Lithium-battery temperature field spatio-temporal modeling method based on EMD-lstm
Through the EMD-LSTM method, KWLTSA dimensionality reduction and EMD algorithm decompose spatiotemporal data, combined with LSTM training, the application difficulties and gradient problems of traditional spatiotemporal modeling in strong nonlinear systems are solved, and high-precision spatiotemporal construction of lithium battery temperature field is realized mold.
Patent Information
- Application Number
- PCT/CN2024/079325
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2023-05-04
- Filing Date
- 2024-02-29
- Publication Date
- 2024-11-07
AI Technical Summary
Traditional spatiotemporal modeling methods are difficult to apply to distributed parameter systems with strong nonlinear unstable conditions, as well as the problems of gradient vanishing and gradient explosion in long time series.
Using the spatiotemporal modeling method of lithium battery temperature field based on EMD-LSTM, through KWLTSA dimensionality reduction, EMD algorithm and LSTM training, the spatiotemporal data are decomposed to obtain the predicted time coefficient and spatial basis function, and the spatiotemporal reconstruction is carried out to construct the temperature field model of lithium battery .
Effectively dealing with strong nonlinear unstable distributed parameter system improves the accuracy of spatial and temporal modeling of the temperature field of lithium batteries, and solves the problems of gradient vanishing and gradient explosion in long time series.
Smart Images

Figure CN2024079325_07112024_PF_FP_ABST
Abstract
Description
A spatiotemporal modeling method for lithium battery temperature field based on EMD-LSTM
[0001] CROSS-REFERENCE TO RELATED APPLICATIONS
[0002] This application claims priority to Chinese patent application number 202310494038.8, filed with the Patent Office of China on May 4, 2023, entitled “A method for spatiotemporal modeling of lithium battery temperature field based on EMD-LSTM,” the entire contents of which are incorporated herein by reference. Technical Field
[0003] The present invention relates to the technical field of distributed parameter system spatiotemporal modeling, and in particular to a spatiotemporal modeling method of a lithium battery temperature field based on EMD-LSTM. Background Art
[0004] Numerous spatiotemporal modeling methods have been studied in recent years. Among them, modeling methods based on spatiotemporal separation have been widely used and proven to be feasible and effective for modeling distributed parameter systems (DPSs). The key idea behind these methods is that spatiotemporal variables can be decomposed into a series of spatial basis functions and corresponding temporal coefficients using Fourier transforms. Therefore, the spatial basis functions and temporal coefficients can be learned separately using either mechanism-based or data-based algorithms.
[0005] Traditional spatiotemporal modeling methods have satisfactory modeling performance for industrial thermal processes, but they are still difficult to apply to strongly nonlinear and non-stationary distributed parameter systems, as well as the problems of gradient vanishing and gradient exploding in long time series.
[0006] Summary of the Invention
[0007] In response to the above-mentioned defects, the present invention proposes a spatiotemporal modeling method for the lithium battery temperature field based on EMD-LSTM, which aims to solve the problems that traditional spatiotemporal modeling methods are difficult to apply to strongly nonlinear and unstable distributed parameter systems, as well as the problems of gradient vanishing and gradient exploding in long time series.
[0008] To achieve this object, the present invention adopts the following technical solutions:
[0009] A spatiotemporal modeling method for lithium battery temperature field based on EMD-LSTM, comprising the following steps:
[0010] Step S1: Use the KWLTSA dimensionality reduction method to convert high-dimensional spatiotemporal data into low-dimensional spatiotemporal data and obtain spatial basis functions;
[0011] Step S2: Project the low-dimensional spatiotemporal data onto the spatial basis function to obtain the corresponding low-order time coefficients;
[0012] Step S3: Decompose the low-order time coefficients using the EMD algorithm to obtain IMF components and residual components;
[0013] Step S4: Perform LSTM training on the IMF component and the residual component respectively to obtain the predicted time coefficient;
[0014] Step S5: reconstruct the predicted time coefficient and the spatial basis function in time and space to obtain a time and space modeling model of the lithium battery temperature field.
[0015] Preferably, in step S1, the KWLTSA dimensionality reduction method specifically includes the following steps:
[0016] Step S11: constructing a local neighborhood;
[0017] Step S12: fitting local coordinates;
[0018] Step S13: globally arranging the local coordinates.
[0019] Preferably, in step S3, the EMD algorithm specifically includes the following sub-steps:
[0020] Step S31: Input the original signal sequence a i (t), calculate a i All the maximum and minimum points of (t) are interpolated using the interpolation method to construct the upper envelope a imax (t), and interpolate all the minimum points to construct the lower envelope a imin (t);
[0021] Step S32: Calculate the upper envelope a imax (t) and the lower envelope a imin (t) the mean curve m i (t), the specific calculation formula is as follows: m i (t)=[a imax (t)+a imin (t)] / 2;
[0022] Step S33: convert the original signal sequence a i (t) and m i (t) is subtracted to obtain a new signal sequence h i (t), the specific calculation formula is as follows: h i (t) = a i (t)-m i (t);
[0023] Step S34: Determine the new signal sequence h i (t) Whether the following two conditions are met:
[0024] The first condition is that in the new signal sequence h i In (t), the sum of the upper and lower extreme points in the new signal sequence is equal to the number of points where the curve crosses the x-axis or differs by at most one;
[0025] The second condition is that for the new signal sequence h i (t), the mean of the upper envelope and the lower envelope must be zero;
[0026] If both conditions are met, the new signal sequence h i (t) is assigned the first IMF component l1;
[0027] Step S35: According to formula a i2 (t) = h i (t)-l1, and get the second decomposed signal sequence a i2 (t), and for a i2 (t) looping through steps S31 to S34 to generate a second IMF component l2;
[0028] Step S36: Loop through steps S31 to S35 for the second IMF component l2 to generate a third IMF component l3. A determination is made as to whether the third IMF component l3 satisfies the threshold screening condition. If so, the signal sequence obtained based on the third IMF component l3 is used as the residual term r(t) of the original signal sequence.
[0029] Preferably, in step S3, a threshold function SD is used k Stop the EMD algorithm process, threshold function SD k as follows:
[0030] Where k is the IMF decomposition number, k is a positive integer greater than 1; t is the number of periods, t = 1, 2, ..., n, n is the sample size; h k (t) represents a new signal sequence; h k-1 (t) is the old signal sequence before decomposition; SD k The range is 0.2-0.3.
[0031] Preferably, step S4 specifically includes the following sub-steps:
[0032] Step S41: IMF component l in period t i (t) and the hidden layer output vector h in period t-1 t-1 After entering the forget gate, it is multiplied by its own weight matrix and added to the forget gate bias vector, and activated by the sigmoid function to obtain the forget gate signal vector f of the tth period. t , the specific calculation formula is as follows:
[0033] f t =σ(W FH h t-1 +W FI l i (t)+b F )
[0034] in, is the weight matrix from hidden layer to forget gate; is the weight matrix from the input layer to the forget gate; is the forget gate bias vector; σ is the sigmoid function; N h is the number of output signals at the previous moment; N i is the number of IMF components;
[0035] Step S42: Calculate the weight parameter i that determines which information to retain t , the specific calculation formula is as follows: i t =σ(W GH h t-1 +W GI l i (t)+b G )
[0036] in, is the weight matrix from hidden layer to input gate; is the weight matrix from the input layer to the input gate; is the input gate bias vector;
[0037] Step S43: Calculate the candidate cell state signal vector of the tth period The specific calculation formula is as follows:
[0038] in, Update the cell state weight matrix from the hidden layer to the input gate; The weight matrix from the input layer to the updated cell state; is the input gate bias vector; tanh is the hyperbolic tangent function;
[0039] Step S44: Calculate the cell state vector C in period t t , the cell state vector C in period t t Used to store historical information. The specific calculation formula is as follows:
[0040] Among them, C t is the cell state vector at period t; The symbol is the Hadamard product of the matrix; f t is the forget gate signal vector of the tth period; C t―1is the cell state vector in period t-1; i t Weight parameters for deciding what information to retain; is the candidate cell state signal vector in period t;
[0041] Step S45: IMF component l in period t i (t) and the hidden layer output vector h in period t-1 t-1 After activation by the output gate activation function, the output information o is generated t , the specific calculation formula is as follows: t =σ(W OH h t-1 +W OI l i (t)+b O )
[0042] Among them, W OH is the weight matrix of the output information; W OI is the weight matrix of IMF component information; b O The paranoid weight for this moment;
[0043] Step S46: Output information o t and the t-th cell state vector C activated by the hidden layer activation function t Perform Hadamard product operation to obtain the hidden layer output vector h of the tth period t , the specific calculation formula is as follows:
[0044] in, The symbol is the Hadamard product of matrices; tanh is the hyperbolic tangent function.
[0045] Preferably, in step S5, the specific calculation formula for spatiotemporal reconstruction is as follows:
[0046] in, Represents the predicted temperature of the lithium battery in time and space dimensions; t k represents the boundary conditions of the lithium-ion battery at time k; S represents the spatial coordinates of the battery; represents the spatial basis function; Represents the prediction time coefficient.
[0047] The technical solutions provided by the embodiments of the present application may have the following beneficial effects:
[0048] This solution uses the EMD algorithm and LSTM to obtain a spatiotemporal model of the lithium battery temperature field. The EMD-LSTM algorithm is more robust to distributed parameter systems with strong nonlinear instabilities, and the resulting spatiotemporal model of the lithium battery temperature field is highly accurate. This solution also uses LSTM to address the problems of vanishing and exploding gradients in long time series. BRIEF DESCRIPTION OF THE DRAWINGS
[0049] Figure 1 is a flowchart of the steps of the spatiotemporal modeling method of lithium battery temperature field based on EMD-LSTM. DETAILED DESCRIPTION
[0050] The embodiments of the present invention are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention and are not to be construed as limiting the present invention.
[0051] A spatiotemporal modeling method for lithium battery temperature field based on EMD-LSTM, characterized by comprising the following steps:
[0052] Step S1: Use the KWLTSA dimensionality reduction method to convert high-dimensional spatiotemporal data into low-dimensional spatiotemporal data and obtain spatial basis functions;
[0053] Step S2: Project the low-dimensional spatiotemporal data onto the spatial basis function to obtain the corresponding low-order time coefficients;
[0054] Step S3: Decompose the low-order time coefficients using the EMD algorithm to obtain IMF components and residual components;
[0055] Step S4: Perform LSTM training on the IMF component and the residual component respectively to obtain the predicted time coefficient;
[0056] Step S5: reconstruct the predicted time coefficient and the spatial basis function in time and space to obtain a time and space modeling model of the lithium battery temperature field.
[0057] This solution uses an EMD-LSTM-based spatiotemporal modeling method for lithium battery temperature fields, as shown in Figure 1. The first step is to use the KWLTSA dimensionality reduction method to convert high-dimensional spatiotemporal data into low-dimensional spatiotemporal data and obtain spatial basis functions. Specifically, because the spatiotemporal distribution of the temperature field in lithium-ion batteries is spatiotemporally coupled, this results in an infinite-dimensional expansion of the temperature field, making it difficult to calculate. This solution uses a spatiotemporal separation method to solve the spatiotemporal distribution of the temperature field. First, the high-dimensional spatiotemporal data generated by the spatiotemporal distribution modeling of the temperature field is simplified to obtain low-dimensional spatiotemporal data. This simplification primarily utilizes the KWLTSA dimensionality reduction method, which is an existing dimensionality reduction method. The spatiotemporal variables are then decomposed into a series of spatial basis functions using the Fourier transform. The second step is to project the low-dimensional spatiotemporal data onto the spatial basis functions to obtain the corresponding low-order time coefficients. Specifically, the Fourier transform not only decomposes the spatiotemporal variables into spatial basis functions, but also produces corresponding low-order time coefficients. The third step is to use the EMD algorithm to decompose the low-order temporal coefficients to obtain IMF components and residual components. Specifically, since lithium-ion batteries may contain some outliers during operation and the data is non-stationary, this solution uses the EMD denoising method to handle non-stationary data. By decomposing the data, a relatively stationary solid-state mode function and a residual term are obtained. The EMD denoising method has better adaptability than traditional denoising methods and no longer requires selecting a basis or setting an effective number of decomposition levels, significantly reducing algorithm complexity and gaining widespread application. The fourth step is to train LSTMs on the IMF components and residual components separately to obtain predicted temporal coefficients. Specifically, LSTM generally refers to a long short-term memory artificial neural network, a type of time-recurrent neural network that can address the problems of vanishing and exploding gradients in long time series. The low-order temporal coefficients obtained through LSTM training can improve the accuracy of the predictive modeling. The fifth step is to perform spatiotemporal reconstruction of the predicted temporal coefficients and spatial basis functions to obtain a spatiotemporal modeling model of the lithium battery temperature field. Specifically, this spatiotemporal modeling model can capture the thermodynamic characteristics of lithium-ion batteries.
[0058] This solution uses the EMD algorithm and LSTM to obtain a spatiotemporal model of the lithium battery temperature field. The EMD-LSTM algorithm is more robust to distributed parameter systems with strong nonlinear instabilities, and the resulting spatiotemporal model of the lithium battery temperature field is highly accurate. This solution also uses LSTM to address the problems of vanishing and exploding gradients in long time series.
[0059] Preferably, in step S1, the KWLTSA dimensionality reduction method specifically includes the following steps:
[0060] Step S11: constructing a local neighborhood;
[0061] Step S12: fitting local coordinates;
[0062] Step S13: globally arranging the local coordinates.
[0063] In this embodiment, by constructing a local domain and fitting local coordinates according to the local domain construction, the local coordinates are globally arranged, so that accurate local coordinates can be obtained, which is beneficial to the positioning of the spatiotemporal modeling model of the lithium battery temperature field.
[0064] Preferably, in step S3, the EMD algorithm specifically includes the following sub-steps:
[0065] Step S31: Input the original signal sequence a i (t), calculate a i All the maximum and minimum points of (t) are interpolated using the interpolation method to construct the upper envelope a imax (t), and interpolate all the minimum points to construct the lower envelope a imin (t);
[0066] Step S32: Calculate the upper envelope a imax (t) and the lower envelope a imin (t) the mean curve m i (t), the specific calculation formula is as follows: m i (t)=[a imax (t)+a imin (t)] / 2 ;
[0067] Step S33: convert the original signal sequence a i (t) and m i (t) is subtracted to obtain a new signal sequence h i (t), the specific calculation formula is as follows: h i (t) = a i (t)-m i (t);
[0068] Step S34: Determine the new signal sequence h i (t) Whether the following two conditions are met:
[0069] The first condition is that in the new signal sequence h i In (t), the sum of the upper and lower extreme points in the new signal sequence is equal to the number of points where the curve crosses the x-axis or differs by at most one;
[0070] The second condition is that for the new signal sequence h i (t), the mean of the upper envelope and the lower envelope must be zero;
[0071] If both conditions are met, the new signal sequence h i (t) is assigned the first IMF component l1;
[0072] Step S35: According to formula a i2 (t) = h i (t)-l1, and get the second decomposed signal sequence a i2 (t), and for a i2 (t) looping through steps S31 to S34 to generate a second IMF component l2;
[0073] Step S36: Loop through steps S31 to S35 for the second IMF component l2 to generate a third IMF component l3. A determination is made as to whether the third IMF component l3 satisfies the threshold screening condition. If so, the signal sequence obtained based on the third IMF component l3 is used as the residual term r(t) of the original signal sequence.
[0074] In this embodiment, the EMD algorithm is used to calculate the low-order time coefficients, that is, the original signal sequence a i (t) is decomposed to obtain the IMF component and the residual component, that is, the new signal sequence h i (t) and the residual term r(t) of the original signal sequence.
[0075] The specific decomposition formula is as follows:
[0076] Where m is the final IMF decomposition number; a i (t) represents the low-order time coefficient of period t, l i (t) represents the IMF component of period t; r(t) represents the residual term of the original signal sequence. Finally, the threshold denoising method is used to denoise the high-frequency part of the time series.
[0077] Preferably, in step S3, a threshold function SD is used k Stop the EMD algorithm process, threshold function SD k as follows:
[0078] Where k is the IMF decomposition number, k is a positive integer greater than 1; t is the number of periods, t = 1, 2, ..., n, n is the sample size; h k (t) represents a new signal sequence; h k-1 (t) is the old signal sequence before decomposition; SD k The range is 0.2-0.3.
[0079] Specifically, since the upper and lower envelopes of the IMF components are almost stacked based on the x-axis, this process is actually a process of eliminating data distortion. Since the upper and lower envelopes are averaged, and the process of taking the average actually loses some information, the stop of the EMD process requires judging whether the difference between two adjacent IMF components is less than a given value. Generally, the threshold function SD is used. k Serves as a threshold function to stop the EMD process.
[0080] Preferably, step S4 specifically includes the following sub-steps:
[0081] Step S41: IMF component l in period t i (t) and the hidden layer output vector h in period t-1 t-1 After entering the forget gate, it is multiplied by its own weight matrix and added to the forget gate bias vector, and activated by the sigmoid function to obtain the forget gate signal vector f of the tth period. t , the specific calculation formula is as follows: f t =σ(W FH h t-1 +W FI l i (t)+b F )
[0082] in, is the weight matrix from hidden layer to forget gate; is the weight matrix from the input layer to the forget gate; is the forget gate bias vector; σ is the sigmoid function; N h is the number of output signals at the previous moment; N i is the number of IMF components;
[0083] Step S42: Calculate the weight parameter i that determines which information to retain t , the specific calculation formula is as follows: i t =σ(W GH h t-1 +W GI l i (t)+b G )
[0084] in, is the weight matrix from hidden layer to input gate; is the weight matrix from the input layer to the input gate; is the input gate bias vector;
[0085] Step S43: Calculate the candidate cell state signal vector of the tth period The specific calculation formula is as follows:
[0086] in, Update the cell state weight matrix from the hidden layer to the input gate; The weight matrix from the input layer to the updated cell state; is the input gate bias vector; tanh is the hyperbolic tangent function;
[0087] Step S44: Calculate the cell state vector C in period t t , the cell state vector C in period t t Used to store historical information. The specific calculation formula is as follows:
[0088] Among them, C t is the cell state vector at period t; The symbol is the Hadamard product of the matrix; f t is the forget gate signal vector of the tth period; C t―1 is the cell state vector in period t-1; i t Weight parameters for deciding what information to retain; is the candidate cell state signal vector in period t;
[0089] Step S45: IMF component l in period t i (t) and the hidden layer output vector h in period t-1 t-1 After activation by the output gate activation function, the output information o is generated t , the specific calculation formula is as follows: t =σ(W OH h t-1 +W OI l i (t)+b O )
[0090] Among them, W OH is the weight matrix of the output information; W OI is the weight matrix of IMF component information; b O The paranoid weight for this moment;
[0091] Step S46: Output information o t and the t-th cell state vector C activated by the hidden layer activation function t Perform Hadamard product operation to obtain the hidden layer output vector h of the tth period t , the specific calculation formula is as follows:
[0092] in, The symbol is the Hadamard product of matrices; tanh is the hyperbolic tangent function.
[0093] In this embodiment, LSTM is used to train the IMF component and the residual component to obtain the hidden layer output vector h of the tth period. t , which is the network output. LSTM is a variant of a recurrent neural network (RNN) that is generated through a system of gated units consisting of an input gate, a forget gate, and an output gate. A "gate" is a structure that selectively controls information, implemented through a sigmoid layer and a point-by-point multiplication operation. Its output value is between 0 and 1, with 0 indicating a complete rejection and 1 indicating a complete pass. LSTM uses the state of internal memory units, or cells, to store historical information. It utilizes different "gates" to allow the network to forget historical information when appropriate and update the cell state based on new input. This solves the problems of vanishing and exploding gradients in recurrent neural networks (RNNs).
[0094] Further explanation: the IMF component l in period t i (t) After passing through three gating networks, part of the information is retained to the cell state vector C in the tth period t In the middle, that is, network memory, the problem that RNN cannot learn data relationships over a long period of time is solved through network memory. The gate structure of LSTM also alleviates the short-term memory problem of the RNN model to a certain extent.
[0095] Preferably, in step S5, the specific calculation formula for spatiotemporal reconstruction is as follows:
[0096] in, Represents the predicted temperature of the lithium battery in time and space dimensions; t k represents the boundary conditions of the lithium-ion battery at time k; S represents the spatial coordinates of the battery; represents the spatial basis function; Represents the prediction time coefficient.
[0097] In this embodiment, by performing spatiotemporal reconstruction on the spatial basis function and the prediction time system, the global spatiotemporal distribution of the predicted lithium battery temperature field can be obtained, which is beneficial to the construction of the spatiotemporal modeling model of the lithium battery temperature field.
[0098] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing module, each unit may exist physically separately, or two or more units may be integrated into a single module. The aforementioned integrated modules may be implemented in the form of hardware or in the form of software functional modules. If the integrated modules are implemented in the form of software functional modules and sold or used as independent products, they may also be stored in a computer-readable storage medium.
[0099] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are exemplary and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.
Claims
1. A spatiotemporal modeling method for lithium battery temperature field based on EMD-LSTM, characterized by: The following steps are involved: Step S1: Use the KWLTSA dimensionality reduction method to convert high-dimensional spatiotemporal data into low-dimensional spatiotemporal data and obtain spatial basis functions; Step S2: Project the low-dimensional spatiotemporal data onto the spatial basis function to obtain the corresponding low-order time coefficients; Step S3: Decompose the low-order time coefficients using the EMD algorithm to obtain IMF components and residual components; Step S4: Perform LSTM training on the IMF component and the residual component respectively to obtain the prediction time coefficient; Step S5: Reconstruct the predicted time coefficients and spatial basis functions in time and space to obtain a time and space modeling model of the lithium battery temperature field.
2. The method for spatiotemporal modeling of lithium battery temperature field based on EMD-LSTM according to claim 1, characterized in that: In step S1, the KWLTSA dimensionality reduction method specifically includes the following steps: Step S11: construct a local neighborhood; Step S12: fitting local coordinates; Step S13: globally arranging the local coordinates.
3. The method for spatiotemporal modeling of lithium battery temperature field based on EMD-LSTM according to claim 1, characterized in that: In step S3, the EMD algorithm specifically includes the following sub-steps: Step S31: Input the original signal sequence a i (t), calculate a i All the maximum and minimum points of (t) are interpolated using the interpolation method to construct the upper envelope a imax (t), and interpolate all the minimum points to construct the lower envelope a imin (t); Step S32: Find the upper envelope a imax (t) and the lower envelope a imin (t) the mean curve m i (t), the specific calculation formula is as follows: m i (t) = [a imax (t)+a imin (t)] / 2; Step S33: convert the original signal sequence a i (t) and m i (t) is subtracted to obtain a new signal sequence h i (t), the specific calculation formula is as follows: i (t) = a i (t)-m i (t); Step S34: Determine the new signal sequence h i (t) Whether the following two conditions are met: The first condition is that in the new signal sequence h i (t), the sum of the upper and lower extreme points in the new signal sequence is equal to the number of points where the curve crosses the x-axis or differs by at most one; The second condition is that for the new signal sequence h i (t), the mean of the upper envelope and the lower envelope must be zero; If both conditions are met, the new signal sequence h i (t) is assigned the value of the first IMF component l1; Step S35: According to formula a i2 (t) = h i (t)-l1, and get the second decomposed signal sequence a i2 (t), and for a i2 (t) looping step S31 to step S34 to generate a second IMF component l2; Step S36: looping steps S31 to S35 for the second IMF component l2 to generate a third IMF component l3, and determining whether the third IMF component l3 satisfies the threshold screening condition. If so, the signal sequence obtained according to the third IMF component l3 is used as the residual term r(t) of the original signal sequence.
4. The method for spatiotemporal modeling of lithium battery temperature field based on EMD-LSTM according to claim 1, characterized in that: In step S3, the threshold function SD is used k Stop the EMD algorithm process, threshold function SD k as follows: Where k is the IMF decomposition number, k is a positive integer greater than 1; t is the number of periods, t = 1, 2, ..., n, n is the sample size; h k (t) represents a new signal sequence; h k-1 (t) is the old signal sequence before decomposition; SD k The range is 0.2-0.
3.
5. The method for spatiotemporal modeling of lithium battery temperature field based on EMD-LSTM according to claim 1, characterized in that: In step S4, the following sub-steps are specifically included: Step S41: IMF component l of period t i (t) and the hidden layer output vector h of the t-1th period t-1 After entering the forget gate, they are multiplied by their respective weight matrices and added to the forget gate bias vector. After activation by the sigmoid function, the forget gate signal vector f of the tth period is obtained. t , the specific calculation formula is as follows: f t =σ(W FH h t-1 +W FI l i (t)+b F ) in, is the weight matrix from hidden layer to forget gate; is the weight matrix from the input layer to the forget gate; is the forget gate bias vector; σ is the sigmoid function; N h N is the number of output signals at the previous moment; i is the number of IMF components; Step S42: Calculate the weight parameter i that determines which information to retain t , the specific calculation formula is as follows: i t =σ(W GH h t-1 +W GI l i (t)+b G ) in, is the hidden layer to input gate weight matrix; is the weight matrix from input layer to input gate; is the input gate bias vector; Step S43: Calculate the candidate cell state signal vector of the tth period The specific calculation formula is as follows: in, Update the cell state weight matrix for the hidden layer to the input gate; The weight matrix from the input layer to the updated cell state; is the input gate bias vector; tanh is the hyperbolic tangent function; Step S44: Calculate the cell state vector C at the tth stage t , the cell state vector C in period t t Used to store historical information. The specific calculation formula is as follows: Among them, C t is the cell state vector at period t; The symbol is the Hadamard product of the matrix; f t is the forget gate signal vector of the tth period; C t―1 is the cell state vector in period t-1; i t Weight parameters for deciding what information to retain; is the candidate cell state signal vector in period t; Step S45: IMF component l of period t i (t) and the hidden layer output vector h of the t-1th period t-1 After being activated by the output gate activation function, the output information o is generated t , the specific calculation formula is as follows: o t =σ(W OH h t-1 +W OI l i (t)+b O ) Among them, W OH is the weight matrix of the output information; W OI is the weight matrix of IMF component information; b O This is the bias weight at this time; Step S46: Output information o t and the cell state vector C of the tth period activated by the hidden layer activation function t Perform Hadamard product operation to obtain the hidden layer output vector h of the tth period t , the specific calculation formula is as follows: in, The symbol is the Hadamard product of matrices; tanh is the hyperbolic tangent function.
6. The method for spatiotemporal modeling of lithium battery temperature field based on EMD-LSTM according to claim 1, characterized in that: In step S5, the specific calculation formula for spatiotemporal reconstruction is as follows: in, Represents the predicted temperature of the lithium battery in time and space dimensions; t k represents the boundary condition of the lithium-ion battery at time k; S represents the spatial coordinate of the battery; represents the spatial basis function; Represents the prediction time coefficient.
Citation Information
Patent Citations
Electric power data credit investigation label application method based on long short-term memory network
CN114418730A
Lithium battery thermal process space-time modeling method based on KWLTSA
CN115392029A
Lithium battery temperature field space-time modeling method based on EMD-LSTM
CN116562008A
Method and apparatus for recognizing sign language or gesture using 3D edm
US20210034846A1
Cited By
Balance control method and system for amplitude and flow of regulating valve for chemical plant
CN119310831A
ConvLSTM model-based sign language recognition method and system
CN119964250A
Maximum load prediction method based on EMD-KPCA-improved BP neural network
CN120124786A
Concrete temperature field reconstruction method and system, electronic equipment and storage medium
CN120234768A
Homogenization cooling control method and system for aluminum alloy extruded part
CN120292817A
