in the signal. In addition, the entropy value is calculated, which measures the complexity and uncertainty of the signal, which is obtained by calculating the probability distribution of the instantaneous amplitude and applying the Shannon entropy formula. Finally, the instantaneous amplitude is linearly fitted to obtain the linear fitting slope, which reflects the trend of the signal over time. These statistics together constitute a multi-dimensional feature description of the aging mode cluster. The mean value, peak factor, kurtosis, entropy value and linear fitting slope obtained by the above calculation are taken as different dimensions of the feature vector, respectively. For example, if the mean value of a certain aging mode cluster is , the peak factor is , and the kurtosis is , the entropy value is , the linear fitting slope is .

[0112] The original feature vector of the aging pattern cluster can be expressed as This original feature vector contains comprehensive feature information of the aging pattern cluster, providing a basis for further analysis and processing.

[0113] In order to eliminate the influence of different feature dimensions and numerical ranges, the original feature vector is standardized by Z-score. For each feature dimension, the standardized value is calculated as where is the mean value of the feature dimension, is the standard deviation, represents the i-th feature dimension. In this way, all feature values are converted to standardized values with a mean of 0 and a standard deviation of 1, making different features comparable and providing a unified scale for subsequent principal component analysis. The standardized original feature vector is subjected to principal component analysis (PCA). PCA is a commonly used dimensionality reduction technique that can retain the main variability of the data while removing redundant information by projecting the original feature vector onto a set of new orthogonal bases (principal components). First, the covariance matrix of the standardized feature vector is calculated, and then the eigenvalues and eigenvectors of the covariance matrix are solved. According to the size of the eigenvalues, the first few principal components are selected, which can explain most of the data variability. Finally, the original feature vector is projected onto these principal components to obtain the reduced statistical feature vector. For example, if the first two principal components are selected, the original feature vector will be projected onto these two principal components to obtain a new two-dimensional statistical feature vector This reduced statistical feature vector not only retains the key feature information of the aging pattern cluster, but also reduces the data dimension, improving the efficiency and accuracy of subsequent processing.

[0114] ​In the embodiments of the present application, representative statistical feature vectors can be extracted from the secondary modulation components of the aging mode cluster. Specifically, by calculating the mean value, peak factor, kurtosis, entropy value and linear fitting slope of the instantaneous amplitude, the strength, peak characteristic, distribution pattern, complexity and trend of change of the signal can be comprehensively reflected; taking these statistical quantities as the dimensions of the original feature vector provides rich information for the feature description of the aging mode; Z-score standardization eliminates the dimensional and numerical range differences between different features, so that the feature vector has a unified scale; principal component analysis further reduces the dimension of the feature vector, retains the key information, and improves the efficiency and accuracy of data processing. This series of operations not only improves the accuracy and efficiency of the aging mode feature extraction, but also provides high-quality input data for the subsequent energy storage prediction model, significantly improves the adaptability and reliability of the energy storage prediction, especially when the battery ages or the electrochemical impedance increases, the prediction accuracy is still high, thereby enhancing the stability and economy of the microgrid operation.

[0115] Optionally, the initial remaining energy evaluation value is obtained by the variable coefficient regression sub-model according to the statistical features of the aging mode cluster, comprising:

[0116] The statistical feature vector of the aging mode cluster is taken as the input of the variable coefficient regression sub-model, and the radial basis kernel is weighted for each dimension of the statistical feature vector to obtain the instantaneous nonlinear coefficient corresponding to each dimension;

[0117] The instantaneous nonlinear coefficient is multiplied by the corresponding feature value of the dimension item by item and accumulated to obtain the initial remaining energy evaluation value corresponding to the aging mode cluster, and the variable coefficient regression sub-model is output.

[0118] Specifically, the statistical feature vector of the aging mode cluster is input into the variable coefficient regression sub-model. For each dimension of the statistical feature vector, a radial basis kernel function is used for weighting processing. The radial basis kernel function usually selects a Gaussian kernel function, which has the form wherein, is the center of the kernel function, is the width parameter of the kernel function, is the calculated kernel weight, and the larger the value is, the closer fᵢ is to the center c, and the higher the weight is. By adjusting and , the nonlinear weighting of each feature dimension can be performed to obtain the instantaneous nonlinear coefficient corresponding to each dimension. For example, if the mean value of the feature dimension is selected as , The instantaneous nonlinear coefficient of each feature dimension can be obtained, which reflects the importance of the feature under the current aging mode. The instantaneous nonlinear coefficient of each feature dimension is multiplied by the corresponding feature value to obtain the weighted feature value . Then, the weighted feature values are accumulated to obtain the initial remaining energy evaluation value corresponding to the aging mode cluster , where n is the number of dimensions in the statistical feature vector. The initial remaining energy evaluation value integrates the weighted contributions of all feature dimensions and can more accurately reflect the remaining energy of the energy storage battery under the current aging mode. Finally, the initial remaining energy evaluation value is output by the output layer of the variable coefficient regression sub-model for subsequent weighted fusion processing.

[0119] In the embodiments of the present application, the initial remaining energy evaluation value is obtained by the variable coefficient regression sub-model. Specifically, the radial basis kernel weighting can perform nonlinear processing on each feature dimension, dynamically adjust the importance of each feature, and make the model better adapt to the feature changes under different aging modes; the operation of multiplying and accumulating can integrate the contributions of all features to obtain a comprehensive and accurate initial remaining energy evaluation value. This process not only improves the adaptability and accuracy of energy storage prediction, but also provides high-quality input data for subsequent weighted fusion, significantly improves the overall performance of energy storage prediction, and especially when the battery ages or the electrochemical impedance increases, it can still maintain high prediction accuracy, thereby enhancing the stability and economy of microgrid operation.

[0120] Optionally, the Ah integration of the instantaneous value sequence of the bus voltage of the microgrid in the current time period within a preset time window is performed to determine the estimated covariance of the statistical mean channel, and a Kalman wave prediction correction cycle is performed according to the bus voltage to obtain the estimated covariance of the Kalman filter channel, including:

[0121] The Ah integration of the instantaneous value sequence is performed according to the preset time window to obtain an integrated charge amount sequence, and the sample variance of the integrated charge amount sequence is determined;

[0122] The sample variance is taken as the estimated covariance of the statistical mean channel;

[0123] The Kalman filter prediction is performed through a state space model to obtain a prior covariance and a prior estimation of the remaining energy;

[0124] The Kalman filter correction is performed according to the instantaneous amplitude of the bus voltage, and the posterior covariance is obtained in combination with the prior covariance;

[0125] ​The posterior covariance is used as the estimated covariance of the Kalman filter channel.

[0126] Specifically, a preset time window is first set, for example, 1 minute or 1 hour. Within each time window, the ampere-hour integral of the instantaneous value sequence (such as the current value) is performed. The calculation formula of the ampere-hour integral is:

[0127] ;

[0128] wherein, is the cumulative charge amount at time , is the instantaneous current value at time , is the sampling time interval. In this way, the integral charge amount sequence composed of the cumulative charge amount in each time window can be obtained. Then, the sample variance of the sequence is calculated as the estimated covariance of the statistical mean channel. The calculation formula of the sample variance is:

[0129] ;

[0130] wherein, is the number of samples within the time window, is the mean value of the integral charge amount sequence, Q(i) is the cumulative charge amount (obtained by ampere-hour integration) corresponding to the i-th sampling point, is the sample variance of the integral charge amount sequence, which is used as the estimated covariance of the statistical mean channel. The sample variance reflects the fluctuation of the integral charge amount sequence and provides a basis for subsequent statistical analysis.

[0131] The sample variance calculated above is directly used as the estimated covariance of the statistical mean channel. The estimated covariance is used to represent the uncertainty of the statistical mean channel. The larger the sample variance, the greater the fluctuation of the integral charge amount sequence, and the higher the uncertainty of the statistical mean channel.

[0132] A state space model is established to describe the change of the remaining energy of the energy storage battery. The state space model usually includes a state equation and an observation equation. The state equation describes the change of the remaining energy over time, for example:

[0133] ;

[0134] wherein, is the state vector (such as the remaining energy) at time , is the state transition matrix, is the control input matrix, is the control input, is the process noise. The observation equation describes the relationship between the bus voltage and the remaining energy, for example:

[0135] ;

[0136] wherein, is the observation value (such as the instantaneous amplitude of bus voltage), is the observation matrix, is the observation noise. Through the Kalman filtering algorithm, the prior covariance and the prior estimation of the residual energy are obtained by predicting according to the state space model . The prior covariance reflects the uncertainty of the residual energy estimation without observation update.

[0137] The Kalman filter correction is performed according to the instantaneous amplitude of the bus voltage . The correction step of the Kalman filter includes calculating the Kalman gain , updating the state estimation , the prior covariance and the posterior covariance . The formula for calculating the Kalman gain is:

[0138] ;

[0139] wherein, is the covariance of the observation noise. The formula for updating the state estimation is:

[0140] ;

[0141] The formula for updating the posterior covariance is:

[0142] ;

[0143] wherein, I is the unit matrix; through these steps, the posterior covariance can be obtained, which reflects the uncertainty of the residual energy estimation after observation update.

[0144] The posterior covariance calculated above is taken as the estimation covariance of the Kalman filter channel, and the posterior covariance is used to represent the uncertainty of the Kalman filter channel. By combining the prior information and the observation information, the reliability of the residual energy estimation can be more accurately reflected.

[0145] In the embodiments of the present application, a reliable estimated covariance is provided for the statistical mean channel through ampere-hour integration and sample variance calculation, reflecting the fluctuation of the integrated charge quantity sequence; and the Kalman filter prediction and correction steps dynamically adjust the estimated value and uncertainty of the remaining energy through a state space model and observation data. This method combining statistical analysis and dynamic filtering not only improves the accuracy and reliability of energy storage prediction, but also better adapts to complex situations such as battery aging or changes in electrochemical impedance, significantly improving the overall performance of energy storage prediction and enhancing the stability and economy of microgrid operation.

[0146] Optionally, the initial remaining energy evaluation value is weighted and fused according to the estimated covariance corresponding to the Kalman filter channel and the statistical mean channel respectively to obtain a final energy evaluation value, including:

[0147] A first weight coefficient is obtained according to the estimated covariance of the Kalman filter channel;

[0148] A second weight coefficient is obtained according to the estimated covariance of the statistical mean channel;

[0149] The sum of the first weight coefficient and the second weight coefficient is taken as a normalization factor to obtain a weight coefficient pair;

[0150] The initial remaining energy evaluation value and the prior estimation of the remaining energy are weighted and summed by the weight coefficient pair to obtain the final energy evaluation value.

[0151] Specifically, the estimated covariance of the Kalman filter channel reflects the uncertainty of this channel. In order to obtain the first weight coefficient , the reciprocal of the estimated covariance can be taken, that is:

[0152] ;

[0153] In this way, the smaller the estimated covariance (i.e. the lower the uncertainty), the larger the first weight coefficient, indicating that the estimated value of this channel should occupy a larger weight in the fusion process.

[0154] The estimated covariance of the statistical mean channel also reflects the uncertainty of this channel. In order to obtain the second weight coefficient , the reciprocal of the estimated covariance is also taken, that is:

[0155] ;

[0156] Similarly to the first weight coefficient, the smaller the estimated covariance, the larger the second weight coefficient, indicating that the estimated value of this channel should occupy a larger weight in the fusion process.

[0157] In order to ensure that the sum of the weight coefficients is 1, the first weight coefficient and the second weight coefficient need to be normalized. The normalization factor is:

[0158] ;

[0159] Then, the normalized weight coefficient pair is calculated:

[0160] ;

[0161] In this way, the weight coefficient pair satisfies , ensuring the rationality of the weighted sum.

[0162] Let the initial residual energy evaluation value be , and the residual energy priori estimate be . Through weighted summation by the weight coefficient pair, the final energy evaluation value is obtained:

[0163]

[0164] The final energy evaluation value integrates the initial residual energy evaluation value and the residual energy priori estimate, and reasonably weights according to the respective uncertainty, thereby obtaining a more accurate and reliable residual energy evaluation value of the energy storage battery.

[0165] In the embodiment of the present application, by calculating the weight coefficient of each channel and performing normalization processing, the rationality and accuracy of the weighted sum are ensured. This method not only integrates the advantages of the two channels, but also dynamically adjusts the weight according to the respective uncertainty, so that the final energy evaluation value can maintain high precision and reliability under different aging degrees and operating states. This process significantly improves the overall performance of energy storage prediction, enhances the stability and economy of microgrid operation, and especially when the battery is aging or the electrochemical impedance is increasing, the prediction accuracy can still be maintained at a high level.

[0166] In combination with the method shown in Figure 2 , the present application provides a kind of energy storage prediction system, comprising:

[0167] Signal acquisition module, for obtaining the bus voltage signal of the microgrid in the current time period by the public connection point of the microgrid;

[0168] Phase perturbation module, for extracting quantum phase perturbation sequence according to the bus voltage signal;

[0169] Modulation component separation module, for identifying and separating processing according to the quantum phase perturbation sequence by adaptive noise shaping algorithm, to obtain the quadratic modulation component of the bus of the microgrid;

[0170] a clustering module configured to perform unsupervised clustering on the secondary modulation components by a density peak clustering algorithm to obtain a plurality of aging pattern clusters;

[0171] a feature extraction module configured to perform feature extraction on the secondary modulation components of the aging pattern clusters to obtain statistical features of the aging pattern clusters;

[0172] an initial evaluation module configured to obtain an initial residual energy evaluation value according to the statistical features of the aging pattern clusters by a variable coefficient regression submodel;

[0173] a covariance estimation module configured to perform ampere-hour integration on a sequence of instantaneous values of the bus voltage of the microgrid in the current time period with a preset time window to determine an estimated covariance of a statistical mean channel, and perform a Kalman wave prediction correction cycle according to the bus voltage to obtain an estimated covariance of a Kalman filtering channel;

[0174] a fusion output module configured to perform weighted fusion on the initial residual energy evaluation value according to the estimated covariances corresponding to the Kalman filtering channel and the statistical mean channel respectively to obtain a final energy evaluation value.

[0175] The energy storage prediction system of the present application has the same advantages as the above-mentioned energy storage prediction method compared with the prior art, which will not be repeated here.

[0176] The electronic device of the present application comprises a memory and a processor;

[0177] The memory is used for storing a computer program;

[0178] The processor is used for implementing the energy storage prediction method as described above when executing the computer program.

[0179] The electronic device of the present application has the same advantages as the above-mentioned energy storage prediction method compared with the prior art, which will not be repeated here.

[0180] Although the present application is disclosed as above, the protection scope of the present application is not limited to this. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present application, and these changes and modifications will fall within the protection scope of the present application.

Claims

1. An energy storage prediction method, characterized in that, The method comprises the following steps: acquiring a bus voltage signal of the micro-grid in a current time period through a public connection point of the micro-grid; performing phase perturbation extraction according to the bus voltage signal to obtain a quantum phase perturbation sequence, specifically comprising: performing Hilbert transform on the bus voltage signal to obtain an instantaneous phase sequence; performing difference operation on the instantaneous phase sequence according to a fundamental phase of the micro-grid to obtain a phase deviation sequence; removing linear drift in the phase deviation sequence through sliding least squares fitting to obtain a residual phase perturbation sequence; and performing quantum noise screening on the residual phase perturbation sequence to generate the quantum phase perturbation sequence; performing identification and separation processing on the quantum phase perturbation sequence through an adaptive noise shaping algorithm to obtain a second modulation component of the bus of the micro-grid, specifically comprising: dividing the quantum phase perturbation sequence into multiple subsequences through a sliding window; performing wavelet packet decomposition on each of the subsequences to obtain multiple layers of detail coefficients and approximation coefficients, and performing adaptive threshold quantization on the multiple layers of detail coefficients and the approximation coefficients according to energy entropy to obtain a reconstructed perturbation subsequence; inputting the reconstructed perturbation subsequence into a noise shaper to perform frequency spectrum shift and shaping on the reconstructed perturbation subsequence to obtain a shaped perturbation subsequence; and performing Hilbert transform on the shaped perturbation subsequence and extracting an instantaneous amplitude envelope to obtain the second modulation component; performing unsupervised clustering on the second modulation component through a density peak clustering algorithm to obtain multiple aging mode clusters; performing feature extraction on the second modulation component of the aging mode cluster to obtain statistical features of the aging mode cluster; obtaining an initial residual energy evaluation value according to the statistical features of the aging mode cluster through a variable coefficient regression submodel; integrating the instantaneous value sequence of the bus voltage of the micro-grid in the current time period with a preset time window to determine the estimated covariance of the statistical mean channel, and performing Kalman wave prediction correction cycle according to the bus voltage to obtain the estimated covariance of the Kalman filtering channel; performing weighted fusion on the initial residual energy evaluation value according to the estimated covariances corresponding to the Kalman filtering channel and the statistical mean channel respectively to obtain a final energy evaluation value.

2. The energy storage prediction method of claim 1, wherein, The method of performing unsupervised clustering on the second modulation component through a density peak clustering algorithm to obtain multiple aging mode clusters comprises: performing short-time Fourier transform on the second modulation component to obtain a time-frequency energy spectrum; calculating the local density and relative distance of each preset time in the current time period based on the time-frequency energy spectrum, and constructing a density-distance decision graph according to the local density and the relative distance; determining a clustering center according to the density-distance decision graph, and performing clustering according to the clustering center to obtain multiple aging mode clusters.

3. The energy storage prediction method of claim 1, wherein, The method of performing feature extraction on the second modulation component of the aging mode cluster to obtain statistical features of the aging mode cluster comprises: The mean value, peak factor, kurtosis, entropy value and linear fitting slope of the instantaneous amplitude of each secondary modulation component of the aging mode cluster are calculated respectively, and the mean value, peak factor, kurtosis, entropy value and linear fitting slope are taken as corresponding dimensions to generate an original feature vector of the aging mode cluster; The original feature vector is subjected to Z-score standardization, and the original feature vector after standardization is subjected to dimension reduction processing through principal component analysis to obtain a statistical feature vector of the aging mode cluster.

4. The energy storage prediction method of claim 3, wherein, The initial residual energy evaluation value is obtained from the statistical features of the aging mode cluster through the variable coefficient regression sub-model, including: The statistical feature vector of the aging mode cluster is taken as the input of the variable coefficient regression sub-model, and the features of each dimension of the statistical feature vector are weighted by a radial basis kernel to obtain the instantaneous nonlinear coefficient corresponding to each dimension; The instantaneous nonlinear coefficient is multiplied by the feature value corresponding to the dimension item by item and accumulated to obtain the initial residual energy evaluation value corresponding to the aging mode cluster, and the variable coefficient regression sub-model is output.

5. The energy storage prediction method of claim 1, wherein, The instantaneous value sequence of the bus voltage of the microgrid in the current time period is integrated in the predetermined time window to determine the estimated covariance of the statistical mean channel, and the Kalman wave prediction correction cycle is performed according to the bus voltage to obtain the estimated covariance of the Kalman filter channel, including: The instantaneous value sequence is integrated in the predetermined time window to obtain an integrated charge quantity sequence, and the sample variance of the integrated charge quantity sequence is determined; The sample variance is taken as the estimated covariance of the statistical mean channel; The Kalman filter prediction is performed through the state space model to obtain the prior covariance and the residual energy prior estimate; The Kalman filter correction is performed according to the instantaneous amplitude of the bus voltage, and the posterior covariance is obtained in combination with the prior covariance; The posterior covariance is taken as the estimated covariance of the Kalman filter channel.

6. The energy storage prediction method of claim 5, wherein, The initial residual energy evaluation value is weighted and fused according to the estimated covariances corresponding to the Kalman filter channel and the statistical mean channel respectively to obtain the final energy evaluation value, including: The first weight coefficient is obtained according to the estimated covariance of the Kalman filter channel; The second weight coefficient is obtained according to the estimated covariance of the statistical mean channel; The sum of the first weight coefficient and the second weight coefficient is taken as a normalization factor to obtain a weight coefficient pair; The initial residual energy evaluation value and the residual energy prior estimate are weighted and summed through the weight coefficient pair to obtain the final energy evaluation value.

7. An energy storage prediction system characterized by, It includes: The signal acquisition module is configured to acquire a bus voltage signal of the microgrid in a current time period through a common connection point of the microgrid; The phase perturbation module is configured to perform phase perturbation extraction on the bus voltage signal to obtain a quantum phase perturbation sequence, and specifically includes: performing Hilbert transform on the bus voltage signal to obtain an instantaneous phase sequence; performing differential operation on the instantaneous phase sequence according to a fundamental phase of the microgrid to obtain a phase deviation sequence; removing linear drift in the phase deviation sequence by sliding least squares fitting to obtain a residual phase perturbation sequence; and performing quantum noise screening on the residual phase perturbation sequence to generate the quantum phase perturbation sequence; The modulation component separation module is configured to perform identification and separation processing on the quantum phase perturbation sequence by an adaptive noise shaping algorithm to obtain a quadratic modulation component of a bus of the microgrid, and specifically includes: dividing the quantum phase perturbation sequence into a plurality of subsequences by sliding windows; performing wavelet packet decomposition on each of the subsequences to obtain a plurality of layers of detail coefficients and approximation coefficients, and performing adaptive threshold quantization on the plurality of layers of detail coefficients and the approximation coefficients according to energy entropy to obtain a reconstructed perturbation subsequence; inputting the reconstructed perturbation subsequence into a noise shaper to perform frequency spectrum shift and shaping on the reconstructed perturbation subsequence to obtain a shaped perturbation subsequence; and performing Hilbert transform on the shaped perturbation subsequence and extracting an instantaneous amplitude envelope to obtain the quadratic modulation component; The clustering module is configured to perform unsupervised clustering on the quadratic modulation component by a density peak clustering algorithm to obtain a plurality of aging mode clusters; The feature extraction module is configured to perform feature extraction on the quadratic modulation component of the aging mode cluster to obtain statistical features of the aging mode cluster; The initial evaluation module is configured to obtain an initial residual energy evaluation value according to the statistical features of the aging mode cluster by a variable coefficient regression submodel; The covariance estimation module is configured to perform integral of instantaneous value sequence of the bus voltage of the microgrid in the current time period with a preset time window to determine an estimated covariance of a statistical mean channel, and perform Kalman wave prediction correction cycle according to the bus voltage to obtain an estimated covariance of a Kalman filter channel; The fusion output module is configured to perform weighted fusion on the initial residual energy evaluation value according to the estimated covariances corresponding to the Kalman filter channel and the statistical mean channel respectively to obtain a final energy evaluation value.

8. An electronic device, comprising: comprises a memory and a processor; the memory is configured to store a computer program; the processor is configured to implement the energy storage prediction method of any one of claims 1-6 when executing the computer program.

Images