Electric vehicle power battery thermal state prediction method, device and equipment and storage medium

Abstract

The application discloses a kind of electric vehicle power battery thermal state prediction method, device, equipment and storage medium, it is related to power battery technical field, comprising: the original sensing signal of power battery is decomposed and reconstructed, and battery reconstruction signal is obtained;The battery reconstruction signal is input into extended long short-term memory network, and temperature prediction value is obtained, wherein, extended long short-term memory network is trained by data fidelity constraint term and physical consistency constraint term combination, solve the technical problem of giving consideration to prediction accuracy and physical consistency under high noise sensing signal and long tail distribution sample condition, through front-end signal purification, the amplification effect of sensing noise in physical constraint term is blocked, and with the aid of physical consistency constraint term, it provides convergent guidance in compliance with the law of thermodynamics for network under extreme working condition sparse sample, so that battery thermal state prediction result has noise resistance and physical rationality under high noise complex working condition of real vehicle.

Description

Technical Field

[0001] This application relates to the field of power battery technology, and in particular to a method, apparatus, device and storage medium for predicting the thermal state of electric vehicle power batteries. Background Technology

[0002] As electric vehicles continue to demand higher driving range and faster charging rates, the energy density and charging / discharging power of power batteries are constantly increasing. This leads to faster heat accumulation and greater temperature fluctuations in batteries under high-load conditions, placing higher technical requirements on the real-time accuracy of battery thermal state prediction and adaptability to extreme conditions.

[0003] Currently, battery thermal state prediction mainly relies on mechanism-driven models or purely data-driven models. Mechanism-driven models suffer from nonlinear drift in physical parameters due to battery aging and environmental changes, making adaptive correction difficult throughout the battery's lifespan. Purely data-driven models exhibit long-tailed distribution failures in scarce sample scenarios such as high temperatures and high-rate fast charging, leading to mean regression and peak clipping in prediction results, making it difficult to accurately capture extreme temperature rise risks. Furthermore, traditional physical information fusion methods amplify noise in physical constraints when processing high-noise sensor signals from real vehicles, causing errors in physical prior injection and compromising network fitting capabilities. These technological limitations prevent existing methods from simultaneously achieving prediction accuracy and physical consistency under complex real-world vehicle conditions characterized by high noise and strong dynamics.

[0004] Therefore, how to balance the accuracy of data fitting with thermodynamic consistency constraints in predicting the thermal state of power batteries is a technical problem that urgently needs to be solved.

[0005] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0006] The main objective of this application is to provide a method, apparatus, device, and storage medium for predicting the thermal state of electric vehicle power batteries, aiming to solve the technical problem of how to balance the accuracy of data fitting with thermodynamic consistency constraints in predicting the thermal state of power batteries.

[0007] To achieve the above objectives, this application proposes a method for predicting the thermal state of an electric vehicle's power battery, the method comprising: The original sensing signals of the power battery are decomposed and reconstructed to obtain the battery reconstructed signal; The battery reconstruction signal is input into an extended long short-term memory network to obtain a temperature prediction value. The extended long short-term memory network is trained by combining data fidelity constraints and physical consistency constraints. In one embodiment, the process of decomposing and reconstructing the original sensing signal of the power battery to obtain a reconstructed battery signal includes: The original sensing signal of the power battery is decomposed to obtain the high-frequency detail coefficient sequence and the low-frequency approximation coefficient sequence; Based on the high-frequency detail coefficient sequence, a threshold is determined; The high-frequency detail coefficient sequence is shrunk according to the threshold to obtain the target high-frequency detail coefficient sequence. The target high-frequency detail coefficient sequence and the low-frequency approximation coefficient sequence are reconstructed to obtain the battery reconstruction signal.

[0008] In one embodiment, the process of decomposing and reconstructing the original sensing signal of the power battery to obtain a reconstructed battery signal includes: The original sensing signal of the power battery is decomposed to obtain the high-frequency detail coefficient sequence and the low-frequency approximation coefficient sequence; Based on the high-frequency detail coefficient sequence, a threshold is determined; The high-frequency detail coefficient sequence is shrunk according to the threshold to obtain the target high-frequency detail coefficient sequence. The target high-frequency detail coefficient sequence and the low-frequency approximation coefficient sequence are reconstructed to obtain the battery reconstruction signal.

[0009] In one embodiment, determining the threshold based on the high-frequency detail coefficient sequence includes: Based on the high-frequency detail coefficient sequence, the target decomposition layer detail coefficient sequence is obtained; Based on the target decomposition layer detail coefficient sequence, the background noise standard deviation is obtained; The threshold is obtained based on the preset time series window length and the background noise standard deviation.

[0010] In one embodiment, before inputting the battery reconstruction signal into the extended long short-term memory network to obtain the temperature prediction value, the method further includes: Obtain the physical weights corresponding to the current training round, the historical battery reconstruction features corresponding to historical sensor signals, the physical parameters to be trained, and the actual historical temperature values; The historical battery reconstruction features are input into a long short-term memory network to obtain historical temperature prediction values; Based on the physical parameters to be trained, the physical weights, the actual historical temperature values, the predicted historical temperature values, and the historical battery reconstruction features, the target physical parameters to be trained and the target long short-term memory network are obtained. The current training round is adjusted according to the preset round increment step size to obtain the target training round; The target physical parameters to be trained are used as the physical parameters to be trained, the target long short-term memory network is used as the long short-term memory network, and the target training round is used as the current training round. The steps of obtaining the physical weights corresponding to the current training round, the historical battery reconstruction features corresponding to the historical sensor signals, the physical parameters to be trained, and the historical actual temperature values ​​are returned to be executed until the current training round reaches the preset round threshold. Then, the target long short-term memory network is used as the extended long short-term memory network.

[0011] In one embodiment, obtaining the target physical parameters to be trained and the target long short-term memory network based on the physical parameters to be trained, the physical weights, the actual historical temperature values, the predicted historical temperature values, and the historical battery reconstruction features includes: Based on the physical parameters to be trained, the historical temperature prediction values, and the historical battery reconstruction features, the theoretical physical value of temperature is obtained. A composite loss function is constructed based on the physical weights, the actual historical temperature values, the predicted historical temperature values, and the theoretical physical temperature values. The optimization objective is to minimize the function value of the composite loss function to obtain the target neural network parameters and the target physical parameters to be trained. The parameters of the neural network to be trained in the long short-term memory network are updated according to the parameters of the target neural network to obtain the target long short-term memory network.

[0012] In one embodiment, the construction of a composite loss function based on the physical weights, the actual historical temperature values, the predicted historical temperature values, and the theoretical physical temperature values, and the optimization objective of minimizing the function value of the composite loss function, yields the target neural network parameters and the target physical parameters to be trained, including: Based on the predicted historical temperature value and the actual historical temperature value, the data fidelity constraint term is determined; Based on the historical temperature predictions and the theoretical temperature values, the physical consistency constraints are determined. Based on the physical weights, the data fidelity constraint and the physical consistency constraint are weighted and combined to obtain a composite loss function; The optimization objective is to minimize the function value of the composite loss function to obtain the target neural network parameters and the target physical parameters to be trained.

[0013] Furthermore, to achieve the above objectives, this application also proposes an electric vehicle power battery thermal state prediction device, which includes: The preprocessing module is used to decompose and reconstruct the original sensing signals of the power battery to obtain the battery reconstruction signal; The output module is used to input the battery reconstruction signal into the extended long short-term memory network to obtain the temperature prediction value, wherein the extended long short-term memory network is trained by combining data fidelity constraint terms and physical consistency constraint terms.

[0014] In addition, to achieve the above objectives, this application also proposes an electric vehicle power battery thermal state prediction device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the electric vehicle power battery thermal state prediction method as described above.

[0015] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the electric vehicle power battery thermal state prediction method described above.

[0016] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the electric vehicle power battery thermal state prediction method described above.

[0017] This application decomposes and reconstructs the original sensing signal of the power battery to obtain a reconstructed battery signal. The reconstructed battery signal is then input into an extended long short-term memory (LSTM) network to obtain a temperature prediction value. The LSM network is trained by a combination of data fidelity constraints and physical consistency constraints. By employing the aforementioned signal purification and dual-stream physical constraint joint training techniques, the technical problem of balancing prediction accuracy and physical consistency under conditions of high-noise sensing signals and long-tailed sample distribution is solved. Front-end signal purification blocks the amplification effect of sensing noise in the physical constraint terms, and the physical consistency constraints provide convergence guidance to the network under extreme operating conditions and sparse samples, conforming to the laws of thermodynamics. This ensures that the battery thermal state prediction results possess noise resistance and physical rationality under complex high-noise conditions in real vehicles. Attached Figure Description

[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating an embodiment of the electric vehicle power battery thermal state prediction method of this application. Figure 2 This is a flowchart of the adaptive wavelet denoising process provided in Embodiment 1 of the electric vehicle power battery thermal state prediction method of this application. Figure 3 This is a flowchart illustrating Embodiment 2 of the method for predicting the thermal state of electric vehicle power batteries in this application. Figure 4 This is a framework diagram for predicting the heat generation of electric vehicle power batteries provided in Embodiment 2 of the present application. Figure 5 This is a flowchart illustrating the execution of the hybrid optimization algorithm under physical constraints provided in Embodiment 2 of the electric vehicle power battery thermal state prediction method of this application. Figure 6 This is a schematic diagram of the module structure of the electric vehicle power battery thermal state prediction device according to an embodiment of this application; Figure 7 This is a schematic diagram of the equipment structure of the hardware operating environment involved in the electric vehicle power battery thermal state prediction method in the embodiments of this application.

[0021] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0022] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0023] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0024] The main solution of this application embodiment is: to decompose and reconstruct the original sensing signal of the power battery to obtain the battery reconstruction signal; to input the battery reconstruction signal into the extended long short-term memory network to obtain the temperature prediction value, wherein the extended long short-term memory network is trained by combining data fidelity constraint terms and physical consistency constraint terms.

[0025] In this embodiment, for ease of description, the following description uses the electric vehicle power battery thermal state prediction system as the execution subject.

[0026] Battery thermal state prediction primarily relies on mechanistic-driven or purely data-driven models. Mechanism-driven models suffer from nonlinear drift in physical parameters due to battery aging and environmental changes, making adaptive correction difficult throughout the battery's lifespan. Purely data-driven models exhibit long-tailed failures in scenarios with scarce samples, such as high temperatures and high-rate fast charging, leading to mean regression and peak clipping in predictions, hindering accurate capture of extreme temperature rise risks. Furthermore, traditional physical information fusion methods amplify noise in physical constraints when processing high-noise sensor signals from real vehicles, resulting in incorrect physical prior injections and compromising network fitting capabilities. These technological limitations prevent existing methods from simultaneously achieving prediction accuracy and physical consistency under complex real-world vehicle conditions characterized by high noise and strong dynamics.

[0027] This application provides a solution that, by employing the aforementioned signal purification and dual-stream physical constraint joint training techniques, solves the technical problem of balancing prediction accuracy and physical consistency under conditions of high-noise sensor signals and long-tailed distributed samples. By using front-end signal purification, the amplification effect of sensor noise in the physical constraint term is blocked, and by using the physical consistency constraint term, the network is provided with convergence guidance that conforms to the laws of thermodynamics under extreme operating conditions and sparse samples. This enables the battery thermal state prediction results to have noise resistance and physical rationality under the high-noise and complex operating conditions of real vehicles.

[0028] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device capable of performing the above functions, such as an electric vehicle power battery thermal state prediction system. The following description uses an electric vehicle power battery thermal state prediction system as an example to illustrate this embodiment and the subsequent embodiments.

[0029] Based on this, embodiments of this application provide a method for predicting the thermal state of an electric vehicle's power battery, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the electric vehicle power battery thermal state prediction method of this application.

[0030] In this embodiment, the electric vehicle power battery thermal state prediction method includes steps S10~S20: Step S10: Decompose and reconstruct the original sensing signal of the power battery to obtain the battery reconstruction signal; It should be noted that a power battery is a rechargeable energy storage device that provides driving energy for electric vehicles; the original sensing signal refers to the time-series signal containing measurement noise, which is collected in real time by temperature and current sensors arranged on the surface of the power battery module; the battery reconfiguration signal refers to the multi-dimensional purified time-series signal that has been decomposed and reconfigured, filtering out high-frequency noise and retaining step change characteristics.

[0031] It is understandable that since the high-frequency measurement noise contained in the original sensing signal will be amplified exponentially in the subsequent calculation of the thermodynamic equation, thereby compromising the accuracy of the physical constraints, step S10 can avoid the interference of noise on the physical information injection process, thereby blocking the physical constraint backlash path and improving the stability of subsequent network training.

[0032] In one feasible implementation, step S10 may include: decomposing the original sensing signal of the power battery to obtain a high-frequency detail coefficient sequence and a low-frequency approximation coefficient sequence; determining a threshold based on the high-frequency detail coefficient sequence; performing a shrinking process on the high-frequency detail coefficient sequence according to the threshold to obtain a target high-frequency detail coefficient sequence; and performing a reconstruction process on the target high-frequency detail coefficient sequence and the low-frequency approximation coefficient sequence to obtain a battery reconstruction signal.

[0033] It should be noted that the high-frequency detail coefficient sequence refers to the set of coefficients representing local abrupt changes and noise components of the original sensor signal at multiple decomposition scales; the low-frequency approximation coefficient sequence refers to the set of coefficients representing the overall trend and gradual change components of the original sensor signal at the coarsest decomposition scale; the threshold is the numerical limit used to distinguish noise components from effective signal components; the target high-frequency detail coefficient sequence refers to the high-frequency detail coefficient sequence after shrinkage processing, in which noise components are suppressed while effective abrupt change features are preserved.

[0034] Specifically, the original sensing signal is input into a discrete wavelet transform module for layer-by-layer decomposition. Each layer of decomposition generates a set of high-frequency detail coefficients and a set of low-frequency approximation coefficients. Noise distribution features are extracted from the high-frequency detail coefficient sequence of the highest decomposition layer, and a threshold is calculated accordingly. The high-frequency detail coefficient sequence of each decomposition layer is then subjected to soft thresholding using this threshold to remove noise components below the threshold, resulting in the target high-frequency detail coefficient sequence. The target high-frequency detail coefficient sequence and the lowest-level low-frequency approximation coefficient sequence are then input into an inverse wavelet transform module to synthesize the battery reconstruction signal. Further, determining the threshold based on the high-frequency detail coefficient sequence includes: obtaining the target decomposition layer detail coefficient sequence based on the high-frequency detail coefficient sequence; obtaining the background noise standard deviation based on the target decomposition layer detail coefficient sequence; and obtaining the threshold based on the preset time series window length and the background noise standard deviation.

[0035] It should be noted that the target decomposition layer detail coefficient sequence refers to the high-frequency detail coefficient sequence of the highest decomposition level, and its main component is regarded as background noise; the background noise standard deviation refers to the statistic obtained by robustly estimating the fluctuation intensity of the target decomposition layer detail coefficient sequence; the preset time series window length refers to the number of continuous sampling points participating in wavelet decomposition.

[0036] Specifically, the high-frequency detail coefficient sequence of the highest decomposition layer is extracted from the multi-scale decomposition results as the detail coefficient sequence of the target decomposition layer; the median absolute deviation of the detail coefficient sequence of the target decomposition layer is calculated, and the background noise standard deviation is derived based on the statistical relationship between the median absolute deviation and the standard deviation; the product of the square root of the preset time series window length and the background noise standard deviation is used as the basic threshold, and a preset relaxation factor is introduced to scale the basic threshold to obtain the threshold threshold.

[0037] like Figure 2 As shown, the original sensing signals include current, voltage, real physical signals, and background measurement noise collected by the Battery Management System (BMS). A Daubechies 4 wavelet basis (db4) with good compact support and asymmetry can be selected, as its waveform characteristics closely match the transient excitation response of the battery. The decomposition level can be set to 3. The Discrete Wavelet Transform (DWT) decomposition process of the original sensing signals can be expressed as follows:

[0038]

[0039] Where j is the wavelet transform level, and k is the wavelet transform translation parameter. For the first High-frequency detail coefficients of the layer, For the first Low-frequency approximation coefficients of the layer; For wavelet mother function, Let x(t) be the scaling function, and let x(t) be the original sensing signal at time t.

[0040] For the high-frequency detail coefficients obtained from the decomposition, an adaptive soft thresholding strategy can be used to remove white noise:

[0041] Where 0.5 is the relaxation factor, and N is the length of the time series window. For threshold, The standard deviation of the background noise is estimated and is calculated as follows:

[0042] in, For the first The high-frequency detail coefficients of the layer are the sequence of detail coefficients of the target decomposition layer. j can be the highest layer, and median represents the absolute deviation of taking the high-frequency detail coefficients of the highest layer.

[0043] Subsequently, a soft thresholding function is used to shrink the high-frequency detail coefficients. The shrinkage formula is as follows:

[0044] in, For the first High-frequency detail coefficients of the layer, sgn represents the threshold, and sgn represents the sign function.

[0045] High-frequency detail coefficients after soft thresholding Together with the retained low-frequency approximation coefficients, they are reconstructed through wavelet inverse transform to obtain the denoised clean current and voltage, which can be used for subsequent physical feature calculations.

[0046] In this embodiment, by adaptively thresholding the high-frequency detail coefficient sequence in the transform domain, the problem of sensor noise being exponentially amplified in subsequent thermodynamic equations is solved.

[0047] The above are merely feasible implementations of step S10 provided in this embodiment. This embodiment does not specifically limit the specific implementation of step S10.

[0048] Step S20: Input the battery reconstruction signal into the extended long short-term memory network to obtain the temperature prediction value, wherein the extended long short-term memory network is trained by a combination of data fidelity constraint terms and physical consistency constraint terms; It should be noted that Extended Long Short-Term Memory (xLSTM) refers to a recurrent neural network that introduces exponential gating and matrix memory mechanisms in forward propagation and accepts joint constraints of data fidelity and physical consistency during training; the temperature prediction value refers to the predicted value of the power battery temperature per unit time output by the extended long short-term memory network; the data fidelity constraint is a loss component constructed based on the deviation between the network's predicted temperature and the measured temperature; the physical consistency constraint is a loss component constructed based on the degree to which the network output violates the thermodynamic equations.

[0049] It is understandable that traditional recurrent neural networks are prone to forgetting early thermal excitation signals when faced with the thermal hysteresis effect of power batteries that lasts for more than half an hour. In addition, conventional activation functions have limited response speed to sudden high current conditions. Therefore, step S20 can quickly capture transient high current excitation features and retain multi-physics coupling information for a long period, thereby avoiding memory decay of early key signals and improving the model's prediction accuracy for extreme temperature rise conditions.

[0050] In one feasible implementation, step S20 may include: obtaining battery reconstruction features based on the battery reconstruction signal; inputting the battery reconstruction features into an extended long short-term memory network to obtain a temperature prediction value, wherein the activation function of the extended long short-term memory network is an exponential function, and the memory cells of the extended long short-term memory network store feature interaction information in the form of a covariance matrix.

[0051] It should be noted that the battery reconfiguration feature refers to the multidimensional time-series feature tensor formed after the battery reconfiguration signal is vectorized and encapsulated; the activation function refers to the function used in a neural network to perform nonlinear transformation on the input; the exponential function refers to the power function with the natural constant as the base, whose function value increases exponentially with the input; the memory unit refers to the internal variable used in a recurrent neural network to store historical state information; the feature interaction information refers to the correlation pattern between different physical quantities in the battery operating state; and the covariance matrix refers to a second-order statistical matrix organized in rows and columns to characterize the pairwise correlation between multidimensional features.

[0052] Specifically, the battery reconstruction signal is divided and arranged according to a preset time window and feature dimension to form battery reconstruction features. After the feature space is purified and multi-dimensional tensor encapsulation is completed, the battery reconstruction features are input into the core prediction backbone network of this scheme—Extended Long Short-Term Memory (LSTM). Due to the extremely long hysteresis inertia of the battery thermal response (continuous acceleration half an hour ago may lead to the current continuous temperature rise), traditional Recurrent Neural Network (RNN) architectures are prone to gradient vanishing due to excessively long backpropagation chains, while standard Long Short-Term Memory (LSTM) networks, due to their scalar hidden state information capacity approaching the upper limit, easily forget important early thermal excitation signals. The Extended Long Short-Term Memory (LSTM) network of this application redefines the gating state using an exponential activation function. For the forget gate and input gate, the calculation method is upgraded as follows:

[0053] in, For the Gate of Oblivion Here, t represents the input gate. It is the weight matrix of the forget gate. The weight matrix of the input gate, It is the battery reconstruction feature of the input at the current time t. It is the hidden state at time t-1.

[0054] To prevent numerical overflow caused by exponential operations, the extended long short-term memory network introduces stabilizer states and normalized states:

[0055] in, This represents the stabilizer state at time t. This represents the normalized state at time t. This represents the stabilizer state at time t-1. Represents the normalized state at time t-1. For the Gate of Oblivion This is the input gate.

[0056] In battery operation, when sudden high-power fast charging or thermal runaway precursors occur, exponential gating can quickly amplify the weight of the signal, enabling the model to reset its memory in a very short time and prioritize the current severe disturbance. This feature effectively improves the model's ability to capture transient conditions.

[0057] The Matrix Long Short-Term Memory (mLSTM) module in xLSTM upgrades the memory unit to matrix memory, allowing the model to store second-order interaction information between features in the form of a covariance matrix. For example, the model can simultaneously encode two states, high current and low temperature, in a matrix. When a similar combination is encountered again, the model can directly retrieve the corresponding high internal resistance heat generation mode through matrix multiplication.

[0058] Its update rule utilizes outer product operations, enabling it to store higher-order association information in the form of key-value pairs:

[0059] in, It is a key vector. These are value vectors, all reconstructed from the input battery features. Derived from a linear transformation; For the Gate of Oblivion Here, t represents time, and T represents the transpose operation; It is a matrix memory unit.

[0060] In this embodiment, by introducing exponential gating and matrix memory mechanisms, the problems of early excitation signals being easily forgotten and slow response to transient conditions in traditional recurrent neural networks under long-term thermal hysteresis are solved.

[0061] The above are merely feasible implementations of step S20 provided in this embodiment. This embodiment does not specifically limit the specific implementation of step S20.

[0062] This embodiment provides a method for predicting the thermal state of an electric vehicle's power battery. The original sensor signal of the power battery is decomposed and reconstructed to obtain a reconstructed battery signal. This reconstructed signal is then input into an extended long short-term memory (LSTM) network to obtain a predicted temperature value. The LSM network is trained using a combination of data fidelity constraints and physical consistency constraints. By employing the aforementioned signal cleansing and dual-stream physical constraint joint training techniques, the technical problem of balancing prediction accuracy and physical consistency under conditions of high-noise sensor signals and long-tailed sample distribution is solved. Front-end signal cleansing blocks the amplification effect of sensor noise in the physical constraint terms, and the physical consistency constraint terms provide convergence guidance for the network under extreme operating conditions and sparse samples, conforming to the laws of thermodynamics. This ensures that the battery thermal state prediction results possess noise resistance and physical rationality under complex high-noise conditions in real vehicles.

[0063] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in the first embodiment described above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 3 The electric vehicle power battery thermal state prediction method further includes steps S11 to S15 before step S20: Step S11: Obtain the physical weights corresponding to the current training round, the historical battery reconstruction features corresponding to the historical sensor signals, the physical parameters to be trained, and the actual historical temperature values. It should be noted that the current training round refers to the current iteration number of the extended long short-term memory network during the training process; physical weights refer to the weighting coefficients used to adjust the proportion of physical consistency constraints in the composite loss function; historical sensor signals refer to the original time-series signals collected by sensors during the actual operation of the power battery in the past; historical battery reconstruction features refer to the multidimensional purified time-series feature tensor obtained after performing decomposition and reconstruction processing on the historical sensor signals; physical parameters to be trained refer to the lumped thermodynamic coefficients that need to be optimized and solved together with the neural network parameters during the training process; and historical actual temperature values ​​refer to the actual temperature values ​​based on the historical sensor signals.

[0064] Specifically, the current training round is obtained, and the corresponding physical weights are determined from the preset annealing scheduling strategy based on the current training round; at the same time, the pre-stored historical sensing signals are extracted from the historical database, and the decomposition and reconstruction processing described in step S10 is performed on the historical sensing signals to obtain historical battery reconstruction features; the physical parameters to be trained are initialized or read after the previous training round is updated, and the historical actual temperature values ​​are read from the historical database.

[0065] It is understandable that, since the intensity of physical constraints needs to change dynamically with the training process, and the physical parameters to be trained need to gradually approach the real physical characteristics in the iteration, performing step S11 can avoid gradient oscillation caused by excessively strong physical constraints in the early stage of training, thereby providing the correct data foundation and initial parameter state for subsequent dual-stream joint optimization.

[0066] Step S12: Input the historical battery reconstruction features into the long short-term memory network to obtain the historical temperature prediction value; It should be noted that the Long Short-Term Memory Network refers to a transient instance of an Extended Long Short-Term Memory Network (LSTM) that includes an exponential gating mechanism and a matrix memory mechanism during the training phase, and its internal weight parameters are the parameters of the neural network to be trained; the historical temperature prediction value refers to the temperature prediction result calculated and output by the LTM based on the historical battery reconstruction features input by the input.

[0067] Specifically, the historical battery reconstruction features are expanded along the time step dimension and sequentially input into the long short-term memory network. At each time step, the network uses exponential gating to selectively memorize and forget the input features, and uses matrix memory units to store and retrieve the second-order interaction information between multi-physics features in the form of covariance. After forward propagation through all time steps, the network output provides a sequence of historical temperature predictions corresponding to the input sequence.

[0068] It is understandable that step S12 is necessary because historical data is needed to drive the network to learn the temporal mapping pattern of battery thermal response. This step provides model prediction output for subsequent loss function calculation, thereby providing a basis for error backpropagation for parameter updates.

[0069] Step S13: Based on the physical parameters to be trained, the physical weights, the actual historical temperature values, the predicted historical temperature values, and the historical battery reconstruction features, obtain the target physical parameters to be trained and the target long short-term memory network.

[0070] It should be noted that the target physical parameters to be trained refer to the lumped thermodynamic coefficients after one round of iterative optimization and update; the target long short-term memory network refers to the neural network instance after one round of iterative optimization and update.

[0071] It is understandable that since relying solely on data-driven training methods can produce prediction results that deviate from thermodynamic laws under long-tailed distribution samples, step S13 can utilize physical consistency constraints to provide the network with convergence guidance that conforms to thermodynamic laws, thereby improving the model's ability to resist divergence extrapolation under scarce conditions such as extreme high temperatures.

[0072] In one feasible implementation, step S13 may include: obtaining a theoretical temperature value based on the physical parameters to be trained, the historical temperature prediction value, and the historical battery reconstruction features; constructing a composite loss function based on the physical weights, the actual historical temperature value, the historical temperature prediction value, and the theoretical temperature value, and solving the problem with minimizing the function value of the composite loss function as the optimization objective to obtain the target neural network parameters and the target physical parameters to be trained; updating the parameters of the neural network to be trained in the long short-term memory network according to the target neural network parameters to obtain the target long short-term memory network.

[0073] It should be noted that the theoretical temperature value refers to the temperature reference value calculated based on the discretized thermodynamic equation and the physical parameters to be trained, which conforms to the constraints of physical laws; the composite loss function refers to the optimization objective function formed by the weighted combination of the data fidelity constraint term and the physical consistency constraint term through the physical weights; and the neural network parameters to be trained refer to the trainable weight matrix and bias vector in the long short-term memory network.

[0074] Further, the step of constructing a composite loss function based on the physical weights, the actual historical temperature values, the predicted historical temperature values, and the theoretical temperature values, and solving for minimizing the function value of the composite loss function to obtain the target neural network parameters and the target physical parameters to be trained, includes: determining the data fidelity constraint term based on the predicted historical temperature values ​​and the actual historical temperature values; determining the physical consistency constraint term based on the predicted historical temperature values ​​and the theoretical temperature values; weighting the data fidelity constraint term and the physical consistency constraint term according to the physical weights to obtain the composite loss function; and solving for minimizing the function value of the composite loss function to obtain the target neural network parameters and the target physical parameters to be trained.

[0075] like Figure 4 As shown, this application proposes a heat generation prediction framework. The input data are current, voltage, and ambient temperature. The current, voltage, and ambient temperature together constitute a battery reconstruction feature tensor, which is input to an extended long short-term memory network. The extended long short-term memory network performs time-series encoding on the input features through an internal exponential gating and matrix memory mechanism, outputs a temperature prediction value, and inputs the temperature prediction value into a data loss calculation module. The deviation between the temperature prediction value and the historical actual temperature value (true temperature) is compared to calculate the data loss, thus obtaining the data fidelity constraint term.

[0076] The current information and the current battery temperature state information are extracted from the battery reconstructed feature tensor and input into the physical constraint calculation module. The physical error calculation module, based on the discretized thermodynamic equations, uses the current information, ambient temperature, predicted temperature, and trainable physical parameters to calculate the theoretical physical value of the temperature change rate that conforms to the law of conservation of energy. The calculation formula is as follows:

[0077] in, This is the theoretical value of the rate of temperature change. and For trainable physical parameters, For current, The predicted temperature value at time i-1 The ambient temperature; The predicted temperature value is approximated by time differentiation to obtain the predicted rate of temperature change, calculated using the following formula: ; in, This is the predicted rate of temperature change. Let i be the predicted temperature value at time i. The predicted temperature value at time i-1; Integrating the physical theoretical value and the predicted value of the temperature change rate over time, we can obtain the temperature change at time i, which is the physical loss (physical consistency constraint).

[0078] The physical loss and data loss are weighted and combined to construct a total loss function. The neural network parameters and the trainable physical parameters in the extended long short-term memory network are updated along the gradient direction of the total loss function through backpropagation, so that the prediction results can fit the measured data while following the constraints of thermodynamic laws.

[0079] More specifically, assume that the rate of change of battery temperature at time t is... According to the law of conservation of energy, it should be determined by both the heat production and heat dissipation terms. By transforming the energy conservation formula, we can obtain a method for calculating the theoretical value of the rate of temperature change. The energy conservation formula is:

[0080] Where m is the battery mass. For specific heat capacity, For heat-generating items, For heat dissipation, I is the current, R is the resistance, T is the battery temperature, U is the overall heat transfer coefficient, and h is the heat transfer coefficient. A represents the ambient temperature, and A represents the heat dissipation area.

[0081] In unsupervised learning scenarios using real-vehicle data, precise internal resistance... Heat transfer coefficient The physical processes are often unknown and change dynamically with the operating conditions. Therefore, this application adopts a semi-empirical and semi-parametric strategy to abstract complex physical processes into lumped parameter models and introduce trainable physical parameters. and And define the physical residual function. The deviation between the predicted temperature change rate and the theoretical temperature change rate can be obtained by integrating the physical residual function, which yields the error between the predicted and theoretical temperature values, i.e., the physical loss.

[0082] in, This represents the predicted battery temperature at time t. This represents the predicted battery temperature at time t-1. This represents the current at time t. Indicates ambient temperature. This represents the change in time from time t-1 to time t.

[0083] It is the rate of temperature change output by xLSTM (approximately automatically differentiated by first-order difference). This represents the heat-generating component dominated by Joule heating. It is the lumped heat production coefficient that the network needs to adaptively learn. This represents the heat dissipation term primarily governed by Newton's law of cooling. In such Figure 4 Within the framework shown, neural networks no longer output a value arbitrarily; they must satisfy certain conditions. The optimal solution is sought under constraints to ensure that the predicted trajectory always operates within the thermodynamically permissible range.

[0084] by Figure 4 The framework shown yields a two-stream network architecture with embedded physical residual constraints. The two-stream network refers to xLSTM and a Physical Information Neural Network (PINN). This architecture features two backpropagation paths for gradients: (1) Data Path: Based on predicted temperature Compared with actual sensor temperature This ensures that the model can fit the measured data and capture subtle nonlinear characteristics that are not covered by the physical formulas (such as parameter drift caused by battery aging).

[0085] (2) Physical Path: Based on physical residuals This path does not rely on the real labels. Even when sensor data is missing or noisy, it still corrects the network parameters by minimizing the residuals, which gives the model noise resistance and generalization ability.

[0086] Based on the two backpropagation paths, a dynamically weighted composite loss function was constructed. :

[0087] in, For composite loss function, For data fidelity constraints, For physical consistency constraints, For physical weights, Let W be the residual, and W be the weight matrix of the neural network.

[0088] Data fidelity constraints The calculation formula is:

[0089] Where N is the sample size. This represents the historical predicted temperature of the battery for the i-th sample. This represents the actual battery temperature of the (i-1)th sample, i.e., the historical actual value of the battery temperature.

[0090] Physical consistency constraints The calculation formula is:

[0091] Where N is the number of samples, and i is the i-th sample within the sample size. This is the physical residual function.

[0092] Physical weight The calculation formula is:

[0093] Where e represents the training round. The maximum regularization coefficient, The length of the preheating stage can be 10.

[0094] In this embodiment, by constructing a dual-stream composite loss function that integrates data fidelity and physical consistency, the problem of data-driven models deviating from the laws of thermodynamics in the long-tail region of extreme operating conditions is solved.

[0095] The above are merely feasible implementations of step S13 provided in this embodiment. This embodiment does not specifically limit the specific implementation of step S13.

[0096] Step S14: Adjust the current training round according to the preset round growth step size to obtain the target training round.

[0097] It should be noted that the preset training round increment refers to the increment of the training round after each iteration; the target training round refers to the new round number after the current training round is incremented.

[0098] Specifically, the preset training increment step size is obtained, and the current training iteration is added to the preset training increment step size to obtain the target training iteration.

[0099] Understandably, since the training process needs to proceed step by step in rounds, and the scheduling of physical weights depends on the precise increment of training rounds, performing step S14 can ensure the orderly progress of the training process and the accurate execution of the physical weight annealing strategy.

[0100] Step S15: Take the target physical parameters to be trained as the physical parameters to be trained, take the target long short-term memory network as the long short-term memory network, take the target training round as the current training round, and return to execute the step of obtaining the physical weights corresponding to the current training round, the historical battery reconstruction features corresponding to the historical sensor signals, the physical parameters to be trained, and the historical actual temperature values, until the current training round reaches the preset round threshold, and take the target long short-term memory network as the extended long short-term memory network.

[0101] It should be noted that the preset round threshold refers to the upper limit of the number of rounds at which the training process will terminate.

[0102] Specifically, it is determined whether the target training round has reached the preset round threshold. If it has not, the target physical parameters to be trained and the target long short-term memory network output in this round are used as the physical parameters to be trained and the long short-term memory network in the next round, respectively. The target training round is used as the current training round in the next round, and the process returns to step S11 to continue iterating. If the preset round threshold has been reached, the training loop is terminated, and the target long short-term memory network obtained at this time is used as the extended long short-term memory network for final deployment.

[0103] like Figure 5 As shown, an annealing strategy was used to dynamically adjust the physical weights in the loss function.

[0104]

[0105] in, The physical weights are defined in each epoch, where epoch represents the training epoch.

[0106] Phase 1 (0-10 Epochs): Using only data loss, the model can quickly learn the statistical patterns of the data and establish a preliminary mapping relationship.

[0107] Phase Two (10-50 Epochs): The value increases linearly with the number of training epochs, growing from 0 to 1.0. At this point, physical constraints gradually come into play, correcting the non-physical behavior of the model.

[0108] Phase Three (50+ Epochs): Full-constraint fine-tuning ensures that the final model strictly adheres to the thermodynamic equations.

[0109] It is understandable that since the physical consistency constraint of the model needs to go through a complete annealing scheduling process to be fully effective, and the increase in the number of iterations can make the network parameters and physical parameters gradually converge to the global optimum, performing step S15 can ensure that the model achieves the optimal balance between data fidelity and physical rationality when training terminates.

[0110] This embodiment provides a method for predicting the thermal state of an electric vehicle's power battery. The method involves obtaining the physical weights corresponding to the current training round, historical battery reconstruction features corresponding to historical sensor signals, the physical parameters to be trained, and historical actual temperature values. The historical battery reconstruction features are input into a Long Short-Term Memory (LSTM) network to obtain historical temperature predictions. Based on the physical parameters to be trained, the physical weights, the historical actual temperature values, the historical temperature predictions, and the historical battery reconstruction features, target physical parameters to be trained and a target LSM network are obtained. The current training round is adjusted according to a preset round increment step size to obtain a target training round. The target physical parameters to be trained are used as the training physical parameters, the target LSM network is used as the LSM network, and the target training round is used as the current training round. The method then returns to the steps of obtaining the physical weights corresponding to the current training round, the historical battery reconstruction features corresponding to historical sensor signals, the physical parameters to be trained, and the historical actual temperature values, until the current training round reaches a preset round threshold. At this point, the target LSM network is used as an Extended Long Short-Term Memory (ESM) network. By employing a training method that combines dual-stream physical constraints with dynamic weighted annealing scheduling, gradient oscillations caused by excessively strong physical constraints in the early stages of training are avoided. This solves the problem of data-driven models deviating from thermodynamic laws in the long-tail region under extreme operating conditions, thereby achieving high-precision and physically consistent stable prediction of battery thermal state under high-noise and complex operating conditions in real vehicles.

[0111] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the electric vehicle power battery thermal state prediction method of this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0112] This application also provides a device for predicting the thermal state of an electric vehicle's power battery. Please refer to [link / reference]. Figure 6 The electric vehicle power battery thermal state prediction device includes: The preprocessing module 10 is used to decompose and reconstruct the original sensing signal of the power battery to obtain the battery reconstruction signal. The output module 20 is used to input the battery reconstruction signal into the extended long short-term memory network to obtain the temperature prediction value, wherein the extended long short-term memory network is trained by combining data fidelity constraint terms and physical consistency constraint terms.

[0113] The electric vehicle power battery thermal state prediction device provided in this application, employing the electric vehicle power battery thermal state prediction method in the above embodiments, can solve the technical problem of how to balance data fitting accuracy and thermodynamic consistency constraints in power battery thermal state prediction. Compared with the prior art, the beneficial effects of the electric vehicle power battery thermal state prediction device provided in this application are the same as those of the electric vehicle power battery thermal state prediction method provided in the above embodiments, and other technical features in the electric vehicle power battery thermal state prediction device are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0114] The preprocessing module 10 is further configured to decompose the original sensing signal of the power battery to obtain a high-frequency detail coefficient sequence and a low-frequency approximation coefficient sequence; determine a threshold based on the high-frequency detail coefficient sequence; perform shrinkage processing on the high-frequency detail coefficient sequence according to the threshold to obtain a target high-frequency detail coefficient sequence; and perform reconstruction processing on the target high-frequency detail coefficient sequence and the low-frequency approximation coefficient sequence to obtain a battery reconstruction signal.

[0115] The preprocessing module 10 is further configured to obtain a target decomposition layer detail coefficient sequence based on the high-frequency detail coefficient sequence; obtain a background noise standard deviation based on the target decomposition layer detail coefficient sequence; and obtain a threshold based on a preset time series window length and the background noise standard deviation.

[0116] The output module 20 is further configured to acquire the physical weights corresponding to the current training round, the historical battery reconstruction features corresponding to the historical sensor signals, the physical parameters to be trained, and the historical actual temperature values; input the historical battery reconstruction features into the long short-term memory network to obtain the historical temperature prediction values; based on the physical parameters to be trained, the physical weights, the historical actual temperature values, the historical predicted temperature values, and the historical battery reconstruction features, obtain the target physical parameters to be trained and the target long short-term memory network; adjust the current training round according to the preset round increment step size to obtain the target training round; use the target physical parameters to be trained as the physical parameters to be trained, the target long short-term memory network as the long short-term memory network, and the target training round as the current training round, and return to execute the steps of acquiring the physical weights corresponding to the current training round, the historical battery reconstruction features corresponding to the historical sensor signals, the physical parameters to be trained, and the historical actual temperature values, until the current training round reaches the preset round threshold, and then use the target long short-term memory network as the extended long short-term memory network.

[0117] The output module 20 is further configured to: obtain a theoretical temperature value based on the physical parameters to be trained, the historical temperature prediction value, and the historical battery reconstruction features; construct a composite loss function based on the physical weights, the actual historical temperature value, the historical temperature prediction value, and the theoretical temperature value; solve for the target neural network parameters and the target physical parameters to be trained based on minimizing the function value of the composite loss function; and update the parameters of the neural network to be trained in the long short-term memory network according to the target neural network parameters to obtain the target long short-term memory network.

[0118] The output module 20 is further configured to: determine the data fidelity constraint term based on the historical temperature prediction value and the historical actual temperature value; determine the physical consistency constraint term based on the historical temperature prediction value and the physical theoretical temperature value; perform a weighted combination of the data fidelity constraint term and the physical consistency constraint term according to the physical weight to obtain a composite loss function; and solve for the target neural network parameters and the target physical parameters to be trained by minimizing the function value of the composite loss function as the optimization objective.

[0119] The output module 20 is further configured to obtain battery reconstruction features based on the battery reconstruction signal; input the battery reconstruction features into an extended long short-term memory network to obtain a temperature prediction value, wherein the activation function of the extended long short-term memory network is an exponential function, and the memory units of the extended long short-term memory network store feature interaction information in the form of a covariance matrix.

[0120] This application provides an electric vehicle power battery thermal state prediction device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the electric vehicle power battery thermal state prediction method in the above embodiment 1.

[0121] The following is for reference. Figure 7 This document illustrates a structural schematic diagram suitable for implementing the electric vehicle power battery thermal state prediction device in the embodiments of this application. The electric vehicle power battery thermal state prediction device in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 7 The electric vehicle power battery thermal state prediction device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0122] like Figure 7As shown, the electric vehicle power battery thermal state prediction device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the electric vehicle power battery thermal state prediction device. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the electric vehicle battery thermal state prediction device to communicate wirelessly or wiredly with other devices to exchange data. Although the figure shows an electric vehicle battery thermal state prediction device with various systems, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems can be implemented alternatively.

[0123] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0124] The electric vehicle power battery thermal state prediction device provided in this application, employing the electric vehicle power battery thermal state prediction method described in the above embodiments, can solve the technical problem of how to balance data fitting accuracy and thermodynamic consistency constraints in power battery thermal state prediction. Compared with the prior art, the beneficial effects of the electric vehicle power battery thermal state prediction device provided in this application are the same as those of the electric vehicle power battery thermal state prediction method provided in the above embodiments, and other technical features in this electric vehicle power battery thermal state prediction device are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0125] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0126] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0127] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, which are used to execute the electric vehicle power battery thermal state prediction method in the above embodiments.

[0128] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0129] The aforementioned computer-readable storage medium may be included in the electric vehicle power battery thermal state prediction device; or it may exist independently and not be assembled into the electric vehicle power battery thermal state prediction device.

[0130] The aforementioned computer-readable storage medium carries one or more programs. When the aforementioned one or more programs are executed by the electric vehicle power battery thermal state prediction device, the electric vehicle power battery thermal state prediction device: decomposes and reconstructs the original sensing signal of the power battery to obtain a battery reconstruction signal; inputs the battery reconstruction signal into an extended long short-term memory network to obtain a temperature prediction value, wherein the extended long short-term memory network is trained by a combination of data fidelity constraint terms and physical consistency constraint terms.

[0131] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0132] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0133] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0134] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described electric vehicle power battery thermal state prediction method. This solves the technical problem of how to balance data fitting accuracy with thermodynamic consistency constraints in power battery thermal state prediction. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the electric vehicle power battery thermal state prediction method provided in the above embodiments, and will not be repeated here.

[0135] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the electric vehicle power battery thermal state prediction method described above.

[0136] The computer program product provided in this application can solve the technical problem of how to balance data fitting accuracy and thermodynamic consistency constraints in predicting the thermal state of power batteries. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the electric vehicle power battery thermal state prediction method provided in the above embodiments, and will not be repeated here.

[0137] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A method for predicting the thermal state of a power battery for an electric vehicle, characterized in that, The method includes: The original sensing signals of the power battery are decomposed and reconstructed to obtain the battery reconstructed signal; The battery reconstruction signal is input into an extended long short-term memory network to obtain a temperature prediction value. The extended long short-term memory network is trained by combining data fidelity constraints and physical consistency constraints.

2. The method as described in claim 1, characterized in that, The process of decomposing and reconstructing the original sensing signal of the power battery to obtain the battery reconstruction signal includes: The original sensing signal of the power battery is decomposed to obtain the high-frequency detail coefficient sequence and the low-frequency approximation coefficient sequence; Based on the high-frequency detail coefficient sequence, a threshold is determined; The high-frequency detail coefficient sequence is shrunk according to the threshold to obtain the target high-frequency detail coefficient sequence. The target high-frequency detail coefficient sequence and the low-frequency approximation coefficient sequence are reconstructed to obtain the battery reconstruction signal.

3. The method as described in claim 2, characterized in that, Determining the threshold based on the high-frequency detail coefficient sequence includes: Based on the high-frequency detail coefficient sequence, the target decomposition layer detail coefficient sequence is obtained; Based on the target decomposition layer detail coefficient sequence, the background noise standard deviation is obtained; The threshold is obtained based on the preset time series window length and the background noise standard deviation.

4. The method as described in claim 1, characterized in that, Before inputting the battery reconstruction signal into the extended long short-term memory network to obtain the temperature prediction value, the method further includes: Obtain the physical weights corresponding to the current training round, the historical battery reconstruction features corresponding to historical sensor signals, the physical parameters to be trained, and the actual historical temperature values; The historical battery reconstruction features are input into a long short-term memory network to obtain historical temperature prediction values; Based on the physical parameters to be trained, the physical weights, the actual historical temperature values, the predicted historical temperature values, and the historical battery reconstruction features, the target physical parameters to be trained and the target long short-term memory network are obtained. The current training round is adjusted according to the preset round increment step size to obtain the target training round; The target physical parameters to be trained are used as the physical parameters to be trained, the target long short-term memory network is used as the long short-term memory network, and the target training round is used as the current training round. The steps of obtaining the physical weights corresponding to the current training round, the historical battery reconstruction features corresponding to the historical sensor signals, the physical parameters to be trained, and the historical actual temperature values ​​are returned to be executed until the current training round reaches the preset round threshold. Then, the target long short-term memory network is used as the extended long short-term memory network.

5. The method as described in claim 4, characterized in that, The step of obtaining the target physical parameters to be trained and the target long short-term memory network based on the physical parameters to be trained, the physical weights, the actual historical temperature values, the predicted historical temperature values, and the historical battery reconstruction features includes: Based on the physical parameters to be trained, the historical temperature prediction values, and the historical battery reconstruction features, the theoretical physical value of temperature is obtained. A composite loss function is constructed based on the physical weights, the actual historical temperature values, the predicted historical temperature values, and the theoretical physical temperature values. The optimization objective is to minimize the function value of the composite loss function to obtain the target neural network parameters and the target physical parameters to be trained. The parameters of the neural network to be trained in the long short-term memory network are updated according to the parameters of the target neural network to obtain the target long short-term memory network.

6. The method as described in claim 5, characterized in that, The method involves constructing a composite loss function based on the physical weights, the historical actual temperature values, the historical predicted temperature values, and the theoretical temperature values. The optimization objective is to minimize the function value of this composite loss function, resulting in the target neural network parameters and the target physical parameters to be trained, including: Based on the predicted historical temperature value and the actual historical temperature value, the data fidelity constraint term is determined; Based on the historical temperature predictions and the theoretical temperature values, the physical consistency constraints are determined. Based on the physical weights, the data fidelity constraint and the physical consistency constraint are weighted and combined to obtain a composite loss function; The optimization objective is to minimize the function value of the composite loss function to obtain the target neural network parameters and the target physical parameters to be trained.

7. The method as described in claim 1, characterized in that, The step of inputting the battery reconstruction signal into the extended long short-term memory network to obtain the temperature prediction value includes: Based on the battery reconfiguration signal, the battery reconfiguration characteristics are obtained; The battery reconstruction features are input into an extended long short-term memory network to obtain temperature prediction values. The activation function of the extended long short-term memory network is an exponential function, and the memory cells of the extended long short-term memory network store feature interaction information in the form of a covariance matrix.

8. A device for predicting the thermal state of an electric vehicle's power battery, characterized in that, The device includes: The preprocessing module is used to decompose and reconstruct the original sensing signals of the power battery to obtain the battery reconstruction signal; The output module is used to input the battery reconstruction signal into the extended long short-term memory network to obtain the temperature prediction value, wherein the extended long short-term memory network is trained by combining data fidelity constraint terms and physical consistency constraint terms.

9. A device for predicting the thermal state of an electric vehicle's power battery, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the electric vehicle power battery thermal state prediction method as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the electric vehicle power battery thermal state prediction method as described in any one of claims 1 to 7.

Images