CN-121995227-A - Storage battery residual life prediction system and method based on physical information neural network algorithm model

CN121995227ACN 121995227 ACN121995227 ACN 121995227ACN-121995227-A

Abstract

The invention relates to the technical field of battery remaining life prediction and discloses a battery remaining life prediction system and method based on a physical information neural network algorithm model. The method comprises the steps of applying multi-frequency-point alternating excitation and pulse current to a storage battery, collecting electrochemical impedance spectrum data, polarized internal resistance data and voltage current temperature data, creating a feature vector based on the electrochemical impedance spectrum data, the polarized internal resistance data and the voltage current temperature data, inputting the feature vector into a physical information neural network model, and outputting a health state predicted value and a residual life predicted value. The invention can guide the physical information neural network algorithm model to learn the mapping relation conforming to the battery degradation mechanism, improves the generalization capability of the model and the physical interpretability of the prediction result, and further improves the accuracy of the prediction of the residual life of the storage battery.

Inventors

YAN SHIQIN
CAI KUNHUA
YUAN HAITAO
WAN MINGMING
Huang Kechuang

Assignees

深圳中鲲智能科技有限公司

Dates

Publication Date: 20260508
Application Date: 20260129

Claims (10)

1. The method for predicting the residual life of the storage battery based on the physical information neural network algorithm model is characterized by comprising the following steps of: Applying multi-frequency point alternating excitation and pulse current to the storage battery, and collecting electrochemical impedance spectrum data, polarized internal resistance data and voltage current temperature data; Creating a feature vector based on the electrochemical impedance spectrum data, the polarized internal resistance data, and the voltage-current-temperature data; And inputting the feature vector into a physical information neural network model, and outputting a health state predicted value and a residual life predicted value.
2. The method for predicting remaining life of a battery based on a physical information neural network algorithm model according to claim 1, wherein applying multi-frequency point alternating excitation and pulse current to the battery, collecting electrochemical impedance spectrum data, polarized internal resistance data and voltage current temperature data, comprises: Applying multi-frequency point alternating excitation and pulse current to the storage battery, and collecting impedance response, pulse voltage response and voltage current temperature data; And performing frequency domain transformation on the impedance response to obtain electrochemical impedance spectrum data, and performing state estimation on the impulse voltage response to obtain polarized internal resistance data.
3. The method for predicting remaining life of a battery based on a physical information neural network algorithm model of claim 1, wherein creating a feature vector based on the electrochemical impedance spectrum data, the polarized internal resistance data, and the voltage-current-temperature data comprises: Carrying out real part and imaginary part decomposition on complex impedance values of all frequency points of the electrochemical impedance spectrum data, extracting ohmic internal resistance, charge transfer impedance and diffusion impedance parameters, and taking the ohmic internal resistance, the charge transfer impedance and the diffusion impedance parameters as a frequency domain impedance characteristic set; Extracting a polarization internal resistance steady-state value, a polarization time constant and a polarization voltage peak value from the polarization internal resistance data, and taking the polarization internal resistance steady-state value, the polarization time constant and the polarization voltage peak value as a time-varying feature set; Intercepting a charging fragment from the voltage-current temperature data, calculating a voltage mean value, a voltage standard deviation, a voltage gradient change rate, a current peak value, a current decay rate and a temperature rise amplitude, and taking the charging fragment as a charging fragment statistical characteristic set; and constructing a feature vector based on the frequency domain impedance feature set, the time-varying feature set and the charging fragment statistical feature set.
4. The method for predicting remaining battery life based on a physical information neural network algorithm model according to claim 1, wherein before inputting the feature vector into the physical information neural network model, the method for predicting remaining battery life based on the physical information neural network algorithm model further comprises: Characterizing the storage battery as a multistage equivalent circuit with anode impedance, electrolyte impedance and cathode impedance connected in series, and constructing a state space equation, wherein state variables of the state space equation comprise solid electrolyte interface membrane impedance, charge transfer impedance and diffusion impedance; Taking the predicted output characteristics of a target neural network as a state variable estimated value of the state space equation, and calculating a physical residual error after substituting the state variable estimated value into the state space equation to obtain a physical constraint loss term, wherein the target neural network comprises a convolutional neural network, a long-term and short-term memory network and a multi-layer perceptron; And constructing a total loss function comprising a data driving loss term, the physical constraint loss term and a boundary condition loss term, and training the target neural network by using the total loss function to obtain a physical information neural network model.
5. The method for predicting remaining life of a battery based on a physical information neural network algorithm model according to claim 4, wherein constructing a total loss function including a data driving loss term, the physical constraint loss term, and a boundary condition loss term, training the target neural network with the total loss function, and obtaining a physical information neural network model, comprises: Constructing a total loss function comprising the data driving loss term, the physical constraint loss term and the boundary condition loss term according to the physical constraint weight coefficient and the boundary condition weight coefficient; Randomly extracting a batch of samples and real labels thereof from a training set, inputting sample vectors into the target neural network, calculating predicted output characteristics, and calculating a total loss function value according to the predicted output characteristics and the real labels; According to the total loss function value, back propagation is carried out to calculate weight parameter gradients of each layer, an optimizer is adopted to update weight parameters, and the learning rate is adjusted according to exponential decay; and monitoring the total loss function value on the verification set, and stopping iteration when the total loss function value is not reduced for N continuous rounds, so as to obtain the physical information neural network model.
6. The method for predicting remaining life of a battery based on a physical information neural network algorithm model according to claim 1, wherein inputting the feature vector into the physical information neural network model, outputting a health state predicted value and a remaining life predicted value, comprises: Dividing the eigenvector into an impedance spectrum eigenvector, a time sequence eigenvector and a statistical eigenvector; Inputting the impedance spectrum feature vector into a convolutional neural network in a physical information neural network model, extracting frequency domain space features, inputting the time sequence feature vector into a long-term and short-term memory network in the physical information neural network model, extracting degradation trend features, inputting the statistical feature vector into a multi-layer perceptron in the physical information neural network model, and extracting nonlinear mapping features; And inputting the frequency domain spatial characteristics, the degradation trend characteristics and the nonlinear mapping characteristics into a fully-connected output layer in a physical information neural network model to perform nonlinear activation transformation, and outputting a health state predicted value and a residual life predicted value.
7. The method for predicting remaining life of a battery based on a physical information neural network algorithm model according to claim 6, wherein inputting the frequency domain spatial feature, the degradation trend feature and the nonlinear mapping feature into a fully connected output layer in the physical information neural network model for nonlinear activation transformation, outputting a health state predicted value and a remaining life predicted value, comprises: splicing the frequency domain spatial feature, the degradation trend feature and the nonlinear mapping feature into a comprehensive feature vector; And inputting the comprehensive feature vector into a fully-connected output layer in a physical information neural network model to perform weighted summation and nonlinear activation transformation, respectively calculating through a first output node to obtain a health state predicted value, and calculating through a second output node to obtain a residual life predicted value.
8. The method for predicting remaining life of a battery based on a physical information neural network algorithm model according to claim 1, wherein after outputting the predicted value of the state of health and the predicted value of remaining life, the method for predicting remaining life of a battery based on a physical information neural network algorithm model further comprises: the storage battery is controlled to execute constant-current discharge through remote nuclear capacity operation, discharge current and discharge time are collected in real time, and integral calculation is carried out on the product of the discharge current and the discharge time to obtain actual discharge capacity; Dividing the actual discharge capacity by the rated capacity of the storage battery and multiplying the actual discharge capacity by one hundred percent to obtain an actual capacity measurement value; Calculating a prediction error value of the health state predicted value and the actual capacity measured value; When the prediction error value exceeds a set threshold value, constructing a fine-tuning data set to perform transfer learning updating on the physical information neural network model, and obtaining an updated physical information neural network model.
9. The method for predicting the remaining life of a storage battery based on a physical information neural network algorithm model according to claim 8, wherein when the prediction error value exceeds a set threshold, constructing a fine-tuning data set to perform migration learning update on the physical information neural network model, and obtaining an updated physical information neural network model, comprises: Combining the newly acquired electrochemical impedance spectrum data, polarized internal resistance data, voltage current temperature data and corresponding actual capacity measurement value with historical training data to obtain a fine tuning data set; Freezing weight parameters of physical constraint branches in the physical information neural network model, setting a first bottom layer weight of a convolutional neural network and a second bottom layer weight of a long-short-term memory network as a first learning rate, setting a first weight of a multi-layer perceptron and a second weight of a fully-connected output layer as a second learning rate, wherein the second learning rate is larger than the first learning rate; And extracting batch samples from the fine adjustment data set, inputting the batch samples into the physical information neural network model to calculate a total loss function, and respectively updating weight parameters by back propagation to obtain an updated physical information neural network model.
10. A storage battery residual life prediction system based on a physical information neural network algorithm model is characterized in that, the steps for implementing the battery remaining life prediction method based on the physical information neural network algorithm model according to any one of claims 1 to 9, comprising: The acquisition module is used for applying multi-frequency-point alternating excitation and pulse current to the storage battery and acquiring electrochemical impedance spectrum data, polarized internal resistance data and voltage current temperature data; a creation module for creating a feature vector based on the electrochemical impedance spectrum data, the polarized internal resistance data, and the voltage-current-temperature data; And the output module is used for inputting the feature vector into the physical information neural network model and outputting a health state predicted value and a residual life predicted value.

Description

Storage battery residual life prediction system and method based on physical information neural network algorithm model Technical Field The invention relates to the technical field of battery remaining life prediction, in particular to a battery remaining life prediction system and method based on a physical information neural network algorithm model. Background In the power industry, a storage battery is used as a core component of an ac/dc power supply system, and the health state of the storage battery is directly related to power supply safety. The traditional storage battery monitoring method mainly collects surface parameters such as voltage, temperature, single-frequency point internal resistance and the like, data are thick and shallow, the aging attenuation state of a battery cannot be effectively estimated, and accurate data support is difficult to provide for storage battery replacement in advance. The existing pure data driving neural network prediction method can learn the statistical rule in the historical data, but lacks constraint of a battery degradation physical mechanism, so that the model has insufficient generalization capability when training samples are limited or no working condition is seen, a prediction result lacks physical interpretability, and unreasonable prediction violating an electrochemical law is easy to occur. Disclosure of Invention The invention mainly aims to provide a storage battery residual life prediction system and method based on a physical information neural network algorithm model, which can guide the physical information neural network algorithm model to learn a mapping relation conforming to a battery degradation mechanism, improve the generalization capability of the model and the physical interpretability of a prediction result, and further improve the accuracy of the residual life prediction of a storage battery. In order to achieve the above purpose, the invention provides a method for predicting the remaining life of a storage battery based on a physical information neural network algorithm model, which comprises the following steps: Applying multi-frequency point alternating excitation and pulse current to the storage battery, and collecting electrochemical impedance spectrum data, polarized internal resistance data and voltage current temperature data; Creating a feature vector based on the electrochemical impedance spectrum data, the polarized internal resistance data, and the voltage-current-temperature data; And inputting the feature vector into a physical information neural network model, and outputting a health state predicted value and a residual life predicted value. Optionally, in a first implementation manner of the first aspect of the present invention, applying multi-frequency point ac excitation and pulse current to the storage battery, collecting electrochemical impedance spectrum data, polarization internal resistance data, and voltage current temperature data includes: Applying multi-frequency point alternating excitation and pulse current to the storage battery, and collecting impedance response, pulse voltage response and voltage current temperature data; And performing frequency domain transformation on the impedance response to obtain electrochemical impedance spectrum data, and performing state estimation on the impulse voltage response to obtain polarized internal resistance data. Optionally, in a second implementation manner of the first aspect of the present invention, creating a feature vector based on the electrochemical impedance spectrum data, the polarized internal resistance data, and the voltage-current-temperature data includes: Carrying out real part and imaginary part decomposition on complex impedance values of all frequency points of the electrochemical impedance spectrum data, extracting ohmic internal resistance, charge transfer impedance and diffusion impedance parameters, and taking the ohmic internal resistance, the charge transfer impedance and the diffusion impedance parameters as a frequency domain impedance characteristic set; Extracting a polarization internal resistance steady-state value, a polarization time constant and a polarization voltage peak value from the polarization internal resistance data, and taking the polarization internal resistance steady-state value, the polarization time constant and the polarization voltage peak value as a time-varying feature set; Intercepting a charging fragment from the voltage-current temperature data, calculating a voltage mean value, a voltage standard deviation, a voltage gradient change rate, a current peak value, a current decay rate and a temperature rise amplitude, and taking the charging fragment as a charging fragment statistical characteristic set; and constructing a feature vector based on the frequency domain impedance feature set, the time-varying feature set and the charging fragment statistical feature set. Optionally, in a third implementation manner