A method for constructing a digital twin model of a power battery

Abstract

The present invention discloses a method for constructing a digital twin model of a power battery, which comprises the following steps: constructing a multi-scale image model of a power battery; constructing a data-driven model; constructing a digital twin model of a power battery based on the multi-scale image model of the power battery and superimposing the data-driven model; processing the data of the digital twin model of the power battery and optimizing the digital twin model of the power battery; through the precise construction of the digital twin model of the service cycle of the power battery, using the data-driven method to extract information from the real-time monitoring data, and supplementing the extracted information data into the multi-scale image model, thereby optimizing the accuracy of the initial multi-scale image model, and studying the twin data processing method of the service cycle of the power battery, through the multi-scale digital twin model components and the optimization processing of the twin data, the twin data of the power battery is expressed with high precision.

Description

Technical Field

[0001] The present invention relates to the technical field of power batteries, and in particular to a method for constructing a digital twin model of a power battery. Background Art

[0002] Power batteries are an important component of new energy vehicles. Since most of the current power batteries are lithium batteries, the life cycle of lithium batteries has a very serious impact on the performance of lithium batteries. The life cycle of lithium batteries can be divided into a stable period, a sharp period and a scrap period.

[0003] In the prior art, at the end of the life cycle of a lithium battery, cascade utilization, recycling and remanufacturing can be selected for recycling according to the remaining capacity of the lithium battery; among them, cascade utilization is mainly aimed at lithium batteries with less capacity attenuation, while recycling refers to crushing and disassembling lithium batteries at the end of their life cycle and recovering the heavy metals therein, and remanufacturing is mainly used for lithium batteries in a critical period, which requires accurate judgment of the stage of the lithium battery.

[0004] In the prior art, the understanding of the life cycle of lithium batteries mainly relies on the test of the remaining capacity of lithium batteries. For large-scale retired lithium batteries, relying solely on the test of the remaining capacity of lithium batteries obviously consumes a lot of manpower and material resources, and it is impossible to select the accurate remanufacturing time domain according to the service cycle of lithium batteries. As a result, a large number of retired lithium batteries can only be recycled through recycling. In order to accurately predict the remanufacturing time domain of lithium batteries according to the service cycle of lithium batteries, it is necessary to construct an electronic model of lithium batteries in advance to reflect the relationship between the service cycle of lithium batteries and their remaining capacity.

[0005] Therefore, a method for constructing a digital twin model of a power battery is proposed to address the above problems. Summary of the invention

[0006] In order to make up for the deficiencies of the prior art and solve at least one of the above problems, the present invention proposes a method for constructing a digital twin model of a power battery.

[0007] A method for constructing a power battery digital twin model, the method comprising the following steps:

[0008] S1: Construction of multi-scale imaging model of power battery;

[0009] S2: Construction of data-driven models;

[0010] S3: Based on the multi-scale image model of the power battery, the data-driven model is superimposed to build a digital twin model of the power battery;

[0011] S4: Process the data of the power battery digital twin model and optimize the power battery digital twin model;

[0012] In S2, data-driven model construction is specifically to extract information from the monitoring data and superimpose it into the multi-scale image model of the power battery.

[0013] Preferably, in said S1, the construction of the multi-scale image model of the power battery includes the construction of the electrochemical-thermal model of the power battery, the thermal-mechanical coupling model of the power battery, and the geometric model from the particle to the electrode, the monomer, the module, and the whole package, respectively; wherein the geometric model corresponding to the real shape of the physical battery is established at the scale of the power battery module and the whole package;

[0014] The electrochemical, thermal and force models of the power battery are expressed by the following formulas:

[0015] Electrochemical Model:

[0016] Lithium concentration gradient creates diffusion-induced stress:

[0017] Stress affects the rate of chemical reactions:

[0018] Thermal Model:

[0019] Temperature differences create thermal stress:

[0020] Mechanical deformation causes heat generation changes: u M =u T (6)

[0021] Force Model:

[0022] Electrochemical parameters that vary with temperature:

[0023] Heat generated by chemical reaction: q = q rev +q irrev (9);

[0024] Preferably, in S3, the multi-scale imaging model of the power battery contains a large number of partial differential equations, which require the input of a large number of parameters. The parameters are obtained through information extraction by a data-driven model of real-time detection data, and are superimposed and supplemented to the multi-scale imaging model of the power battery to optimize the digital twin model of the power battery.

[0025] Preferably, in S4, the data of the power battery digital twin model includes monitoring data of the physical layer and simulation data of the virtual layer; the data of the power battery digital twin model is processed, including a power battery twin data quality assessment and training data recovery model based on Deep-LSTM.

[0026] Preferably, the Deep-LSTM-based power battery twin data quality assessment specifically includes extracting external characteristic variables of monitoring data, including voltage, current, temperature, and external characteristic variables of simulation data, including SOC, SOH, RUL, from the power battery twin data as input vectors of the Deep-RNN model;

[0027] Input matrix FF t,L It is expressed as:

[0028]

[0029] Where T1, T2…T n Represents the time series of external characteristic variables, FF t,L is the input variable of the Deep-RNN network, and L is the length of the input variable;

[0030] At the same time, the STM unit is used to improve the performance of the Deep-RNN temporal feature extraction model, and three different gates are deployed in the LSTM unit to control the information flow in the Deep-RNN network;

[0031] The time series of the battery terminal voltage is selected as the output of the network, and the correct battery voltage state is predicted based on the time series features extracted from the time series of the battery terminal voltage. A fully connected regression layer is set on top of the Deep-LSTM network, and the entire network is trained as a sequence-to-sequence regression model. The time series of the battery terminal voltage in the regression model defines the following model training objectives:

[0032] J t,L =[U(tL)...U(n)...U(t)] (11)

[0033] Where U is the battery terminal voltage value observed at time t; after the Deep-LSTM model is fully trained, its inter-sequence regression error is used to evaluate the battery data quality in real time:

[0034]

[0035] Use L samples in the battery historical operation data sequence as model input to estimate the battery terminal voltage The difference between it and the observed value U(t+1) is used to evaluate the data quality.

[0036] Preferably, the training data recovery model includes using a trained Deep-LSTM network to extract the time features of the battery external feature data; using a three-layer structure DAE consisting of an input layer, a hidden layer and an output layer to process the model features in the battery external feature data; wherein the front end of the network is used as an encoder to extract the model features, and the fully connected layer is used as a decoder to reconstruct and repair bad samples, and the noise-free normal battery data is used as the output to train the data recovery model.

[0037] The present invention is beneficial in that:

[0038] 1. The present invention uses a data-driven method to extract information from real-time monitoring data through precise components of the digital twin model of the service cycle of power batteries, and supplements the extracted information data into the multi-scale image model, thereby optimizing the accuracy of the initial multi-scale image model, as well as research on the twin data processing method of the service cycle of power batteries. Through the multi-scale digital twin model components and the optimization processing of twin data, the power battery twin data is expressed with high precision. Through the components of the digital twin model of power batteries, the relationship between the service cycle of lithium batteries and their remaining capacity is easier to obtain, so that the time domain of lithium battery remanufacturing can be accurately judged without testing the remaining capacity of the battery, thereby selecting different recycling methods to recycle the lithium battery. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0040] Figure 1 A flowchart of an embodiment of the present invention;

[0041] Figure 2 This is a diagram of the electrochemical-thermal-mechanical coupling framework structure of a power battery in one embodiment of the present invention;

[0042] Figure 3 A fusion diagram of a multi-scale image model and a data-driven model of a power battery in an embodiment of the present invention;

[0043] Figure 4 A topological structure diagram of a Deep-RNN network in an embodiment of the present invention;

[0044] Figure 5 An information flow diagram in an LSTM unit in one embodiment of the present invention;

[0045] Figure 6 This is a diagram of a battery data recovery model based on a feature fusion method in an embodiment of the present invention. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in 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.

[0047] See also Figure 1-Figure 6 As shown, a method for constructing a power battery digital twin model includes the following steps:

[0048] S1: Construction of multi-scale imaging model of power battery;

[0049] S2: Construction of data-driven models;

[0050] S3: Based on the multi-scale image model of the power battery, the data-driven model is superimposed to build a digital twin model of the power battery;

[0051] S4: Process the data of the power battery digital twin model and optimize the power battery digital twin model;

[0052] In S2, data-driven model construction is specifically to extract information from the monitoring data and superimpose it into the multi-scale image model of the power battery.

[0053] As an embodiment of the present invention, in S1, the construction of a multi-scale image model of a power battery includes the construction of a power battery electrochemical-thermal model, a power battery thermal-mechanical coupling model, and a geometric model from particles to electrodes, monomers, modules, and whole packages; wherein a geometric model corresponding to the real shape of a physical battery is established at the power battery module and whole package scales; when constructing a multi-scale image model of a power battery, it is necessary to comprehensively analyze the full-factor parameter information of the power battery under the coupling state of multiple physical fields such as electrochemical field, temperature field, and stress field, and construct a multi-scale image model of a power battery from particles to electrodes, monomers, modules, and whole packages. In-depth analysis of the parameter transfer and interaction laws between different physical fields and different hierarchical models of power batteries, adding a data-driven model on the basis of the multi-scale image model, and establishing a power battery digital twin model provide a basic model for the study of the cycle aging and decline mechanism of power batteries; traditional experimental methods can only monitor the apparent characteristics of batteries, such as voltage, temperature, capacity, etc., but it is difficult to obtain the electrochemical, mechanical and thermal characteristics information from the microscopic to the macroscopic inside of the battery, resulting in the inability to accurately evaluate the comprehensive performance of power batteries. To address the above issues, an electrochemical-thermal-mechanical coupling model of power batteries is constructed from power battery particles to electrodes and monomers, such as Figure 2 As shown in the figure; at the particle scale, the diffusion-induced stress generated by the lithium concentration gradient is considered, and the electrochemical-mechanical model of the power battery is constructed based on the P2D model; at the electrode scale, according to the energy conservation equation, an electrochemical-thermal model is established on the basis of the P2D model; at the monomer scale, the thermal stress and thermal expansion behavior caused by the temperature gradient during the battery operation are considered to construct a thermal-mechanical coupling model of the power battery; from the power battery module and the whole package scale, a geometric model corresponding to the real shape of the physical battery is established; the industrial modeling software is used to construct a three-dimensional CAD model of the power battery to accurately express its key structural parameters, the constraints and positioning relationships between monomers and modules; Pixyz Studio is further used to optimize the geometric model through surface subdivision, topological correction, extraction, and repair of the three-dimensional CAD model;

[0054] The electrochemical, thermal and force models of the power battery are expressed by the following formulas:

[0055] Electrochemical Model:

[0056] Lithium concentration gradient creates diffusion-induced stress:

[0057] Stress affects the rate of chemical reactions:

[0058] Thermal Model:

[0059] Temperature differences create thermal stress:

[0060] Mechanical deformation causes heat generation changes: u M =u T (6)

[0061] Force Model:

[0062] Electrochemical parameters that vary with temperature:

[0063] Heat generated by chemical reaction: q = q rev +q irrev (9);

[0064] As an embodiment of the present invention, in S3, the multi-scale image model of the power battery contains a large number of partial differential equations, which require the input of a large number of parameters. The parameters are obtained by information extraction from the data-driven model of the real-time detection data, and are superimposed and supplemented to the multi-scale image model of the power battery to optimize the power battery digital twin model; in the process of using the multi-scale image model of the power battery, it is inevitable that the model parameters will change with the application conditions or environment, resulting in inaccurate models and other problems. It is difficult for a single model to conduct in-depth mining of the collected data and to update the model parameters in real time. The multi-scale image model of the power battery contains a large number of partial differential equations, which require the input of a large number of parameters for calculation, and some internal parameters of the battery cannot be accurately obtained in real time. In order to make the battery digital twin modeling more accurate, a data-driven model is added on the basis of the multi-scale digital twin modeling of the power battery, such as Figure 3 As shown in the figure, a data-driven method is used to extract information from real-time monitoring data, and the extracted data is supplemented into the multi-scale image model to construct a multi-scale digital twin model of the power battery and improve the modeling accuracy.

[0065] As an embodiment of the present invention, in S4, the data of the power battery digital twin model includes monitoring data of the physical layer and simulation data of the virtual layer; the data of the power battery digital twin model is processed, including a power battery twin data quality assessment and training data recovery model based on Deep-LSTM.

[0066] As an embodiment of the present invention, the Deep-LSTM-based power battery twin data quality assessment specifically includes extracting external characteristic variables of monitoring data, including voltage, current, temperature, and external characteristic variables of simulation data, including SOC, SOH, and RUL, from the power battery twin data as input vectors of the Deep-RNN model;

[0067] Input matrix FF t,L It is expressed as:

[0068]

[0069] Where T1, T2…T n Represents the time series of external characteristic variables, FF t,L is the input variable of the Deep-RNN network, and L is the length of the input variable, as shown in the attached Figure 4 As shown;

[0070] At the same time, the STM unit is used to improve the performance of the Deep-RNN time series feature extraction model. Three different gates are deployed in the LSTM unit to control the information flow in the Deep-RNN network, as shown in the attached figure. Figure 5 As shown;

[0071] The time series of the battery terminal voltage is selected as the output of the network, and the correct battery voltage state is predicted based on the time series features extracted from the time series of the battery terminal voltage. A fully connected regression layer is set on top of the Deep-LSTM network, and the entire network is trained as a sequence-to-sequence regression model. The time series of the battery terminal voltage in the regression model defines the following model training objectives:

[0072] J t,L =[U(tL)...U(n)...U(t)] (11)

[0073] Where U is the battery terminal voltage value observed at time t; after the Deep-LSTM model is fully trained, its inter-sequence regression error is used to evaluate the battery data quality in real time:

[0074]

[0075] Use L samples in the battery historical operation data sequence as model input to estimate the battery terminal voltage The difference between it and the observed value U(t+1) is used to evaluate the data quality.

[0076] As an embodiment of the present invention, the training data recovery model includes using a trained Deep-LSTM network to extract the time characteristics of the battery external feature data; using a DAE with a three-layer structure consisting of an input layer, a hidden layer and an output layer to process the model characteristics in the battery external feature data; wherein the front end of the network is used as an encoder to extract the model characteristics, and a fully connected layer is used as a decoder to reconstruct and repair the bad samples, and the normal battery data without noise is used as the output, and the training data recovery model is as shown in the attached Figure 6 As shown;

[0077] Working principle: Through the precise components of the digital twin model of the power battery service cycle, a data-driven method is used to extract information from real-time monitoring data, and the extracted information data is supplemented into the multi-scale image model, which optimizes the accuracy of the initial multi-scale image model. As well as the research on the twin data processing method of the power battery service cycle, the power battery twin data is expressed with high precision through the multi-scale digital twin model components and the optimization processing of the twin data.

[0078] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0079] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements all fall within the scope of the present invention to be protected.

Claims

1. A method for constructing a power battery digital twin model, characterized in that: The construction method includes the following steps: S1: First, construct a multi-scale image model of the power battery; S2: Construction of data-driven models; S3: Based on the multi-scale image model of the power battery, the data-driven model is superimposed to build a digital twin model of the power battery; S4: Process the data of the power battery digital twin model and optimize the power battery digital twin model; S5: Reflect the service life of power batteries based on the power battery digital twin model; In S2, data-driven model construction is specifically to extract information from real-time monitoring data and superimpose and supplement it into the multi-scale image model of the power battery; In the S1, a multi-scale image model of a power battery is constructed, including a particle scale, an electrode scale, a cell scale, a module scale, and a whole package scale. An electrochemical-thermal-mechanical coupling model is constructed from the power battery particle, motor, and monomer scales. From the particle scale, the diffusion-induced stress generated by the lithium concentration gradient is combined to establish an electrochemical-mechanical model of the power battery; from the electrode scale, an electrochemical-thermal model is established on the basis of the P2D model according to the energy conservation and moisture-proof principle; from the monomer scale, a thermal-mechanical coupling model of the power battery is constructed in combination with the thermal stress and thermal expansion behavior caused by the temperature gradient during the battery operation process; from the power battery module and the whole package scale, a geometric model corresponding to the real shape of the physical battery is established; The electrochemical, thermal and force models of the power battery are expressed by the following formulas: Electrochemical Model: Lithium concentration gradient creates diffusion-induced stress: Stress affects the rate of chemical reactions: Thermal Model: Temperature differences create thermal stress: Mechanical deformation causes heat generation changes: u M =u T (6) Force Model: Electrochemical parameters that vary with temperature: Heat generated by chemical reaction: q = q rev +q irrev (9); In S3, the multi-scale image model of the power battery contains a large number of partial differential equations, which require a large number of parameters to be input. The parameters are obtained by information extraction from the data-driven model of the real-time detection data, and are superimposed and supplemented to the multi-scale image model of the power battery to optimize the digital twin model of the power battery; In S4, the data of the power battery digital twin model includes monitoring data of the physical layer and simulation data of the virtual layer; the data of the power battery digital twin model is processed, including a power battery twin data quality assessment and training data recovery model based on Deep-LSTM; The Deep-LSTM-based power battery twin data quality assessment specifically includes extracting external characteristic variables of monitoring data, including voltage, current, temperature, and external characteristic variables of simulation data, including SOC, SOH, and RUL, from the power battery twin data as input vectors of the Deep-RNN model; Input matrix FF t,L It is expressed as: Where T1, T2…T n Represents the time series of external characteristic variables, FF t,L is the input variable of the Deep-RNN network, L is the length of the input variable; Ti is the i-th external characteristic variable, including temperature, current and SOC; n is the total number of external characteristic variable types; Ti(t) is the time series of the external characteristic variable at time t, and Ti(tL) is the time series of the external characteristic variable at time (tL); At the same time, the STM unit is used to improve the performance of the Deep-RNN temporal feature extraction model, and three different gates are deployed in the LSTM unit to control the information flow in the Deep-RNN network; The time series of the battery terminal voltage is selected as the output of the network, and the correct battery voltage state is predicted based on the time series features extracted from the time series of the battery terminal voltage. A fully connected regression layer is set on top of the Deep-LSTM network, and the entire network is trained as a sequence-to-sequence regression model. The time series of the battery terminal voltage in the regression model defines the following model training objectives: J t,L =[U(t-L)...U(n)...U(t)] (11) Where U is the battery terminal voltage value observed at time t; after the Deep-LSTM model is fully trained, its inter-sequence regression error is used to evaluate the battery data quality in real time: Use L samples in the battery historical operation data sequence as model input to estimate the battery terminal voltage The difference between it and the observed value U(t+1) is used to evaluate the data quality.

2. The method for constructing a power battery digital twin model according to claim 1, characterized in that: The training data recovery model includes using a trained Deep-LSTM network to extract the time features of the battery external feature data; using a DAE with a three-layer structure consisting of an input layer, a hidden layer and an output layer to process the model features in the battery external feature data; wherein the front end of the network is used as an encoder to extract the model features, and a fully connected layer is used as a decoder to reconstruct and repair bad samples, and the noise-free normal battery data is used as the output to train the data recovery model.

Images