Online Fast Prediction Method, Device and Storage Medium for the Service Life of Vehicle Power Batteries

Abstract

The present invention relates to an online rapid prediction method, device and storage medium for the life of a vehicle power battery integrating a mechanism model and a neural network. The method includes: obtaining internal parameters of the battery and determining key electrochemical parameters; constructing an estimation model of battery electrochemical parameters based on a neural network, using the load current and terminal voltage as features and the key electrochemical parameters as labels for training; constructing a lithium-ion battery life prediction model for online rapid life prediction: using the battery electrochemical parameter estimation model to process the data set to obtain the parameter change trend of the electrochemical model; classifying the degradation modes of the battery according to the change trend of the key electrochemical parameters, as well as the inflection points and different life lengths, and respectively establishing battery life prediction models; integrating different battery life prediction models to obtain a comprehensive online life prediction model. Compared with the prior art, the present invention has the advantages of high prediction accuracy and simple operation.

Description

Technical Field

[0001] The present invention relates to the field of power battery management for new energy vehicles, and particularly to an online fast prediction method, device and storage medium for the life of vehicle power batteries that integrates a mechanism model and a neural network. Background Art

[0002] The exhaustion of traditional energy and environmental pollution problems have drawn worldwide attention to the new energy field. Among various new energy batteries, lithium-ion batteries are widely used in multiple industries and fields including new energy vehicles due to their comprehensive performance characteristics such as small size, high energy density, long life cycle, and wide operating temperature range. However, in practical applications, system failures and safety accidents caused by battery life failure pose a great threat to people's lives and property safety.

[0003] The cascade utilization of lithium-ion batteries is also one of the hotspots in current electric vehicle research. Through cascade recycling, it is beneficial to reduce costs and achieve the full utilization of battery resources. However, a profound understanding and prediction ability of battery aging behavior are prerequisites for evaluating the overall cost of the system and the economic benefits of battery reuse.

[0004] There are usually two technical routes for the life prediction method of lithium-ion batteries, namely the model-based method and the data-driven method. The model-based method is usually based on an electrochemical model or an equivalent circuit model, and combines a particle filter algorithm to observe and update the internal state of the battery and predict the change trend of the battery capacity. This method has strong short-term prediction ability but poor long-term prediction effect, and the calculation process is relatively complex, and the actual application scenarios of the battery cannot provide sufficient computing resources. In recent years, with the development of big data of electric vehicles and the rise of some intelligent algorithms, the data-driven battery life prediction algorithm is gradually becoming a research hotspot for life prediction. The data-driven method obtains information from the current and terminal voltage data during the charge and discharge cycle of the battery, and uses methods such as neural networks and Gaussian regression to learn the potential relationship between this information and the final life, so as to achieve the effect of life prediction. The difficulty of the data-driven method lies in how to select the features with the highest correlation with the final life from the massive lithium-ion battery data and realize the prediction of the lithium-ion battery life as early as possible. At present, the problem with this method is that when the selected features are inappropriate or the amount of feature data is insufficient, the prediction accuracy of the data-driven method will decrease to a large extent. Summary of the Invention

[0005] The purpose of the present invention is to provide an online fast prediction method, device and storage medium for the life of vehicle power batteries that integrates a mechanism model and a neural network, which while retaining the high prediction accuracy of the mechanism model, combines the neural network model to simplify the prediction process and improve the prediction speed.

[0006] The object of the present invention can be achieved by the following technical solutions:

[0007] An online rapid prediction method for the life of vehicle power batteries by integrating a mechanism model and a neural network, comprising the following steps:

[0008] Step 1) Obtain internal battery parameters and determine key electrochemical parameters:

[0009] Step 11) Establish an electrochemical model that accurately reflects the internal potential of the battery and the distribution of lithium-ion concentration, and solve the electrochemical model using numerical methods;

[0010] Step 12) Conduct a sensitivity analysis on the parameters of the electrochemical model, and determine the required key electrochemical parameters according to the results of the sensitivity analysis;

[0011] Step 2) Construct and train a battery electrochemical parameter estimation model based on a neural network:

[0012] Step 21) Modify the values of the key electrochemical parameters to simulate the aging of the battery with cycling, solve the electrochemical model under different parameter conditions, obtain the output terminal voltage of the electrochemical model under different load currents, and generate a dataset for training the battery electrochemical parameter estimation model;

[0013] Step 22) Use the load current and terminal voltage as the features of the model, and the key electrochemical parameter values as the labels to train the neural network. Consider the physical constraints brought by the differential equations during the network training process to improve the network accuracy;

[0014] Step 3) Construct a lithium-ion battery life prediction model and use it for online rapid life prediction:

[0015] Step 31) Use the publicly available lithium-ion battery dataset as the training basis for the lithium-ion battery life prediction model, and process the publicly available dataset using the battery electrochemical parameter estimation model established in Step 2) to obtain the parameter change trend of the electrochemical model;

[0016] Step 32) Classify the degradation modes of the battery according to the different change trends, inflection points and life lengths of different key internal electrochemical parameters of the battery, and establish battery life prediction models for different categories respectively;

[0017] Step 33) Integrate the battery life prediction models of different degradation modes to obtain a comprehensive online life prediction model with a small scale and fast operation speed for battery life prediction.

[0018] In the above Step 11), an electrochemical model of a lithium-ion battery is established based on the pseudo-two-dimensional (P2D) model.

[0019] In step 12), the sensitivity analysis of the electrochemical model parameters is realized based on the one-at-a-time (OAT) method, and the sensitivity calculation formula for each parameter is as follows:

[0020]

[0021] where SI represents the sensitivity index of the parameter, N s represents the number of selections of the parameter within its effective range, and V j is the magnitude of the terminal voltage at the j-th value of the parameter, is the average value of the terminal voltage at the j-th value of the parameter.

[0022] The battery electrochemical parameter estimation model is trained by using the Physics-Informed Neural Network (PINN) method.

[0023] In step 31), using the battery electrochemical parameter estimation model established in step 2) to process the public dataset specifically means: preprocessing the public dataset by using the battery electrochemical parameter estimation model to obtain the time series of key electrochemical parameters changing with battery aging.

[0024] In step 32), the Bacon-Watts (BW) method is used to model and fit the battery nonlinear degradation curve to obtain its inflection point, and its expression is:

[0025] Y = α0 + α1(x - x1) + α2(x - x1)tanh((x - x1) / γ) + Z

[0026] where α0 is related to the initial capacity of the curve, α1 and α2 are related to the degradation rates of the two segments of the curve, x1 is the position where the inflection point of the curve appears, γ is the curve turning degree coefficient, the smaller this coefficient is, the faster the curve turning speed is, Y is the capacity of the vehicle-used power battery, x is the number of cycles of the vehicle-used power battery, and Z is the noise generated during capacity measurement.

[0027] In step 32), the degradation modes of different batteries are classified by the k-nearest neighbor algorithm.

[0028] In step 32), the battery life prediction model is established based on the Long-Short Time Memory (LSTM) network. The input is the sequence of electrochemical parameters changing with time, and the output is the remaining service life and inflection point value of the battery. After training the battery life prediction model, it is used to estimate the remaining service life and inflection point of the battery.

[0029] In step 33), the multiple LSTM networks obtained in step 32) are integrated into a smaller-scale and faster-running sub-network.

[0030] An on-line rapid prediction device for the life of vehicle power batteries integrating a mechanism model and a neural network, comprising a memory, a processor, and a program stored in the memory, and when the processor executes the program, the method as described above is implemented.

[0031] A storage medium, on which a program is stored, and when the program is executed, the method as described above is implemented.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] (1) The on-line prediction method for battery life of the present invention uses a battery mechanism model to obtain features with a high correlation with the degree of battery aging, and uses the changes of these features in the early stage of the battery life cycle to classify the lithium-ion degradation mode, predict the remaining service life and inflection point. The feature dimension is higher and the prediction accuracy is high;

[0034] (2) Compared with the single data-driven method, the feature construction of the present invention is more interpretable;

[0035] (3) Compared with the single model-based method, the present invention has a faster operation speed, more accurate results and stronger adaptability. Description of the Drawings

[0036] Figure 1 It is the flowchart of the method of the present invention.

[0037] Figure 2 It is a schematic diagram of the P2D battery electrochemical model.

[0038] Figure 3 It is a schematic diagram of the method for obtaining the inflection point of the battery capacity attenuation curve. Detailed Embodiments

[0039] The present invention will be described in detail below with reference to the drawings and specific embodiments. This embodiment is implemented on the premise of the technical solution of the present invention, and gives the detailed implementation manner and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0040] This embodiment provides an on-line rapid prediction method for the life of vehicle power batteries integrating a mechanism model and a neural network, as Figure 1 shown, including the following steps:

[0041] Step 1) Obtain the internal parameters of the battery and determine the key electrochemical parameters.

[0042] Step 11) Establish an electrochemical model that accurately reflects the internal electric potential and lithium-ion concentration distribution of the battery, and use numerical methods to solve the electrochemical model.

[0043] In this embodiment, an electrochemical model of a lithium-ion battery is established based on the P2D model. The P2D model is as follows: Figure 2 As shown, the battery is divided into a current collector, an anode, a separator, and a cathode in the x direction. Each part is controlled by the following coupled differential equations, which together describe the potential distribution, lithium-ion concentration distribution, and electrode overpotential in the solid phase and electrolyte phase inside the lithium-ion battery:

[0044] Lithium-ion concentration distribution in the electrolyte phase:

[0045]

[0046] where ε e is the electrolyte volume fraction, c e is the lithium-ion concentration in the electrolyte phase of the battery, is the equivalent diffusion coefficient of the electrolyte, is the cation transference number, F is the Faraday constant, and j Li is the electrode current density.

[0047] Lithium-ion concentration distribution in the solid phase:

[0048]

[0049] where, c s is the lithium-ion concentration in the solid phase, and D s is the solid-phase diffusion coefficient.

[0050] Electrolyte phase potential distribution:

[0051]

[0052] where κ eff is the effective conductivity of the electrolyte, R is the gas constant, T is the temperature, and f is the mean molar activity coefficient.

[0053] Solid-phase potential distribution:

[0054]

[0055] where σ eff is the effective conductivity of the solid phase.

[0056] The electrode surface current density couples these partial differential equations. The current density j Li is described by the Bulter-Volmer equation:

[0057]

[0058] where is the specific surface area of the electrode, i0 is the exchange current density, α a and αc The charge transfer coefficients of the anode and cathode, respectively.

[0059] The battery is divided into multiple control volumes in the x - direction using the finite volume method, and the derivatives in the differential equations are replaced by differences. Finally, the numerical solution of the terminal voltage of the P2D equation is obtained according to the applied current condition.

[0060] Step 12): Conduct a sensitivity analysis on the parameters of the electrochemical model, and determine the required key electrochemical parameters according to the results of the sensitivity analysis.

[0061] In this embodiment, the value ranges of many parameters in the P2D model are estimated through relevant literature, and then the sensitivity of each parameter is analyzed by the OAT method, that is, only one parameter is changed each time and its impact on the final output is observed. The method of changing parameters is to take values at equal intervals within the value range of each parameter; the final output selects the terminal voltage of the battery. The sensitivity calculation method for each parameter is as follows:

[0062]

[0063] where SI represents the sensitivity index of the parameter, N s represents the number of selections of the parameter within its reasonable range, V j is the magnitude of the terminal voltage at the j - th value of the parameter, is the average value of the terminal voltage at the j - th value of the parameter.

[0064] By calculating the influence of different parameter changes on the magnitude of the terminal voltage, the sensitivity of each parameter can be obtained. Combining the roles of each parameter during battery aging, three parameters that have a relatively high impact on the terminal voltage value and have a certain correlation with battery aging can be finally selected as the observation objects of the battery electrochemical parameter estimation model, that is, the key electrochemical parameters.

[0065] Step 2): Construct a neural - network - based battery electrochemical parameter estimation model and train it.

[0066] Step 21): Modify the values of the key electrochemical parameters to simulate the aging of the battery during cycling. Solve the electrochemical model under different parameter conditions to obtain the output terminal voltage of the electrochemical model under different applied currents, and generate a data set for training the battery electrochemical parameter estimation model.

[0067] During the cyclic aging process of the battery, as the internal components of the battery age, it will cause a certain degree of change in the response of its terminal voltage under the condition of the same applied current. Through the analysis in Step 12), the change in the terminal voltage caused by battery aging can be attributed to the changes in three key parameters.

[0068] Since the existing datasets lack the annotation of the parameters of the battery electrochemical model, that is, the existing data is insufficient to train the battery electrochemical parameter estimation model. Therefore, it is necessary to solve the battery electrochemical model obtained in step 11) to generate a dataset. During the process of generating the dataset, the key parameters obtained in step 12) should be reasonably modified so that the final terminal voltage response result is as close as possible to the response result of the aged battery in the real dataset.

[0069] In step 22), the load current and the terminal voltage are used as the features of the model, and the key electrochemical parameter values are used as the labels to train the neural network. During the network training process, the physical constraints brought by the differential equations are considered to improve the network accuracy.

[0070] After obtaining the dataset, the PINN method is used to train the battery electrochemical parameter estimation model. The PINN method essentially belongs to a type of neural network, but the difference is that it will consider the physical constraint information brought by the differential equations during the construction of the loss function. Ordinary neural networks usually use mean square error or cross-entropy error, etc. when constructing the loss function. However, this kind of loss function only considers the fitting degree of the neural network output to the real data. For real physical systems such as batteries, whose dynamic behavior is described by multiple partial differential equations, the training speed and network prediction accuracy of the neural network can be effectively improved by adding the constraints of the derivative terms brought by the partial differential equations to the loss function.

[0071] Step 3) Construct a lithium-ion battery life prediction model and use it for online rapid life prediction.

[0072] In step 31), the publicly available dataset of lithium-ion batteries is used as the training basis for the lithium-ion battery life prediction model, and the battery electrochemical parameter estimation model established in step 2) is used to process the publicly available dataset to obtain the trend of parameter changes of the electrochemical model.

[0073] Specifically, in this step, a time series of key electrochemical parameters changing with battery aging will be obtained.

[0074] In step 32), according to the different trends of key electrochemical parameter changes inside different batteries, as well as the different inflection points and life lengths, the degradation modes of the batteries are classified, and battery life prediction models are established for different categories respectively.

[0075] The ultimate goal of the present invention is to online predict the remaining service life and inflection points of lithium-ion batteries, where the remaining service life is generally defined as how many cycles the battery capacity decays to 80% of the initial capacity. This information can be directly obtained from the publicly available lithium-ion battery aging dataset. The battery inflection point is generally defined as: before the inflection point, the battery capacity shows a linear decline, while after the inflection point, the battery capacity shows a non-linear rapid decline.

[0076] The inflection point of the battery capacity decay trajectory is obtained by modeling and fitting the non-linear decay curve of the battery using the Bacon-Watts method, and its expression is:

[0077] Y = α0 + α1(x - x1) + α2(x - x1)tanh((x - x1) / γ) + Z

[0078] where α0 is related to the initial capacity of the curve, α1 and α2 are related to the decay rates of the two segments of the curve, x1 is the position where the inflection point of the curve appears, γ is the curve turning degree coefficient, the smaller this coefficient, the faster the curve turns, Y is the capacity of the vehicle power battery, x is the number of cycles of the vehicle power battery, and Z is the noise generated during capacity measurement.. The schematic diagram of fitting the battery capacity decay curve by the Bacon-Watts (BW) method is as Figure 3 shown.

[0079] The non-linear function fitting method uses the Levenberg-Marquart (LM) method to obtain the parameters in the Bacon-Watts model, that is, to obtain the inflection point. The LM method uses the mean square error as the optimization objective function, and the update vector h lm in its optimization process is as follows:

[0080]

[0081] where is the model function provided by the Bacon-Watts method, Q is the actual battery capacity data, p = [α0, α1, x1, α2] T is the parameter of the BW model function, W is the weight matrix, is the Jacobian matrix of the model function with respect to the parameters. Using this method, the inflection point value of each battery decay trajectory can be obtained.

[0082] Combining step 31) and step 32), the input-output data set of the battery life prediction model is N is the number of training samples; where x n = [θ1, θ2, θ3] T is the time series of internal high-sensitivity parameter changes of the nth battery, and y n = [h, kp] T is the life and inflection point value of the capacity sequence of the nth battery.

[0083] Different battery usage conditions will result in different battery life decay trends. According to different life lengths and inflection point values, clustering is performed using the k-nearest neighbor algorithm, and the lithium-ion battery capacity trajectories are divided into three decay modes: short, medium, and long. For these three categories, a battery life prediction model based on the LSTM network is established respectively to obtain more accurate prediction results. The input of this network is the time series of highly sensitive electrochemical parameters of the battery in the previous 100 cycles, and the input is the remaining useful life and inflection point with confidence. The calculation of the confidence interval is achieved by setting the fractional loss:

[0084]

[0085] Step 33) Integrate the battery life prediction models of different degradation modes, that is, integrate the three LSTM network models in Step 32) to obtain a comprehensive online life prediction model with a small scale and fast operation speed for battery life prediction.

[0086] If the above functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0087] The preferred specific embodiments of the present invention have been described in detail above. It should be understood that those of ordinary skill in the art can make many modifications and variations based on the concept of the present invention without creative labor. Therefore, all technical solutions that can be obtained by those skilled in the art in the technical field of the present invention based on the concept of the present invention through logical analysis, reasoning, or limited experiments on the basis of the prior art should be within the protection scope determined by the claims.

Claims

1. An online rapid prediction method for the life of vehicle power batteries that integrates a mechanism model and a neural network, characterized in that, Including the following steps: Step 1) Obtain the internal parameters of the battery and determine the key electrochemical parameters: Step 11) Establish an electrochemical model that accurately reflects the internal potential of the battery and the distribution of lithium-ion concentration, and solve the electrochemical model using numerical methods; Step 12) Conduct a sensitivity analysis on the parameters of the electrochemical model, and determine the required key electrochemical parameters according to the results of the sensitivity analysis; Step 2) Construct a battery electrochemical parameter estimation model based on a neural network and train it: Step 21) Modify the values of the key electrochemical parameters to simulate the aging of the battery with cycling, solve the electrochemical model under different parameter conditions, obtain the output terminal voltage of the electrochemical model under different load currents, and generate a dataset for training the battery electrochemical parameter estimation model; Step 22) Use the load current and terminal voltage as the features of the model, and the key electrochemical parameter values as the labels to train the neural network. Consider the physical constraints brought by the differential equations during the network training process to improve the network accuracy; Step 3) Construct a lithium-ion battery life prediction model and use it for online rapid life prediction: Step 31) Use the publicly available lithium-ion battery dataset as the training basis for the lithium-ion battery life prediction model, and use the battery electrochemical parameter estimation model established in Step 2) to process the publicly available dataset to obtain the change trend of the parameters of the electrochemical model; Step 32) Classify the degradation modes of the battery according to the different change trends, inflection points and life lengths of different key internal electrochemical parameters of the battery, and establish battery life prediction models for different categories respectively; Step 33) Integrate the battery life prediction models of different degradation modes to obtain a comprehensive online life prediction model with a small scale and fast operation speed for battery life prediction.

2. The online rapid prediction method for the service life of a vehicle power battery integrating a mechanism model and a neural network according to claim 1, wherein In the above Step 11), an electrochemical model of the lithium-ion battery is established based on the pseudo-two-dimensional P2D model.

3. An online rapid prediction method for the service life of vehicle power batteries integrating a mechanism model and a neural network, characterized in that, In the above Step 12), the sensitivity analysis of the electrochemical model parameters is realized based on the one-by-one analysis method, and the sensitivity calculation formula for each parameter is: Among them, SI represents the sensitivity index of the parameter, N s represents the number of choices of the parameter within its valid range, V j is the magnitude of the terminal voltage at the j-th value of the parameter, is the average value of the terminal voltage at the j-th value of the parameter.

4. A method for online rapid prediction of the service life of vehicle power batteries integrating a mechanism model and a neural network, characterized in that, The battery electrochemical parameter estimation model is trained using the PINN method.

5. An online rapid prediction method for the service life of vehicle power batteries integrating a mechanism model and a neural network according to claim 1, characterized in that, In the above Step 31), using the battery electrochemical parameter estimation model established in Step 2) to process the publicly available dataset specifically means: using the battery electrochemical parameter estimation model to preprocess the publicly available dataset to obtain the time series of the key electrochemical parameters changing with battery aging.

6. The online rapid prediction method for the service life of vehicle power batteries integrating a mechanism model and a neural network according to claim 1, characterized in that In the above Step 32), the Bacon-Watts method is used to model and fit the non-linear degradation curve of the battery to obtain its inflection point, and its expression is: Y = α0 + α1(x - x1) + α2(x - x1)tanh((x - x1) / γ) + Z Where, α0 is related to the initial capacity of the curve, α1 and α2 are related to the degradation rates of the two segments of the curve, x1 is the position where the inflection point of the curve appears, γ is the turning degree coefficient of the curve, the smaller this coefficient is, the faster the turning speed of the curve, Y is the capacity of the vehicle power battery, x is the number of cycles of the vehicle power battery, and Z is the noise generated during capacity measurement.

7. The online rapid prediction method for the service life of vehicle power batteries integrating a mechanism model and a neural network according to claim 1, characterized in that, In the above Step 32), the k-nearest neighbor algorithm is used to classify the degradation modes of different batteries.

8. An online rapid prediction method for the service life of vehicle power batteries integrating a mechanism model and a neural network, characterized in that, In step 32), the battery life prediction model is established based on a long short-term memory network. The input is a sequence of electrochemical parameters varying with time, and the output is the remaining service life and inflection point value of the battery. After training the battery life prediction model, it is used to estimate the remaining service life and inflection point of the battery.

9. An on-line rapid prediction device for the life of vehicle power batteries integrating a mechanism model and a neural network, comprising a memory, a processor, and a program stored in the memory, characterized in that, When the processor executes the program, it implements the method according to any one of claims 1-8.

10. A storage medium having a program stored thereon, characterized in that, When the program is executed, it implements the method according to any one of claims 1-8.

Images