Lithium ion battery temperature prediction model based on space-time model

Abstract

The invention discloses a lithium ion battery temperature prediction model based on a space-time model. The lithium ion battery temperature prediction model comprises the following steps: firstly, separating battery temperature data into an orthogonal space basis function and a time coefficient by using an orthogonal locality preserving projection (OLPP) method; then, taking the current, the voltage and the separated time coefficient at the past moment as input, taking the predicted time coefficient as output, constructing N extreme learning machine (ELM) models, and obtaining N predicted time coefficient estimated values; and designing a set of credibility evaluation rules according to a Z score theory to eliminate uncredible estimated values. And taking the average value of the residual estimation results after screening as the final output of the model. And finally, reconstructing original temperature distribution by using orthogonality of a space basis function through a space-time synthesis method. And a method is provided for accurately predicting the lithium ion temperature.

Description

Technical Field

[0001] The invention relates to a lithium ion battery temperature prediction model, belonging to the technical field of battery application. Background Art

[0002] Lithium-ion battery Accurate temperature prediction is beneficial to improving the accuracy of thermal monitoring and thermal control of battery thermal management systems .However Lithium-ion battery The thermal process belongs to a typical parabolic distributed parameter system, which has spatiotemporal coupling, strong nonlinearity and time-varying characteristics. Therefore, it is extremely difficult to establish an accurate temperature prediction model for lithium-ion batteries.

[0003] In recent years, data-driven spatiotemporal modeling methods have been widely used Lithium-ion battery Temperature prediction. The modeling process of this type of method is divided into three steps. First, based on the temperature distribution data collected by the sensor, the temperature variable is separated into spatial basis functions and time coefficients using a dimensionality reduction method. Among them, the Kalkonen-Loewe (KL) method is one of the most commonly used dimensionality reduction methods. However, the KL method is a global linear method that ignores the local manifold information between battery temperature data and easily causes the loss of spatiotemporal distribution information. To solve this problem, some scholars have combined local linear embedding (LLE) and Laplace eigenmap (LE ) The local manifold dimensionality reduction method is introduced into the spatiotemporal modeling framework. However, LLE and LE cannot provide a suitable transformation function for the reduced data, so they have certain limitations. After obtaining the time coefficient, the second step of the spatiotemporal modeling method is to use machine learning methods, such as extreme learning machine, least squares support vector machine, etc., to establish a revealing Lithium-ion battery The time series model of the relationship between the input and the time coefficient function. However, methods such as the ELM and the Least Squares Support Vector Machine do not consider that the random weights will add uncertainty to the output results. Finally, after obtaining the time series model, the third step of the spatiotemporal modeling method can reconstruct the original battery temperature distribution through spatiotemporal synthesis.

[0004] In order to solve the problems existing in the traditional data-driven spatiotemporal modeling method, a spatiotemporal modeling method based on orthogonal local preserving projection (OLPP) is proposed. First, the battery temperature data is separated into spatial basis functions and time coefficients using OLPP. OLPP is a manifold learning method that can preserve the local manifold structure of temperature data, so it can overcome the nonlinear information loss problem caused by the KL method. In addition, OLPP provides a set of orthogonal basis functions to realize the transformation of temperature from the spatiotemporal domain to the time domain, overcoming the problem that LLE and LE cannot provide a suitable conversion function for the data. Then, a time series model of the integrated extreme learning machine (EnELM) is constructed to overcome the problem of random weights in traditional methods such as extreme learning machines and least squares support vector machines that increase the uncertainty of the results. Finally, the original temperature distribution is reconstructed using the spatiotemporal synthesis method. Summary of the invention

[0005] The purpose of the present invention is to provide a lithium ion battery temperature prediction model based on a spatiotemporal model to solve the problems existing in the above-mentioned prior art.

[0006] To achieve the above object, the present invention provides the following solution: The present invention provides a lithium ion battery temperature prediction model based on a spatiotemporal model, and the method steps are as follows: S1. Based on the lithium battery temperature distribution data collected by the sensor, the temperature variable is separated into the spatial basis function and the time coefficient by using the dimensionality reduction method; S2, using the collected current and voltage data and the separated time coefficient as inputs of the EnELM model, and obtaining the predicted time coefficient as output; S3. Perform spatiotemporal synthesis of the separated spatial basis functions and the predicted time coefficients to obtain reconstructed lithium battery temperature distribution data.

[0007] Preferably, the S1 comprises the following steps: S1-1. Collect lithium battery temperature data , where x, y are spatial coordinates, n is the number of sensors, t is time, and L0 is the total duration; S1-2, based on OLPP, the lithium battery temperature data is separated in time and space with a model order of N. (1) OLPP obtains the first i Orthogonal basis functions : , (2) in, X =[ T (:,1), T (:,2),…,T (:, L 0)] is the temperature data matrix; For the front i- 1 spatial basis function matrix; D The diagonal elements are The diagonal matrix of ; S ij is the similarity matrix S No. i Line j Column elements; L =DS is the Laplace matrix. Generally speaking, the similarity matrix is ​​calculated using the following formula:

[0008] Here, you can use the threshold θ ,like , then it is believed that T (:, i )and T (:, j )neighbors.

[0009] Defining the Matrix for:

[0010] Then, formula (2) can be solved by the following process: 1) Calculate spatial basis functions : Pair Matrix Perform eigendecomposition and use the eigenvector with the smallest eigenvalue as the first spatial basis function ; 2) Calculate spatial basis functions : Defining the Matrix M i for:

[0011] in, I is the identity matrix. Then, M i Perform eigendecomposition and take the eigenvector with the smallest eigenvalue as the i Spatial basis functions ; 3) Repeat step 2) until i = N ; The spatial basis functions obtained by the above process are orthogonal, but in order to accurately reconstruct the space-time variables, the spatial basis functions need to be normalized. Based on the unit orthogonal property of the spatial basis functions, combined with formula (1), the time coefficient can be calculated by the following formula: :

[0012] Preferably, S2 comprises the following steps: S2-1. Construct an ELM model. For a single ELM model, its hidden layer output expression can be expressed as:

[0013] In the above formula, h i (x) represents the output of the ith hidden layer node of ELM, g(·) represents the activation function, and w i represents the random input weight, b i Represents the hidden neuron bias, and the output Y of ELM can be expressed as:

[0014] In the above formula, L represents the number of hidden layer nodes, β=[β1,…, β L ] T is the connection weight between the hidden layer and the output layer, H=[H1(x), …, H L (x)] is the hidden layer output. The learning goal of ELM is to find the output weight 𝛽 that minimizes the training error. The optimal 𝛽 can be calculated by the following formula:

[0015] In the above formula, H + is the generalized inverse matrix of H; S2-2, using the collected current and voltage data and the separated time coefficients as inputs of N single ELM models to obtain N predicted time coefficient outputs; S2-3. Based on the Z-score theory, a set of “credibility evaluation rules” was designed to eliminate unreliable estimates.

[0016]

[0017] In the above formula, X is the output result of a single ELM model, μ is the mean of the output results of N ELM models, and σ is the standard deviation of the output results of N ELM models. It is considered that every result with a Z value in the interval (-3, 3) is credible, and vice versa. The average of all credible results is calculated and output as the final result.

[0018] Preferably, S3 comprises the following steps: S3-1. Prediction time coefficient obtained by EnELM model And the spatial basis function obtained by the OLPP model Combined, we get the reconstructed original temperature distribution .

[0019] BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A basic flow chart for conceiving a lithium-ion battery temperature prediction model based on a spatiotemporal model according to the present invention; Figure 2 This is the distribution map of lithium-ion battery sensors; Specific implementation method, the following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below in combination with the drawings and specific implementation methods.

[0021] Reference Figure 1 The present invention provides a lithium ion battery temperature prediction model based on a spatiotemporal model, comprising the following steps: S1. Based on the lithium battery temperature distribution data collected by the sensor, the temperature variable is separated into the spatial basis function and the time coefficient by using the dimensionality reduction method; S1-1. Collect lithium battery temperature data , where x, y are spatial coordinates, n is the number of sensors, t is time, and L0 is the total duration; S1-2, based on OLPP, the lithium battery temperature data is separated in time and space with a model order of N=5. (1) OLPP obtains the first i Orthogonal basis functions : , (2) in, X =[ T (:,1), T (:,2),…, T (:, L 0)] is the temperature data matrix; For the front i-1 spatial basis function matrix; D The diagonal elements are The diagonal matrix of ; S ij is the similarity matrix S No. i Line j Column elements; L =DS is the Laplace matrix. Generally speaking, the similarity matrix is ​​calculated using the following formula:

[0022] Here, you can use the threshold θ = 0.01, if , then it is believed that T (:, i )and T (:, j )neighbors.

[0023] Defining the Matrix for:

[0024] Then, formula (2) can be solved by the following process: 1) Calculate spatial basis functions : Pair Matrix Perform eigendecomposition and use the eigenvector with the smallest eigenvalue as the first spatial basis function ; 2) Calculate spatial basis functions : Defining the Matrix M i for:

[0025] in, I is the identity matrix. Then, M i Perform eigendecomposition and take the eigenvector with the smallest eigenvalue as the i Spatial basis functions ; 3) Repeat step 2) until i = N ; The spatial basis functions obtained by the above process are orthogonal, but in order to accurately reconstruct the space-time variables, the spatial basis functions need to be normalized. Based on the unit orthogonal property of the spatial basis functions, combined with formula (1), the time coefficient can be calculated by the following formula: :

[0026] S2, using the collected current and voltage data and the separated time coefficient as inputs of the EnELM model, and obtaining the predicted time coefficient as output; S2-1. Construct an ELM model. For a single ELM model, its hidden layer output expression can be expressed as:

[0027] In the above formula, h i (x) represents the output of the ith hidden layer node of ELM, g(·) uses the Sigmoid function, and w i With b i Using uniform distribution to randomly generate in the interval [-1,1], the output Y of ELM can be expressed as:

[0028] In the above formula, L represents the number of hidden layer nodes, β=[β1,…, β L ] T is the connection weight between the hidden layer and the output layer, H=[H1(x), …, H L (x)] is the hidden layer output. The learning goal of ELM is to find the output weight 𝛽 that minimizes the training error. The optimal 𝛽 can be calculated by the following formula:

[0029] In the above formula, H + is the generalized inverse matrix of H; S2-2, using the collected current and voltage data and the separated time coefficients as inputs of N=200 single ELM models, and obtaining N predicted time coefficient outputs; S2-3. Based on the Z-score theory, a set of “credibility evaluation rules” was designed to eliminate unreliable estimates.

[0030]

[0031] In the above formula, X is the output result of a single ELM model, μ is the mean of the output results of N ELM models, and σ is the standard deviation of the output results of N ELM models. It is considered that every result with a Z value in the interval (-3, 3) is credible, and vice versa. The average of all credible results is calculated and output as the final result.

[0032] S3. Perform spatiotemporal synthesis of the separated spatial basis functions and the predicted time coefficients to obtain reconstructed lithium battery temperature distribution data.

[0033] S3-1. Prediction time coefficient obtained by EnELM model And the spatial basis function obtained by the OLPP model Combined, we get the reconstructed original temperature distribution .

[0034]

[0035] The embodiments described above are only descriptions of the preferred modes of the present invention, and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A lithium-ion battery temperature prediction model based on a spatiotemporal model, characterized in that: It includes the following steps: S1. Based on the lithium battery temperature distribution data collected by the sensor, the temperature variable is separated into the spatial basis function and the time coefficient by using the dimensionality reduction method; S2, using the collected current and voltage data and the separated time coefficient as inputs of an integrated extreme learning machine (EnELM) model, and obtaining the predicted time coefficient as output; S3. Perform spatiotemporal synthesis of the separated spatial basis functions and the predicted time coefficients to obtain reconstructed lithium battery temperature distribution data.

2. The lithium-ion battery temperature prediction model based on the spatiotemporal model according to claim 1, characterized in that: The S1 comprises the following steps: S1-1. Collect lithium battery temperature data , where x, y are spatial coordinates, n is the number of sensors, t is time, and L0 is the total duration; S1-2, based on OLPP, the lithium battery temperature data is separated in time and space with a model order of N. （1） OLPP obtains the first i Orthogonal basis functions : , （2） in, X =[ T (:,1), T (:,2),…, T (:, L 0)] is the temperature data matrix; For the front i- 1 spatial basis function matrix; D The diagonal elements are The diagonal matrix of ; S ij is the similarity matrix S No. i Line j Column elements; L =DS is the Laplace matrix. Generally speaking, the similarity matrix is ​​calculated using the following formula: , here you can use the threshold setting θ ,like , then it is believed that T (:, i )and T (:, j ) neighbors. Define the matrix for: , then the following process can be used to solve formula (2): 1) Calculate spatial basis functions : Pair matrix Perform eigendecomposition and use the eigenvector with the smallest eigenvalue as the first spatial basis function ; 2) Calculate the spatial basis function : Define the matrix M i for: ,in I is the identity matrix. Then, M i Perform eigendecomposition and take the eigenvector with the smallest eigenvalue as the i Spatial basis functions ;3) Repeat step 2) until i = N ; The spatial basis functions obtained by the above process are orthogonal, but in order to accurately reconstruct the space-time variables, the spatial basis functions need to be normalized. Based on the unit orthogonal property of the spatial basis functions, combined with formula (1), the time coefficient can be calculated by the following formula :

3. The lithium-ion battery temperature prediction model based on the spatiotemporal model according to claim 1, characterized in that: The S2 comprises the following steps: S2-1. Construct an ELM model. For a single ELM model, its hidden layer output expression can be expressed as: , where h i (x) represents the output of the ith hidden layer node of ELM, g(·) represents the activation function, and w i represents the random input weight, b i Represents the hidden neuron bias, and the output Y of ELM can be expressed as: , where L represents the number of hidden layer nodes, β=[β1,…, β L ] T is the connection weight between the hidden layer and the output layer, H=[H1(x), …, H L (x)] is the hidden layer output. The learning goal of ELM is to find the output weight 𝛽 that minimizes the training error. The optimal 𝛽 can be calculated by the following formula: , H in the above formula + is the generalized inverse matrix of H; S2-2, using the collected current and voltage data and the separated time coefficients as inputs of N single ELM models to obtain N predicted time coefficient outputs; S2-3. Based on the Z-score theory, a set of "credibility evaluation rules" is designed to eliminate unreliable estimates: , where X is the output of a single ELM model, μ is the mean of the output of N ELM models, and σ is the standard deviation of the output of N ELM models. It is considered that every Z value in the interval (-3, 3) is credible, otherwise it is not credible. The average of all credible results is calculated and output as the final result.

4. The lithium-ion battery temperature prediction model based on the spatiotemporal model according to claim 1, characterized in that: The S3 comprises the following steps: S3-1. Prediction time coefficient obtained by EnELM model And the spatial basis function obtained by the OLPP model Combined, we get the reconstructed original temperature distribution :

Images