A temperature modeling method for energy storage lithium battery based on kernel isometric feature mapping and sliding window reduced kernel extreme learning machine
Through the nuclear isometric feature mapping and sliding window reduction core over-limit learning machine, a highly adaptable lithium battery temperature model was built, which solved the problem that the lithium battery temperature model in the prior art is difficult to reflect nonlinear characteristics and real-time requirements. Achieve efficient temperature prediction performance.
Patent Information
- Application Number
- CN202310937924.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-07-27
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2043-07-27
AI Technical Summary
The existing lithium battery temperature model is difficult to accurately predict the spatial and temporal changes of battery temperature, especially in complex internal components and external environments of the battery, and cannot meet the needs of real-time thermal management systems. Traditional data-driven methods cannot effectively reflect the non-battery temperature. Linear features.
The global time information is extracted based on kernel isometric feature mapping (KISOMAP), and the low-order timing model and temperature reconstruction model are constructed in combination with sliding window reduction core overlimit learning machine (SW-RKELM), and the model is updated through kernel functions and sliding windows. , adapt to different working environments and reflect the nonlinear characteristics of battery temperature.
Accurate global nonlinear modeling and real-time prediction of lithium battery temperature is realized, adapted to different working environments, improved prediction performance, and able to effectively predict battery temperature under complex conditions.
Smart Images

Figure CN118586262B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of lithium battery temperature prediction, and in particular to a temperature modeling method for energy storage lithium batteries based on kernel isometric feature mapping and a sliding window reduced kernel extreme learning machine. Background Art
[0002] With the rise of new energy vehicles, lithium battery life, performance, safety and other issues have received widespread attention. Battery temperature is a key factor affecting the above issues. Accurately predicting battery temperature is a necessary prerequisite for achieving battery thermal management. However, lithium battery temperature has the characteristics of time-space coupling. It is not only related to time, but also affected by spatial distribution. In addition, due to the frequent changes in boundary conditions, the battery temperature has both time-varying and strong nonlinear characteristics, which brings great difficulties to accurately establish a temperature distribution model.
[0003] Existing lithium battery temperature models can be roughly divided into two categories: mechanism-based temperature prediction models and data-driven temperature prediction models. The mechanism-based model is a temperature prediction model established based on the electrochemical reaction inside the battery, multiphase heat transfer and other mechanisms. Commonly used methods include electro-thermal coupling model, electrochemical-thermodynamic coupling model, etc. Although the mechanism-based model has high accuracy and can well predict the changes in battery temperature in time and space, this type of method requires prior knowledge of the battery temperature partial differential equation and boundary conditions. In practical applications, due to the complex internal components of the battery and the external environment, the temperature partial differential equation and boundary conditions are difficult to obtain accurately. In addition, the mechanism model generally uses numerical calculation methods such as computational fluid dynamics and finite difference method with high time complexity to solve the partial differential equation. Therefore, it cannot be applied to thermal management systems with real-time requirements.
[0004] To solve this problem, some scholars have proposed a temperature prediction model based on data-driven. Compared with the mechanism model, the data-driven model does not need to understand the electrochemical and heat transfer mechanisms inside the battery. The temperature prediction model can be established only through the sensor acquisition system input and temperature data. Traditional data-driven methods include spatiotemporal modeling methods based on Kallen-Loewe (KL) and spatiotemporal modeling methods based on manifold learning, such as local linear embedding method (LLE) and Laplace eigenmap (LE) method. However, the KL method is a linear modeling method and cannot reflect the nonlinear characteristics of battery temperature well. Although LLE and LE are nonlinear methods, they use local linear methods to approximate nonlinear characteristics and cannot obtain global nonlinear characteristics. Summary of the invention
[0005] The main purpose of the present invention is to provide a temperature modeling method for energy storage lithium batteries based on kernel equidistant feature mapping and sliding window reduced kernel extreme learning machine to overcome the above problems.
[0006] To achieve the above objectives, the technical solution adopted by the present invention is: a temperature modeling method of a lithium battery for energy storage based on kernel isometric feature mapping and sliding window reduced kernel extreme learning machine, which specifically includes the following steps:
[0007] Step 1: Use KISOMAP to extract global time information and transform the high-dimensional battery temperature variable T(Z,t) into a low-dimensional time variable a(t), where Z is the spatial coordinate of the battery and t is time;
[0008] Step 2: Use SW-RKELM to build a low-order timing model and establish the system input u(t) = [E(t), I(t)] T The functional relationship between the low-dimensional time variable a(t) and the low-dimensional time variable a(t) obtained in step 1 is used to solve the output weights in the low-order timing model to obtain the model of the time variable a(t), where I(t) and E(t) are current and voltage respectively;
[0009] Step 3: Use SW-RKELM to construct a battery temperature reconstruction model, and use the model of the time variable a(t) obtained in step 2 to solve the output weights of the battery temperature reconstruction model to obtain the battery temperature reconstruction model.
[0010] Preferably, step one specifically includes the following steps:
[0011] Step 11: Set t i The temperature sample measured by the sensor at the moment is Where N Z is the number of sensors;
[0012] Step 12: Use Euclidean distance to find the K nearest neighbors of the sample point and use them to construct a neighborhood graph; if t j The sample at time T(:,t j ) belongs to T(:,t i )’s K nearest neighbors, then T(:,t j ) and T(:,t i ) are connected on the neighborhood graph, and the distance on the neighborhood graph is the Euclidean distance between the two samples; otherwise, if T(:,t j ) does not belong to T(:,t i )’s K nearest neighbors, then the two samples are not connected and the distance is ∞; if there are N t If the temperature sensor data at a certain moment is obtained, then step 12 is repeated;
[0013] Step 13: After constructing the neighborhood graph, calculate the shortest distance d between each pair of samples. a (T(:,t i ),T(:,t j ));
[0014] Step 14. KISOMAP will ensure that the shortest distance of the data before and after dimensionality reduction is consistent by defining the following optimization objectives:
[0015]
[0016] Among them, a(t i ) and a(t j ) are T(:,t i ) and T(:,t j ) The time variable after dimension reduction, d b (a(t i ),a(t j )) is a(t i ) and a(t j )'s shortest distance;
[0017] Step 15. Define D a is the distance matrix of the high-dimensional temperature sample, and the element in the i-th row and j-th column is the sample d a (T(:,t i ),T(:,t j )) Similarly, define D b is the shortest distance matrix of the sample after dimensionality reduction, then, formula (1) can be further transformed into the following objective function:
[0018]
[0019] Among them, τ(D a )=-HA a H / 2,τ(D b )=-HA b H / 2, matrix I is the identity matrix, e=[1,…,1] T , matrix A a Element A a(i,j) =d a (T(:,t i ),T(:,t j ), matrix A b Element A b(i,j) =d b (a(t i ),a(t j )), is the L2 norm of the matrix;
[0020] Step 16: Set λ i is the matrix τ(D a )’s ith eigenvalue, V i p is the corresponding τ(D a)'s p-th element of the ith eigenvector, according to the multidimensional scaling method, the low-dimensional a(t i )'s pth element a p (t i ) is calculated as:
[0021]
[0022] Step 17: Let τ(D a ) has all eigenvalues a(t i ) can be calculated using formula (4):
[0023]
[0024] η≥0.99;
[0025] Step 18: Based on KISOMAP, assume that there is the following mapping function between T(:, t) and a(t):
[0026]
[0027] Among them, κ(.,.) is the kernel function of two sample points, α i is the i-th output weight vector;
[0028] Step 19: Let A = [a(t1),…,a(t Nt )] is from t1 to t Nt The time variable matrix at each moment, T = [T(:, t1),…,T(:, t Nt )] is the corresponding temperature sample matrix, then the matrix form of formula (5) can be expressed as:
[0029] A=κα, (6)
[0030] Among them, κ is the kernel matrix, and its element in the i-th row and j-th column is [κ] ij =κ(T(:,t i ),T(:,t j )), is the input weight matrix;
[0031] Step 110: According to formula (6), α is calculated using the following formula:
[0032]
[0033] in, is the pseudo-inverse of the matrix κ; I is the identity matrix; C is the regularization parameter that needs to be set by the user;
[0034] Step 111, by substituting a(t) obtained by formula (4) into formula 7, calculate α;
[0035] Step 112: Substitute the obtained α into formula (5) to obtain the model of a(t).
[0036] Preferably, step 2 specifically includes the following steps:
[0037] Step 21: The SW-RKELM algorithm first extracts the data set S samples are randomly selected from the support vector set Among them U i are randomly selected samples, i = 1,…,S, U(t) = [a(t-1), u(t-1)] and a(t) as the input and output of the SW-RKELM model, u(t) = [E(t), I(t)] T , I(t) and E(t) are current and voltage respectively;
[0038] Step 22: Through the kernel function κ(·,·), SW-RKELM expresses the model input U(t) and the model output a(t) as:
[0039]
[0040] in, Based on support vector set The hidden layer output of , β=[β1,…,β S ] T is the output weight;
[0041] Step 23: Express formula (8) as:
[0042] κ0β=A0, (9)
[0043] in,
[0044] Step 24: Calculate β in formula (9) using the following formula:
[0045]
[0046] in,
[0047] Step 25: Substitute a(t) obtained by formula (5) into formula (10) to calculate β, and substitute β into formula (8) to obtain the a(t) model.
[0048] Preferably, the current data set is Step 3 specifically includes the following steps:
[0049] Step 31: SW-RKELM from the dataset Randomly select samples, as the support vector set where a i is a randomly selected sample,
[0050] Step 32: Dataset The compact form of is:
[0051]
[0052] in, is the output matrix; is the output weight;
[0053] Step 33: It is expressed by the following formula:
[0054]
[0055] in,
[0056] Step 34: Replace T(:,t i ) into formula (12), we can calculate The calculated Substituting into formula (11), we can get the temperature reconstruction model Used to predict the temperature of lithium batteries.
[0057] Compared with the prior art, the present invention has the following beneficial effects:
[0058] 1) The adopted KISOMAP can obtain the global nonlinear time information of temperature data;
[0059] 2) SW-RKELM can use a sliding window to update the timing model and temperature reconstruction model according to new data, which enables this patent to adapt to the battery operation status under different working environments and has high prediction performance. BRIEF DESCRIPTION OF THE DRAWINGS
[0060] Figure 1 is a schematic diagram according to a preferred embodiment of the present invention;
[0061] Figure 2 This is the working principle diagram of SW-RKELM;
[0062] Figure 3 This is the layout diagram of the sensor on the lithium battery in the experimental example;
[0063] Figure 4 are the current and voltage in the lithium battery;
[0064] Figure 5 It is the predicted temperature distribution diagram of lithium battery. DETAILED DESCRIPTION
[0065] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art may think of other obvious variations.
[0066] A temperature modeling method for energy storage lithium batteries based on kernel isotropic feature mapping (KISOMAP) and sliding window reduced kernel extreme learning machine (SW-RKELM) specifically includes the following steps:
[0067] Step 1: Use KISOMAP to extract global time information and transform the high-dimensional battery temperature variable T(Z,t) (Z is the spatial coordinate of the battery, t is time) into a low-dimensional time variable a(t);
[0068] Step 2: Use SW-RKELM to build a low-order timing model and establish the system input u(t) = [E(t), I(t)] T (I(t) and E(t) are current and voltage respectively) and the functional relationship between the time variable a(t);
[0069] Step 3: Use SW-RKELM to build a battery temperature reconstruction model, reconstruct the original battery temperature distribution, and restore the battery temperature spatial distribution information.
[0070] Step 1 specifically includes the following steps:
[0071] Step 11: Set t i The temperature sample measured by the sensor at the moment is Where N Z is the number of sensors;
[0072] Step 12: Use Euclidean distance to find the K nearest neighbors of the sample point and use them to construct a neighborhood graph; if t j The sample at time T(:,t j ) belongs to T(:,t i )’s K nearest neighbors, then T(:,t j ) and T(:,t i ) are connected on the neighborhood graph, and the distance on the neighborhood graph is the Euclidean distance between the two samples; otherwise, if T(:,t j ) does not belong to T(:,t i )’s K nearest neighbors, then the two samples are not connected and the distance is ∞; if there are N t If the temperature sensor data at each moment is obtained, the neighborhood graph between all samples can be constructed by repeating the above process.
[0073] Step 13: After constructing the neighborhood graph, use the Dijkstra algorithm to calculate the shortest distance between two samples.
[0074] Step 14. KISOMAP will ensure that the shortest distance of the data before and after dimensionality reduction is consistent by defining the following optimization objectives:
[0075]
[0076] Among them, a(t i ) and a(t j ) are T(:,t i ) and T(:,t j ) The time variable after dimension reduction, d b (a(t i ),a(t j )) is a(t i ) and a(t j )'s shortest distance;
[0077] Step 15. Define D a is the distance matrix of the high-dimensional temperature sample, and the element in the i-th row and j-th column is the sample d a (T(:,t i ),T(:,t j )) Similarly, define D b is the shortest distance matrix of the sample after dimensionality reduction, then, formula (1) can be further transformed into the following objective function:
[0078]
[0079] Among them, τ(D a )=-HA a H / 2,τ(D b )=-HA b H / 2, matrix I is the identity matrix, e=[1,…,1] T , matrix A a Element A a(i,j) =d a (T(:,t i ),T(:,t j ), matrix A b Element A b(i,j) =d b (a(t i ),a(t j )), is the L2 norm of the matrix;
[0080] Step 16: Set λ i is the matrix τ(D a)’s ith eigenvalue, V i p is the corresponding τ(D a ), then According to the multidimensional scaling method, the low-dimensional a(t i )'s pth element a p (t i ) can be calculated as:
[0081]
[0082] Step 17: Let τ(D a ) has all eigenvalues a(t i ) can be calculated using formula (4):
[0083]
[0084] In general, η≥0.99;
[0085] Step 18: Formulas (1) to (3) are only applicable to constructing t i =t1,…,t Nt Transformation between time T(:, t) and a(t). To solve the problem of converting temperature samples at other times to time variables, based on KISOMAP, the following mapping function is assumed between T(:, t) and a(t):
[0086]
[0087] Among them, κ(.,.) is the kernel function of the two sample points, which can generally be selected as a Gaussian kernel function; α i is the i-th output weight vector;
[0088] Step 19: Let A = [a(t1),…,a(t Nt )] is from t1 to t Nt The time variable matrix at each moment, T = [T(:, t1),…,T(:, t Nt )] is the corresponding temperature sample matrix, then the matrix form of formula (5) can be expressed as:
[0089] A=κα, (6)
[0090] Among them, κ is the kernel matrix, and its element in the i-th row and j-th column is [κ] ij =κ(T(:,t i ),T(:,t j )), is the input weight matrix;
[0091] Step 110, in formula (6), the only unknown α can be obtained using the following formula:
[0092]
[0093] in, is the pseudo-inverse of the matrix κ; I is the identity matrix; C is the regularization parameter that needs to be set by the user;
[0094] Step 111, by substituting a(t) obtained by formula (4) into formula 7, α can be calculated;
[0095] Step 112: Substitute the obtained α into formula (5) to obtain the model of a(t). Formula (5) can be used to calculate the time variable a(t) at any time.
[0096] For step 2, after obtaining the time variable, SW-RKELM can be used to establish a low-order time series model to reflect the mapping relationship between the input signals u(t) and a(t). SW-RKELM uses a sliding window with a length of L and a sliding step of l for modeling. Figure 2 As shown, U(t) = [a(t-1), u(t-1)] and a(t) are used as the input and output of the model, u(t) = [E(t), I(t)] T , I(t) and E(t) are current and voltage respectively, the solid line is the data stored in the current sliding window, and it can be seen that there are L sets of data in total. When KISOMAP generates a(t) at l time points, the sliding window will slide once with a sliding step of l, and the data stored in the window will become the dotted line part.
[0097] The step 2 specifically includes the following steps: To facilitate the description of the modeling steps of SW-RKELM, assume that the current internal data is
[0098] Step 21: The SW-RKELM algorithm first extracts the data set S samples are randomly selected from the support vector set Among them U i is a sample randomly selected from the data set U, i=1,…,S;
[0099] Step 22: Through the kernel function κ(·,·) (such as Gaussian kernel function), SW-RKELM expresses the model input U(t) and the model output a(t) as:
[0100]
[0101] in, Based on support vector set The hidden layer output of , β=[β1,…,βS ] T is the output weight;
[0102] Step 23: To facilitate analysis, based on The compact form of formula (8) can be expressed as:
[0103] κ0β=A0, (9)
[0104] in,
[0105] Step 24: Since S samples are randomly selected to calculate the hidden layer output, unlike formula (7), β in formula (9) is calculated using the following formula:
[0106]
[0107] in,
[0108] Step 25: Substitute the obtained a(t) into formula (10) to calculate β, and substitute β into formula (8) to obtain the a(t) model.
[0109] It should be noted that the a(t) model in formula (5) can only calculate a(t) at the past moment, while the a(t) model in formula (8) can predict a(t) at the future moment. The a(t) obtained by formula (5) is used to calculate β in formula (8) when modeling.
[0110] Every time the sliding window slides once, the output weight β needs to be updated according to formulas (8) to (10).
[0111] For step 3, SW-RKELM is also used to nonlinearly construct the temperature reconstruction model, assuming that the data set inside the current sliding window is The specific steps include:
[0112] Step 31: SW-RKELM from the dataset Randomly select samples, as the support vector set where a i is a randomly selected sample,
[0113] Step 32: Dataset The compact form of is:
[0114]
[0115] in, is the output matrix; is the output weight;
[0116] Step 33: It is expressed by the following formula:
[0117]
[0118] in,
[0119] Step 34: Replace T(:,t i ) into formula (12), we can calculate The calculated Substituting into formula (11), we can get the temperature reconstruction model It is used to predict the temperature of lithium batteries. As can be seen from step 2, the input of the final model of the present invention is current and voltage, and the output is temperature. As long as the current and voltage of the lithium battery are obtained, the temperature can be predicted.
[0120] When the sliding window slides once, the output weights need to be updated according to formulas (11) to (12)
[0121] In short, the general process of modeling of the present invention can be summarized as follows: first, the actual data of the data to be predicted is detected by sensors, these actual data are input into the model to obtain the parameters of the model, and the obtained parameters are substituted into the model to complete the modeling.
[0122] Experimental example
[0123] Take the thermal process of Li(NiCoMn)O2 ternary lithium battery as an example. The dimensions of the battery are 0.24m long, 0.156m wide, and 0.008m thick. The standard capacity, rated voltage, charge cut-off voltage, and discharge cut-off voltage are 32Ah, 3.7V, 4.2V, and 2.75V, respectively. Figure 3 As shown in Figure 1, a total of 30 thermocouple sensors (S1–S30) are evenly distributed on the surface of the lithium battery (5 rows and 6 columns) to collect spatiotemporal temperature data. In the experiment, the battery was placed in a thermal chamber to maintain the ambient temperature at 25°C. Then, a 2A constant current discharge mode was used to stimulate the battery thermal process. Figure 4 shows the current and resulting voltage, Figure 5 This is the predicted temperature distribution at 620 seconds according to the present invention.
[0124] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions only describe the principles of the present invention. The present invention may be subject to various changes and improvements without departing from the spirit and scope of the present invention. These changes and improvements fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the attached claims and their equivalents.
Claims
1. A temperature modeling method for energy storage lithium batteries based on kernel isometric feature mapping and sliding window reduced kernel extreme learning machine, specifically comprising the following steps: Step 1: Use KISOMAP to extract global time information and transform the high-dimensional battery temperature variable T(Z,t) into a low-dimensional time variable a(t), where Z is the spatial coordinate of the battery and t is time; Step 2: Use SW-RKELM to build a low-order timing model and establish the system input u(t) = [E(t), I(t)] T The functional relationship between the low-dimensional time variable a(t) and the low-dimensional time variable a(t) obtained in step 1 is used to solve the output weights in the low-order time series model to obtain the model of the time variable a(t), where I(t) and E(t) are current and voltage respectively; Step 3: Use SW-RKELM to construct a battery temperature reconstruction model, and use the model of the time variable a(t) obtained in step 2 to solve the output weights of the battery temperature reconstruction model to obtain the battery temperature reconstruction model; Step 1 specifically includes the following steps: Step 11: Set t i The temperature sample measured by the sensor at the moment is Where N Z is the number of sensors; Step 12: Use Euclidean distance to find the K nearest neighbors of the sample point and use them to construct a neighborhood graph; if t j The sample at time T(:,t j ) belongs to T(:,t i )’s K nearest neighbors, then T(:,t j ) and T(:,t i ) are connected on the neighborhood graph, and the distance on the neighborhood graph is the Euclidean distance between the two samples; otherwise, if T(:,t j ) does not belong to T(:,t i )’s K nearest neighbors, then the two samples are not connected and the distance is ∞; if there are N t If the temperature sensor data at a certain moment is obtained, then step 12 is repeated; Step 13: After constructing the neighborhood graph, calculate the shortest distance d between each pair of samples. a (T(:,t i ),T(:,t j )); Step 14. KISOMAP will ensure that the shortest distance of the data before and after dimensionality reduction is consistent by defining the following optimization objectives: in, a(t i ) and a(t j ) are T(:,t i ) and T(:,t j ) The time variable after dimension reduction, d b (a(t i ),a(t j )) is a(t i ) and a(t j )'s shortest distance; Step 15. Define D a is the distance matrix of the high-dimensional temperature sample, and the element in the i-th row and j-th column is the sample d a (T(:,t i ), T(:, t j )) Similarly, define D b is the shortest distance matrix of the sample after dimensionality reduction, then formula (1) can be further transformed into the following objective function: Among them, τ(D a )=-HA a H / 2,τ(D b )=-HA b H / 2, matrix I is the identity matrix, e = [1, ..., 1] T , matrix A a Element A a(i,j) =d a (T(:,t i ), T(:, t j ), matrix A b Element A b(i,j) =d b (a(t i ), a(t j )), is the L2 norm of the matrix; Step 16: Set λ i is the matrix τ(D a )’s ith eigenvalue, V i p is the corresponding τ(D a )'s p-th element of the ith eigenvector, according to the multidimensional scaling method, the low-dimensional a(t i )'s pth element a p (t i ) is calculated as: Step 17: Let τ(D a ) has all eigenvalues a(t i ) can be calculated using formula (4): η≥0.99; Step 18: Based on KISOMAP, assume that there is the following mapping function between T(:, t) and a(t): Among them, κ(.,.) is the kernel function of two sample points, α i is the i-th output weight vector; Step 19: Let A = [a(t1), ..., a(t Nt )] is from t1 to t Nt The time variable matrix at the moment, T = [T(:,t1),…,T(:,t Nt )] is the corresponding temperature sample matrix, then the matrix form of formula (5) can be expressed as: A=κα, (6) Among them, κ is the kernel matrix, and its element in the i-th row and j-th column is [κ] ij =κ(T(:,t i ),T(:,t j )), is the input weight matrix; Step 110: According to formula (6), α is calculated using the following formula: in, is the pseudo-inverse of the matrix κ; I is the identity matrix; C is the regularization parameter that needs to be set by the user; Step 111, by substituting a(t) obtained by formula (4) into formula 7, calculate α; Step 112, substituting the obtained α into formula (5) to obtain the model of a(t); Step 2 specifically includes the following steps: Step 21: The SW-RKELM algorithm first extracts the data set S samples are randomly selected from the dataset, where L is the number of groups of data in the dataset, as the support vector set Among them U i are randomly selected samples, i = 1,…,S, U(t) = [a(t-1), u(t-1)] and a(t) as the input and output of the SW-RKELM model, u(t) = [E(t), I(t)] T , I(t) and E(t) are current and voltage respectively; Step 22: Through the kernel function κ(·,·), SW-RKELM expresses the model input U(t) and the model output a(t) as: in, Based on support vector set The hidden layer output of , β = [β1,,β S ] T is the output weight; Step 23: Express formula (8) as: κ0β=A0, (9) in, Step 24: Calculate β in formula (9) using the following formula: in, Step 25: Substitute a(t) obtained by formula (5) into formula (10) to calculate β, and substitute β into formula (8) to obtain the a(t) model.
2. The modeling method according to claim 1, characterized in that: Assume the current data set is Step 3 specifically includes the following steps: Step 31: SW-RKELM from the dataset Randomly select samples, as the support vector set where a i is a randomly selected sample, Step 32: Dataset The compact form of is: in, is the output matrix; is the output weight; Step 33: It is expressed by the following formula: in, Step 34: Replace T(:,t i ) into formula (12), we can calculate The calculated Substituting into formula (11), we can get the temperature reconstruction model Used to predict the temperature of lithium batteries.
