Lithium battery internal state estimation method based on physical information neural network

Abstract

The invention relates to the technical field of battery health state management, and discloses a lithium battery internal state estimation method based on a physical information neural network, and the method comprises the steps: obtaining a loss function for carrying out the optimization training of the internal parameters of a neural network model through an e-SPM model, enabling the internal parameters of the neural network model to be subjected to the optimization training through the loss function, and achieving the estimation of the internal state of a lithium battery. And internal state estimation of the lithium battery is realized. According to the invention, the current, voltage and temperature of the lithium battery are directly mapped to the internal electrochemical state of the lithium battery, so that the output result of the physical neural network is more explanatory and is easy to understand and analyze, and the output state quantity is substituted into the simplified physical model of the voltage by adopting an unsupervised learning mode. In the training process, a large amount of labeled data is not needed, and the experimental data acquisition difficulty is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of battery health status management, and in particular to a method for estimating the internal state of a lithium battery based on a physical information neural network. Background Art

[0002] Electric Vehicles (EVs) are generally regarded as an effective way to cope with the challenges of excessive carbon emissions and the gradual depletion of fossil fuels. Among them, the energy storage system, as a core component of electric vehicles, largely determines its performance and commercialization potential. Due to its advantages such as high energy density and long life, lithium-ion batteries (LIBs) dominate the energy storage field of electric vehicles. In order to meet the high voltage and high capacity requirements of electric vehicles, multiple single cells are usually used to form a battery system. Therefore, a battery management system (Battery Management System, BMS) is needed to monitor and control these battery cells to maintain the safety and efficient operation of the entire system. State estimation is the key to obtaining dynamic information inside the battery and is also a prerequisite for the efficient implementation of other BMS functions.

[0003] A RNN for online SOC and health state estimation of electric vehicle batteries is proposed in the prior art. Compared with the traditional multi-layer perceptron, the network shows better performance; the LSTM network proposed in the prior art is used for SOC and SOH estimation under different ambient temperatures, where LSTM shows better performance than the standard RNN; a combined convolutional neural network (CNN)-LSTM for lithium-ion battery SOC estimation in the prior art can extract spatial and temporal features from input data, and its performance is better than CNN and LSTM themselves.

[0004] Although these data-driven methods have achieved satisfactory results in SOC or SOH estimation, no data-driven methods based on physical models have been proposed to track the dynamic changes of unmeasurable electrochemical states inside lithium-ion batteries. Moreover, a large amount of training data is required to optimize the loss function and network parameters, and it is impossible to accurately estimate the internal state of the battery that is difficult to obtain experimentally. Summary of the invention

[0005] 1. Technical issues to be resolved In view of the shortcomings of the prior art, the present invention provides a method for estimating the internal state of a lithium battery based on a physical information neural network, which has the advantages of simplifying data labeling, and solves the problem that it is difficult to obtain the internal state of a lithium battery, thereby affecting the estimation of its internal state.

[0006] (II) Technical solution To achieve the above object, the present invention provides the following technical solution: a method for estimating the internal state of a lithium battery based on a physical information neural network, comprising the following steps: S1. Simplify the P2D model into an e-SPM model to obtain a simplified physical model that can represent the voltage by the solid phase lithium ion concentration and the liquid phase lithium ion concentration; S2, inputting the actually detected voltage U, current I and temperature T into the neural network model to obtain the output state quantity; S3. Substitute the output state quantity into the simplified physical model of the voltage to obtain the estimated terminal voltage value and the estimated terminal voltage of the battery body. , and calculate the mean square error with the actual detection voltage value and the SOC of the actual detection end to obtain the loss function; S4. The neural network model optimizes its internal parameters through the loss function to achieve internal state estimation of the lithium battery.

[0007] Preferably, the simplified physical model of the voltage in S2 is obtained by: Assume that in the two electrodes, the solid concentration The x dimension remains constant, so the liquid phase lithium ion concentration model is: ; The solid phase lithium ion concentration model is: ; Among them, in the two electrodes, the solid phase concentration Keeping the x dimension constant, represents the initial lithium ion concentration in the electrolyte, , and denote the electrolyte volume fractions of the negative electrode, separator and positive electrode, respectively, , and Respectively represent the length of the negative electrode, separator and positive electrode, S represents the cross-sectional area of ​​the battery cell, and represent the initial lithium ion concentrations of the negative electrode and the positive electrode, respectively, and represent the volume fractions of the negative electrode and the positive electrode, respectively; When the battery temperature When the change in the x dimension is small and only changes with time t, the solid phase lithium ion concentration model and the liquid phase lithium ion concentration model are input into the neural network model to output four voltage term models: ; ; ; ; Therefore, the simplified physical model of voltage is: ; in, represents the estimated equilibrium potential, The first part is the ohmic potential drop due to electrolyte conductivity, and the second part is the electrolyte concentration overpotential estimate, represents the voltage on the thin film resistor, represents the overpotential, and represents the equilibrium potential of the positive and negative electrodes, and represent the surface lithium ion concentration of the negative electrode and the positive electrode respectively, , as well as They represent the effective reaction rate constants of the negative electrode, the separator and the positive electrode respectively, I represents the charge and discharge current, R represents the universal gas constant, and F represents the Faraday constant. represents the lithium ion migration number, and Represent the positive and negative film resistances respectively, and Represent the interface area of ​​the positive and negative spherical particles, α represents the charge transfer coefficient of the electrode, and represent the exchange current density of the positive and negative electrodes respectively.

[0008] Preferably, the loss function in S4 is obtained by: Input voltage U, current I and temperature T into the neural network model to output three output state quantities , and , which are the electrode surface concentration, the electrode average concentration and the electrolyte concentration, respectively. Substituting the three output state quantities into the estimated terminal voltage model, four simplified voltage term models are obtained: ; ; ; ; Thus the estimated terminal voltage is calculated: ; in, , , and They are four voltage term models, G is the simplified physical model of voltage, represents the simplified equilibrium potential estimate, The first part is the simplified ohmic potential drop due to electrolyte conductivity, and the second part is the simplified potential estimate of the electrolyte concentration, represents the simplified voltage on the thin film resistor, represents simplified overpotential; Calculate the estimated end of the battery : ; in, represents the average concentration at the anode electrode, and are the maximum and minimum lithium insertion amounts of the anode within the operating voltage range, Indicates the maximum lithium ion concentration of the anode.

[0009] The estimated end and estimated voltage value The actual detection end SOC and actual detection voltage value The mean square error is calculated to obtain the loss function: ; in, represents a time sample, Indicates the actual detection voltage value at time t, represents the estimated terminal voltage value at time t, Indicates the estimated end at time t , Represents the SOC of the actual detection end at time t.

[0010] (III) Beneficial effects Compared with the prior art, the present invention provides a method for estimating the internal state of a lithium battery based on a physical information neural network, which has the following beneficial effects: 1. This method of estimating the internal state of a lithium battery based on a physical information neural network uses the e-SPM model to obtain a loss function for optimizing the training of the internal parameters of the neural network model, directly mapping the lithium battery current, voltage, and temperature to the internal electrochemical state of the lithium battery, making the output results of the physical neural network (PINNs) more interpretable and easy to understand and analyze.

[0011] 2. This method of estimating the internal state of a lithium battery based on a physical information neural network is different from the traditional data-driven method which requires a large amount of known data for training. In this method, an unsupervised learning method is used to substitute the output state quantity into a simplified physical model of the voltage and embed it into the loss function of the neural network as a constraint. During this training process, a large amount of labeled data is not required, which reduces the difficulty of obtaining experimental data. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a schematic diagram of the process of the present invention; Figure 2 This is the e-SPM model diagram of the present invention; Figure 3 This is the P2D model diagram of the present invention; Figure 4 This is a structural block diagram of the physical neural network (PINNs) of the present invention. DETAILED DESCRIPTION

[0013] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0014] like Figure 1-4 As shown, a method for estimating the internal state of a lithium battery based on a physical information neural network comprises the following steps: S1. Simplify the P2D model into an e-SPM model to obtain a simplified physical model that can represent the voltage by the solid phase lithium ion concentration and the liquid phase lithium ion concentration; S2, inputting the actually detected voltage U, current I and temperature T into the neural network model to obtain the output state quantity; S3. Substitute the output state quantity into the simplified physical model of the voltage to obtain the estimated terminal voltage value and the estimated terminal voltage of the battery body. , and calculate the mean square error with the actual detection voltage value and the SOC of the actual detection end to obtain the loss function; S4. The neural network model optimizes its internal parameters through the loss function to achieve internal state estimation of the lithium battery.

[0015] In this embodiment, Figure 1 As shown, the solid phase in the cathode and anode consists of two radii, and The particle representation of S2 is as follows: Assume that in the two electrodes, the solid concentration The x dimension remains constant, so the liquid phase lithium ion concentration model is: ; The solid phase lithium ion concentration model is: ; Among them, in the two electrodes, the solid phase concentration Keeping the x dimension constant, represents the initial lithium ion concentration in the electrolyte, , and denote the electrolyte volume fractions of the negative electrode, separator and positive electrode, respectively, , and Respectively represent the length of the negative electrode, separator and positive electrode, S represents the cross-sectional area of ​​the battery cell, and represent the initial lithium ion concentrations of the negative electrode and the positive electrode, respectively, and represent the volume fractions of the negative electrode and the positive electrode, respectively; When the battery temperature When the change in the x dimension is small and only changes with time t, the solid phase lithium ion concentration model and the liquid phase lithium ion concentration model are input into the neural network model to output four voltage term models: ; ; ; ; Therefore, the simplified physical model of voltage is: ; in, represents the estimated equilibrium potential, The first part is the ohmic potential drop due to electrolyte conductivity, and the second part is the electrolyte concentration overpotential estimate, represents the voltage on the thin film resistor, represents the overpotential, and represents the equilibrium potential of the positive and negative electrodes, and represent the surface lithium ion concentration of the negative electrode and the positive electrode respectively, , as well as They represent the effective reaction rate constants of the negative electrode, the separator and the positive electrode respectively, I represents the charge and discharge current, R represents the universal gas constant, and F represents the Faraday constant. Represents the lithium ion migration number, and Represent the positive and negative film resistances respectively, and Represent the interface area of ​​the positive and negative spherical particles, α represents the charge transfer coefficient of the electrode, and represent the exchange current density of the positive and negative electrodes respectively.

[0016] In this embodiment, the loss function is obtained in S4: Input voltage U, current I and temperature T into the neural network model to output three output state quantities , and , which are the electrode surface concentration, the electrode average concentration and the electrolyte concentration, respectively. Substituting the three output state quantities into the estimated terminal voltage model, four simplified voltage term models are obtained: ; ; ; ; Thus the estimated terminal voltage is calculated: ; in, , , and They are four voltage term models, G is the simplified physical model of voltage, represents the simplified equilibrium potential estimate, The first part is the simplified ohmic potential drop due to electrolyte conductivity, and the second part is the simplified potential estimate of the electrolyte concentration, represents the simplified voltage on the thin film resistor, represents simplified overpotential; Calculate the estimated end of the battery : ; in, represents the average concentration at the anode electrode, and are the maximum and minimum lithium insertion amounts of the anode within the operating voltage range, Indicates the maximum lithium ion concentration of the anode.

[0017] The estimated end and estimated voltage value The actual detection end SOC and actual detection voltage value The mean square error is calculated to obtain the loss function: ; in, represents a time sample, Indicates the actual detection voltage value at time t, represents the estimated terminal voltage value at time t, Indicates the estimated end at time t , Represents the SOC of the actual detection end at time t.

[0018] It should be noted that the Adam algorithm is used in this method to optimize the loss function and update parameters. Due to the use of momentum terms and adaptive learning rates, it has the advantage of efficient computing and is suitable for solving optimization problems with large amounts of data and parameters. The loss function can converge quickly with the Adam optimization algorithm.

[0019] Among them, the internal optimization parameters of the neural network model in S4 are as follows: The loss function is embedded into the neural network model as a constraint, so that the neural network parameters are updated as follows: ; ; ; ; ; ; in, is the learning rate, and are two exponential decay rates from the estimate, For parameters The loss function is and are biased first-order distance estimates and biased second-order distance estimates, respectively. is the update parameter at the tth time step, , and The initial values ​​of are set to 0.001, 0.9, and 0.999, respectively. To avoid the denominator being 0, .

[0020] In addition, in order to better estimate the internal state of the lithium battery, the side reaction overpotential is also one of the important monitoring values. If it is less than zero, a series of reactions such as lithium deposition may occur inside the battery. Therefore, the side reaction overpotential The definition is as follows: ; in, represents the equilibrium potential of the side reaction, and it is assumed that, The most critical value of appears at the junction of the anode and the diaphragm; assuming , combined with the simplified e-SPM equation, the estimated value of the side reaction overpotential is It can be described by the following formula: .

[0021] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for estimating the internal state of a lithium battery based on a physical information neural network, characterized in that: The following steps are involved: S1. Simplify the P2D model into an e-SPM model to obtain a simplified physical model that can represent the voltage by the solid phase lithium ion concentration and the liquid phase lithium ion concentration; S2, inputting the actually detected voltage U, current I and temperature T into the neural network model to obtain the output state quantity; S3. Substitute the output state quantity into the simplified physical model of the voltage to obtain the estimated terminal voltage value and the estimated terminal voltage of the battery body. , and calculate the mean square error with the actual detection voltage value and the SOC of the actual detection end to obtain the loss function; S4. The neural network model optimizes its internal parameters through the loss function to achieve internal state estimation of the lithium battery.

2. A lithium battery internal state estimation method based on physical information neural network according to claim 1, characterized in that: The simplified physical model of the voltage in S2 is obtained as follows: Assume that in the two electrodes, the solid concentration The x dimension remains constant, so the liquid phase lithium ion concentration model is: ； The solid phase lithium ion concentration model is: ； Among them, in the two electrodes, the solid phase concentration Keeping the x dimension constant, represents the initial lithium ion concentration in the electrolyte, , and denote the electrolyte volume fractions of the negative electrode, separator and positive electrode, respectively, , and Respectively represent the length of the negative electrode, separator and positive electrode, S represents the cross-sectional area of ​​the battery cell, and represent the initial lithium ion concentrations of the negative electrode and the positive electrode, respectively, and represent the volume fractions of the negative electrode and the positive electrode, respectively; When the battery temperature When the change in the x dimension is small and only changes with time t, the solid phase lithium ion concentration model and the liquid phase lithium ion concentration model are input into the neural network model to output four voltage term models: ； ； ； ； Therefore, the simplified physical model of voltage is: ； in, represents the estimated equilibrium potential, The first part is the ohmic potential drop due to electrolyte conductivity, and the second part is the electrolyte concentration overpotential estimate, represents the voltage on the thin film resistor, represents the overpotential, and represents the equilibrium potential of the positive and negative electrodes, and represent the surface lithium ion concentration of the negative electrode and the positive electrode respectively, , as well as They represent the effective reaction rate constants of the negative electrode, the separator and the positive electrode respectively, I represents the charge and discharge current, R represents the universal gas constant, and F represents the Faraday constant. Represents the lithium ion migration number, and Represent the positive and negative film resistances respectively, and Represent the interface area of ​​the positive and negative spherical particles, α represents the charge transfer coefficient of the electrode, and represent the exchange current density of the positive and negative electrodes respectively.

3. The method for estimating the internal state of a lithium battery based on a physical information neural network according to claim 1, characterized in that: The loss function in S4 is obtained by: Input voltage U, current I and temperature T into the neural network model to output three output state quantities , and , which are the electrode surface concentration, the electrode average concentration and the electrolyte concentration, respectively. Substituting the three output state quantities into the estimated terminal voltage model, four simplified voltage term models are obtained: ； ； ； ； Thus the estimated terminal voltage is calculated: ； in, , , and They are four voltage term models, G is the simplified physical model of voltage, represents the simplified equilibrium potential estimate, The first part is the simplified ohmic potential drop due to electrolyte conductivity, and the second part is the simplified potential estimate of the electrolyte concentration, represents the simplified voltage on the thin film resistor, represents simplified overpotential; Calculate the estimated end of the battery : ； in, represents the average concentration at the anode electrode, and are the maximum and minimum lithium insertion amounts of the anode within the operating voltage range, Indicates the maximum lithium ion concentration at the anode; The estimated end and estimated voltage value The actual detection end SOC and actual detection voltage value The mean square error is calculated to obtain the loss function: ； in, represents a time sample, Indicates the actual detection voltage value at time t, represents the estimated terminal voltage value at time t, Indicates the estimated end at time t , Represents the SOC of the actual detection end at time t.

Images