Data-driven lithium ion battery voltage prediction modeling method fusing physical information

Abstract

The invention relates to a data-driven lithium ion battery voltage prediction modeling method fusing physical information, and belongs to the technical field of batteries. According to the method, a multi-working-condition battery operation database is constructed, a terminal voltage prediction model fusing a physical model and a neural network is designed, a mathematical constraint relation is established for equivalent circuit parameters and neural network intermediate variables, and data driving characteristics and physical parameters are jointly optimized by using a gradient descent method. The method specifically comprises the steps of collecting battery current, voltage and temperature data; constructing an input feature and an output tag; outputting an intermediate variable through a parameter layer and associating an equivalent circuit parameter; and introducing physical loss function constraint model training to ensure the physical rationality of parameters. The method has high precision of data driving and interpretability of a physical model, improves multi-working-condition adaptability through temperature compensation and parameter constraint, and can be widely applied to battery state estimation and health management of electric automobiles and energy storage systems.

Description

Technical Field

[0001] The present invention belongs to the technical field of batteries and relates to a data-driven lithium-ion battery voltage prediction modeling method integrating cyber-physical information. Background Art

[0002] Due to advantages such as long lifespan, high charge-discharge power, and large energy density, lithium-ion batteries have been widely used in various fields such as electric vehicles, energy storage, and consumer electronics. During the charge-discharge process of lithium-ion batteries, the terminal voltage changes dynamically with the change of current load, and its terminal voltage response is jointly determined by factors such as battery materials, state of charge, polarization state, and current load. In order to ensure that the battery operates in a safe and efficient voltage range, it is necessary to model the lithium-ion battery to predict the battery voltage response under different current load inputs. Common lithium-ion battery models include equivalent circuit models, electrochemical models, data-driven models, etc. The equivalent circuit model relies on accurate parameter identification methods, and the identified parameters are often only applicable to specific working conditions and need to be updated regularly to adapt to battery aging; the construction of the electrochemical model deeply depends on the material parameters of lithium-ion batteries, and the acquisition of relevant parameters is difficult, the identification accuracy is low, the model calculation complexity is large, and the use cost is high in actual application scenarios; the data-driven model regards the battery as a black box and uses neural network methods to map the non-linear relationship between system inputs and outputs, which has certain application prospects, but traditional methods have poor interpretability and low robustness and cannot reflect the physical characteristics of battery operation.

[0003] In view of the above problems, there has not yet been an effective lithium-ion battery voltage prediction modeling method integrating cyber-physical information to accurately predict the terminal voltage of lithium-ion batteries. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a data-driven lithium-ion battery voltage prediction modeling method integrating cyber-physical information, which can realize the modeling of lithium-ion batteries considering cyber-physical information and accurate prediction and estimation of voltage.

[0005] To achieve the above purpose, the present invention provides the following technical solutions:

[0006] A data-driven lithium-ion battery voltage prediction modeling method integrating cyber-physical information, the method comprising the following steps:

[0007] S1: Conduct charge-discharge experiments on lithium-ion batteries under different working conditions, collect data such as battery current, terminal voltage, operating temperature, etc., and establish a multi-condition operation database for lithium-ion batteries;

[0008] S2: According to the battery physical model and physical parameter information, establish a lithium-ion battery terminal voltage prediction model integrating physical model information and neural network methods, and use the gradient descent method to train the model;

[0009] S3: Predict the terminal voltage response of the lithium-ion battery based on the trained model.

[0010] Optionally, S1 is specifically:

[0011] S11: Conduct charge and discharge experiments on lithium-ion batteries under different working conditions, and perform constant current or dynamic charge and discharge tests on the batteries at different temperatures and different current rates;

[0012] S12: Collect the test data of the lithium-ion battery, including charge and discharge current, battery terminal voltage, test temperature, and time, and establish a multi-condition operation database for the lithium-ion battery;

[0013] Optionally, S2 is specifically:

[0014] S21: Extract the measured terminal voltage U of the battery at the current moment t,k as Feature 1;

[0015] S22: Set the floating-point number 1.0 as Feature 2;

[0016] S23: Extract the battery current I at the next moment k+1 as Feature 3;

[0017] S24: Extract the battery current I at the current moment k as Feature 4;

[0018] S25: Extract the measured terminal voltage U of the battery at the next moment t,k+1 as the output label of the model;

[0019] S26: Based on the feedforward neural network, establish a lithium-ion battery voltage prediction model that integrates physical information. The first layer of the model is the input layer, and Features 1 to 4 are used as the inputs In 1 ~In 4 . After the input layer, add several fully connected layers as hidden layers for extracting high-dimensional features. The number of hidden layers and the number of their neurons can be adjusted according to requirements; finally, connect a fully connected layer after the hidden layer as the equivalent circuit model parameter layer. The number of neurons in this layer is 4, and each neuron represents an intermediate variable with physical information. These intermediate variables θ 1 ~θ 4 have a mathematical derivation relationship with the ohmic internal resistance R int , dynamic characteristic internal resistance R p , dynamic characteristic capacitance C p , open-circuit voltage U ocv :

[0020]

[0021] where Δt is the sampling time and U oc is the open-circuit voltage at the current moment. This model structure endows the neuron parameters in the neural network model with physical meanings;

[0022] S27: According to the physical definition of the first-order equivalent circuit model of the lithium-ion battery, the terminal voltage output of the deterministic model of the lithium-ion battery is:

[0023]

[0024] where In = [U t,k , 1, I k+1 , I k , θ = [θ 1 , θ 2 , θ 3 , θ 4 = [α, (1 - α)U ocv , -R int , αR int - (1 - α)R p ;

[0025] S28: Define the physical loss L of the model as the absolute value of the difference between U t,k+1 and . Iteratively train the neural network model according to L and the gradient descent algorithm until L drops to a preset value or there is no longer a downward trend;

[0026] Optionally, the specific content of S3 is as follows:

[0027] S31: Real-time collect the current state of the battery, including the measured terminal voltage U k of the battery at the current moment as In1, extract the battery current I k+1 at the next moment as feature 2, extract the battery current I k at the current moment as feature 3, and set the floating-point number 1.0 as feature 4;

[0028] S33: Use features 1 to 4 as the model input In 1 ~In 4 , and input them into the trained model, and the model outputs the predicted value of the terminal voltage at the next moment.

[0029] The beneficial effects of the present invention are as follows:

[0030] (1) By deeply coupling the physical mechanism of the equivalent circuit model with the data-driven characteristics of the neural network, it not only overcomes the defect of insufficient accuracy of the traditional equivalent circuit model under dynamic conditions but also avoids the problem of lack of physical interpretability of the pure data-driven model, achieving the unity of high-precision prediction and physical interpretability.

[0031] (2) Physical constraints ensure parameter rationality: Introduce the mathematical constraint relationship of the equivalent circuit parameters in the neural network parameter layer, and embed physical laws into the training process through differentiable programming to ensure that the intermediate variables output by the model conform to the actual battery electrochemical characteristics and avoid non-physical parameter deviations that may occur in traditional black-box models.

[0032] (3) By collecting data under multiple temperature conditions and using it as input features, dynamically correct the temperature dependence of the equivalent circuit parameters, significantly improving the prediction robustness of the model in a wide temperature range environment and solving the problem of cumulative prediction errors caused by traditional models ignoring temperature effects.

[0033] (4) Construct a multi-condition database based on dynamic charge and discharge experiments, covering constant current, variable current, and different temperature scenarios, enabling the model to learn the non-linear dynamic characteristics within the entire life cycle of the battery and enhancing the generalization ability for complex actual conditions (such as rapid acceleration / braking of electric vehicles).

[0034] (5) Through the design of a lightweight feedforward neural network, combined with feature inputs at fixed time steps (such as voltage and current at the current moment), achieve millisecond-level terminal voltage prediction, meet the requirements of the battery management system (BMS) for real-time state estimation, and provide high-refresh-rate data support for battery safety warning and energy optimization control.

[0035] (6) A modular-designed data acquisition, model construction, and prediction system support adapting to different battery types by replacing the equivalent circuit model (such as first-order RC, second-order RC, etc.) or adjusting the neural network structure, providing a general solution for the health management of various energy storage devices such as lithium-ion batteries and solid-state batteries.

[0036] Other advantages, objectives, and features of the present invention will be described to some extent in the subsequent specification, and to some extent, will be obvious to those skilled in the art based on the study of the following text, or can be learned from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following specification.

[0037] Other advantages, objectives, and features of the present invention will be described to some extent in the subsequent specification, and to some extent, will be obvious to those skilled in the art based on the study of the following text, or can be learned from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following specification. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be described in detail preferably with reference to the accompanying drawings, where:

[0039] Figure 1 is the overall method flowchart of the present invention;

[0040] Figure 2 is the overall framework diagram of the method of the embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0041] The following uses specific specific examples to illustrate the embodiments of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the drawings provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner. Without conflict, the following embodiments and the features in the embodiments can be combined with each other.

[0042] Among them, the accompanying drawings are only for illustrative purposes, showing only schematic diagrams, not physical diagrams, and should not be construed as a limitation to the present invention; in order to better illustrate the embodiments of the present invention, some components in the accompanying drawings will be omitted, enlarged or reduced, which do not represent the size of the actual product; for those skilled in the art, it is understandable that some well-known structures and their descriptions in the accompanying drawings may be omitted.

[0043] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components; in the description of the present invention, it should be understood that if there are terms such as "upper", "lower", "left", "right", "front", "rear", etc. indicating the orientation or positional relationship, it is based on the orientation or positional relationship shown in the accompanying drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, the terms describing the positional relationship in the accompanying drawings are only for illustrative purposes and should not be construed as a limitation to the present invention. For those of ordinary skill in the art, the specific meanings of the above terms can be understood according to specific circumstances.

[0044] The technical solution provided by the embodiment of the present invention can build a lithium-ion battery voltage prediction modeling method integrating physical information based on the operation data of the lithium-ion battery.

[0045] Please refer to Figure 1 , a data-driven lithium-ion battery voltage prediction modeling method integrating physical information can be divided into the following steps:

[0046] S1: Conduct lithium-ion battery charge and discharge experiments under different working conditions, collect data such as battery current, terminal voltage, and operating temperature, and establish a multi-condition operation database for lithium-ion batteries;

[0047] S2: According to the battery physical model and physical parameter information, establish a lithium-ion battery terminal voltage prediction model that integrates physical model information and neural network methods, and use the gradient descent method to train the model;

[0048] S3: Predict the lithium-ion battery terminal voltage response based on the trained model;

[0049] As an alternative embodiment, the complete technical roadmap of this solution is as Figure 2 shown.

[0050] As an alternative embodiment, the above S1 specifically includes S11 - S12:

[0051] S11: Conduct lithium-ion battery charge and discharge experiments under different working conditions, and perform constant current or dynamic charge and discharge tests on the battery at different temperatures and different current rates;

[0052] S12: Collect the test data of the lithium-ion battery, including charge and discharge current, battery terminal voltage, test temperature, and time, and establish a multi-condition operation database for lithium-ion batteries.

[0053] As an alternative embodiment, the above S2 specifically includes S21 - S28:

[0054] S21: Extract the measured terminal voltage U of the battery at the current moment t,k as Feature 1;

[0055] S22: Set the floating-point number 1.0 as Feature 2:

[0056] S23: Extract the battery current I at the next moment k+1 as Feature 3:

[0057] S24: Extract the battery current I at the current moment k as Feature 4;

[0058] S25: Extract the measured terminal voltage U of the battery at the next moment t,k+1 as the output label of the model;

[0059] S26: Based on the feedforward neural network, establish a lithium-ion battery voltage prediction model that integrates physical information. The first layer of the model is the input layer, and Features 1 - 4 are used as the inputs In 1 ~In 4. After the input layer, several fully connected layers are added as hidden layers for extracting high-dimensional features. The number of hidden layers and the number of neurons in each layer can be adjusted according to requirements. Finally, a fully connected layer is connected after the hidden layers as the equivalent circuit model parameter layer. The number of neurons in this layer is 4, and each neuron represents an intermediate variable with physical information. These intermediate variables θ 1 ~θ 4 have a mathematical derivation relationship with the ohmic internal resistance R int , dynamic characteristic internal resistance R p , dynamic characteristic capacitance C p , open-circuit voltage U ocv :

[0060]

[0061] where Δt is the sampling time, and U oc is the open-circuit voltage at the current moment. This model structure endows the neuron parameters in the neural network model with physical meanings;

[0062] S27: According to the physical definition of the first-order equivalent circuit model of lithium-ion batteries, the terminal voltage output of the lithium-ion battery deterministic model is:

[0063]

[0064] where,

[0065]

[0066] θ = [θ 1 , θ 2 , θ 3 , θ 4 = [α, (1 - α)U ocv , -R int , αR int - (1 - α)R p

[0067] S28: Define the physical loss L of the model as the absolute value of the difference between U t,k+1 and . Iteratively train the neural network model according to L and the gradient descent algorithm until L drops to a preset value or there is no longer a downward trend;

[0068] As an optional embodiment, the above S3 specifically includes S31 - S32:

[0069] S31: Real-time collect the current state of the battery, including the measured terminal voltage U t,k of the battery at the current moment as feature 1, set the floating-point number 1.0 as feature 2, and extract the battery current I at the next moment​k+1 As Feature 3 and extract the battery current I at the current moment k As Feature 4;

[0070] S32: Use Features 1 to 4 as the model input In 1 ~In 4 , and input it into the trained model, and the model outputs the predicted value of the terminal voltage at the next moment;

[0071] Example 1: Implementation of the lithium-ion battery terminal voltage prediction method

[0072] 1. Data collection and preprocessing

[0073] Test object: Select a ternary lithium-ion battery (capacity 60Ah, rated voltage 3.7V), and control the temperature range from -10°C to 45°C in an incubator.

[0074] Charge and discharge protocol:

[0075] Constant current charge and discharge: Charge and discharge at 0.5C, 1C, and 2C rates until the cut-off voltage (4.2V / 2.5V).

[0076] Dynamic working condition: Simulate the UDDS driving cycle of an electric vehicle, and the current rate changes dynamically between -3C (regenerative braking) and 3C.

[0077] Data recording: Collect the current I, terminal voltage U t , and temperature T at a sampling frequency of 1Hz to build a multi-condition database containing 100,000 data points.

[0078] 2. Model construction and training

[0079] Input feature design:

[0080] Feature 1: Terminal voltage U at the current moment t,k (Normalized to [0,1]).

[0081] Feature 2: Fixed constant 1.0 (used for bias term modeling).

[0082] Feature 3: Current U at the next moment k+1 (Obtained 1 second in advance to simulate the BMS control command).

[0083] Feature 4: Current I at the current moment k .

[0084] Neural network structure:

[0085] Input layer: 4 neurons (corresponding to Features 1 - 4).

[0086] Hidden layer: 2 fully connected layers, with 64 neurons in each layer, and the activation function is ReLU.

[0087] Parameter layer: 4 neurons output intermediate variable θ 1 ~θ 4 .

[0088] Physical parameter mapping: According to the formula in Claim 3, map θ 1 ~θ 4 to equivalent circuit parameters R int , R p , C p , U ocv , and constrain R int > 0, R p > 0, C p > 0.

[0089] Loss function and training:

[0090] Physical loss:

[0091] Parameter constraint loss: L con = λ(max(0, -R int ) + max(0, -R p ) + max(0, -C p ))), where λ = 0.1.

[0092] Optimizer: Adam algorithm, learning rate 0.001, converges after 500 rounds of training.

[0093] 3. Prediction and verification

[0094] Test scenario: In a low - temperature environment of - 10°C, dynamically discharge at a rate of 2C, and compare the prediction errors of the model of the present invention and the first - order RC equivalent circuit model.

[0095] Results:

[0096] Model of the present invention: Mean Absolute Error (MAE) is 8 mV, maximum error ≤ 25 mV.

[0097] Traditional RC model: MAE is 35 mV, maximum error reaches 80 mV.

[0098] Temperature compensation effect: Under high - temperature conditions of 45°C, the model automatically corrects R int and U ocv , and MAE is stabilized within 10 mV.

[0099] Example 2: System implementation (in - vehicle BMS integration)

[0100] 1. Hardware configuration:

[0101] Data acquisition module: High-precision current sensor (±0.1% accuracy), voltage acquisition chip (ADS131M08), temperature sensor (NTC thermistor).

[0102] Calculation unit: Embedded GPU (NVIDIA Jetson Nano), deploying the trained neural network model.

[0103] 2. Real-time prediction process:

[0104] Collect U once every 100 ms t,k 、I k 、T, and receive the current command I for the next moment sent by the BMS k+1 .

[0105] The input features are sent to the model after being standardized, and the predicted value of the output terminal voltage is provided for the BMS to perform SOC estimation and overvoltage / undervoltage protection judgment.

[0106] 3. Effect verification:

[0107] During the actual vehicle road test, compared with the BMS system without using the present invention, the voltage prediction delay is reduced to within 10 ms, and the SOC estimation error under extreme conditions is reduced from 5% to 1.5%.

[0108] Example 3: Extended application (variant of the second-order RC model)

[0109] Replace the first-order RC model with a second-order RC model, adding a polarization resistance R p2 and a capacitor C p2 , correspondingly expanding the neural network parameter layer to 6 intermediate variables, and adjusting the physical constraint formula.

[0110] Experimental results: Under high-frequency pulse conditions, the MAE of the second-order model variant is further reduced to 5 mV, which is suitable for energy storage frequency modulation scenarios with higher requirements for transient response.

[0111] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the purpose and scope of the present technical solution, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A data-driven lithium-ion battery voltage prediction modeling method integrating physical information, characterized in that: The method comprises the following steps: S1: Establish a multi-operating database for lithium-ion batteries, and collect battery current, terminal voltage, and operating temperature data through charge and discharge experiments; S2: Constructing a lithium-ion battery terminal voltage prediction model that integrates a physical model and a neural network, wherein the model is trained by a gradient descent method, wherein physical information is incorporated into the model through a mathematical relationship between equivalent circuit model parameters and intermediate variables of the neural network; S3: Based on the trained model, input the real-time battery status data and predict the terminal voltage response at the next moment.

2. The data-driven lithium-ion battery voltage prediction modeling method integrating physical information according to claim 1 is characterized in that: The S1 is specifically: S11: Perform constant current or dynamic charge and discharge tests on the battery at different temperatures and current rates; S12: Collect charging and discharging current, terminal voltage, temperature and time data to build a multi-operating condition database.

3. The data-driven lithium-ion battery voltage prediction modeling method integrating physical information according to claim 1 is characterized in that: The S2 is specifically: S21: Extract the battery measurement terminal voltage U at the current moment t,k As feature 1; S22: Set the floating point number 1.0 as feature 2; S23: Extract the battery current I at the next moment k+1 As feature 3; S24: Extract the current battery current I k As feature 4; S25: Extract the battery measurement terminal voltage U at the next moment t,k+1 As the output label of the model; S26: Based on a feedforward neural network, a lithium-ion battery voltage prediction model integrating physical information is established; the first layer of the model is the input layer, and features 1 to 4 are used as inputs In1 to In4 of the model respectively; after the input layer, several fully connected layers are added as hidden layers for extracting high-dimensional features, and the number of hidden layers and their neurons can be adjusted according to needs; finally, a fully connected layer is connected after the hidden layer as an equivalent circuit model parameter layer, and the number of neurons in this layer is 4, each neuron represents an intermediate variable with physical information, and these intermediate variables θ1 to θ4 are related to the ohmic internal resistance R in the equivalent circuit model. int , Dynamic characteristics internal resistance R p , dynamic characteristic capacitance C p , open circuit voltage U ocv With mathematical derivation relationship: Where Δt is the sampling time, U oc is the open circuit voltage at the current moment, which makes the neuron parameters in the neural network model have physical meaning; S27: According to the physical definition of the first-order equivalent circuit model of lithium-ion batteries, the terminal voltage output of the lithium-ion battery deterministic model for: in, In=[U t,k ,1,I k+1 ,I k ] θ=[θ1,θ2,θ3,θ4]=[α,(1-α)U ocv ,-R int ,αR int -(1-α)R p ] S28: Define the physical loss L of the model as U t,k+1 and The absolute value of the difference is taken, and the neural network model is iteratively trained according to L and the gradient descent algorithm until L drops to a preset value or there is no longer a downward trend.

4. The data-driven lithium-ion battery voltage prediction modeling method integrating physical information according to claim 1 is characterized in that: The S3 is specifically: S31: Real-time acquisition of the current state of the battery, including the battery measurement terminal voltage U at the current moment t,k As feature 1, set the floating point number 1.0 as feature 2, extract the battery current I at the next moment k+1 As feature 3 and extracting the current battery current I k As feature 4; S32: Features 1 to 4 are used as model inputs In1 to In4, and input into the trained model, and the model outputs the predicted value of the terminal voltage at the next moment.

5. The data-driven lithium-ion battery voltage prediction modeling method integrating physical information according to claim 1 is characterized in that: The operating temperature data collected in S1 is used as an additional input feature to correct the temperature dependence of the equivalent circuit model parameters.

6. The data-driven lithium-ion battery voltage prediction modeling method integrating physical information according to claim 1 is characterized in that: The operating temperature data collected in S1 is used as an additional input feature to correct the temperature dependence of the equivalent circuit model parameters.

7. The data-driven lithium-ion battery voltage prediction modeling method integrating physical information according to claim 3 is characterized by: The intermediate variables θ1 to θ4 of the parameter layer constrain the weights of the neural network so that the ohmic internal resistance R in the equivalent circuit model during the training process int , Dynamic characteristics internal resistance R p , dynamic characteristic capacitance C p , satisfying the physical rationality condition: R int >0, R p >0,C p >0.

8. The data-driven lithium-ion battery voltage prediction modeling method integrating physical information according to claim 3 is characterized by: The equivalent circuit model is a first-order RC model, and its terminal voltage dynamic response is described by the following formula:

Images