A method for reconstructing the spatiotemporal distribution of temperature of cylindrical lithium-ion batteries under fast charging conditions over a wide temperature range

Abstract

The present invention discloses a method for reconstructing the spatiotemporal distribution of temperature of a cylindrical lithium-ion battery under wide temperature range and fast charging conditions, including constructing a full-life spatiotemporal distribution reconstruction model of temperature based on a physical information neural network and a cross-domain reconstruction of the thermal field of a lithium battery based on the method. The present invention is applied to the field of lithium-ion battery management. By integrating heat generation and the physical laws of thermodynamics into the neural network, a full-life spatiotemporal distribution reconstruction model of temperature with physical significance is constructed, which can effectively improve the adaptability and accuracy of the model. It can not only reconstruct the temperature of the lithium-ion battery, but also infer time-varying physical quantities such as thermal conductivity coefficient, convection heat transfer coefficient, and open circuit voltage. At the same time, the thermal field of lithium batteries under different ambient temperatures and fast charging conditions is reconstructed through transfer learning, thereby improving the generalization ability of the model in different application scenarios, which is of great significance for thermal abuse warning and improving battery thermal control performance.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium-ion battery management, and in particular to a method for reconstructing the spatiotemporal distribution of temperature of a cylindrical lithium-ion battery under fast-charging conditions over a wide temperature range. Background Art

[0002] Lithium-ion batteries, with their high energy density, strong power performance and long cycle life, have become the core power source for new energy equipment such as highly mobile vehicles and various space equipment. However, the rapid development of new energy technologies has put forward higher requirements on the charging capacity of lithium-ion batteries in various environments, and the limitation of charging capacity has become a key bottleneck restricting its further widespread application. Fast charging technology is seen as an effective strategy to solve this problem, shorten energy replenishment time, and alleviate endurance anxiety. However, due to the extreme service environment and complex working conditions, the non-uniformity of temperature distribution and local overheating have become increasingly significant in the actual application of lithium battery fast charging technology, which in turn has a serious negative effect on the performance and safety of the equipment, and may even induce the risk of thermal runaway.

[0003] At present, the battery management system has significant limitations in temperature monitoring, and it is difficult to effectively support the decision-making and control of thermal management, which seriously hinders the realization of the goal of safe and efficient operation of the battery system. Given that the charging rate is the core indicator of the charging rate, its improvement is limited by the characteristics of the battery material, but this will sacrifice the cycle life and safety performance. In addition, the fast charging performance is not only constrained by the performance of the bulk material, but also affected by multiple factors. Temperature has a significant effect on the electrochemical reaction rate of lithium batteries. Low temperature slows down the charging and discharging rate and aggravates self-discharge, while high temperature is prone to cause side reactions, shorten the cycle life and increase the risk of thermal runaway. Therefore, in the fast charging scenario, the development of a high-precision and high-efficiency lithium-ion battery temperature spatiotemporal distribution prediction model is of great academic value and practical significance for ensuring the safe and efficient operation of the battery in a wide temperature range.

[0004] The battery thermal process is a typical distributed parameter system (DPS), whose characteristics are manifested in dynamic changes in two dimensions: time and space. In actual battery management system (BMS) applications, temperature monitoring often relies on temperature sensors installed on the surface of the battery, and the measured temperature is regarded as the average temperature of the battery. However, regardless of the sampling frequency, surface-mounted temperature sensors often have difficulty capturing rapid fluctuations in the temperature inside the battery. To overcome this limitation, a feasible solution is to embed micro-temperature sensors inside the battery cell or use electrochemical impedance spectroscopy for measurement. However, this method faces additional manufacturing difficulties and instrument requirements, which significantly increases the cost and complexity of the system, limiting its feasibility in practical applications.

[0005] In view of this, researchers turned to explore the combination of electrochemical models and thermal models, combining battery parameters such as current, voltage, and internal resistance, and estimating the internal temperature of the battery through specific algorithms. However, the implementation of this method requires a deep understanding of the internal working mechanism and reaction kinetics of each battery, and then constructing a complex mathematical model. This not only increases the difficulty of model construction, but also brings complex solution problems. In response to the above problems, data-driven methods have emerged. This method does not rely on the physical properties of the system, but uses data collected from the system to build an analysis model. It is worth noting that the modeling accuracy of this method depends largely on the quantity and quality of the sampled data. In the case of insufficient data, the long-term prediction accuracy of the method will be affected and the generalization performance will also decrease. However, in actual operation, it is often a challenge to obtain a large amount of high-quality data. In addition, although this method relies on a large amount of labeled data and empirical observations, the understanding of the background process may be limited or completely unknown. It can be seen that in the field of dynamic thermal property reconstruction during battery operation, data-driven methods and physical model-based methods each show unique advantages. In order to improve the efficiency and accuracy of temperature spatiotemporal distribution reconstruction, a few studies have proposed hybrid reconstruction models that integrate physical information and neural networks to simulate the complex electrothermal dynamic behavior of batteries. Under this framework, the model can not only generate reconstruction results that can be explained according to physical laws, thereby ensuring the generalization performance of the reconstruction, but also further improve the accuracy of the reconstruction by virtue of the powerful nonlinear mapping ability of the data-driven model. This method is widely recognized because it can effectively integrate physical mechanisms and experimental observation data, showing a high degree of flexibility and making up for the insurmountable gap of traditional methods. However, there is currently a lack of research to verify the accuracy of battery thermal field reconstruction under fast charging conditions. In addition, during the aging process of the battery, its heat generation characteristics will change accordingly, which requires that appropriate methods must be selected in the aging cycle test to accurately reconstruct the battery thermal field, so as to ensure that the battery management system can accurately monitor the system status. Furthermore, in practical applications, lithium-ion batteries often need to operate at a variety of ambient temperatures, which leads to significant differences in data distribution between training temperature and test temperature, which may cause model performance degradation. And in real application scenarios, obtaining accurate and reliable labeled data also faces many difficulties. Therefore, developing a method to reconstruct the spatiotemporal distribution of temperature of full-life lithium-ion batteries for fast charging conditions over a wide temperature range has far-reaching significance for improving the adaptability and accuracy of the model.

[0006] In view of the current challenges in the field of lithium battery temperature spatiotemporal distribution estimation, combined with the characteristics of lithium batteries under wide temperature range fast charging conditions, the present invention proposes a method for reconstructing the spatiotemporal distribution of lithium-ion battery temperature over the entire life cycle that is suitable for wide temperature range fast charging conditions. The method includes a pre-training stage and a migration stage. In the pre-training stage, for fast charging application scenarios, we integrate heat generation and thermodynamic physical laws into the gated recurrent unit (GRU) neural network model, aiming to learn and characterize the dynamic behavior of the underlying physical system through data fitting. And introduce an attention mechanism to effectively capture long-term dependencies in historical data, accurately observe the temperature fluctuations of the battery in aging cycle tests, and then construct a full life cycle temperature spatiotemporal distribution reconstruction model with physical significance. In the migration learning stage, the method uses a domain adaptive migration learning method to migrate the knowledge learned in the pre-training stage, aiming to reconstruct the thermal field of lithium batteries across different ambient temperatures and fast charging conditions, thereby improving the generalization ability of the model in different application scenarios. Summary of the invention

[0007] In view of the deficiencies in the above-mentioned prior art, the present invention provides a method for reconstructing the spatiotemporal distribution of temperature of a cylindrical lithium-ion battery under fast charging conditions over a wide temperature range. The method integrates heat generation and the physical laws of thermodynamics into a neural network to construct a full-life temperature spatiotemporal distribution reconstruction model with physical significance, which can effectively improve the adaptability and accuracy of the model, and reconstruct the thermal field of lithium batteries under different ambient temperatures and fast charging conditions through transfer learning, thereby improving the generalization ability of the model in different application scenarios.

[0008] To achieve the above object, the present invention provides a method for reconstructing the spatiotemporal distribution of temperature of a cylindrical lithium-ion battery under wide temperature range and fast charging conditions, comprising the following steps:

[0009] Step 1, constructing a reconstruction model of the spatiotemporal distribution of the source domain temperature of the lithium battery throughout its life, including a neural network and a thermodynamic model of the lithium battery, wherein the thermodynamic model includes a heat generation model, a heat transfer model, and a heat dissipation model;

[0010] Step 2, training the source domain full life temperature spatiotemporal distribution reconstruction model, including:

[0011] Inputting the BMS monitoring data collected in the source domain into the neural network, so that the neural network outputs the network reconstructed temperature field of the lithium battery and the model parameters of the thermodynamic model;

[0012] Substituting the model parameters and the network reconstruction temperature field into the thermodynamic model, determining whether they conform to the laws of thermodynamics, and calculating the loss function;

[0013] Based on the loss function, iteratively updating the network parameters of the neural network and the model parameters through a back propagation algorithm until the training converges;

[0014] Step 3, based on the trained source domain full-life temperature spatiotemporal distribution reconstruction model, reconstruct the lithium battery temperature field in the source domain temperature scenario to obtain the spatiotemporal change of temperature during the operation of the lithium battery in the source domain;

[0015] Step 4: Using the trained source domain full life temperature spatiotemporal distribution reconstruction model as the target domain full life temperature spatiotemporal distribution reconstruction model, and using the BMS monitoring data collected in the target domain to train the target domain full life temperature spatiotemporal distribution reconstruction model;

[0016] Step 5: Based on the trained target domain full-life temperature spatiotemporal distribution reconstruction model, the lithium battery temperature field is reconstructed in the target domain temperature scenario to obtain the spatiotemporal changes of temperature during the operation of the lithium battery in the target domain.

[0017] Compared with the prior art, the present invention has the following beneficial technical effects:

[0018] 1. The present invention integrates heat generation and thermodynamic laws into neural networks to construct a full-life temperature spatiotemporal distribution reconstruction model with physical significance, which can effectively improve the adaptability and accuracy of the model. It can not only reconstruct the temperature of lithium-ion batteries, but also infer time-varying physical quantities such as thermal conductivity, convection heat transfer coefficient, and open circuit voltage;

[0019] 2. The present invention reconstructs the thermal field of lithium batteries under different ambient temperatures and fast charging conditions through transfer learning, thereby improving the generalization ability of the model in different application scenarios, which is of great significance for thermal abuse warning and improving battery thermal control performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] 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 the structures shown in these drawings without paying creative work.

[0021] Figure 1 This is a flow chart of a method for reconstructing the spatiotemporal distribution of temperature of a cylindrical lithium-ion battery under wide temperature range and fast charging conditions according to an embodiment of the present invention;

[0022] Figure 2 It is a training framework diagram of the full life temperature spatiotemporal distribution reconstruction model in an embodiment of the present invention;

[0023] Figure 3 It is a migration framework diagram of the full life temperature spatiotemporal distribution reconstruction model in an embodiment of the present invention;

[0024] Figure 4 A schematic diagram of a coordinate system in an embodiment of the present invention;

[0025] Figure 5 This is a temperature distribution diagram of the battery during charging in an embodiment of the present invention.

[0026] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0027] 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.

[0028] In addition, the technical solutions between the various embodiments of the present invention can be combined with each other, but it must be based on the fact that ordinary technicians in the field can implement it. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0029] This embodiment discloses a method for reconstructing the spatiotemporal distribution of temperature of a cylindrical lithium-ion battery under wide temperature range and fast charging conditions, which embeds a thermodynamic model that follows the laws of thermodynamics into a neural network architecture to construct a full-life temperature spatiotemporal distribution reconstruction model with physical significance. Among them, the neural network preferably adopts a gated recurrent unit (GRU) neural network, so as to effectively capture the long-term dependencies in historical data by utilizing its powerful time series processing capabilities. At the same time, an attention mechanism method is introduced into the neural network to accurately capture abnormal temperature fluctuations during battery aging, and then dynamically update relevant physical parameters based on measured data, thereby improving the accuracy and adaptability of the model. Finally, in order to enhance the generalization ability of the model under different working conditions, a domain adaptive transfer learning method is introduced to extract domain invariant features from unknown working condition data, so that the model can reconstruct the battery temperature field without retraining, and improve the applicability and robustness of the model under various boundary conditions and initial states.

[0030] The method for reconstructing the spatiotemporal distribution of temperature of cylindrical lithium-ion batteries under wide temperature range fast charging conditions in this embodiment mainly includes two aspects in the specific application process: constructing a full-life temperature spatiotemporal distribution reconstruction model of a physical information neural network and a cross-domain reconstruction of the thermal field of a lithium battery based on . Among them, the pre-training of the full-life temperature spatiotemporal distribution reconstruction model is to establish a mapping from BMS monitoring data to the spatiotemporal distribution thermal field of the battery temperature field in accordance with the physical laws through a GRU neural network and a basic physical model, and to capture long sequence dependencies and aggregation features through an attention mechanism, so that it can use surface and core temperature sensors to achieve accurate reconstruction of the thermal field and learning and updating of related physical parameters. And use limited target domain measurement data to fine-tune some parameters of the model while maintaining its compliance with physical laws and efficient thermal field reconstruction capabilities. Finally, the migration model can achieve high-precision reconstruction of the battery temperature field and real-time updating of dynamic physical parameters in a wide temperature range and under different aging conditions, thereby improving the generalization ability and practical application value of the model.

[0031] refer to Figures 1 to 3 The method for reconstructing the spatiotemporal distribution of temperature of a cylindrical lithium-ion battery under wide temperature range and fast charging conditions in this embodiment specifically includes the following steps:

[0032] Step 1, constructing a reconstruction model of the temperature spatiotemporal distribution of the source domain of the lithium battery throughout its life cycle, including a neural network and a thermodynamic model of the lithium battery;

[0033] Step 2: training the reconstruction model of the spatiotemporal temperature distribution of the source domain over the entire life cycle, including:

[0034] The BMS monitoring data collected in the source domain is input into the neural network. After passing through the gated recurrent unit, attention mechanism module and fully connected layer of the neural network, the network reconstructed temperature field of the lithium battery and the model parameters of the thermodynamic model are output. The BMS monitoring data is specifically multivariate information of the lithium-ion battery during the charging and discharging process under the source domain temperature scenario, including but not limited to voltage, current, position, time, SOC, charging capacity and ambient temperature.

[0035] Substitute the model parameters and network reconstruction temperature field into the thermodynamic model to determine whether the prediction results of the neural network conform to the physical laws of thermodynamics, and calculate the loss function based on the error;

[0036] Based on the loss function, the network parameters and model parameters of the neural network are iteratively updated through the back propagation algorithm until the training converges;

[0037] Step 3: Based on the trained source domain full-life temperature spatiotemporal distribution reconstruction model, the lithium battery temperature field is reconstructed under the source domain temperature scenario to obtain the spatiotemporal changes of temperature during the operation of the lithium battery under the source domain;

[0038] Step 4: Use the trained spatiotemporal distribution reconstruction model of the full life temperature in the source domain as the spatiotemporal distribution reconstruction model of the full life temperature in the target domain, and use the BMS monitoring data collected in the target domain to train the spatiotemporal distribution reconstruction model of the full life temperature in the target domain, wherein the composition of the BMS monitoring data collected in the target domain is the same as that in the source domain. In specific applications, the spatiotemporal distribution reconstruction model of the full life temperature in the target domain can also be re-established, and the model parameters and network parameters of the trained spatiotemporal distribution reconstruction model of the full life temperature in the source domain are used as the initial values ​​of the spatiotemporal distribution reconstruction model of the full life temperature in the target domain;

[0039] Step 5: Based on the trained target domain full-life temperature spatiotemporal distribution reconstruction model, the lithium battery temperature field is reconstructed under the target domain temperature scenario to obtain the spatiotemporal changes of temperature during the operation of the lithium battery in the target domain.

[0040] During the charge and discharge process of a lithium-ion battery, heat generation, heat transfer and heat dissipation are the main factors affecting the battery temperature. Therefore, this embodiment infers the temporal and spatial changes of the battery by constructing a heat generation model, a heat transfer model and a heat dissipation model of lithium ions.

[0041] For the heat generation model in the thermal model, this embodiment simulates the uneven heat generation of the battery by calculating the heat generation through the equivalent internal resistance of the battery, specifically:

[0042] (1)

[0043] The first term on the right side of the equation describes the irreversible heat generated by the internal resistance dissipation, and the second term is the reversible entropy heat related to the battery reaction. is the rate of heat generation or consumption, is the battery voltage, is the battery current, is the battery temperature, is the entropy thermal coefficient, is the open circuit voltage, which can be obtained experimentally.

[0044] For heat transfer models in thermal models, refer to Figure 4 , since the cylindrical battery is an axisymmetric figure, the battery heat transfer model is composed of a two-dimensional transient cylinder heat transfer model, namely:

[0045] (2)

[0046] in, is the battery density, is the specific heat capacity of the battery, is the axial heat transfer coefficient, is the heat transfer coefficient, , are the axial and radial coordinates of the battery in two-dimensional space, is the battery volume;

[0047] The heat dissipation model is:

[0048] (3)

[0049] in, is the ambient temperature of the battery, is the convective heat transfer coefficient.

[0050] The temporal and spatial distribution of the temperature of a lithium-ion battery depends on the amount of electrical energy and environmental parameters. In this embodiment, the battery BMS monitoring information is selected as the input feature of the physical information neural network, namely:

[0051] (4)

[0052] in, Represents the location information of the battery in two-dimensional space, τ Indicates the monitoring information obtained from the BMS system. t For test time, I Represents the battery current, U Indicates the battery voltage, Indicates the ambient temperature of the battery.

[0053] In addition, it can be seen from the evolution mechanism of the thermal field that temperature rise is a process of cumulative gradual evolution over time. The historical information of the battery thermal extension process is obtained through a gated recurrent neural network. And the attention mechanism is used to capture subtle changes caused by battery aging, operating conditions and other factors, thereby improving the battery's reconstruction ability over the entire life cycle. In the process of battery temperature field reconstruction, transient thermal behavior is not only related to performance parameters such as specific heat capacity, thermal conductivity, and density, but also closely related to experimental conditions such as convection coefficient and ambient temperature. When the environment changes, it is often necessary to calibrate the relevant parameters. Therefore, this embodiment simultaneously optimizes the unknown model parameters in the thermodynamic model and the network parameters of the neural network, so that the full-life temperature spatiotemporal distribution reconstruction model has the ability to identify unknown values ​​in the physical model. Experimental tests have shown that during the entire life cycle of the battery, the battery's specific heat capacity changes little, while the thermal conductivity changes greatly, which significantly affects the accuracy of temperature field reconstruction. Therefore, the model parameters in this embodiment include , , ,Right now:

[0054] (5)

[0055] in, are network parameters, are model parameters, is the loss function of the reconstruction model of the spatiotemporal distribution of temperature over the entire life cycle, is the optimal network parameter, are the optimal model parameters.

[0056] The heat dissipation equation constraint can be written as:

[0057] (6)

[0058] in, is the boundary of the computational domain, is the boundary loss function, and its specific calculation formula is:

[0059] (7)

[0060] in, is the location of the sampling point, is the temperature value corresponding to the sampling point in the network reconstructed temperature field, i ={ z =0, z = z 0, r = r 0}.

[0061] In the PINN (Physics-Informed Neural Networks) framework of this embodiment, the temperature is approximated using the GRU neural network, where the input is and τ , the output is the approximate solution of temperature, and the loss function of PINN is defined as:

[0062] (8)

[0063] in, is the boundary condition loss term, is the heat conduction residual loss term, , is the weight coefficient.

[0064] In the process of training the source domain full life time and space distribution reconstruction model, this embodiment selects sampling points in the boundary and time and space domains, and each loss term in the loss function is calculated according to its respective sampling point in each training iteration. Specifically:

[0065] The boundary condition loss term is:

[0066] (9)

[0067] in, is the number of sampling points of the boundary condition loss term, For the border k The axial and radial coordinates of the sampling points in two-dimensional space, For the border k BMS monitoring data corresponding to each sampling point, For the k The boundary constraint loss function of the sampling points is is the convective heat transfer coefficient corresponding to the sampling point, is the ambient temperature corresponding to the sampling point;

[0068] The heat conduction residual loss term is:

[0069] (10)

[0070] in, is the number of sampling points of the heat conduction residual loss term, is the heat conduction residual loss term k The axial and radial coordinates of the sampling points in two-dimensional space, is the heat conduction residual loss term k BMS monitoring data corresponding to each sampling point, is the heat conduction residual loss term k The temperature value corresponding to each sampling point in the network reconstructed temperature field, is the heat conduction residual loss term k The heat conduction constraint loss function of the sampling points.

[0071] In the lifetime temperature spatiotemporal distribution reconstruction model of this embodiment, the input is n The hidden state is obtained by layer GRU, and the multi-head attention mechanism projects the hidden state into the linear space, decodes and weights the information of the long-time step sequence input, and then obtains the two-dimensional output tensor through full connection. The surface temperature of the battery is used as the observation data to optimize the model parameters and network parameters. The physical battery model is encoded into the network structure by adding the residual of the physical equation to the loss function. The derivation of the output relative to the input is achieved by automatic differentiation to meet the physical requirements. In the back propagation, the LBFGS optimizer is used to minimize the total loss.

[0072] Furthermore, when additional labeled data are available, a data correction loss term can be added to the total loss of the full-life temperature spatiotemporal distribution reconstruction model, i.e.:

[0073] (11)

[0074] in, is the data correction loss term, is the weight coefficient;

[0075] The data correction loss items are as follows:

[0076] (12)

[0077] in, is the number of sampling points for the data correction loss term, The data correction loss item k The axial and radial coordinates of the sampling points in two-dimensional space, The data correction loss item k BMS monitoring data corresponding to each sampling point, The data correction loss item k The temperature value corresponding to each sampling point in the network reconstructed temperature field, The data correction loss item k The actual temperature value measured by the temperature sensor corresponding to the sampling point.

[0078] Due to the nature of PINN, both physical laws and data are soft constraints of the model, which means that the loss will usually decrease during training, but will not be completely zero. Since the constraints are not strictly met, the labeled data can be arbitrarily merged with the physical laws without causing over-constraint problems. Adding data-based losses can be used as auxiliary data to guide the training of the full-life temperature spatiotemporal distribution reconstruction model, thereby accelerating training, and can also identify unknown parameters and discover physical laws. However, it is extremely difficult to measure the internal temperature of the battery body without destroying its own structure. Therefore, the internal heat conduction information contained in the battery measurement data is extremely limited, which increases the difficulty of estimating related parameters. Therefore, constraints that meet physical phenomena are introduced to improve the accuracy of temperature field reconstruction and parameter estimation. According to actual measurement data, it can be seen that under natural convection heat dissipation, the temperature change from the center to the surface of the battery satisfies the known convex trend, and during the charging process, the temperature gradually decreases from the positive pole to the negative pole in the axial direction. Such as Figure 5 As shown, this embodiment further adds a monotonicity loss term to the loss function of the full life temperature spatiotemporal distribution reconstruction model, namely:

[0079] (13)

[0080] in, is the monotonicity loss term, is the weight coefficient;

[0081] The monotonicity loss term is:

[0082] (14)

[0083] (15)

[0084] (16)

[0085] in, is the radial heat conduction monotonicity loss function, is the monotonic loss function of axial heat conduction, is the number of sampling points for radial heat conduction monotonicity constraint, The radial heat conduction monotonicity constraint k The axial and radial coordinates of the sampling points in two-dimensional space, The radial heat conduction monotonicity constraint k BMS monitoring data corresponding to each sampling point, The radial heat conduction monotonicity constraint k The temperature value corresponding to each sampling point in the network reconstructed temperature field, is the number of sampling points for the monotonicity constraint of axial heat conduction, The monotonicity constraint for axial heat conduction is k The axial and axial coordinates of the sampling points in two-dimensional space, The monotonicity constraint for axial heat conduction is k BMS monitoring data corresponding to each sampling point, The monotonicity constraint for axial heat conduction is k The temperature value corresponding to each sampling point in the network reconstructed temperature field.

[0086] It is worth noting that this embodiment includes multiple target losses when jointly estimating the temperature field and thermal parameters, so a multi-task learning method is used to weigh the losses of each individual task. The weights in the loss function directly determine the speed and degree of convergence. The losses of the control equations, constraints, and data are trained synchronously, which means that they share the same gradient descent strategy. Therefore, it is difficult to find a set of weights that best suit all loss terms. In order to maintain a balance between the losses of the control equations, constraints, and sensor data, an adaptive balancing algorithm is used to make these weight coefficients trainable. After giving the initial value of each weight coefficient, the weight coefficients in the loss function are adaptively balanced at each training iteration, specifically:

[0087] (17)

[0088] in, To iterate to Step 1 The weight coefficient of the loss, To iterate to Step 1 Item loss, To iterate to Step 1 Item loss, is the number of loss items.

[0089] The above description is only a preferred embodiment of the present invention, and does not limit the protection scope of the present invention. All equivalent structural changes made by using the contents of the present invention specification and drawings under the inventive concept of the present invention, or directly / indirectly applied in other related technical fields are included in the protection scope of the present invention.

Claims

1. A method for reconstructing the spatiotemporal distribution of temperature of a cylindrical lithium-ion battery under wide temperature range and fast charging conditions, characterized in that: The steps include: Step 1, constructing a reconstruction model of the spatiotemporal distribution of the source domain temperature of the lithium battery throughout its life, including a neural network and a thermodynamic model of the lithium battery, wherein the thermodynamic model includes a heat generation model, a heat transfer model, and a heat dissipation model; Step 2, training the source domain full life temperature spatiotemporal distribution reconstruction model, including: Inputting the BMS monitoring data collected in the source domain into the neural network, so that the neural network outputs the network reconstructed temperature field of the lithium battery and the model parameters of the thermodynamic model; Substituting the model parameters and the network reconstruction temperature field into the thermodynamic model, determining whether they conform to the laws of thermodynamics, and calculating the loss function; Based on the loss function, iteratively updating the network parameters of the neural network and the model parameters through a back propagation algorithm until the training converges; Step 3, based on the trained source domain full-life temperature spatiotemporal distribution reconstruction model, reconstruct the lithium battery temperature field in the source domain temperature scenario to obtain the spatiotemporal change of temperature during the operation of the lithium battery in the source domain; Step 4: Using the trained source domain full life temperature spatiotemporal distribution reconstruction model as the target domain full life temperature spatiotemporal distribution reconstruction model, and using the BMS monitoring data collected in the target domain to train the target domain full life temperature spatiotemporal distribution reconstruction model; Step 5: Based on the trained target domain full-life temperature spatiotemporal distribution reconstruction model, the lithium battery temperature field is reconstructed in the target domain temperature scenario to obtain the spatiotemporal changes of temperature during the operation of the lithium battery in the target domain.

2. The method for reconstructing the spatiotemporal distribution of temperature of a cylindrical lithium-ion battery under wide temperature range and fast charging conditions according to claim 1, characterized in that: The neural network is a gated recurrent unit neural network with an attention mechanism.

3. The method for reconstructing the spatiotemporal distribution of temperature of a cylindrical lithium-ion battery under wide temperature range and fast charging conditions according to claim 1, characterized in that: In step 1, the heat generation model is: in, is the rate of heat generation or consumption, is the battery current, is the battery voltage, is the open circuit voltage, is the battery temperature, is the entropy thermal coefficient; The heat transfer model is: in, is the battery density, is the specific heat capacity of the battery, is the axial heat transfer coefficient, is the radial heat transfer coefficient, , are the axial and radial coordinates of the battery in two-dimensional space, is the battery volume; The heat dissipation model is: in, is the ambient temperature of the battery, is the convective heat transfer coefficient; The model parameters of the thermodynamic model include , , .

4. The method for reconstructing the spatiotemporal distribution of temperature of a cylindrical lithium-ion battery under wide temperature range and fast charging conditions according to claim 1, 2 or 3, characterized in that: The loss function is: in, is the loss function, is the boundary condition loss term, is the heat conduction residual loss term, is the data correction loss term, is the monotonicity loss term, , , , is the weight coefficient.

5. The method for reconstructing the spatiotemporal distribution of temperature of a cylindrical lithium-ion battery under wide temperature range and fast charging conditions according to claim 4, characterized in that: In the process of training the source domain full life temperature spatiotemporal distribution reconstruction model, selecting sampling points in the boundary and spatiotemporal domains; At each training iteration, each loss term in the loss function is calculated based on its respective sampling points, specifically: The boundary condition loss term is: in, is the number of sampling points of the boundary condition loss term, For the border k The axial and radial coordinates of the sampling points in two-dimensional space, For the border k BMS monitoring data corresponding to each sampling point, For the k The boundary constraint loss function of the sampling points is is the convective heat transfer coefficient corresponding to the sampling point, is the ambient temperature corresponding to the sampling point; The heat conduction residual loss term is: in, is the number of sampling points of the heat conduction residual loss term, is the heat conduction residual loss term k The axial and radial coordinates of the sampling points in two-dimensional space, is the heat conduction residual loss term k BMS monitoring data corresponding to each sampling point, is the heat conduction residual loss term k The temperature value corresponding to each sampling point in the network reconstructed temperature field, is the heat conduction residual loss term k Heat conduction constraint loss function for each sampling point; The data correction loss term is: in, is the number of sampling points for the data correction loss term, The data correction loss item k The axial and radial coordinates of the sampling points in two-dimensional space, The data correction loss item k BMS monitoring data corresponding to each sampling point, The data correction loss item k The temperature value corresponding to each sampling point in the network reconstructed temperature field, The data correction loss item k The actual temperature value measured by the temperature sensor corresponding to each sampling point; The monotonicity loss term is: in, is the radial heat conduction monotonicity loss function, is the monotonic loss function of axial heat conduction.

6. The method for reconstructing the spatiotemporal distribution of temperature of a cylindrical lithium-ion battery under wide temperature range and fast charging conditions according to claim 5, characterized in that: In the monotonicity loss term: in, is the number of sampling points for the radial heat conduction monotonicity constraint, The radial heat conduction monotonicity constraint k The axial and radial coordinates of the sampling points in two-dimensional space, The radial heat conduction monotonicity constraint k BMS monitoring data corresponding to each sampling point, The radial heat conduction monotonicity constraint k The temperature value corresponding to each sampling point in the network reconstructed temperature field, is the number of sampling points for the monotonicity constraint of axial heat conduction, The monotonicity constraint for axial heat conduction is k The axial and axial coordinates of the sampling points in two-dimensional space, The monotonicity constraint for axial heat conduction is k BMS monitoring data corresponding to each sampling point, The monotonicity constraint for axial heat conduction is k The temperature value corresponding to each sampling point in the network reconstructed temperature field.

7. The method for reconstructing the spatiotemporal distribution of temperature of a cylindrical lithium-ion battery under wide temperature range and fast charging conditions according to claim 4, characterized in that: In each training iteration, each weight coefficient in the loss function is adaptively balanced, specifically: in, To iterate to Step 1 The weight coefficient of the loss, To iterate to Step 1 Item loss, To iterate to Step 1 Item loss, is the number of loss items.

Images