A battery temperature prediction method and system based on neural ordinary differential equations

Abstract

The application provides a battery temperature prediction method and system based on a neural ordinary differential equation, and the method comprises the following steps: acquiring multi-source operation data in the operation process of a battery system; performing time interpolation processing on the data, uniformly mapping data with different sampling frequencies to a continuous time dimension, and obtaining continuous time input features; constructing global features and discriminative features based on the input features, wherein the discriminative features are used to represent key thermal influence factors such as battery heat generation and cooling medium flow; fusing the global features, the discriminative features and the current temperature state of the battery to construct comprehensive features; inputting the comprehensive features into a thermal dynamics model based on a neural ordinary differential equation, solving the continuous evolution process of the battery temperature through numerical integration, and realizing single-cell temperature prediction. The application can not only follow the continuous evolution law of thermodynamics, but also take into account the spatial coupling characteristics and multi-working-condition robustness.

Description

Technical Field

[0001] This invention relates to the field of battery temperature prediction technology, and in particular to a battery temperature prediction method and system based on the neural ordinary differential equation. Background Technology

[0002] Battery systems undergo complex thermophysical evolution during charging and discharging, and exhibit strong spatial thermal coupling between individual cells. Accurate temperature prediction is the cornerstone of battery management systems for achieving thermal runaway early warning and lifetime management. Currently, battery thermal modeling mainly follows two approaches:

[0003] One type is analytical or numerical models based on physical equations, such as finite element methods and lumped parameter models. These models have strong physical interpretability, but they are highly dependent on the accurate calibration of key parameters such as heat capacity and thermal resistance. Under the complex and variable operating conditions in actual operation, the nonlinear drift of physical parameters makes it difficult for the model to maintain high accuracy over a long period of time, and the large-scale consumption of computing resources limits its online application.

[0004] Second, there are data-driven statistical or machine learning models, such as LSTM and CNN. These models are good at capturing nonlinear features, but they are essentially black-box mappings of time steps, lacking the characterization of continuous thermodynamic evolution. When faced with abrupt changes in input signals or non-uniform sampling, the robustness and temporal continuity of the models are poor.

[0005] In recent years, the emergence of Neural ODEs has provided a new approach to continuous-time dynamics modeling. Neural ODEs, by fitting vector fields with neural networks, transform the hierarchical structure of deep learning into a continuous time evolution process, theoretically better aligning with the physical essence of battery thermodynamics. However, directly applying Neural ODEs to model power battery systems still faces the following challenges:

[0006] High-dimensional state characterization is difficult: the thermal state space of large-scale battery arrays has extremely high dimensionality, resulting in a heavy computational burden for solving ordinary differential equations.

[0007] Lack of spatiotemporal coupling mechanism: Existing models are unable to take into account both the continuous evolution in the time dimension and the complex heat conduction coupling in the spatial dimension;

[0008] Insufficient generalization and stability: Under a wide range of operating conditions such as dynamic current and variable environment, the model is prone to unstable time interpolation and prediction drift.

[0009] Therefore, there is an urgent need for a battery temperature prediction method and system based on the neural ordinary differential equation, which can both follow the continuous evolution of thermodynamics and take into account the spatial coupling characteristics and multi-condition robustness. Summary of the Invention

[0010] To address the aforementioned shortcomings of existing technologies, the purpose of this invention is to provide a battery temperature prediction method and system based on Neural ODEs, aiming to solve the technical problems of complex high-dimensional state solutions, inaccurate spatiotemporal coupling characterization, and poor generalization across multiple operating conditions when Neural ODEs are applied to battery thermal modeling.

[0011] To achieve the above objectives, in a first aspect, the present invention provides a battery temperature prediction method based on the neural ordinary differential equation, the steps of which include:

[0012] S1. Acquire multi-source operational data during the operation of the battery system;

[0013] S2. Perform time interpolation processing on the multi-source operating data to uniformly map data with different sampling frequencies and unequal time intervals to the continuous time dimension to obtain continuous time input features;

[0014] S3. Based on the continuous time input features, construct global features and discriminant features respectively, wherein the discriminant features are used at least to characterize the thermal influencing factors corresponding to the battery heat generation and the cooling medium flow rate;

[0015] S4. The global features, discrimination features and the current temperature state of the battery are fused together to construct a comprehensive feature for describing the thermodynamic state of the battery;

[0016] S5. Input the comprehensive features into the thermodynamic model constructed based on the neural network differential equation, and solve the continuous evolution process of battery temperature over time by numerical integration to obtain the cell-by-cell temperature prediction results of the battery system at the target time.

[0017] As a further improvement to the above scheme, the multi-source operating data includes at least sampling time, ambient temperature, heat generation of individual battery cells, battery operating current, cooling medium flow rate, and cooling medium inlet temperature.

[0018] As a further improvement to the above scheme, the thermodynamic model uses a neural network to parameterize the change function of battery temperature state with respect to time, so as to form a neural ordinary differential equation model.

[0019] As a further improvement to the above scheme, the time interpolation process uses an interpolation algorithm to map the original discrete sampling data into a continuous time series, so as to eliminate the impact of inconsistent sampling periods of different sensors on the modeling accuracy.

[0020] As a further improvement to the above scheme, the thermodynamic model includes a heating model and a heat dissipation model;

[0021] The heating model is used to describe the internal heating effect of the battery, while the heat dissipation model is used to describe the influence of external cooling conditions.

[0022] As a further improvement to the above scheme, the outputs of the heating model and the heat dissipation model are fused by weighting parameters to characterize the temperature change trend under different operating conditions.

[0023] Specifically, the temperature prediction formula is as follows:

[0024] ;

[0025] in, The temperature prediction output represents the predicted temperature distribution of all cells in the battery pack at a specific moment. For dynamic weight parameters, ∈[0,1], automatically learned through a neural network; This is a heating function used to simulate the heating effect generated by the heat source inside the battery. This is a heat dissipation function used to simulate the heat loss of the battery system to the external environment; The comprehensive features are the input.

[0026] As a further improvement to the above scheme, the neural ordinary differential equation model is solved by a numerical integration algorithm to obtain the continuous change of battery temperature over time.

[0027] Preferably, the fourth-order Runge-Kutta method (RK4) is used for the numerical integration of Neural ODE; during the training phase, the adjoint sensitivity method is used to calculate the gradient to reduce memory usage and support long sequence modeling.

[0028] Secondly, the present invention also provides a battery temperature prediction system based on the neural ordinary differential equation, employing the method provided in the first aspect, the system comprising:

[0029] The data acquisition module is used to collect multi-source operating data of the battery system;

[0030] The data preprocessing module is used to perform time interpolation processing on the multi-source operating data to construct continuous time input features;

[0031] The feature extraction module includes a global feature extraction submodule and a discriminative feature extraction submodule, which are used to generate feature vectors that characterize the thermal state of the battery;

[0032] The thermodynamic modeling module is used to construct a battery thermodynamic model based on the neural network differential equation and to make continuous-time predictions of battery temperature.

[0033] The output module is used to output the predicted temperature of each cell in the battery system at the target time.

[0034] As a further improvement to the above scheme, the thermodynamic modeling module includes a heating module, a heat dissipation module, and a fusion module, which are used to model the internal heating effect and the external cooling effect of the battery respectively and perform weighted fusion.

[0035] As a further improvement to the above scheme, a model training module is also included, which is used to pre-train the battery thermodynamic model based on the neural network differential equation using historical data or simulation data.

[0036] Because the present invention adopts the above technical solutions, the beneficial effects of this application are as follows:

[0037] This invention provides a battery temperature prediction method based on the neural network constant differential equation (NCE). First, it models the thermodynamic process using a data-driven approach, reducing reliance on explicit thermal parameter modeling. Specifically, it parametrically models the battery thermodynamic process using NCE, achieving temperature prediction without explicit modeling of physical parameters such as heat capacity and thermal resistance. This reduces model construction complexity and improves applicability in practical engineering scenarios. Second, this invention improves the model's adaptability under different operating conditions. Specifically, because this invention employs continuous-time modeling and combines time interpolation with feature separation and fusion structures, the model can naturally reflect the thermal behavior characteristics under different operating conditions by reflecting changes in input features when operating conditions change. This alleviates, to some extent, the problem of insufficient generalization ability under multiple operating conditions in existing technologies.

[0038] Furthermore, in step S2, time interpolation is performed on multi-source operating data with different sampling frequencies and inconsistent time intervals. This ensures that various input parameters participate in modeling under the same continuous time axis, thereby avoiding modeling biases introduced by asynchronous sampling times and improving the continuity and stability of the description of battery temperature change processes. Time interpolation unifies data with different sampling frequencies to a continuous time dimension, eliminating the accumulated errors caused by inconsistent sampling in traditional discrete models. Combining the continuous-time modeling characteristics of neural network differential equations, the solution of high-dimensional temperature states is transformed into a numerical integration problem of a continuous dynamic system, avoiding the computational complexity of independent calculations for each time step in traditional methods.

[0039] In step S3, global features and discriminant features are constructed separately. The discriminant features focus on modeling parameters highly correlated with thermal behavior, such as battery heat generation and cooling medium flow rate. This helps to strengthen the characterization of key thermal factors under high-dimensional input conditions and prevents important thermal information from being weakened during the overall feature modeling process. By constructing global and discriminant features separately, the coupling relationship between multiple parameters is preserved while strengthening the specific characterization of key thermal parameters. The feature fusion mechanism combines prior physical knowledge with data-driven modeling, effectively capturing the spatial thermal coupling relationship between individual battery cells. In step S4, global features, discriminant features, and the current temperature state of the battery are combined to construct a comprehensive feature, enabling the model to consider both changes in operating conditions and historical state information when making temperature predictions, thus helping to maintain the temporal consistency of the temperature evolution process. In step S5, a neural network constant differential equation is used to describe the relationship between battery temperature and time, and the temperature trajectory is solved through numerical integration. Compared with traditional discrete-time prediction models, this can directly obtain temperature prediction results at any time point, making it more suitable for describing the continuous change characteristics of battery thermal state under complex operating conditions. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0041] Figure 1 This is a flowchart illustrating a battery temperature prediction method based on neural ordinary differential equations disclosed in Embodiment 1 of the present invention.

[0042] Figure 2 This is a partial structural schematic diagram of a battery temperature prediction system based on the neural ordinary differential equation disclosed in Embodiment 2 of the present invention.

[0043] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0044] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0045] It should be noted that the technical solutions of the various embodiments of the present invention can be combined with each other, but only if they are based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0046] Example 1

[0047] See Figure 1 This invention provides a battery temperature prediction method based on the neural ordinary differential equation. It is designed to address the temperature prediction requirements of battery systems during actual operation and to solve the problem of unstable temperature prediction accuracy caused by inconsistent sampling, complex high-dimensional states, and insufficient spatiotemporal coupling in existing technologies.

[0048] Specifically, the steps of the battery temperature prediction method include:

[0049] S1. Acquisition of multi-source operational data:

[0050] First, acquire multi-source operational data of the battery system during operation. This multi-source operational data includes, but is not limited to, parameters such as sampling time, ambient temperature, heat generation of individual battery cells, battery operating current, cooling medium flow rate, and cooling medium inlet temperature.

[0051] By simultaneously collecting operational data reflecting both the internal heating characteristics of the battery and the external cooling conditions, the subsequent temperature prediction process can simultaneously consider changes in the battery's own heat source and changes in heat dissipation conditions, providing a relatively complete data foundation for thermodynamic modeling.

[0052] S2. Temporal interpolation processing of multi-source data:

[0053] Because the aforementioned multi-source operational data often exhibits different sampling frequencies and inconsistent sampling times during actual acquisition, directly using it for modeling can easily introduce time alignment errors. In this embodiment, time interpolation processing is performed on the acquired multi-source operational data to uniformly map data with different sampling frequencies and unequal time intervals to a continuous time dimension, thereby obtaining continuous-time input features.

[0054] By using time interpolation, various operating parameters can be modeled on the same time scale, which helps to reduce modeling bias caused by asynchronous sampling, thereby improving the continuity and stability of the description of battery temperature change process.

[0055] S3. Construction of global and discriminative features:

[0056] After completing the time interpolation process, global features and discriminant features are constructed based on the continuous-time input features. The global features are used to characterize the overall trend of multi-source operational data over continuous time; the discriminant features are used to characterize at least the thermal influencing factors corresponding to battery heat generation and cooling medium flow rate.

[0057] Since battery heat generation and cooling medium flow rate are highly correlated with temperature changes, targeted feature modeling of the above parameters helps to enhance the characterization of key thermal influencing factors under high-dimensional input conditions and avoids the weakening of important thermal information in the overall feature modeling process.

[0058] S4. Construction of comprehensive features:

[0059] In this embodiment, the constructed global features and discrimination features are fused with the current temperature state of the battery to form a comprehensive feature for describing the thermodynamic state of the battery.

[0060] By incorporating the current temperature state of the battery during the feature construction stage, the model can simultaneously consider changes in operating conditions and historical temperature state information when predicting subsequent temperature changes, which helps maintain the continuity and consistency of the battery temperature evolution process in the time dimension.

[0061] S5. Temperature prediction based on the constant differential equation:

[0062] The comprehensive features are input into a thermodynamic model based on the constant differential equations of the neutron, and the continuous evolution of battery temperature over time is solved by numerical integration to obtain the cell-by-cell temperature prediction results of the battery system at the target time.

[0063] Compared to traditional discrete-time prediction models, this modeling approach directly describes the continuous change of temperature state with respect to time through a neural ordinary differential equation. It can predict the battery temperature at any point in time without relying on explicit physical parameters such as thermal capacity and thermal resistance for modeling, making it more suitable for describing the continuous change characteristics of battery thermal state under complex operating conditions.

[0064] In a preferred embodiment, the thermodynamic model uses a neural network to parameterize the change function of battery temperature state with respect to time, thereby constructing a neural ordinary differential equation model to describe the continuous evolution of battery temperature over time.

[0065] During the operation of a battery system, the temperature change of a single cell conforms to the law of conservation of energy, and its thermal equilibrium relationship can be expressed as:

[0066] ;

[0067] in, Indicates the battery at time temperature state, The equivalent heat capacity of the battery. This indicates the heat generated by the battery during operation. This represents the heat lost through convection, conduction, and radiation. In traditional thermal models, parameters such as heat capacity and thermal resistance usually need to be obtained through experimental calibration, which is difficult to determine accurately under complex operating conditions or in large-scale battery systems.

[0068] To avoid explicit modeling of the aforementioned physical parameters, this embodiment introduces a neural network to parameterize the battery temperature change rate. Specifically, the battery system at time... The temperature state of each individual cell is represented as a state vector:

[0069] ;

[0070] in, This represents the number of individual battery cells. The input variables reflecting operating conditions are represented as follows:

[0071] ;

[0072] These correspond to sampling time, ambient temperature, heat generation of individual battery cells, battery operating current, cooling medium flow rate, and cooling medium inlet temperature, respectively.

[0073] Based on this, the temperature state as a function of time is parameterized using a neural network, and a neural ordinary differential equation model of the following form is constructed:

[0074] ;

[0075] in, Let the continuous-time dynamic function be represented by a neural network. These are the learnable parameters of the network. Through this method, the rate of change of battery temperature no longer depends on explicitly given heat capacity or thermal resistance parameters, but is automatically learned by the neural network during training based on historical operating data.

[0076] In some preferred embodiments, the neural network is further divided into a heating subnetwork for describing the internal heating effect of the battery and a heat dissipation subnetwork for describing the influence of external cooling conditions, thereby expressing the above-mentioned neural network ordinary differential equation as:

[0077] ;

[0078] in, Used to depict the temperature change trend caused by internal battery heating. Used to characterize the effects of factors such as cooling medium flow rate, inlet temperature, and environmental conditions on temperature changes; and The weight parameters are learned by the network and are used to adjust the relative contributions of heating and heat dissipation effects to temperature changes under different operating conditions.

[0079] In the model inference stage, the above-mentioned ordinary differential equation is solved by numerical integration to obtain the evolution trajectory of battery temperature in any time interval. The prediction process can be expressed as follows:

[0080] ;

[0081] By employing the aforementioned modeling method based on neural network ordinary differential equations, this embodiment can describe the dynamic changes in battery temperature over continuous time, enabling the model to maintain good temporal consistency even when faced with unequal time sampling and varying operating conditions. Furthermore, by implicitly mapping complex thermal mechanisms into the neural network parameters, it helps reduce the reliance on explicit thermal parameter modeling and improves the model's engineering applicability in large-scale battery systems.

[0082] In a preferred embodiment, the time interpolation process uses an interpolation algorithm to map the original discrete sampling data into a continuous time series, so as to eliminate the influence of inconsistent sampling periods of different sensors on the accuracy of thermodynamic modeling.

[0083] In actual battery system operation, different operating parameters are typically collected by different types of sensors, and the sampling periods and sampling times of each sensor are often inconsistent. For example, parameters such as battery current and cooling medium flow rate may be sampled at a higher frequency, while parameters such as ambient temperature and cooling medium inlet temperature may be sampled at a relatively lower frequency. At the same time, the sampling time points of various sensors may also have time offsets. If the above discrete sampling data is directly used as model input, it is easy to introduce discontinuities or misalignments in the time dimension, thereby affecting the stability and consistency of temperature prediction.

[0084] Therefore, before proceeding with subsequent feature construction and thermodynamic modeling, time interpolation is performed on the multi-source discrete sampling data. Specifically, let a certain operating parameter at discrete sampling times be:

[0085] ;

[0086] The corresponding sample value is:

[0087] ;

[0088] By using an interpolation algorithm, the discrete sequence is mapped to a continuous function defined over a continuous time interval. .

[0089] In one implementation, the interpolation process can employ linear interpolation, the expression of which is:

[0090] ;

[0091] Using the above method, it is possible to achieve any target time point. The estimated value of the operating parameter is obtained. For other operating parameters, the corresponding interpolation method is used to process them, thereby mapping all types of original discrete sampling data onto the same continuous time axis, forming a continuous-time input feature sequence. Through the above time interpolation processing, this implementation method can effectively reduce the impact of inconsistent sampling periods of multiple sensors on the modeling process, which is beneficial to improving the ability to maintain time continuity in battery temperature prediction, and thus enhancing the stability and engineering applicability of the thermodynamic model under complex operating conditions.

[0092] In a preferred embodiment, the thermodynamic model further includes a heating model and a heat dissipation model, wherein the heating model is used to describe the internal heating effect of the battery, and the heat dissipation model is used to describe the influence of external cooling conditions on the battery temperature change.

[0093] During battery system operation, battery temperature changes are typically determined by two types of thermal effects: internal heat generation and external heat dissipation. Internal heat generation mainly originates from factors such as electrochemical reaction heat and ohmic heat, while external heat dissipation is closely related to factors such as cooling medium flow rate, cooling medium inlet temperature, and environmental conditions. If these different thermal effects are modeled uniformly, it is easy to mask the relative impact of various thermal effects under different operating conditions, which is not conducive to accurately depicting temperature change trends.

[0094] Therefore, in this embodiment, the thermodynamic model is divided into a heating model and a heat dissipation model. The heating model is used to model the temperature rise effect caused by the internal heat source of the battery, and the heat dissipation model is used to model the temperature drop effect caused by the battery system dissipating heat to the external environment. By distinguishing the internal heating mechanism and the external cooling mechanism at the model structure level, it is beneficial to separately characterize the contribution of different thermal effects to temperature changes.

[0095] Based on this, the dynamic function in the neural network's constant differential equation can be further expressed as a weighted combination of the heating and cooling models. The temperature change rate calculated using this weighted combination is then used to obtain the temperature prediction result at the corresponding time point through the numerical integration process of the neural network's constant differential equation, thus characterizing the combined effect of temperature change trends under different operating conditions. Specifically, the predicted output of the battery temperature can be expressed as:

[0096] ;

[0097] in, The temperature prediction output represents the predicted temperature distribution of all cells in the battery pack at a specific moment. For dynamic weight parameters, It is automatically learned through a neural network and is used to adjust the relative weights of heating and heat dissipation effects under different operating conditions. This is a heating function used to simulate the heating effect generated by the heat source inside the battery; This is a heat dissipation function used to simulate the heat loss of the battery system to the external environment; The input is a comprehensive feature, which consists of operating condition features and the current temperature state of the battery.

[0098] By introducing dynamic weight parameters to fuse the heating and heat dissipation models, the weight of the heating model can be increased when internal battery heating is dominant, and the weight of the heat dissipation model can be increased when changes in cooling conditions have a more significant impact on temperature. This structure enables the model to adaptively adjust the contribution ratio of different thermal effects according to changes in operating conditions, thereby more reasonably describing the temperature change process.

[0099] In a preferred embodiment, the neural ordinary differential equation model is solved by a numerical integration algorithm to obtain the continuous evolution result of battery temperature over time.

[0100] In the process of constructing a battery thermodynamic model based on the constant differential equation, the relationship between the battery temperature state and time can be expressed as:

[0101] ;

[0102] in, Indicates the battery system at time The temperature state vector of each individual unit. This indicates the operating condition input. Let be the temperature change rate function parameterized by the neural network. To obtain the trajectory of battery temperature change over a given time interval, the above neural network ordinary differential equation needs to be solved numerically.

[0103] In this embodiment, the fourth-order Runge-Kutta method (RK4) is preferably used for numerical integration of the neural ordinary differential equation. Let the initial value problem be expressed as:

[0104] ;

[0105] Then at a time step of Under the given conditions, the iterative process of RK4 can be expressed as:

[0106] ;

[0107] ;

[0108] ;

[0109] ;

[0110] The battery temperature status update result at the next time step is as follows:

[0111] ;

[0112] Through the numerical integration process described above, the evolution trajectory of battery temperature over time can be gradually obtained in a continuous time dimension. Compared with low-order numerical integration methods, the fourth-order Runge-Kutta method has higher integration accuracy while ensuring numerical stability, which is beneficial for accurately characterizing the dynamic change process of battery temperature under large time steps, thereby improving the reliability of continuous-time prediction results.

[0113] During the model training phase, in order to optimize the parameters in the neural network's ordinary differential equation model, it is necessary to calculate the gradient of the prediction error with respect to the network parameters. In this embodiment, the adjoint method is preferably used for gradient calculation.

[0114] Specifically, the loss function is expressed as:

[0115] ;

[0116] in, This is the temperature prediction result at the termination time obtained through numerical integration. The adjoint sensitivity method introduces adjoint variables. The corresponding adjoint equation is solved in reverse time, thereby enabling the calculation of the gradient of the model parameters without having to save all intermediate states during the forward integration process.

[0117] By employing the adjoint sensitivity method, this implementation can significantly reduce the memory requirements for storing intermediate states during training while ensuring the correctness of gradient calculation. This enables the model to support modeling of longer time series or higher-dimensional battery temperature states, which is beneficial to improving the trainability and engineering applicability of the model in large-scale battery systems.

[0118] Example 2

[0119] See Figure 2 This invention also provides a battery temperature prediction system based on the neural network constant differential equation, employing the method described in Example 1, for continuous-time prediction of the temperature of each individual cell in a battery system. Specifically, the prediction system includes:

[0120] The data acquisition module is used to collect multi-source operational data of the battery system during operation. This multi-source operational data includes, but is not limited to, parameters such as sampling time, ambient temperature, individual cell heat generation, battery operating current, cooling medium flow rate, and cooling medium inlet temperature. By collecting operational data that simultaneously reflects the battery's internal heating characteristics and external cooling conditions, the system can comprehensively consider different thermal influencing factors during subsequent modeling, providing a more complete input information foundation for battery temperature prediction.

[0121] The data preprocessing module performs time interpolation on the multi-source operational data acquired by the data acquisition module to construct continuous-time input features. Through interpolation, all input variables (including time, tamb, cellHeat, current, flowrate, and inletTemp) and the output 4160 battery temperature sequences are uniformly mapped to continuous time points, satisfying the requirements of the neural network's ordinary differential equations for continuous-time dynamics. After interpolation, the input matrix X and output vector Y in the continuous-time sense are obtained, providing a consistent data foundation for subsequent feature extraction and dynamic modeling. Since different operational parameters are often acquired by different sensors, their sampling periods and sampling times may differ. The data preprocessing module uses a time interpolation algorithm to uniformly map the original discrete sampled data to the continuous-time dimension, allowing various operational parameters to participate in subsequent processing at the same time scale. This approach helps reduce the impact of inconsistent sampling times on the modeling process, thereby improving the continuity and consistency of the battery temperature change process in the time dimension.

[0122] The feature extraction module comprises a global feature extraction submodule and a discriminative feature extraction submodule, used to generate feature vectors characterizing the battery's thermal state from continuous-time input features. The global feature extraction submodule models the overall trend of multi-source operational data over continuous time to obtain global features reflecting the overall thermal state of the battery system. The discriminative feature extraction submodule extracts targeted features from operational parameters highly correlated with thermal behavior, such as battery heat generation and cooling medium flow rate. Specifically, the global feature extraction submodule receives the interpolated complete input variable X. This module utilizes a deep neural network structure to uniformly model and abstract the coupling relationships and short-term and long-term correlation features of six types of input parameters—time, tamb, cellHeat, current, flowrate, and inletTemp—over continuous time. This module outputs a 128-dimensional global feature vector reflecting the overall trend of the battery's thermal state over time.

[0123] To further enhance the characterization of key parameters that have the greatest impact on thermodynamics, a discriminative feature extraction submodule is set up, which specifically processes only the two physically most relevant variables, flowrate and cellHeat. The discriminative module aims to strengthen the model's sensitivity to changes in cooling capacity (flow rate) and the battery's own heat generation capacity, thereby improving its ability to capture nonlinear thermal behaviors such as thermal abrupt changes and temperature rise rates. This module outputs another set of features, which, together with the global features, participate in subsequent modeling.

[0124] Features from the global feature extraction submodule and the discriminative feature extraction submodule are fused and then concatenated with the interpolated 4160-dimensional output temperature vector to form a comprehensive feature vector containing global battery state information, key parameter discrimination information, and neighboring state information. This provides sufficient information input for the subsequent thermal effect network. By combining global features with discriminative features, it is beneficial to highlight key thermal influencing factors under high-dimensional input conditions and avoid weakening important thermal information during feature modeling.

[0125] The thermodynamic modeling module is used to construct a battery thermodynamic model based on the neural network's ordinary differential equations (NDEs) and to make continuous-time predictions of the battery temperature based on these models. In this embodiment, the thermodynamic model uses a neural network to parameterize the change function of the battery temperature state with respect to time, forming a NDE model, and then solves for the temperature evolution over time using numerical integration. This continuous-time modeling approach enables the system to smoothly describe the dynamic changes in battery temperature under conditions of unequal sampling times and varying operating conditions, thereby improving the stability of the temperature prediction results.

[0126] The output module is used to output the temperature prediction results of each cell in the battery system at a target time. The temperature prediction results can be the instantaneous temperature distribution of a single cell, which can be used for subsequent thermal management strategy formulation, abnormal temperature rise monitoring, or safety control decisions.

[0127] Through the modular structure described above, the system in this embodiment can effectively organize the battery temperature prediction process while ensuring a clear overall structure, which is beneficial for the deployment and maintenance of the system in actual engineering environments.

[0128] In a preferred embodiment, the thermodynamic modeling module further includes a heating module, a heat dissipation module, and a fusion module, which are used to model the internal heating effect and the external cooling effect of the battery respectively, and to perform weighted fusion of the modeling results of the two to form a thermodynamic expression for battery temperature prediction.

[0129] During the operation of a battery system, battery temperature changes are typically influenced by both internal heat generation and external heat dissipation. Internal heat generation mainly originates from factors such as battery electrochemical reactions and conductivity losses, while external heat dissipation is related to factors such as cooling medium flow rate, cooling medium inlet temperature, and environmental conditions. Failing to distinguish between these different thermal effects in modeling can weaken the ability to characterize temperature change mechanisms under complex operating conditions.

[0130] Therefore, in this embodiment, the heating module is used to model the internal heating effect of the battery. The heating module takes the comprehensive features output by the feature extraction module as input to model the temperature rise trend reflecting changes in the internal heat source of the battery, thereby obtaining the corresponding heating effect output. By independently modeling the internal heating effect, the model's responsiveness to changes in the battery's own heat generation is enhanced.

[0131] The heat dissipation module is used to model the impact of external cooling conditions on battery temperature changes. The module also uses comprehensive features as input to model the heat loss effect caused by factors such as cooling medium flow rate, cooling medium inlet temperature, and ambient temperature, thereby obtaining the corresponding heat dissipation effect output. Independently modeling the external heat dissipation effect helps to accurately reflect the impact of changes in cooling conditions on the temperature change process.

[0132] The fusion module is used to weight and fuse the outputs of the heating module and the heat dissipation module. In one embodiment, the fusion module introduces dynamic weight parameters to combine the heating effect and the heat dissipation effect, and its temperature prediction expression can be expressed as:

[0133] ;

[0134] in, The temperature prediction output represents the predicted temperature distribution of each cell in the battery system at the target time. For dynamic weight parameters, It is automatically learned by a neural network and is used to adjust the relative weights of heating and heat dissipation effects under different operating conditions. This represents the output function of the heating module. This represents the output function of the heat dissipation module; The comprehensive features are the input.

[0135] By introducing a fusion module and weighting the heating and heat dissipation effects, when the internal heating effect of the battery dominates, the fusion module can correspondingly increase the weight of the heating module's output; when external cooling conditions have a more significant impact on temperature changes, the fusion module can correspondingly increase the weight of the heat dissipation module's output. This structure enables the thermodynamic modeling module to adaptively adjust the contribution ratio of different thermal effects to temperature changes according to changes in operating conditions, thereby more reasonably representing the trend of battery temperature changes.

[0136] By setting up the heating module, heat dissipation module, and fusion module, this embodiment can perform structured modeling of the internal heating effect and external cooling effect of the battery while maintaining a clear model structure. This is beneficial to improving the adaptability of the thermodynamic model to complex operating conditions and providing a more stable and consistent dynamic basis for continuous-time temperature prediction based on the neural ordinary differential equation.

[0137] In a preferred embodiment, the system further includes a model training module for pre-training the battery thermodynamic model based on the neural network differential equation using historical data or simulation data.

[0138] In battery temperature prediction applications, the amount of high-quality labeled data that can be directly obtained in actual systems is usually limited due to constraints in the operating environment, operating condition combinations, and data acquisition conditions. If model training relies solely on a small amount of real-world data, it can easily lead to insufficient convergence of model parameters or inadequate generalization ability. Therefore, this implementation introduces a model training module into the system architecture to pre-train the neural network ordinary differential equation model in the thermodynamic modeling module.

[0139] The model training module first acquires historical operating data or simulation data generated based on the battery thermal model. This simulation data covers different load conditions, cooling conditions, and ambient temperature ranges, reflecting the temperature change characteristics of the battery system under various operating scenarios. By introducing simulation data, the model can be exposed to a wider range of operating conditions in the early stages of training, which is beneficial for forming a stable thermodynamic characterization.

[0140] In one embodiment, the model training module trains the battery thermodynamic model's parameters based on the aforementioned historical or simulation data. During training, the model calculates the predicted temperature trajectory using numerical integration and compares the predicted results with the corresponding real or simulated temperatures to construct a loss function and update the model parameters. Through this pre-training process, the model learns the fundamental evolutionary laws governing battery temperature changes over time.

[0141] By pre-training with historical or simulation data, the model training module can provide a more reasonable initial parameter state for the battery thermodynamics model before deployment. This reduces the model's reliance on large amounts of real-world data in practical applications, improving its stability and engineering applicability under different operating conditions. Simultaneously, this pre-training method provides a foundation for subsequent model adjustments based on a small amount of real-world data, helping to shorten the model training cycle and improve the overall system deployment efficiency.

[0142] The above are merely preferred embodiments of the present invention and do not limit the patent scope of the present invention. All equivalent structural transformations made using the contents of the present invention's specification and drawings under the inventive concept of the present invention, or direct / indirect applications in other related technical fields, are included within the patent protection scope of the present invention.

Claims

1. A battery temperature prediction method based on the ordinary differential equation, characterized in that, The steps include: S1. Acquire multi-source operational data during the operation of the battery system; S2. Perform time interpolation processing on the multi-source operating data to uniformly map data with different sampling frequencies and unequal time intervals to the continuous time dimension to obtain continuous time input features; S3. Based on the continuous time input features, construct global features and discriminant features respectively, wherein the discriminant features are used at least to characterize the thermal influencing factors corresponding to the battery heat generation and the cooling medium flow rate; S4. The global features, discrimination features and the current temperature state of the battery are fused together to construct a comprehensive feature for describing the thermodynamic state of the battery; S5. Input the comprehensive features into a thermodynamic model constructed based on the neural network constant differential equation. The thermodynamic model includes a heating model and a heat dissipation model. The heating model describes the internal heating effect of the battery, and the heat dissipation model describes the influence of external cooling conditions. The outputs of the heating model and the heat dissipation model are fused through weight parameters to characterize the temperature change trend under different operating conditions. The continuous evolution process of battery temperature over time is solved by numerical integration to obtain the cell-by-cell temperature prediction result of the battery system at the target time. The temperature prediction formula is as follows: ； in, For temperature prediction output; For dynamic weight parameters, ∈[0,1], automatically learned through a neural network; For heating function, This is the heat dissipation function; The comprehensive features are the input.

2. The battery temperature prediction method based on the neural ordinary differential equation according to claim 1, characterized in that, The thermodynamic model uses a neural network to parameterize the change function of battery temperature state with respect to time, thereby forming a neural ordinary differential equation model.

3. The battery temperature prediction method based on the neural ordinary differential equation according to claim 1, characterized in that, The time interpolation process uses an interpolation algorithm to map the original discrete sampled data into a continuous time series, so as to eliminate the impact of inconsistent sampling periods of different sensors on the modeling accuracy.

4. The battery temperature prediction method based on the neural ordinary differential equation according to claim 2, characterized in that, The aforementioned constant differential equation model is solved using a numerical integration algorithm to obtain the continuous change of battery temperature over time.

5. A battery temperature prediction system based on the ordinary differential equation, employing the method described in any one of claims 1-4, characterized in that, The system includes: The data acquisition module is used to collect multi-source operating data of the battery system; The data preprocessing module is used to perform time interpolation processing on the multi-source operating data to construct continuous time input features; The feature extraction module includes a global feature extraction submodule and a discriminative feature extraction submodule, which are used to generate feature vectors that characterize the thermal state of the battery; The thermodynamic modeling module is used to construct a battery thermodynamic model based on the neural network differential equation and to make continuous-time predictions of battery temperature. The output module is used to output the predicted temperature of each cell in the battery system at the target time.

6. A battery temperature prediction system based on a neural ordinary differential equation according to claim 5, characterized in that, The thermodynamic modeling module includes a heating module, a heat dissipation module, and a fusion module, which are used to model the internal heating effect and the external cooling effect of the battery respectively and perform weighted fusion.

7. A battery temperature prediction system based on a neural ordinary differential equation according to claim 5, characterized in that, It also includes a model training module for pre-training the battery thermodynamics model based on the neural ordinary differential equation using historical data or simulation data.

Images