CN-117590267-B - Early warning method and system for performance degradation of energy storage battery

CN117590267BCN 117590267 BCN117590267 BCN 117590267BCN-117590267-B

Abstract

The invention discloses an early warning method for performance degradation of an energy storage battery, which comprises the steps of obtaining a capacity degradation data set of the energy storage battery and preprocessing the capacity degradation data set to obtain a training data set, carrying out wavelet analysis to obtain multidimensional intrinsic components under different frequencies, carrying out principal component analysis to obtain a core frequency signal, constructing an early warning initial model for the performance degradation of the energy storage battery and training the initial model by adopting the core frequency signal to obtain an early warning model for the performance degradation of the energy storage battery, and carrying out actual early warning for the performance degradation of the energy storage battery by adopting the early warning model for the performance degradation of the energy storage battery. The invention also discloses a system for realizing the early warning method for the performance degradation of the energy storage battery. The invention not only can realize early warning of the performance degradation of the energy storage battery, but also has higher reliability, better accuracy and better universality.

Inventors

ZHANG XINGWEI
TIAN JIAXIANG
Ma Qiongwu
XU ZHIQIANG
XIE YUXIANG
CHEN ZHONGWEI
HE LI
Liao Yingao
YANG DONGMEI
YANG XIAO
ZHANG JIANLIANG

Assignees

国网湖南省电力有限公司
国网湖南省电力有限公司经济技术研究院
国家电网有限公司

Dates

Publication Date: 20260512
Application Date: 20231123

Claims (9)

1. An early warning method for performance degradation of an energy storage battery comprises the following steps: s1, acquiring a capacity degradation data set of an energy storage battery; s2, carrying out data preprocessing on the data obtained in the step S1 to obtain a training data set; s3, carrying out wavelet analysis on the training data set obtained in the step S2 to obtain multidimensional intrinsic components under different frequencies; s4, carrying out principal component analysis on the intrinsic component data information obtained in the step S3 to obtain a core frequency signal; s5, constructing an early warning initial model of energy storage battery performance degradation based on Informer networks, wherein the method specifically comprises the following steps: Based on Informer network, constructing an early warning initial model of energy storage battery performance degradation; the model comprises an encoder, a decoder and a full connection layer; the encoder encodes the input sequence using a probability self-attention mechanism and inputs the encoded data to the decoder; the decoder decodes the input coded data by adopting a one-time generation type prediction mode, and uploads the decoded data to the full-connection layer; The full connection layer is used for obtaining a final output result according to the input decoding data; S6, training the initial model constructed in the step S5 by adopting the core frequency signal obtained in the step S4 to obtain an early warning model of the performance degradation of the energy storage battery; and S7, performing actual early warning of the performance degradation of the energy storage battery by adopting the early warning model of the performance degradation of the energy storage battery obtained in the step S6.
2. The method for early warning of performance degradation of an energy storage battery according to claim 1, wherein the acquiring of the capacity degradation dataset of the energy storage battery in step S1 specifically comprises the following steps: Acquiring a capacity degradation data set of an energy storage battery in the following way; acquiring a group of energy storage batteries; fully charging the energy storage battery in a set normal temperature environment by adopting a mode specified by the energy storage battery, placing the energy storage battery in a measured environment for a set time, discharging the energy storage battery to a cut-off voltage by adopting a 1C multiplying power, and recording capacity data of the energy storage battery in the discharging process; Repeating the above process for several times, recording capacity data of each time, and taking the average value to obtain the final capacity degradation data set of the energy storage battery.
3. The early warning method for performance degradation of an energy storage battery according to claim 2, wherein the step S3 of performing wavelet analysis on the training data set obtained in the step S2 to obtain multidimensional intrinsic components under different frequencies specifically comprises the following steps: A. selecting a wavelet basis function suitable for the capacity degradation data set of the energy storage battery, wherein the wavelet basis function is expressed as: In the middle of Is a wavelet basis function; T is a time domain variable of the function; Is the scale of wavelet analysis; the shift amount for wavelet analysis; B. Wavelet basis function to be selected Capacity degradation data with an energy storage battery to be analyzed Doing inner product and calculating coefficient C for representing And Is a degree of similarity of (2); C. shifting the wavelet basis function by k units to the right to obtain a second wavelet basis function Repeating step B until Ending; D. expanding the wavelet basis function to obtain a third wavelet basis function Repeating steps B and C; E. Repeating the step D to expand the wavelet basis function until the condition is set; finally, the overall formula for the continuous wavelet transform is: In the middle of Is the overall function of the continuous wavelet transform; And obtaining multidimensional intrinsic components under different frequencies according to a total formula of continuous wavelet transformation.
4. The early warning method for performance degradation of an energy storage battery according to claim 3, wherein the main component analysis is performed on the intrinsic component data information obtained in step S3 in step S4 to obtain a core frequency signal, and specifically comprises the following steps: Acquiring intrinsic component data information data obtained in the step S3, wherein the number of the frequency signals of the wavelet component is n, each frequency signal sequence comprises m index variables, and the j index of the i frequency signal is expressed as ; The various indices were normalized using the following formula: In the middle of A j index which is the normalized i frequency signal; The sample mean value of the j index; standard deviation of the j index; the correlation coefficient matrix R is calculated using the following equation: In the middle of The correlation coefficient of the ith index and the jth index; Calculating m eigenvalues of a correlation coefficient matrix R And corresponding feature vector Wherein the j-th feature vector Represented as , ; According to the eigenvector The m index variables are calculated using the following equation: In the middle of Is the m-th principal component; is the nth standardized index variable; calculating information contribution rates and accumulated contribution rates of m eigenvalues: In the middle of As the j-th principal component Is a ratio of information contribution of (a); As the j-th principal component Is a cumulative contribution rate of (a); According to the accumulated contribution rate of m eigenvalues, the following judgment is carried out: If it is The main components corresponding to the first p characteristic values are selected as core frequency signals; Is a set threshold.
5. The method for early warning of performance degradation of an energy storage battery according to claim 4, wherein the input sequence specifically comprises the following contents: taking the core frequency signal obtained in the step S4 as an input multivariable sequence, and expressing as follows: In the middle of To input a multivariate sequence; is the length of the current input sequence; Is that Time of day input sequence A plurality of points; Vector dimensions for each sequence point; The model adopts a point multiplication self-attention mechanism and adopts a time stamp as position information coding, and the time stamp calculation comprises a local time stamp PE and a global time stamp SE; the local timestamp PE is calculated using the following equation: In the middle of Is a local timestamp; Is position information; , Is a rounding operation; the feature dimension is the feature dimension after input; after the time stamp coding is aligned with the input dimension, an input representation vector of the model is obtained: In the middle of For data that is ultimately input into the encoder; To balance the factors of size between scalar mapping and local/global embedding; As a feature scalar quantity, ; Is a local timestamp; is a global timestamp; Representing a total of p types of global timestamps.
6. The method for early warning of performance degradation of an energy storage battery according to claim 5, wherein the encoder comprises the following specific contents: self-attention is given to the input of the receiving tuple The method comprises the steps of defining and executing a dot product after scaling, wherein Q is a Query Value, K is a Key Key Value, and V is a Value; obtained by linear transformation of an input matrix X and expressed as: In the middle of The method comprises the steps of obtaining a query parameter matrix to be trained; the key parameter matrix to be trained; The value parameter matrix to be trained; The self-attention mechanism is obtained: In the middle of A calculation formula for an attention mechanism; Is a normalized exponential function; Is the dimension of the input vector; By using Represents the i-th row of Q, The i-th row of the symbol K is indicated, Note attention, representing line i of V, that gets the ith Query is defined as a kernel smoothed probability form: In the middle of Attention for the ith Query; is an asymmetric exponential kernel function, and ; Line j of V; is a kernel smoothed probability form of attention; According to the above formula, attention attention of the ith Query to all Key keys is defined as a probability distribution , ; The KL-divergence is used to measure the similarity between the distributions p and q: In the middle of KL-divergence between the distributions p and q; Is the vector length; after removing the constant, the sparsity measure of the ith query is defined as: In the middle of Sparsity measure for the ith query; according to the KL-divergence value, each key is made to pay attention to only u queries to realize probability sparse self-attention, expressed as: In the middle of Is a sparse matrix with the same size as q and only comprises sparse metrics The next most important u queries; thus, the approximation of the query sparsity metric is obtained as: In the middle of A calculation result of the query sparsity measure; As a natural consequence of the probabilistic self-attention mechanism, the feature map obtained by the encoder has redundant combinations of values, the core features are preserved using a distillation operation, and the progress of the distillation operation from the j-th layer to the j+1th layer is expressed as: In the middle of Is the characteristic diagram of the j+1th layer; A max-pooling downsampling function; is an activation function; is a one-dimensional convolution function; is the feature map of the j-th layer; Is a fundamental operation of probability self-attention; In order to enhance the robustness of the distillation operation, a plurality of stacks of encoders are built in the encoder architecture, each stack being an independent sub-encoder, one layer at a time is discarded as the number of stacks increases, thereby reducing the number of distillation operation layers, eventually aligning the output dimensions, and finally the outputs of all stacks are concatenated to obtain the final representation of the encoder Encoder.
7. The method for early warning of performance degradation of an energy storage battery according to claim 6, wherein the decoder comprises the following specific contents: the Decoder comprises a probability sparse self-attention and multi-head self-attention module; the decoder produces the long sequence output by a forward process: In the middle of Is the data which is input into the Decoder after being processed; splicing the two groups of vectors; Is a Start token; placeholders for target sequences to be predicted; the length of the sampling start token sequence is the length of the sampling start token sequence; A predicted sequence length for the decoder; Data feature dimensions; by selecting the mask size, each bit in the X-sequence is prevented from focusing on future positions, thus avoiding autoregressions.
8. The method for early warning of performance degradation of an energy storage battery according to claim 7, wherein the training in step S6 obtains an early warning model of performance degradation of the energy storage battery, and specifically comprises the following steps: during training, logCosh loss functions are used for training the model.
9. A system for realizing the early warning method of the performance degradation of the energy storage battery according to one of claims 1 to 8, which is characterized by comprising a data acquisition module, a data processing module, a wavelet analysis module, a principal component analysis module, a model construction module, a model training module and an early warning module; the system comprises a data acquisition module, a data processing module, a wavelet analysis module, a principal component analysis module, a model construction module, a model training module and an early warning module which are sequentially connected in series, wherein the data acquisition module is used for acquiring a capacity degradation data set of an energy storage battery and uploading data to the data processing module, the data processing module is used for preprocessing the data according to the received data to obtain a training data set and uploading the data to the wavelet analysis module, the wavelet analysis module is used for carrying out wavelet analysis on the training data set according to the received data to obtain multidimensional intrinsic components under different frequencies and uploading the data to the principal component analysis module, the principal component analysis module is used for carrying out principal component analysis on intrinsic component data information according to the received data to obtain a core frequency signal and uploading the data to the model construction module, the model construction module is used for constructing an early initial model of the performance degradation of the energy storage battery based on Informer network, the model training module is used for carrying out training on the constructed initial model according to the received data to obtain an early warning model of the performance degradation of the energy storage battery, the early warning module is used for uploading the data to obtain the early warning model of the performance degradation of the energy storage battery according to the received data, early warning of the performance degradation of the actual energy storage battery is carried out.

Description

Early warning method and system for performance degradation of energy storage battery Technical Field The invention belongs to the field of electrical automation, and particularly relates to an early warning method and system for performance degradation of an energy storage battery. Background Along with the development of economic technology and the improvement of living standard of people, electric energy becomes an indispensable secondary energy source in the production and living of people, and brings endless convenience to the production and living of people. Therefore, ensuring stable and reliable supply of electric energy becomes one of the most important tasks of the electric power system. At present, new energy power generation systems are beginning to be largely integrated into the power grid for power generation. However, the randomness of the output of the new energy power generation system brings great operating pressure to the power system. In order to better eliminate new energy output and simultaneously to better cut peaks and fill valleys, the power system ensures stable and reliable operation, and the power system is gradually provided with an energy storage battery system. After the energy storage battery system is connected with the power system, great assurance is brought for the stable and reliable operation of the power system. Therefore, it is very important to predict and pre-warn the performance of the energy storage battery. At present, conventional early warning schemes for performance degradation of an energy storage battery often adopt various types of neural network schemes for prediction, such as AlexNet network architecture, LSTM network architecture or WaveNet related scheme. However, such schemes still have the disadvantage of low reliability and poor accuracy. Disclosure of Invention The invention aims to provide an early warning method for the performance degradation of an energy storage battery, which has high reliability, good accuracy and good universality. The second objective of the present invention is to provide a system for implementing the early warning method for performance degradation of the energy storage battery. The early warning method for the performance degradation of the energy storage battery provided by the invention comprises the following steps: s1, acquiring a capacity degradation data set of an energy storage battery; s2, carrying out data preprocessing on the data obtained in the step S1 to obtain a training data set; s3, carrying out wavelet analysis on the training data set obtained in the step S2 to obtain multidimensional intrinsic components under different frequencies; s4, carrying out principal component analysis on the intrinsic component data information obtained in the step S3 to obtain a core frequency signal; s5, constructing an early warning initial model of energy storage battery performance degradation based on Informer networks; s6, training the initial model constructed in the step S5 by adopting the core frequency signal obtained in the step S4 to obtain an early warning model of the performance degradation of the energy storage battery; S7, performing actual early warning of the performance degradation of the energy storage battery by adopting the early warning model of the performance degradation of the energy storage battery obtained in the step S6. The step S1 of acquiring the capacity degradation data set of the energy storage battery specifically comprises the following steps: Acquiring a capacity degradation data set of an energy storage battery in the following way; acquiring a group of energy storage batteries; fully charging the energy storage battery in a set normal temperature environment by adopting a mode specified by the energy storage battery, placing the energy storage battery in a measured environment for a set time, discharging the energy storage battery to a cut-off voltage by adopting a 1C multiplying power, and recording capacity data of the energy storage battery in the discharging process; Repeating the above process for several times, recording capacity data of each time, and taking the average value to obtain the final capacity degradation data set of the energy storage battery. The step S3 of carrying out wavelet analysis on the training data set obtained in the step S2 to obtain multidimensional intrinsic components under different frequencies specifically comprises the following steps: A. selecting a wavelet basis function suitable for the capacity degradation data set of the energy storage battery, wherein the wavelet basis function is expressed as: Wherein, psi a,b (t) is a wavelet basis function, psi (t) is a wavelet function before transformation, t is a time domain variable of the function, a is a scale of wavelet analysis, b is a translation amount of wavelet analysis; B. Taking the selected wavelet basis function psi a,b (t) as an inner product with capacity degradation data f (t) of the ener