Battery hybrid modeling method
By combining current preprocessing and battery hybrid modeling, and leveraging the advantages of multiple models, the problems of battery modeling accuracy and efficiency as well as unstable current measurement were solved. This enabled high-precision prediction and enhanced stability of battery terminal voltage, making it suitable for electrical equipment such as electric vehicles and aviation battery systems.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NORTHWESTERN POLYTECHNICAL UNIV
- Filing Date
- 2026-02-06
- Publication Date
- 2026-05-12
AI Technical Summary
Existing battery modeling methods struggle to balance modeling accuracy and computational efficiency under complex operating conditions, and current measurements are susceptible to sensor noise and abnormal data, leading to unstable voltage predictions.
By combining current preprocessing and battery hybrid modeling, and integrating physical mechanism model, circuit equivalent model and data-driven model, a battery terminal voltage prediction model is constructed. Current preprocessing is used to correct and reconstruct the current signal, and the advantages of multiple models are combined to improve the voltage prediction accuracy and robustness.
High-precision prediction of battery terminal voltage was achieved under different operating conditions, which enhanced the applicability and stability of the model, reduced the impact of operating condition changes and noise on the prediction results, and provided a reliable modeling foundation.
Smart Images
Figure CN122017587A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of electrical engineering, and specifically relates to a hybrid modeling method for batteries. Background Technology
[0002] Power batteries are widely used in various electrical equipment such as electric vehicles and aviation battery systems. During operation, the battery terminal voltage is affected by multiple factors, including input current, operating environment, and internal electrochemical processes, exhibiting significant nonlinear and time-varying characteristics. Existing battery modeling methods mainly include electrochemical models, equivalent circuit models, and data-driven models. Different models have their own advantages and limitations in terms of physical mechanism description capabilities, computational complexity, and nonlinear modeling accuracy. A single model cannot simultaneously achieve both modeling accuracy and computational efficiency under complex operating conditions. Furthermore, in actual operation, battery current measurement is easily affected by sensor noise, signal interference, or abnormal data. If abnormal current is directly used as model input, it can easily lead to unstable voltage prediction results or even large deviations. Therefore, there is an urgent need for a high-precision modeling method for normal battery terminal voltage that integrates the advantages of multiple models and is applicable to various application scenarios while ensuring the validity of current input. Summary of the Invention
[0003] To address the aforementioned issues, this invention provides a battery hybrid modeling method, comprising current preprocessing and battery hybrid model modeling. By preprocessing the battery current signal and correcting and reconstructing the current data in conjunction with the vehicle's operating status, a normal current input that reflects the battery's true operating state is obtained. Based on this, a normal voltage prediction model is constructed, which is formed by fusing a physical mechanism model, a circuit equivalent model, and a data-driven model, for accurately predicting the battery terminal voltage under different operating conditions.
[0004] The battery current preprocessing addresses issues such as sensor malfunctions and data transmission errors that cause abnormal currents collected by vehicle-mounted sensors. It constructs an abnormal current value preprocessing model to ensure that the input current of the coupled model is a normal value.
[0005] The vehicle status is divided into two states: driving and charging. Different current processing procedures are used in different operating states.
[0006] During operation, the battery current is preprocessed based on the motor bus current and the motor efficiency equation: Calculate and determine whether the total current error value is within the normal range. The total current error value is calculated according to the following formula: In the formula, Let be the battery current at time t. Let be the actual motor bus current at time t. Let t be the current in the vicinity at time t.
[0007] When the total current error is within the normal range, let the battery current at time t be taken as the processed battery current at time t: If the total current error is not within the normal range, continue to determine whether the difference between the measured value and the estimated value of the motor bus current is within the normal range. The difference between the estimated value and the actual value of the motor bus current is expressed as: In the formula, The value of the motor bus current at time t is an estimate, calculated using the motor efficiency equation. Let t be the actual motor bus current at time t.
[0008] The equation for motor efficiency is: In the formula, It's the motor voltage. It is the output torque of the motor. It is the motor output speed. This represents the actual motor bus current.
[0009] The estimated value of the motor bus current at time t is: In the formula, It is the motor output torque at time t. It is the motor output speed at time t. It is the motor voltage at time t.
[0010] Difference between estimated and measured values of motor bus current Within the normal range, the sum of the actual motor bus current and the accessory current is used as the processed current: Otherwise, the sum of the estimated motor bus current and the accessory current is used as the processed current: During charging, the battery current is preprocessed based on the calibrated charging current.
[0011] The total current error value is calculated according to the following formula: In the formula, The charging current is calibrated at time t. Let t be the battery current at time t.
[0012] If the total current error value is within the normal range, the battery current will be used as the processed current. If the total current error value is outside the normal range, the calibrated charging current will be used as the processed current. The current collected by the vehicle's sensors is processed by the current preprocessing model (CPM) to replace abnormal currents with normal currents, so that the hybrid model can accurately predict the normal output voltage.
[0013] The above current and processing process addresses the current measurement deviations caused by sensor failures, signal interference, or abnormal data transmission during actual battery operation. It identifies and corrects abnormal values in the collected battery current signal, and reconstructs the abnormal current based on the battery's operating state to obtain a normal current input that truly reflects the battery's working state.
[0014] The battery hybrid modeling is specifically as follows: Based on the single-particle model (SPM), the battery input current is... With output voltage The relationship between them can be described as follows: In the formula, and These are the positive and negative equilibrium potentials, respectively. and These are the overpotentials of the positive and negative electrodes, respectively. and These represent the lithium-ion concentrations on the surface of the positive and negative electrode solid particles, respectively. and These represent the reaction kinetic rates at the positive and negative electrodes, respectively. and These represent the molar fluxes at the surfaces of positive and negative particles, respectively. and These are the membrane resistances between the positive and negative solid-phase electrolytes, respectively. and These are the specific interface areas of the positive and negative electrodes, respectively. and These represent the thicknesses of the positive and negative electrodes, respectively.
[0015] The normal voltage prediction model fully leverages the advantages of the three types of models by coupling electrochemical models, equivalent circuit models, and data-driven models, thereby improving prediction accuracy, robustness, and engineering applicability.
[0016] By integrating the parts of the formula relating battery input current to output voltage, we can obtain: In the formula, U is the terminal voltage. This is the equilibrium potential between the positive and negative electrodes. It is the difference in overpotential between the positive and negative electrodes. It is the membrane potential difference between solid electrolyte phases.
[0017] First, calculations were performed using electrochemical models, equivalent circuit models, and data-driven models, respectively. , and Then, the terminal voltage U is calculated. The coupled three-model approach fully leverages the advantages of each: the electrochemical model ensures accuracy; the equivalent circuit model reduces computational load; and the data-driven model can calculate components that are difficult to estimate using the electrochemical and equivalent circuit models, and it incorporates voltage noise through training with real-vehicle data.
[0018] Electrochemical model was used to calculate the interphase membrane potential difference of solid electrolytes. Simultaneously, an electrochemical model was used to calculate the surface molar flux of positive and negative particles. and As Input: In the formula, The current is the result of processing by the current processing module CPM, where F is the Faraday constant and A is the surface area of the electrode. and These are the specific interface areas of the positive and negative electrodes, respectively. and These represent the thicknesses of the positive and negative electrodes, respectively.
[0019] Estimate using Thevenin model It is expressed by the following formula: In the formula, This is the battery open-circuit voltage.
[0020] According to Kirchhoff's laws, the circuit equations of the Thevenin model are as follows: In the formula, , , , I and I represent polarization voltage, internal resistance, polarization resistance, polarization capacitance, and current, respectively.
[0021] Radial basis function neural network fitting is used to calculate .
[0022] Input positive and negative particle surface molar flux and The system maps to a high-dimensional space using radial basis functions, and then computes the results through a linear output layer. .
[0023] The transfer function between the input layer and the RBF hidden layer is: In the formula, exp() is an exponential function. is the width parameter of the radial basis function.
[0024] The formula for calculating the output layer is: In the formula, m is the number of neurons in the hidden layer. These are the weights of the hidden layer and the output layer.
[0025] Based on mean squared error, define the loss function for the RBF hidden layer: In the formula, This represents the true value of the difference between the positive and negative overpotentials at time t. This is the predicted value of the difference between the positive and negative overpotentials at time t.
[0026] Gradient descent is used to simultaneously optimize the parameters in the loss function of the RBF hidden layer. In the formula, α is the learning rate.
[0027] The above battery hybrid model is used to construct a normal voltage prediction model integrating a physical mechanism model, a circuit equivalent model, and a data-driven model. The physical mechanism model is used to describe the internal electrochemical processes of the battery and its voltage variation law related to operating conditions. The circuit equivalent model is used to characterize the voltage response corresponding to the battery's open-circuit voltage and polarization characteristics. The data-driven model is used to compensate for the nonlinear voltage deviation that the physical mechanism model and the circuit equivalent model cannot accurately describe under complex operating conditions. By fusing the output voltage components of the above models, the prediction result of the normal terminal voltage of the battery based on normal current input is obtained, thus achieving accurate modeling of the battery voltage behavior.
[0028] This invention introduces a current preprocessing mechanism before battery modeling to perform anomaly analysis and correction on battery current signals collected during actual operation, improving the consistency and reliability of the model's input current and thus providing a stable input foundation for subsequent battery voltage modeling. Based on this, by integrating physical mechanism models, circuit equivalent models, and data-driven models, a battery terminal voltage prediction model for different operating conditions is constructed. This improves the accuracy and stability of voltage prediction under complex operating conditions, enhances the model's robustness and applicability, and provides a reliable modeling foundation for subsequent analysis and application of the battery system. This method can fully utilize actual battery operating data, effectively reducing the impact of operating condition changes, current disturbances, and noise on the model's prediction results, enhancing the model's adaptability to multiple operating conditions, and exhibiting good engineering applicability and scalability. Attached Figure Description
[0029] Figure 1 Battery current pretreatment method; Figure 2 Battery hybrid modeling method. Detailed Implementation
[0030] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.
[0031] like Figure 2 The battery hybrid modeling method shown includes two steps: battery current data preprocessing and battery hybrid model modeling. Step 1: Current data preprocessing: To address issues such as sensor malfunctions and data transmission errors leading to abnormal current readings from vehicle-mounted sensors, a preprocessing model for abnormal current values is constructed to ensure that the input current of the coupled model is within normal range. The calculation process is as follows: Figure 1 As shown.
[0032] Different current processing procedures are adopted under different operating conditions: in driving condition, the battery current is preprocessed based on the motor bus current and motor efficiency equation; in charging condition, the battery current is preprocessed based on the calibrated charging current.
[0033] While the vehicle is in motion, calculate and determine whether the total current error value is within the normal range. The total current error value is calculated according to the following formula: In the formula, Let be the battery current at time t. Let be the actual motor bus current at time t. Let t be the current in the vicinity at time t.
[0034] When the total current error is within the normal range, let the battery current at time t be taken as the processed battery current at time t: If the total current error is not within the normal range, continue to determine whether the difference between the measured value and the estimated value of the motor bus current is within the normal range. The difference between the estimated value and the actual value of the motor bus current is expressed as: In the formula, The value of the motor bus current at time t is an estimate, calculated using the motor efficiency equation. Let t be the actual motor bus current at time t.
[0035] The equation for motor efficiency is: In the formula, It's the motor voltage. It is the output torque of the motor. It is the motor output speed. This represents the actual motor bus current.
[0036] The estimated value of the motor bus current at time t is: In the formula, It is the motor output torque at time t. It is the motor output speed at time t. It is the motor voltage at time t.
[0037] Difference between estimated and measured values of motor bus current Within the normal range, the sum of the actual motor bus current and the accessory current is used as the processed current: Otherwise, the sum of the estimated motor bus current and the accessory current is used as the processed current: During charging, the total current error value is calculated according to the following formula: In the formula, The charging current is calibrated at time t. Let t be the battery current at time t.
[0038] If the total current error value is within the normal range, the battery current will be used as the processed current. If the total current error value is outside the normal range, the calibrated charging current will be used as the processed current. The current collected by the vehicle's sensors is processed by the current preprocessing model (CPM) to replace abnormal currents with normal currents, so that the hybrid model can accurately predict the normal output voltage.
[0039] Step 2: Battery Hybrid Modeling: Based on the preprocessed battery current, individual cell voltage, state of charge, and temperature, an electrochemical model, an equivalent circuit model, and a data-driven model are constructed. The electrochemical model is used to calculate voltage components related to the battery's inherent characteristics, the equivalent circuit model is used to estimate state variables such as the battery's open-circuit voltage online, and the data-driven model is used to compensate for nonlinear voltage components under complex operating conditions. The outputs of each model are integrated to obtain the predicted normal terminal voltage of the battery under the current operating conditions.
[0040] Based on the single-particle model (SPM), the relationship between the battery input current I(t) and the output voltage U(t) is described as follows: In the formula, and These are the positive and negative equilibrium potentials, respectively. and These are the overpotentials of the positive and negative electrodes, respectively. and These represent the lithium-ion concentrations on the surface of the positive and negative electrode solid particles, respectively. and These represent the reaction kinetic rates at the positive and negative electrodes, respectively. and These represent the molar fluxes at the surfaces of positive and negative particles, respectively. and These are the membrane resistances between the positive and negative solid-phase electrolytes, respectively. and These are the specific interface areas of the positive and negative electrodes, respectively. and These represent the thicknesses of the positive and negative electrodes, respectively.
[0041] The normal voltage prediction model fully leverages the advantages of the three types of models by coupling electrochemical models, equivalent circuit models, and data-driven models, thereby improving prediction accuracy, robustness, and engineering applicability.
[0042] By integrating the parts of the formula relating battery input current to output voltage, we can obtain: In the formula, U is the terminal voltage. This is the equilibrium potential between the positive and negative electrodes. It is the difference in overpotential between the positive and negative electrodes. It is the membrane potential difference between solid electrolyte phases.
[0043] First, calculations were performed using electrochemical models, equivalent circuit models, and data-driven models, respectively. , and Then, the terminal voltage U is calculated. The coupled three-model approach fully leverages the advantages of each: the electrochemical model ensures accuracy; the equivalent circuit model reduces computational load; and the data-driven model can calculate components that are difficult to estimate using the electrochemical and equivalent circuit models, and it incorporates voltage noise through training with real-vehicle data.
[0044] Electrochemical model was used to calculate the interphase membrane potential difference of solid electrolytes. Simultaneously, an electrochemical model was used to calculate the surface molar flux of positive and negative particles. and As Input: In the formula, The current is the result of processing by the current processing module CPM, where F is the Faraday constant and A is the surface area of the electrode. and These are the specific interface areas of the positive and negative electrodes, respectively. and These represent the thicknesses of the positive and negative electrodes, respectively.
[0045] Estimate using Thevenin model It is expressed by the following formula: In the formula, This is the battery open-circuit voltage.
[0046] According to Kirchhoff's laws, the circuit equations of the Thevenin model are as follows: In the formula, , , , I and I represent polarization voltage, internal resistance, polarization resistance, polarization capacitance, and current, respectively.
[0047] Radial basis function neural network fitting is used to calculate .
[0048] Input positive and negative particle surface molar flux and The system maps to a high-dimensional space using radial basis functions, and then computes the results through a linear output layer. .
[0049] The transfer function between the input layer and the RBF hidden layer is: In the formula, exp() is an exponential function. is the width parameter of the radial basis function.
[0050] The formula for calculating the output layer is: In the formula, m is the number of neurons in the hidden layer. These are the weights of the hidden layer and the output layer.
[0051] Based on mean squared error, define the loss function for the RBF hidden layer: In the formula, This represents the true value of the difference between the positive and negative overpotentials at time t. This is the predicted value of the difference between the positive and negative overpotentials at time t.
[0052] Gradient descent is used to simultaneously optimize the parameters in the loss function of the RBF hidden layer. In the formula, α is the learning rate.
[0053] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A battery hybrid modeling method, comprising current preprocessing and battery hybrid modeling, characterized in that, The current preprocessing includes: detecting abnormal values in the battery current signal, and replacing or reconstructing the abnormal current when an abnormal current is detected, based on the battery's operating state. The operating states include at least vehicle driving state and charging state, and different current correction methods are used in different operating states; The hybrid modeling includes: based on the normal current input obtained from the current preprocessing, constructing a normal voltage prediction model formed by the fusion of a physical mechanism model, a circuit equivalent model, and a data-driven model, which is used to accurately predict the battery terminal voltage under different operating conditions.
2. According to the battery hybrid modeling method of claim 1, the battery current is preprocessed based on the motor bus current and the motor efficiency equation under driving conditions: Calculate and determine whether the total current error value is within the normal range. The total current error value is calculated according to the following formula: In the formula, Let be the battery current at time t. Let be the actual motor bus current at time t. Let t be the current in the vicinity; When the total current error is within the normal range, let the battery current at time t be the processed battery current at time t: If the total current error is not within the normal range, continue to determine whether the difference between the measured value and the estimated value of the motor bus current is within the normal range. The difference between the estimated value and the actual value of the motor bus current is expressed as: In the formula, The value of the motor bus current at time t is an estimate, calculated using the motor efficiency equation. Let t be the actual motor bus current at time t; The equation for motor efficiency is: In the formula, It's the motor voltage. It is the output torque of the motor. It is the motor output speed. This represents the actual motor bus current. The estimated value of the motor bus current at time t is: In the formula, It is the motor output torque at time t. It is the motor output speed at time t. It is the motor voltage at time t; Difference between estimated and measured values of motor bus current Within the normal range, the sum of the actual motor bus current and the accessory current is used as the processed current: Otherwise, the sum of the estimated motor bus current and the accessory current is used as the processed current: 。 3. The battery hybrid modeling method according to claim 1, wherein the battery current is preprocessed based on the calibrated charging current during the charging state: The total current error value is calculated according to the following formula: In the formula, The charging current is calibrated at time t. Let t be the battery current. If the total current error value is within the normal range, the battery current will be used as the processed current. If the total current error value is outside the normal range, the calibrated charging current will be used as the processed current. 。 4. In the battery hybrid modeling method according to claim 1, after the current collected by the vehicle sensors is processed by the current preprocessing model, the abnormal current is replaced with the normal current so that the hybrid model can accurately predict the normal output voltage.
5. The battery hybrid modeling method according to claim 1, wherein the hybrid modeling specifically comprises: Based on the single-particle model, the battery input current With output voltage The relationship between them can be described as follows: In the formula, and These are the positive and negative equilibrium potentials, respectively. and These are the overpotentials of the positive and negative electrodes, respectively. and These represent the lithium-ion concentrations on the surface of the positive and negative electrode solid particles, respectively. and These represent the reaction kinetic rates at the positive and negative electrodes, respectively. and These represent the molar fluxes at the surfaces of positive and negative particles, respectively. and These are the membrane resistances between the positive and negative solid-phase electrolytes, respectively. and These are the specific interface areas of the positive and negative electrodes, respectively. and These are the thicknesses of the positive and negative electrodes, respectively. By integrating the parts of the formula relating battery input current to output voltage, we can obtain: In the formula, U is the terminal voltage. This is the equilibrium potential between the positive and negative electrodes. It is the difference in overpotential between the positive and negative electrodes. It is the potential difference between the solid electrolyte phase membranes; Calculation of interphase membrane potential difference in solid electrolytes using an electrochemical model Meanwhile, electrochemical models were used to calculate the surface molar flux of positive and negative particles. and As Input: In the formula, The current is the result of processing by the current processing module CPM, where F is the Faraday constant and A is the surface area of the electrode. and These are the specific interface areas of the positive and negative electrodes, respectively. and These are the thicknesses of the positive and negative electrodes, respectively. Estimate using Thevenin model It is expressed by the following formula: In the formula, This is the battery open-circuit voltage; According to Kirchhoff's laws, the circuit equations of the Thevenin model are as follows: In the formula, , , , I and I represent polarization voltage, internal resistance, polarization resistance, polarization capacitance, and current, respectively. Radial basis function neural network fitting is used to calculate ; Input positive and negative particle surface molar flux and The system maps to a high-dimensional space using radial basis functions, and then computes the results through a linear output layer. ; The transfer function between the input layer and the RBF hidden layer is: In the formula, exp() is an exponential function. is the width parameter of the radial basis functions; The formula for calculating the output layer is: In the formula, m is the number of neurons in the hidden layer. These are the weights of the hidden layer and the output layer; Based on mean squared error, define the loss function for the RBF hidden layer: In the formula, This represents the true value of the difference between the positive and negative overpotentials at time t. This is the predicted value of the difference between the positive and negative overpotentials at time t; Gradient descent is used to simultaneously optimize the parameters in the loss function of the RBF hidden layer. In the formula, α is the learning rate.