Tensor Tucker decomposition-based soft package lithium ion battery temperature field prediction method and system

Abstract

The invention discloses a tensor Tucker decomposition-based soft package lithium ion battery temperature field prediction method and system, and the method comprises the steps: carrying out the time-space separation of the two-dimensional temperature field data of the surface of a battery through employing a tensor Tucker decomposition technology, and then building a neural network prediction model for estimating the future low-order time characteristics of a nominal system; and the prediction information of the future temperature field of the battery is reconstructed through Tucker decomposition inverse transformation. According to the method, a model simplification method based on tensor Tucker decomposition is adopted, the complex calculation problem that high-order data are directly modeled one by one is avoided, meanwhile, the two-dimensional space-time characteristics of the battery surface temperature field are considered, and the prediction modeling difficulty is greatly reduced while the model precision is guaranteed. The method disclosed by the invention is simple and easy to implement, is particularly suitable for the temperature field prediction problem of the large-scale soft package lithium ion battery, can better adapt to the modeling requirement of a computing resource limited system, and has a wide application prospect.

Description

Technical Field

[0001] The present invention belongs to the technical field of lithium-ion battery thermal management, and relates to a method for predicting the surface temperature field of a soft-pack lithium-ion battery, and specifically to a method and system for predicting the surface temperature field of a soft-pack lithium-ion battery based on tensor Tucker decomposition. Background Art

[0002] As energy storage and conversion components, batteries play an important role in industries such as aerospace, electronic equipment, and new energy vehicles. Among various batteries, lithium-ion batteries are widely used due to their high energy density and long cycle life. Temperature effects are the main factor limiting the performance of lithium-ion batteries. Battery thermal management technology is an emerging solution based on multiple disciplines and fields such as materials science, electrochemistry, heat transfer, and molecular dynamics. It aims to address thermal dissipation or thermal runaway that occurs when batteries operate under extreme temperature conditions (such as too high or too low), thereby optimizing and enhancing the overall performance and safety of the battery.

[0003] Predicting the battery surface temperature field is a key technology in battery thermal management. Traditional prediction methods use thermocouple probes embedded near the battery tabs to collect battery temperature data. Under different operating conditions, based on the impact of temperature on battery performance and combining the battery's electrochemical characteristics and heat generation mechanism, an equivalent circuit model is used to simplify the temperature field model into a lumped parameter model. However, due to the degradation of battery consistency caused by environmental factors such as battery aging, this lumped parameter model cannot accurately calculate the actual changes in the battery temperature field.

[0004] With the introduction of technologies like infrared thermal imaging, it's easier to set up multiple temperature measurement points on the battery surface, thereby obtaining more comprehensive information about the battery surface temperature field. Currently, research has been conducted on data-driven battery temperature field prediction methods. However, these studies focus on simplifying the battery thermal system into a one-dimensional system. These models cannot directly characterize the two-dimensional spatial temperature characteristics of the lithium-ion battery surface, resulting in low prediction efficiency. Summary of the Invention

[0005] This invention provides a method and system for predicting the temperature field of soft-pack lithium-ion batteries based on tensor Tucker decomposition. This method uses tensor Tucker decomposition to reduce the order of two-dimensional battery surface temperature field data. A neural network prediction model based on a nominal system is then established to predict future low-dimensional eigenvalues. This method then reconstructs spatiotemporal data to predict future information about the battery surface temperature field. While maintaining model prediction accuracy, this method significantly reduces the complexity of the prediction model and improves its computational efficiency.

[0006] The purpose of the present invention is achieved through the following technical solutions:

[0007] A method for predicting the temperature field of a soft-pack lithium-ion battery based on Tucker decomposition includes the following steps:

[0008] Step (1) arranging a sufficient number of sensors on the battery surface, performing a charge and discharge experiment on the lithium-ion battery over a period of time, using the sensors to collect temperature field data on the battery surface, and preprocessing the acquired temperature data to obtain a tensor form of the temperature field spatiotemporal data;

[0009] Step (2) performs Tucker decomposition with an initial rank of (1,1,1) on the spatiotemporal data of the lithium-ion battery temperature field to obtain a spatial modal decomposition factor matrix, a temporal modal factor matrix and a core tensor, and judges whether the inverse transformation result of the Tucker decomposition and the reconstruction error of the original data meet the accuracy requirements of the system; if not, discard the decomposition result and increase the rank of the tensor decomposition, repeat the above Tucker decomposition steps until the reconstruction error meets the accuracy requirements, extract the spatial modal factor matrix as the spatial characteristic matrix of the original data, and the temporal modal factor matrix as the low-order temporal characteristic matrix;

[0010] Step (3) constructs a neural network prediction model based on GRU (Gated Recurrent Unit), which consists of a recurrent network module and a feature selection module, wherein:

[0011] The recurrent network module includes two GRU layers, and a Hardswish activation function is connected after each GRU layer;

[0012] The feature selection module consists of a linear layer and an activation function. The linear layer performs a weighted summation on the output of the recurrent network module, converting the features extracted by the recurrent network module into the final prediction value. Finally, the activation function is used to further improve the nonlinear expression ability of the model.

[0013] Step (4) Given the current value at the current moment, the low-order time characteristic matrix obtained by Tucker decomposition and the constructed neural network prediction model are applied, and the external input variable sequence is combined for prediction modeling to predict the low-order time characteristic matrix at the future moment, and it is reconstructed with the spatial characteristic matrix and the core tensor by dimensionality increase to obtain the high-order temperature field prediction tensor at the future moment.

[0014] Compared with the prior art, the present invention has the following advantages:

[0015] 1. The model building process of the present invention is divided into three stages: model reduction, predictive modeling, and order-raising reconstruction. In the model reduction stage, as the rank selected by Tucker decomposition increases, the number of columns of the decomposition factor matrix continues to increase. By using the method of checking the reconstruction error, as the rank increases, the decomposition algorithm can better capture the spatial and temporal characteristics of the system, thereby improving the accuracy of the neural network prediction model. The neural network prediction model is trained based on low-order time characteristics and external input variable sequences. Finally, the predicted low-order time characteristic matrix is ​​reconstructed by the inverse transformation of Tucker decomposition to obtain a high-dimensional temperature field prediction tensor.

[0016] 2. The present invention adopts a model simplification method based on tensor Tucker decomposition, avoiding the complex calculation problem of directly modeling high-order data one by one. At the same time, it takes into account the two-dimensional spatiotemporal characteristics of the battery surface temperature field, greatly reducing the difficulty of predictive modeling while ensuring model accuracy.

[0017] 3. The method of the present invention is simple and easy to implement, and is particularly suitable for the temperature field prediction problem of large soft-pack lithium-ion batteries. Both theoretical analysis and experimental results have proved that the predictive modeling method of the present invention is a highly efficient temperature field modeling method that can better adapt to the modeling requirements of systems with limited computing resources and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a diagram of tensor Tucker decomposition;

[0019] Figure 2 It is a diagram of the neural network structure for time series prediction;

[0020] Figure 3 is the computational graph of the gated recurrent unit;

[0021] Figure 4 It is the battery system control process;

[0022] Figure 5 is the heat transfer coefficient h(t);

[0023] Figure 6 is the charge and discharge current i(t);

[0024] Figure 7 The original tensor is sliced ​​and visualized along the time dimension;

[0025] Figure 8 It is the visualization of the error slice of the tensor Tucker decomposition reconstruction;

[0026] Figure 9 It is the visualization of the time modal decomposition factor matrix;

[0027] Figure 10 It is the prediction effect diagram of time mode t1;

[0028] Figure 11 It is the prediction effect diagram of time mode t2;

[0029] Figure 12 It is the prediction effect diagram of time mode t3;

[0030] Figure 13 is a slice visualization of the modeling error. DETAILED DESCRIPTION

[0031] The technical solution of the present invention is further described below with reference to the accompanying drawings, but is not limited thereto. Any modification or equivalent replacement of the technical solution of the present invention that does not depart from the spirit and scope of the technical solution of the present invention should be included in the scope of protection of the present invention.

[0032] The present invention provides a method for predicting the temperature field of soft-pack lithium-ion batteries based on tensor Tucker decomposition. First, the two-dimensional temperature field data of the battery surface is temporally and spatially separated using tensor Tucker decomposition technology. Then, a neural network prediction model is established to estimate the future low-order time characteristics of the nominal system. Finally, the predicted information of the battery's future temperature field is reconstructed through the inverse Tucker decomposition transform. The method specifically includes the following steps:

[0033] Step (1) Arrange a sufficient number of sensors on the battery surface, conduct a charge and discharge experiment on the lithium-ion battery over a period of time, use the sensors to collect temperature field data on the battery surface, and pre-process the acquired temperature data to obtain the tensor form of the temperature field spatiotemporal data. The specific steps are as follows:

[0034] Step (11) performs a charge and discharge experiment on the lithium ion battery over a period of time. Assume that there are enough temperature sensors arranged on the battery surface. Without loss of generality, assume that m×n temperature sensors are evenly arranged on the x-axis and y-axis, and the total number of steps in the charge and discharge process is k. Given the charge and discharge current is Battery surface heat transfer coefficient The temperature data set collected from the m×n temperature sensors of the battery at k moments is T temp (x,y,t), m is the number of temperature sensors on the x-axis, and n is the number of temperature sensors on the y-axis.

[0035] Step (12) is to calculate the temperature data set T temp (x, y, t) performs data preprocessing, which includes the following three steps:

[0036] First, check whether there are any missing values ​​in the data. If so, fill them with the mean of the temperature matrix at the current moment.

[0037] Secondly, check whether the elements in the tensor contain outliers. Since the temperature field has a hysteresis characteristic, the temperature point data between adjacent moments and adjacent positions will not change suddenly. Therefore, if the data contains noise data greater than 3 times the standard deviation, it will be replaced with the mean of the temperature matrix at the current moment.

[0038] Finally, in order to make the Tucker decomposition process and the back propagation process in the prediction model more stable, the data needs to be normalized to the maximum and minimum values ​​so that the data range is scaled between 0 and 1, and the tensor form of the temperature data is obtained. The maximum-minimum normalization formula is:

[0039]

[0040] Among them, max(T temp ) and min(T temp ) represent the data sets T temp The maximum and minimum values ​​in (x,y,t).

[0041] Step (2) performs Tucker decomposition with an initial rank of (1,1,1) on the spatiotemporal data of the lithium-ion battery temperature field to obtain the spatial modal decomposition factor matrix, the time modal factor matrix and the core tensor, and judges whether the inverse transformation result of the Tucker decomposition and the reconstruction error of the original data meet the accuracy requirements of the system; if not, discard the decomposition result and increase the rank of the tensor decomposition, repeat the above Tucker decomposition steps until the reconstruction error meets the accuracy requirements, extract the spatial modal factor matrix as the spatial characteristic matrix of the original data, and the time modal factor matrix as the low-order time characteristic matrix. The specific steps are as follows:

[0042] Step (21) According to the tensor Tucker decomposition theory, the rank r (r x ×r y ×r t )∈N + , where r x 、r y and r t is the rank on the three modes.

[0043] Step (22) is based on the initial rank r x =r y =r t = 1 to start Tucker decomposition, and judge whether the reconstruction error of the inverse transform result of Tucker decomposition and the original data meets the accuracy requirements of the system; if not, discard the decomposition result and increase the rank of tensor decomposition, repeat the above Tucker decomposition steps until the reconstruction error meets the accuracy requirements, and obtain the final r x 、r y and rt , then the high-order space-time tensor Z(x,y,t) can be decomposed into core tensors The product of the decomposition factor matrix on each mode is shown in the Tucker decomposition diagram. Figure 1 Therefore, the expression of Tucker decomposition of the third-order tensor Z(x,y,t) is:

[0044]

[0045] in, Represents the approximate tensor obtained by Tucker decomposition and reconstruction of the original tensor Z(x, y, t); g is called the core tensor, and each element of it represents the degree of interaction between different components; and Represents the decomposition factor matrix of the original tensor Z(x, y, t) under three modes. The decomposition factor matrix is ​​usually orthogonal and can be regarded as the principal component along the corresponding mode.

[0046] In formula (1), “× n The "N-module product" operation is a tensor-to-matrix multiplication operation, which is used to represent the product between a tensor and a matrix. Its mathematical definition is to multiply each slice of the tensor on the n-th mode by the matrix, that is:

[0047]

[0048] Where, represents the rth time modal factor matrix T t A quantity, The operation represents the outer product, which is an operation that combines two vectors into a higher-dimensional tensor. The elements of the outer product of a tensor are represented as follows:

[0049]

[0050] Step (23) extracts the spatial modal factor matrix as the spatial characteristic matrix of the original data, and the temporal modal factor matrix as the low-order temporal characteristic matrix.

[0051] Step (3) constructs a neural network prediction model based on GRU (Gated Recurrent Unit).

[0052] Step (4) Given the current value at the current moment, the low-order time characteristic matrix obtained by Tucker decomposition and the constructed neural network prediction model are applied, and the external input variable sequence is combined for prediction modeling to predict the low-order time characteristic matrix at the future moment, and it is reconstructed with the spatial characteristic matrix and the core tensor by dimensionality increase to obtain the high-order temperature field prediction tensor at the future moment.

[0053] In the present invention, the process of predictive modeling is to construct a lookback window of size w at each sampling moment k, and to construct a neural network prediction model based on the true value of the time component from the kth moment to the previous kw moments and the external factors of the system. The purpose is to enable the model to obtain a single-step predicted temperature field tensor based on the system input and the current state of the controlled object.

[0054] Assume that the value of the time modal factor matrix T(t,r) at point (p,γ) is t γ (p), where γ represents the γth component in the T(t,r) matrix and p represents the pth time step, then 1≤p≤k and 1≤γ≤r t The prediction process takes into account the charge and discharge current i(t) of the battery system, the battery surface heat transfer coefficient h(t), and the rth order of the low-order time characteristic matrix. t Quantity Constitute the following external input sequence data set samples

[0055]

[0056] Where w is the size of the lookback window. Establish a time series prediction data set, let w<p<k, then the feature window constructed at time p is u(p)=[t γ (p)...t γ (pw),i(p)...i(pw),h(p)...h(pw)]. Prediction model through neural network Modeling, we get the following functional relationship:

[0057]

[0058] Where, It is a set of neural network prediction models obtained through time series prediction, representing the use of time characteristic matrix components t γ The predicted value of the time characteristic matrix component at the next moment predicted by the previous w historical values ​​in the external input variable sequence All r t of Aggregate and merge into a predicted low-order time matrix Obtained Expands to the following form:

[0059]

[0060] Finally, the high-order temperature prediction tensor for the future moment is obtained by Tucker's inverse transform. Expressed as:

[0061]

[0062] For the prediction tensor Perform inverse normalization to obtain high-order temperature field data on the original feature space The predicted value at the next p+1 time. Inverse normalization formula:

[0063]

[0064] In the present invention, the spatiotemporal modeling process can be described as:

[0065] 1. Tucker decomposition time and space separation:

[0066] In this step, Tucker decomposition can be written in the following matrix form:

[0067]

[0068] Where Z (n) Represents the matrixization of the tensor Z in the nth mode. For n-order tensors n=1,2,3,..., the matrixization of tensors is also called tensor expansion or tensor flattening, which is the process of expanding a high-order tensor into a matrix along a certain mode. For example, Z (1) Expand the first modulus of Z, keeping the first dimension of the tensor as the row index of the matrix, and the second and third dimensions as the column matrix of the matrix. The resulting matrix The operation represents the Kronecker product, also known as the direct product or tensor product. For a vector of shape t×r t Matrix and y×r y Matrix The Kronecker product results in a shape of ty×r t r y The matrix is ​​defined as:

[0069]

[0070] Since the ultimate goal of Tucker decomposition is to reconstruct the tensor The reconstruction error between the original tensor Z is as small as possible, so the process of solving the decomposition factor can be written as an optimization problem:

[0071]

[0072] Here, ||·|| represents the norm of the tensor, which is similar to the Frobenius norm of a matrix, that is, taking the square root of the sum of all squared elements:

[0073]

[0074] In the above formula, the tensor E is the reconstruction tensor The error representation obtained by subtracting the original tensor Z, (E) αβγ is the element representation of the tensor E.

[0075] To iteratively solve the decomposition factor matrix using the optimization method, it is necessary to first convert the minimum optimization problem into a maximum optimization problem and write the objective function (12) in square form:

[0076]

[0077] The core tensor g of the definition (2) decomposed by Tucker satisfies:

[0078] g=Z×1X T ×2Y T ×3T T (15)

[0079] Substituting the core tensor g into the result of formula (14), we can obtain:

[0080]

[0081] Since ||Z|| is a constant, the minimum optimization objective of Equation (12) can be redefined as a series of sub-problems containing the following maximization problems:

[0082] max g,X,Y,T ||Z×1X T ×2Y T ×3T T || (17)

[0083] Furthermore, the above formula can be written in matrix form:

[0084]

[0085] The above optimization problem is solved using the alternating least squares method. The initial solution of the decomposition factor matrix can be determined by singular value decomposition. It can be solved by simply defining the decomposition factors as the first r left singular vectors of W. Next, the factor matrix of each mode is updated in turn until convergence. This step is cyclic, that is, in each iteration, the factor matrix of the first mode is updated, followed by the second mode, and so on, until the factor matrices of all modes are updated. Finally, the core tensor g is determined by Equation (15). This method will converge to a solution so that the objective function no longer decreases.

[0086] 2. Low-order temporal characteristics prediction modeling

[0087] The present invention adopts Figure 2 Constructed neural network prediction model Identify the low-order time characteristics of the temperature system. In the Tucker decomposition stage, the decomposition results include the low-order time characteristic matrix is the rth element in the low-order time characteristic matrix T(t,r) t column vectors, construct a lookback window of size w, then The prediction modeling process can be described as: combining the charge and discharge current i(t), the battery surface heat transfer coefficient h(t) and the rth order of the low-order time characteristic matrix t Quantity Construct an external input sequence data set sample u(t) and train a neural network prediction model with a recurrent network module and a feature selection module The prediction process expression is:

[0088]

[0089] Neural network prediction model It consists of two modules: a recurrent network module and a feature selection module. The recurrent network module contains two GRU layers, and a Hardswish activation function is connected after each GRU layer. The recurrent network module introduces a gating mechanism through the GRU layer to solve the gradient vanishing and gradient exploding problems in traditional recurrent neural networks and provide a method for effectively capturing long-term dependencies. Combined with the smoothing gradient feature of the Hardswish activation function, the low-order model in the present invention can better capture long-term dependencies when processing long sequence data. Hardswish function calculation formula:

[0090]

[0091] The feature selection module consists of a linear layer and an activation function. The features output by the GRU layer are high-dimensional, and the network contains multiple hidden layers with large hidden units. Ultimately, these high-dimensional features need to be mapped to the dimensions of a single-step prediction. Secondly, the output of the GRU layer contains rich temporal information and features, but this information is often distributed across multiple time steps and hidden layer units. The linear layer integrates these dispersed temporal features by performing a weighted summation of the output of the recurrent network module, providing a comprehensive output. In other words, the linear layer can be viewed as a "decision layer" that transforms the features extracted by the recurrent network module into the final prediction value. Finally, the activation function is used to further increase the model capacity.

[0092] The calculation process of the GRU layer is as follows Figure 3 As shown, it contains two key gates: the update gate λ and the reset gate z. The update gate determines how much old information the model needs to remember and integrate in the current step, while receiving new information. The calculation formula of the update gate is:

[0093]

[0094] Among them, x t is the input vector of the tth time step, that is, the input sequence The tth component of , it will undergo a linear transformation, that is, with the weight matrix W (z) Multiply. Hidden state h t-1 The information stored is the information from the previous time step t-1, which also undergoes a linear transformation. The updater adds these two parts of information and inputs them into the Sigmoid activation function, thus compressing the result between 0 and 1. The formula of the Sigmoid activation function is:

[0095]

[0096] The reset gate determines how much old information the model needs to ignore and reset the hidden state in the current step. t-1 and x t First, it undergoes a linear transformation, then adds and inputs the Sigmoid activation function to output the activation value, and resets the calculation formula of the gate:

[0097]

[0098] In the use of the reset gate, the new memory content will use the reset gate to store the past related information. t With h t-1 The Hadamard product of the corresponding elements is the product of the corresponding elements. Since the reset gate calculated in the above formula is a vector consisting of 0-1, it will measure the size of the gate opening. The Hadamard product will determine the previous information to be retained and the current memory content:

[0099] h t '=tanh(x t W x +(λ t ⊙h t-1 )W h +b) (24)

[0100] According to the above formula, if z = 0, the new hidden state h t 'Depends only on the input x t , that is, the past hidden state will be completely ignored and no past memory will be retained. The calculation of the final memory requires the use of the update gate λ to determine the current memory content h t ' and the previous time step hidden state h t-1 The information that needs to be collected. This process can be expressed as:

[0101] h t =λ t ⊙ht-1 +(1-λ t )⊙h t ' (25)

[0102] Among them, λ t To update the activation result of the gate, it controls the inflow of information in the form of gating. t With h t-1 The Hadamard product represents the information retained in the final memory from the previous time step. The newly added information is weighted, and this information plus the information retained in the final memory equals the final GRU layer output. The GRU layer does not clear previous information over time; it retains relevant information and passes it to the next unit. Therefore, it utilizes all information and avoids the vanishing gradient problem.

[0103] Since the time series prediction method used in the present invention is a single-step prediction of multiple input variables, the input features are n=2 (n depends on the number of external input variables considered in the prediction model), the number of features of the GRU layer prediction output depends on the size of the hidden layer, and the output of the prediction model in the single-step prediction is a single value. Therefore, the model structure diagram is as follows Figure 2 As shown in Figure 1, it is necessary to connect a feature selection module consisting of linear layers after the recurrent network module to perform feature reduction, so as to obtain the output of single-step prediction while increasing the nonlinear expression ability of the model F.

[0104] In the present invention, the definitions of the above parameters are shown in Table 1:

[0105] Table 1

[0106]

[0107]

[0108] The present invention also provides a soft-pack lithium-ion battery temperature field prediction system based on tensor Tucker decomposition for implementing the above method. The system includes a data acquisition module, a neural network prediction model construction module, a temperature field prediction module, a time-space coupling module, and a battery charge and discharge control module, wherein:

[0109] The data acquisition module is used to acquire battery surface temperature field data and perform preprocessing to obtain the battery temperature field spatiotemporal tensor;

[0110] The neural network prediction model construction module is used to construct a neural network prediction model;

[0111] The temperature field prediction module is used to predict the newly given input current using the established neural network prediction model to obtain the future low-order time characteristic prediction matrix;

[0112] The spatiotemporal coupling module is used to perform tensor Tucker decomposition and inverse transformation of temperature field data to obtain future temperature field data;

[0113] The battery charge and discharge control module is used to control the battery charge and discharge process.

[0114] Example:

[0115] This embodiment provides a predictive modeling method based on a tensor Tucker decomposition-gated loop structure, the method comprising the following steps:

[0116] Step (1) collects a lithium-ion battery surface temperature data set, preprocesses the data, and organizes it into a tensor form.

[0117] Step (2) performs Tucker decomposition with an initial rank of (1,1,1) on the spatiotemporal data of the lithium-ion battery temperature field to obtain a spatial modal decomposition factor matrix, a temporal modal factor matrix and a core tensor, and determines whether the reconstruction error between the inverse transformation result of the Tucker decomposition and the original data meets the accuracy requirements of the system; if not, discard the decomposition result and increase the rank of the tensor decomposition, repeat the above steps until the reconstruction error meets the accuracy requirements, extract the two spatial modal factor matrices as the spatial characteristic matrix of the original data, and the temporal modal factor matrix as the low-order temporal characteristic matrix.

[0118] Step (3) uses the low-order time characteristic matrix obtained by Tucker decomposition and combines it with the external input variable sequence to train the neural network prediction model, thereby predicting the low-order time characteristic matrix of the system in the future.

[0119] Step (4) reconstructs the predicted low-order time characteristic matrix, the spatial modal factor matrix, and the core tensor by increasing the dimension to obtain a high-order temperature prediction tensor for the future moment. Finally, the predicted tensor is compared with the true value tensor of the temperature field to verify the accuracy of the modeling method.

[0120] To demonstrate the effectiveness of the modeling method based on tensor Tucker decomposition combined with a gated loop structure in this embodiment, a large soft-pack lithium-ion battery is used as an example to study the temperature field control problem during the charging and discharging process under different current conditions in a variable thermal coefficient environment, illustrating that this method has the advantages of high efficiency and stability in dealing with such problems.

[0121] The test conditions are described as follows:

[0122] Consider a two-dimensional temperature field system on the surface of a large square soft-pack lithium-ion battery with a size of 180mm*240mm, simulated by a pseudo-two-dimensional model. During the experiment, the initial charge of the battery was 0.1C and the initial temperature was 20°C. The finite element analysis method was used to solve the time domain for 3600s, with a period of 1s. A total of 20×20 temperature probes were arranged on the battery surface. The battery system control process is as follows: Figure 4 As shown, a bidirectional electronic load is connected to the lithium-ion battery, and the battery surface temperature is measured by a thermal imager. After compensation, the temperature is input into a data acquisition and processing device for modeling, and the current i(t) is finally obtained and fed back to the bidirectional electronic load.

[0123] Perform charge and discharge experiments on the battery, and give the battery surface heat transfer coefficient h(t), such as Figure 5 As shown, given the charge and discharge current i(t), as Figure 6 As shown, the battery is numerically simulated by the finite element method to obtain the two-dimensional temperature field data T temp (x,y,t), which is preprocessed and organized into third-order tensor data Slice the obtained Z(x,y,t) tensor along the time dimension, and the visualization result is as follows Figure 7 shown.

[0124] Result analysis:

[0125] First, given the accuracy requirement that the reconstruction error is less than 0.9°, the original tensor Z(x, y, t) is subjected to tensor Tucker decomposition, and the rank is determined by the inspection reconstruction error method adopted by the present invention. is (4, 4, 3), that is, r x =4, r y =4, r t =3, and its reconstruction error is 0.15℃, which just meets the accuracy requirement. The reconstruction error slice obtained by inverse transformation and dimensionality increase after Tucker decomposition is visualized as follows Figure 8 As shown. Figure 8 Analysis shows that although the Tucker algorithm is a linear algorithm, the reconstruction accuracy still meets the requirements, with a maximum reconstruction error of 0.15°C. The error is concentrated at the lower left boundary point in spatial location and at the time point of the temperature surge in temporal variation.

[0126] Secondly, the decomposition factor matrix T(t,r) of Tucker decomposition on the time mode is as follows Figure 9 As shown in Figure 2, the algorithm captures the overall temporal trend of all probe temperatures across the entire space by observing the first component of the time decomposition factor matrix. Components 2 and 3, on the other hand, show the spatial temperature trends as input current and thermal convection coefficients change.

[0127] Then, the three time modal components t1, t2 and t3 are modeled separately to obtain the predicted low-order time matrix The results of time series prediction for the three time components are as follows Figure 10-12 As shown in the experimental results, it can be clearly seen that the model designed by the present invention The time modal factor matrix obtained by Tucker decomposition was effectively predicted. The red dotted line in the figure represents the prediction result of the training set, the blue dotted line represents the prediction result of the test set, and the black solid line represents the true value. Judging from the performance of the training set, the red dotted line and the black solid line are basically consistent, indicating that the model effectively captured the main features of the time series during the training process and successfully learned the dynamic changes of the time modal factor matrix. In the prediction results of the test set, although the blue dotted line has certain deviations and fluctuations from the true value (black solid line) in some time periods, especially in some intervals with drastic short-term changes, the overall trend is still relatively consistent. This shows that the model Not only does it perform well on the training data, it also demonstrates good generalization capabilities on the test set.

[0128] Although there are some fluctuations on the test set, the model It can still capture the long-term trend of the data well, especially on a larger time scale, the predicted value of the model maintains a high degree of consistency with the true value. This is an important advantage of the prediction model structure in this invention, which can effectively capture long-term dependencies through the gating mechanism and show good robustness when dealing with possible noise in the time series. In this experiment, the prediction model The reason why it can achieve good performance is mainly due to the following reasons: First, the present invention regulates the flow of information through its gating mechanism, which can effectively avoid the problem of gradient disappearance, thereby maintaining the memory of information for longer time steps, which is particularly important when processing spatiotemporal data with long-term dependencies; Second, the present invention uses the Bayesian optimization method to systematically tune the hyperparameters to ensure that the model can be trained under the optimal combination of various hyperparameters. The Bayesian hyperparameter optimizer searches for the key parameters of the model, such as the learning rate, hidden layer dimension, and regularization coefficient, and makes fine adjustments, thereby effectively improving the convergence speed of the model and the final prediction performance; Third, in the model The Hardswish activation function used in [1] is a new type of activation function that has been widely used in recent years. It not only has the nonlinear advantages of the ReLU activation function, but also has lower computational cost and better smoothing properties. This function can provide a more stable gradient flow during model training, making the model more robust when learning complex data.

[0129] Finally, the three models Time component of the forecast as well as Merge into a predicted low-order time matrix And perform outer product with the spatial decomposition factor matrix X, Y and core tensor g obtained by Tucker decomposition before, and the predicted high-order data can be obtained The spatiotemporal error function ζ, RMSE (Root Mean Square Error), and MER (Mean Error Rate) are defined as follows:

[0130]

[0131] MER is based on the ratio of the maximum absolute error between the predicted tensor and the original tensor to the numerical range of the original tensor. This indicator pays special attention to the performance of the model in the worst case. In this embodiment, the MER of the algorithm prediction tensor and the true value tensor is only 4.90031%. It proves that even considering the data points with the greatest modeling difficulty, the model can still make good predictions. In the battery temperature field prediction task, the prediction results do not produce phenomena that violate the hysteresis characteristics, such as sudden changes in temperature distribution. RMSE is one of the commonly used indicators for evaluating the accuracy of prediction models. It measures the degree of deviation between the predicted value and the true value. In this embodiment, The root mean square error between the predictions and Z(x,y,t) is 0.05823, which proves that the prediction error at each point is very small and the overall modeling accuracy is high.

[0132] Model prediction tensor The slice visualization of the error tensor ζ(x,y,t) obtained by subtracting the true value tensor Z(x,y,t) is as follows Figure 13 As shown. By observing Figure 13 The mean square error (MSE) shows a maximum error of only 0.8°C, and the complete algorithm runs for only 98 seconds, precisely meeting the modeling method's design objectives. Both theoretical analysis and experimental results demonstrate the method's high computational efficiency and excellent prediction accuracy, making it ideal for online modeling of spatially distributed processes in systems with limited computational resources.

Claims

1. A method for predicting the temperature field of soft-pack lithium-ion batteries based on tensor Tucker decomposition, characterized in that The method comprises the following steps: Step (1) arranging a sufficient number of sensors on the battery surface, performing a charge and discharge experiment on the lithium-ion battery over a period of time, using the sensors to collect temperature field data on the battery surface, and preprocessing the acquired temperature data to obtain a tensor form of the temperature field spatiotemporal data; Step (2) performs Tucker decomposition with an initial rank of (1,1,1) on the spatiotemporal data of the lithium-ion battery temperature field to obtain a spatial modal decomposition factor matrix, a temporal modal factor matrix and a core tensor, and determines whether the inverse transformation result of the Tucker decomposition and the reconstruction error of the original data meet the accuracy requirements of the system; if not, discard the decomposition result and increase the rank of the tensor decomposition, repeat the above Tucker decomposition steps until the reconstruction error meets the accuracy requirements, extract the spatial modal factor matrix as the spatial characteristic matrix of the original data, and the temporal modal factor matrix as the low-order temporal characteristic matrix; Step (3) constructs a neural network prediction model based on GRU, wherein the neural network prediction model consists of a recurrent network module and a feature selection module, wherein: The recurrent network module includes two GRU layers, and a Hardswish activation function is connected after each GRU layer; The feature selection module is composed of a linear layer and an activation function. The linear layer performs weighted summation on the output of the recurrent network module, converts the features extracted by the recurrent network module into the final prediction value, and finally further improves the nonlinear expression ability of the model through the activation function. Step (4) Given the current value at the current moment, the low-order time characteristic matrix obtained by Tucker decomposition and the constructed neural network prediction model are applied, and the external input variable sequence is combined for prediction modeling to predict the low-order time characteristic matrix at future moments, and it is reconstructed with the spatial characteristic matrix and the core tensor by dimensionality increase to obtain the high-order temperature field prediction tensor at future moments.

2. The method for predicting the temperature field of a soft-pack lithium-ion battery based on tensor Tucker decomposition according to claim 1 is characterized in that The specific steps of step (1) are as follows: Step (11) Assume that m×n temperature sensors are evenly arranged on the x-axis and y-axis, the total number of steps in the charge and discharge process is k, and the given charge and discharge current is Battery surface heat transfer coefficient The temperature data set collected from m×n temperature sensors of the battery at k moments is T temp (x,y,t), m is the number of temperature sensors on the x-axis, n is the number of temperature sensors on the y-axis; Step (12) performs temperature data set T temp (x,y,t) for data preprocessing: First, check whether there are any missing values ​​in the data. If there are any, fill them with the mean value of the temperature matrix at the current moment; Secondly, check whether the elements in the tensor contain outliers. If the data contains noise data greater than 3 times the standard deviation, replace it with the mean of the temperature matrix at the current moment; Finally, the data is normalized to the maximum and minimum values ​​so that the data range is scaled between 0 and 1 to obtain the tensor form of the temperature data.

3. The method for predicting temperature field of soft-pack lithium-ion batteries based on tensor Tucker decomposition according to claim 1 is characterized in that The specific steps of step (2) are as follows: Step (21) According to the tensor Tucker decomposition theory, the rank r (r x × y × t )∈N + , where r x 、r y and r t is the rank on the three modes; Step (22) is based on the initial rank r x =r y =r t = 1 to start Tucker decomposition, and determine whether the inverse transformation result of Tucker decomposition and the reconstruction error of the original data meet the accuracy requirements of the system; if not, discard the decomposition result and increase the rank of tensor decomposition, repeat the above Tucker decomposition steps until the reconstruction error meets the accuracy requirements, and get the final r x 、r y and r t , then Z(x,y,t) is decomposed into core tensors Multiplication with the factorization matrix on each mode; Step (23) extracts the spatial modal factor matrix as the spatial characteristic matrix of the original data, and the temporal modal factor matrix as the low-order temporal characteristic matrix.

4. The method for predicting temperature field of soft-pack lithium-ion batteries based on tensor Tucker decomposition according to claim 3 is characterized in that In step (22), the Tucker decomposition expression of Z(x, y, t) is: in, Represents the approximate tensor obtained by Tucker decomposition and reconstruction of the original tensor Z(x,y,t); and Represents the decomposition factor matrix of the original tensor Z(x,y,t) in three modes; "× n The "operation is a tensor n-module product, which is used to represent the product between a tensor and a matrix. Its mathematical definition is to multiply each slice of the tensor on the nth mode by the matrix, that is: In the formula, represents the rth time modal factor matrix T t Quantity, The operation represents the outer product, and the elements of the tensor outer product are represented as follows:

5. The method for predicting temperature field of soft-pack lithium-ion batteries based on tensor Tucker decomposition according to claim 1, characterized in that The specific steps of step (4) are as follows: Step (41) Assume that the value of the time modal factor matrix T(t,r) at point (p,γ) is t γ (p), where γ represents the γth component in the T(t,r) matrix and p represents the pth time step, then 1≤p≤k and 1≤γ≤r t ; Step (42) considers the charge and discharge current i(t) of the battery system, the heat transfer coefficient h(t) of the battery surface, and the rth element of the low-order time characteristic matrix t Quantity The following external input sequence data set samples are constructed Where w is the size of the lookback window; Step (43) establishes a time series prediction data set, and sets w<p<k. Then the feature window constructed at time p is u(p)=[t γ (p)...t γ (pw),i(p)...i(pw),h(p)...h(pw)]; Step (44) is modeled by a neural network prediction model to obtain the following modeling function relationship: In the formula, It is a neural network prediction model, which represents the use of time characteristic matrix component t γ The predicted value of the time characteristic matrix component at the next moment predicted by the previous w historical values ​​in the external input variable sequence Step (45) convert all r t The predicted sequence of components Aggregate and merge into a predicted low-order time matrix What you get Expands to the following form: Step (46) reconstructs the high-order temperature prediction tensor at the future time by up-dimensionality through Tucker's inverse transform It is expressed as: Step (47) performs the prediction tensor Perform inverse normalization to obtain high-order temperature field data on the original feature space The predicted value at time p+1 in the future.

6. A soft-pack lithium-ion battery temperature field prediction system based on tensor Tucker decomposition that implements the method described in any one of claims 1 to 5, characterized in that The system includes a data acquisition module, a neural network prediction model construction module, a temperature field prediction module, a time-space coupling module and a battery charge and discharge control module, wherein: The data acquisition module is used to acquire the battery surface temperature field data and perform preprocessing to obtain the battery temperature field spatiotemporal tensor; The neural network prediction model building module is used to build a neural network prediction model; The temperature field prediction module is used to predict the newly given input current using the established neural network prediction model to obtain the future low-order time characteristic prediction matrix; The spatiotemporal coupling module is used to perform tensor Tucker decomposition and inverse transformation of temperature field data to obtain future temperature field data; The battery charge and discharge control module is used to control the battery charge and discharge process.

Images