Battery system multi-step temperature prediction method, system and device based on DLSTM model

Abstract

The invention discloses a battery system multi-step temperature prediction method, system and device based on a DLSTM model, and the method comprises the steps: obtaining the operation data of a battery system, and decomposing the operation data into a trend term and a residual term through a moving average kernel; respectively inputting the trend item and the residual item into the two LSTM networks to obtain a trend prediction feature and a residual prediction feature; and adding the trend prediction feature and the residual prediction feature to obtain the predicted temperature of the battery system. According to the battery system multi-step temperature prediction method based on the DLSTM model, robustness and accuracy are greatly improved, the battery system temperature prediction result based on the DLSTM model is combined with a related outlier identification method, an abnormal heating temperature sensor is positioned in advance, safe operation of the battery system is ensured, and thermal runaway accidents are avoided.

Description

Technical Field

[0001] The present invention belongs to the field of energy storage technology, and in particular relates to a multi-step temperature prediction method, system and device for a battery system based on a DLSTM model. Background Art

[0002] Lithium-ion batteries are widely used in various fields including new energy vehicles due to their advantages in energy storage. With the popularization of new energy vehicles, the safety of new energy vehicle battery systems has gradually become the focus of social attention. The safety of battery systems is particularly affected by temperature. Therefore, accurately predicting the battery system temperature and giving drivers thermal runaway warnings are crucial to safe driving.

[0003] A lot of research has been done on battery system temperature prediction, such as: Kalman filter model, prediction model combining long short-term memory neural network and convolutional neural network, extreme gradient boosting method, etc. Although the existing research has made significant progress, there are still some limitations: some temperature-based diagnostic techniques can only work after a thermal runaway fault occurs; some temperature prediction models have a limited prediction range; and although some methods can achieve multi-step temperature prediction, they usually use autoregressive or iterative multi-step prediction technology, and as the number of prediction steps increases, there will be a significant error accumulation effect. Patent 202311670876.2 uses a DLinear-based direct multi-step temperature prediction model to predict the battery system temperature, but the prediction accuracy and robustness need to be improved. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides a multi-step temperature prediction method, system and device for a battery system based on a DLSTM model.

[0005] The multi-step temperature prediction method of the battery system based on the DLSTM model comprises the following steps: Acquire the operating data of the battery system, clean the operating data, the DLSTM model uses a moving average kernel and two LSTM networks for modeling, uses the moving average kernel to decompose the operating data into a trend item and a residual item, and performs a filling operation; The trend term and the residual term are input into two LSTM networks respectively to obtain the trend prediction feature and the residual prediction feature; The trend prediction feature and the residual prediction feature are added together to obtain the predicted temperature of the battery system.

[0006] As an implementable embodiment, the operating data includes: total current, total voltage, sensor temperature, insulation resistance and state of charge, and the operating data is expressed as:

[0007] in, Indicates the number of features related to temperature; Indicates the number of temperature sensors in the battery system; Indicates the length of the running data; , , , and They represent the current, SOC, total voltage, insulation resistance and temperature of the battery system respectively.

[0008] As an implementable embodiment, the temperature of the battery system predicted by the DLSTM model is expressed as:

[0009] in, represents the temperature of the battery system predicted by the DLSTM model, Indicates the length of the input historical running data segment, represents the step size of the battery system temperature predicted by the length of the input historical operation data segment, and T represents the temperature of the battery system.

[0010] As an implementable method, the moving average kernel is used to decompose the operation data into a trend term and a residual term, and the decomposition formula is expressed as:

[0011]

[0012] in, , denote the trend term and residual term of decomposition respectively, is the moving average function, Indicates operating data.

[0013] As an implementable method, before using the moving average kernel to decompose the running data into trend terms and residual terms, the DLSTM model needs to be trained first. The specific training steps include: Acquire the normal operating data of the battery system, clean the operating data, and obtain a normal battery system data set; The normal battery system data set is divided into a training set, a test set and a validation set in proportion, and the multi-step temperature prediction model of the battery system based on the DLSTM model is trained to obtain a trained multi-step temperature prediction model of the battery system based on the DLSTM model.

[0014] As an implementable method, the following steps are included: The multi-step temperature prediction method of the battery system based on the DLSTM model is used to predict the temperature changes of the sensors in the future period of the battery system. Use the PCA algorithm to reduce the dimension of the predicted multi-step temperature to prevent the data dimension from being too high and reducing the performance of the anomaly detection algorithm; The DBSCAN anomaly detection algorithm is used to locate abnormal heating temperature sensors in advance.

[0015] As an implementable embodiment, the diagnosis and early warning method of the abnormal temperature sensor is: The matrix Input into the DBSCAN algorithm and get the evaluation index:

[0016] In the formula, 0 and -1 respectively indicate that the temperature sensor has no abnormality and has an abnormality.

[0017] A multi-step temperature prediction system for a battery system based on a DLSTM model, comprising: a data decomposition module, an LSTM processing module, and a temperature prediction module. A data decomposition module is used to obtain the operating data of the battery system and clean the operating data. The DLSTM model includes a moving average kernel and two LSTM networks. The moving average kernel is used to decompose the data into a trend item and a residual item. An LSTM processing module is used to input the trend item and the residual item into two LSTM networks respectively to obtain the trend prediction feature and the residual prediction feature; The temperature prediction module is used to add the trend prediction feature and the residual prediction feature to obtain the predicted temperature of the battery system.

[0018] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following method is implemented: Acquire the operating data of the battery system, clean the operating data, the DLSTM model uses a moving average kernel and two LSTM networks for modeling, uses the moving average kernel to decompose the operating data into a trend item and a residual item, and performs a filling operation; The trend term and the residual term are input into two LSTM networks respectively to obtain the trend prediction feature and the residual prediction feature; The trend prediction feature and the residual prediction feature are added together to obtain the predicted temperature of the battery system.

[0019] A multi-step temperature prediction of a battery system based on a DLSTM model, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor implements the following method when executing the computer program: Acquire the operating data of the battery system, clean the operating data, the DLSTM model uses a moving average kernel and two LSTM networks for modeling, uses the moving average kernel to decompose the operating data into a trend item and a residual item, and performs a filling operation; The trend term and the residual term are input into two LSTM networks respectively to obtain the trend prediction feature and the residual prediction feature; The trend prediction feature and the residual prediction feature are added together to obtain the predicted temperature of the battery system.

[0020] In summary, the multi-step temperature prediction method, system and device for a battery system based on a DLSTM model described in this application have the following advantages: 1. The present invention adopts a multi-step temperature prediction method for a battery system based on a DLSTM model. The operating data is first decomposed into a trend item and a residual item, and the trend item and the residual item are respectively input into an LSTM network for processing. Finally, the two processing results are added together to obtain the predicted temperature of the final battery system. The results show that the multi-step temperature prediction method for a battery system based on a DLSTM model has greatly improved robustness and accuracy. 2. The present invention combines the battery system temperature prediction results based on the DLSTM model with relevant outlier identification methods to locate abnormal heating temperature sensors in advance, ensure the safe operation of the battery system, and avoid thermal runaway accidents. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0022] The following drawings are examples of specific implementation methods of the present application: Figure 1 It is a structural schematic diagram of the DLSTM model of the present invention; Figure 2 It is a schematic diagram of the structure of the LSTM network described in the present invention; Figure 3 It is a flow chart of the multi-step temperature prediction method of the battery system based on the DLSTM model of the present invention; Figure 4 It is a structural diagram of a multi-step temperature prediction system for a battery system based on a DLSTM model according to the present invention; Figure 5 It is the relative error of battery system temperature prediction based on DLinear direct multi-step temperature prediction model; Figure 6is the relative error of the multi-step temperature prediction of the battery system based on the DLSTM model described in the present invention; Figure 7 It is a schematic diagram of the battery system thermal runaway fault diagnosis and early warning process constructed by the present invention; Figure 8 It is the percentage variance of the different numbers of principal components output by PCA; Fig. 9 It is the diagnosis result of DBSCAN algorithm at the 3772th iteration; Fig.10 is the reliability factor ε of the DBSCAN algorithm under different Eps. DETAILED DESCRIPTION

[0023] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings 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 should fall within the scope of protection of the present invention.

[0024] Embodiment 1: A multi-step temperature prediction method for battery system based on DLSTM (Decomposition Long Short-Term Memory) model, such as Figure 3 As shown, the following steps are included: S100, obtaining the operating data of the battery system, cleaning the operating data, the DLSTM model adopts a moving average kernel and two LSTM networks for modeling, and uses the moving average kernel to decompose the data into a trend item and a residual item; S200, inputting the trend term and the residual term into two LSTM networks respectively to obtain trend prediction features and residual prediction features; S300: Add the trend prediction feature and the residual prediction feature to obtain a predicted temperature of the battery system.

[0025] The embodiment of the present invention proposes a multi-step temperature prediction method for a battery system based on a DLSTM model for realizing temperature prediction of a battery system. The structural schematic diagram of the DLSTM model is shown in FIG. Figure 1As shown, it includes a moving average kernel and two LSTM networks. The multi-step temperature prediction method of the battery system mainly consists of three parts: the first part uses the DLSTM model to decompose the operating data into trend items and residual items; the second part inputs the trend items and residual items into two LSTM networks respectively to obtain trend prediction features and residual prediction features; the third part adds the obtained trend prediction features and residual prediction features to obtain the final battery system predicted temperature.

[0026] The structural diagram of the LSTM network is as follows Figure 2 As shown in the figure, the repetitive module in the LSTM network is composed of three sigmoid activation functions and one activation function, and a memory cell with the same shape as the hidden state is introduced to record additional information. In order to control the memory cell, other gates are introduced, among which the output gate is used to output entries from the memory cell; the input gate is used to decide whether to read data into the memory cell; the update gate is used to control how much state information of the hidden state at the previous moment is transferred to the current hidden state; the forget gate is used to reset the content of the memory cell. This gate uses a dedicated mechanism to decide when to ignore or remember the input in the hidden state. The working principle of each gate is introduced in detail below.

[0027] Forget Gate The forget gate mainly determines what information to discard from the previous memory element. This function is implemented by the sigmoid fully connected layer. The forget gate looks at the output of the previous node. and the current node input , is the previous memory state Output a number between 0 and 1 for each information in, 1 means completely retain, and 0 means completely delete:

[0028] in, and Respectively represent the weight parameters and bias parameters of the forget gate, Indicates the state of the forget gate of the current node, Represents the output of the previous node, Represents the input of the current node.

[0029] (1) Input gate The sigmoid fully connected layer of the input gate determines the degree of information update; the tanh fully connected layer is used to generate candidate state quantities added to the memory element. . Current node input And the previous node output After passing through the input gate, the new state created by the two activation functions can be expressed as follows:

[0030]

[0031] In the formula, and Respectively represent the weight parameters of the input gate; and They represent the bias parameters of the input gate respectively; and Respectively represent the new state quantities generated by the sigmoid activation function and the tanh activation function, Represents the output of the previous node, Represents the input of the current node.

[0032] (2) Update gate The update gate does not use any activation function. It iterates the output state of the forget gate and the input gate through multiplication and sum operations, which can be expressed as follows:

[0033] in, and Respectively represent the new state quantities generated by the sigmoid activation function and the tanh activation function, Indicates the state of the forget gate of the current node, Represents the previous memory state, Indicates the current state.

[0034] From the above formula, we can see that the update mechanism of the update gate is to linearly superimpose the state quantities generated by the forget gate and the input gate. If the input of the previous node is not acted upon by the activation function, but only passes through the hidden layer by simple summation, it is called long-term information. No gradient attenuation occurs during the transmission process, and the information can be stored for a long time; on the contrary, the information that passes through the fully connected layer of the activation function is called short-term information.

[0035] (3) Output gate The output gate will first input the current node Same as the previous node output The integrated state quantity uses the sigmoid activation function to extract the information, and then uses the tanh activation function to update the current node memory state. Finally, the results of the two activation functions are multiplied to get the output:

[0036]

[0037] in, and denote weight parameters and bias parameters respectively, Represents the output of the current node, represents the output of the output gate, Indicates the current node input, Indicates the output of the previous node.

[0038] In this embodiment, in step S100, the operating data of the battery system is obtained, the operating data is cleaned, and the data is decomposed into trend items and residual items using a moving average kernel. The specific steps include: S101 obtains the operating data of the battery system, the operating data including: total current, total voltage, sensor temperature, insulation resistance and state of charge. The operating data is expressed as:

[0039] in, is the number of features related to temperature, is the number of temperature sensors in the battery system, To enter the length of the historical run data segment, , , , and They are current, SOC, total battery system voltage, insulation resistance and temperature.

[0040] S102 cleans the operating data. In this embodiment, the operating data is cleaned using a Savitzky-Golay filter to reduce the noise of each charging and discharging segment.

[0041] S103 uses a moving average kernel to decompose the data into a trend term and a residual term.

[0042] The moving average kernel is used to decompose the data into trend terms and residual terms, and the decomposition formula is:

[0043]

[0044] in, , Represents the extracted trend term and residual term, and uses A moving average is implemented and a padding operation is performed to keep the data length unchanged, and the kernel size is set to 25.

[0045] S200, trend item and the residual Input into the LSTM network respectively to obtain the prediction features and :

[0046]

[0047] in, Represents the trend prediction characteristics, represents the residual prediction features.

[0048] S300, adding the trend prediction feature and the residual prediction feature to obtain the predicted temperature of the battery system. The temperature of the battery system predicted by the DLSTM model is expressed as:

[0049] in, is the step size of the predicted battery system temperature.

[0050] The present invention adopts a multi-step temperature prediction method for a battery system based on the DLSTM model. The operating data is first decomposed into a trend item and a residual item, and the trend item and the residual item are respectively input into the LSTM network for processing. Finally, the two processing results are added together to obtain the predicted temperature of the final battery system. The results show that the multi-step temperature prediction method for a battery system based on the DLSTM model has greatly improved in robustness and accuracy.

[0051] Example 2: A battery system thermal runaway fault diagnosis and early warning method based on the DLSTM model The battery system thermal runaway fault diagnosis and early warning method based on the DLSTM model is as follows: Figure 6 As shown, the following steps are included: S1, using the battery system multi-step temperature prediction method based on the DLSTM model to predict the temperature change of the sensor in the future period of the battery system; S2, use the PCA algorithm to reduce the dimension of the predicted multi-step temperature to prevent the data dimension from being too high and reducing the performance of the anomaly detection algorithm; S3, use the DBSCAN anomaly detection algorithm to locate the abnormal heating temperature sensor in advance.

[0052] The present invention uses two sets of battery system data of the same specifications, one of which is the first normal battery system operation data, which does not have any faults during operation. The data of this battery system is used to train the DLSTM model, and the other set is the second thermal runaway fault battery system operation data, and its data is used to verify the feasibility of the proposed thermal runaway fault diagnosis and early warning method. However, under complex operating conditions, the battery system contains a lot of noise in its data. For example, duplication or omission will inevitably occur during data transmission, and the temperature sensor has low resolution, which brings great challenges to accurately obtaining battery status information. In order to ensure the reliability of algorithm training, it is very important to clean the original data, including missing value filling and outlier correction, and the Savitzky-Golay filter is used to reduce the noise of each data segment.

[0053] (1) Training of the DLSTM model The DLSTM network is trained using the operating data of the first normal battery system after cleaning. The operating data of the first normal battery system is obtained, and the operating data is cleaned to obtain a normal battery system data set, which is divided into a training set, a test set, and a validation set in a ratio of 7:1:2. The battery system based on the DLSTM model is trained and validated to obtain a battery system based on the DLSTM model.

[0054] (2) Multi-step temperature prediction of thermal runaway fault battery system

[0055] After the DLSTM model is trained on the operating data of the first normal battery system, the operating data of the second thermal runaway fault battery system is used to predict the temperature change of the battery system before the thermal runaway occurs. The input HL, output PL and input HL moving step are set to 60, 30 and 1 respectively. This means that the historical operating data of the battery system within 5 minutes is used to predict the temperature change of the battery system within the next 3 minutes, and every time a new data is generated by the battery system operation, the input HL is kept unchanged, the old historical operating data is discarded, and the new operating data is included, and then a direct multi-step temperature prediction is performed, and it is iterated until the battery system stops sampling. There are 4200 data of the second thermal runaway fault battery system selected, and a total of 4141 direct multi-step predictions are performed, generating 12430 prediction data. In order to measure the error between the predicted data and the actual data, the predicted data are selected at time intervals, and the mean mean square error (MMSE) between the true value and the predicted value is calculated. The prediction of the temperature of the second thermal runaway fault battery system by the DLSTM model is shown in the attached figure. Figure 6 It can be clearly observed from the figure that the change trend of the predicted value curve of the DLSTM network is similar to the true value curve, and the MMSE of the multi-step temperature prediction value is 0.0251. The predicted value, true value and relative error curve of the DLinear network layer on the test set are as follows Figure 5 As shown in the figure, it can be observed that the shape trends of the predicted value and the true value curve are similar, but the relative error curve fluctuates greatly, and the maximum value is close to 1.5%. From the DLSTM prediction curve ( Figure 6) It can be seen that the predicted value curve of the DLSTM network is very consistent with the true value curve, the maximum relative error is less than 0.5%, and its fluctuation range is small. On the same test set, the root mean square error predicted by the DLinear model is 0.0491, and the root mean square error predicted by the DLSTM model is 0.0322, which confirms that the robustness and accuracy of the neural network are enhanced after replacing the DLinear network layer with the LSTM layer.

[0056] (3) PCA dimensionality reduction analysis The PCA algorithm performs a dimensionality reduction method on the predicted multi-step temperature features, specifically including: The zero mean matrix of is represented as:

[0057] In the formula, Represents temperature sensor In the The zero mean temperature for the time step is given by:

[0058] The matrix is ​​calculated by the following formula The covariance matrix of is:

[0059] Calculate and sort the matrix The eigenvalue of and the corresponding eigenvector :

[0060]

[0061] Select Matrix Before The eigenvectors form a matrix , then the matrix after dimensionality reduction is It can be expressed as:

[0062] The diagnosis and early warning method of the abnormal temperature sensor: The matrix Input into the DBSCAN algorithm and get the evaluation index:

[0063] In the formula, 0 and -1 respectively indicate that the temperature sensor has no abnormality and has an abnormality.

[0064] If the predicted temperature in the next 30 steps is directly input into the DBSCAN algorithm, the excessively high data dimension will cause the dimensionality disaster and reduce the performance of anomaly detection. Therefore, it is necessary to reduce the dimension of the predicted high-dimensional temperature data. The present invention adopts the PCA dimensionality reduction algorithm to extract the principal component components of the 30-step temperature data predicted each time. In order to determine the appropriate number of principal component components output by PCA, it is necessary to calculate the percentage of the explained variance of each principal component to the total variance. This value can reflect the amount of information contained in each principal component of the predicted temperature in the next 30 steps. Appendix Figure 7 The percentage variance of the output of different numbers of principal components was compared, and the percentage variance of the first three principal components was 31.6175%, 0.03645% and 0.001793% respectively. The number of principal components after dimensionality reduction was selected as 3 to retain 99.9982% of the predicted temperature information.

[0065] (4) DBSCAN algorithm detects abnormal temperature sensors Based on the direct multi-step temperature prediction results of the DLSTM network for the historical operating data of the second thermal runaway fault battery system, the PCA dimensionality reduction algorithm is used to reduce the dimension of the future 30-step temperature changes predicted in each iteration to 3 principal components, and then input them into the DBSCAN anomaly detection algorithm to locate the abnormal temperature sensor in advance. The main parameters of the DBSCAN algorithm include the neighborhood radius Eps and the minimum number of data points within the neighborhood radius MinPts. When using the DBSCAN algorithm to identify abnormal temperature sensors, first randomly select a three-dimensional data point, and count the number of other data points within the neighborhood radius Eps of the data point. If the number of other data points is not less than MinPts, the data points within the neighborhood radius Eps belong to the same cluster; if the number of other data points is less than MinPts, and the data point is not within the neighborhood radius Eps of other data points, the data point is determined to be an outlier. MinPts is usually set to twice the data dimension, that is, Eps is set to 0.11 in the present invention. At the 3772nd iteration, temperature sensor #13 was diagnosed as abnormal for the first time. Attached Figure 8 The fault diagnosis at the 3772th iteration is clearly shown. It can be observed that temperature sensor #13 and other normal sensors have obvious clustering. The 3772th iteration prediction corresponds to the historical operation data of sampling points 3772 to 3831, which shows that temperature sensor #13 was detected at 3:40:21 on June 28, 2020, when the temperature rose out of the group trend. The fault warning time was about 50 minutes earlier than the BMS alarm time.

[0066] (5) Neighborhood radius Eps sensitivity analysis In order to reduce false alarms, it is necessary to analyze the impact of different neighborhood radii Eps on the abnormal temperature sensor diagnosis results. The Eps interval is selected as 0.05~0.2 and the sampling interval is 0.01. Define the reliability factor of the DBSCAN algorithm If the algorithm has no false alarms before the 3722nd iteration (when the temperature sensor #13 is first detected to be abnormal), and can accurately identify the abnormal temperature sensor after the 3722nd iteration, then , indicating that the algorithm has no false alarms or missed alarms and has high reliability under the neighborhood radius Eps; if the algorithm falsely alarms before the 3722nd iteration or misses an alarm after the 3722nd iteration, then , the algorithm reliability is low. Reliability factors under different field radius Eps The results are attached Fig. 9 As shown. Fig. 9 It can be seen that when Eps is 0.1~0.15, the reliability factor Both are 1, indicating that the DBSCAN algorithm can accurately locate abnormal temperature sensors in this range. Therefore, the optimal value of Eps is 0.1~0.15.

[0067] Embodiment 3: A multi-step temperature prediction system for a battery system based on a DLSTM model, such as Figure 4 As shown, it includes: a data decomposition module 100, an LSTM processing module 200, and a temperature prediction module 300. The data decomposition module 100 is used to obtain the operating data of the battery system and clean the operating data. The DLSTM model includes a moving average kernel and two LSTM networks. The moving average kernel is used to decompose the data into a trend item and a residual item. The LSTM processing module 200 is used to input the trend item and the residual item into two LSTM networks respectively to obtain the trend prediction feature and the residual prediction feature; The temperature prediction module 300 is used to add the trend prediction feature and the residual prediction feature to obtain the predicted temperature of the battery system.

[0068] Various changes and modifications can be made without departing from the spirit and scope of the present invention, and all equivalent technical solutions also belong to the scope of the present invention.

[0069] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0070] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, devices or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0071] The present invention is described with reference to the flowcharts and / or block diagrams of the method, terminal device (system), and computer program product according to the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0072] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0073] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable terminal device to implement the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0074] It should be noted that: The "one embodiment" or "embodiment" mentioned in the specification means that the specific features, structures or characteristics described in conjunction with the embodiment are included in at least one embodiment of the present invention. Therefore, the phrases "one embodiment" or "embodiment" appearing in various places throughout the specification do not necessarily refer to the same embodiment.

[0075] In addition, it should be noted that the shapes and names of the parts of the specific embodiments described in this specification may be different. Any equivalent or simple changes made based on the structure, features and principles described in the patent concept of the present invention are included in the protection scope of the patent of the present invention. The technicians in the technical field of the present invention can make various modifications or supplements to the specific embodiments described or replace them in a similar manner, as long as they do not deviate from the structure of the present invention or exceed the scope defined by the claims, they should all fall within the protection scope of the present invention.

Claims

1. A multi-step temperature prediction method for a battery system based on a DLSTM model, characterized in that: The following steps are involved: Acquire the operating data of the battery system, clean the operating data, the DLSTM model uses a moving average kernel and two LSTM networks for modeling, uses the moving average kernel to decompose the operating data into a trend item and a residual item, and performs a filling operation; The trend term and the residual term are input into two LSTM networks respectively to obtain the trend prediction feature and the residual prediction feature; The trend prediction feature and the residual prediction feature are added together to obtain the predicted temperature of the battery system.

2. The multi-step temperature prediction method for a battery system based on a DLSTM model according to claim 1, characterized in that: The operation data includes: total current, total voltage, sensor temperature, insulation resistance and state of charge. The operation data is expressed as: in, represents the number of features related to temperature, Indicates the number of temperature sensors in the battery system, Indicates the length of the running data, , , , and They represent the current, SOC, total voltage, insulation resistance and temperature of the battery system respectively.

3. The multi-step temperature prediction method for a battery system based on a DLSTM model according to claim 2, characterized in that: The temperature of the battery system predicted by the DLSTM model is expressed as: in, represents the temperature of the battery system predicted by the DLSTM model, Indicates the length of the input historical running data segment, represents the step size of the predicted battery system temperature, and T represents the temperature of the battery system.

4. The multi-step temperature prediction method for a battery system based on a DLSTM model according to claim 1, characterized in that: The moving average kernel is used to decompose the operating data into a trend term and a residual term, and the decomposition formula is expressed as: in, , denote the trend term and residual term of decomposition respectively, is the moving average function, Indicates operating data.

5. The multi-step temperature prediction method for a battery system based on a DLSTM model according to claim 1, characterized in that: Before using the moving average kernel to decompose the running data into trend terms and residual terms, the DLSTM model needs to be trained. The specific training steps include: Acquire the normal operating data of the battery system, clean the operating data, and obtain a normal battery system data set; The normal battery system data set is divided into a training set, a test set and a validation set in proportion, and the multi-step temperature prediction model of the battery system based on the DLSTM model is trained to obtain a trained multi-step temperature prediction model of the battery system based on the DLSTM model.

6. A battery system thermal runaway fault diagnosis and early warning method based on claim 1, characterized in that: The following steps are involved: The multi-step temperature prediction method of the battery system based on the DLSTM model is used to predict the temperature changes of the sensors in the future period of the battery system. Use the PCA algorithm to reduce the dimension of the predicted multi-step temperature to prevent the data dimension from being too high and reducing the performance of the anomaly detection algorithm; The DBSCAN anomaly detection algorithm is used to locate abnormal heating temperature sensors in advance.

7. A battery system thermal runaway fault diagnosis and early warning method according to claim 6, characterized in that: The diagnosis and early warning method of the abnormal temperature sensor: The matrix Input into the DBSCAN algorithm and get the evaluation index: In the formula, 0 and -1 respectively indicate that the temperature sensor has no abnormality and has an abnormality.

8. A multi-step temperature prediction system for a battery system based on a DLSTM model, characterized in that: include: Data decomposition module, LSTM processing module, temperature prediction module, A data decomposition module is used to obtain the operating data of the battery system and clean the operating data. The DLSTM model includes a moving average kernel and two LSTM networks. The moving average kernel is used to decompose the data into a trend item and a residual item. An LSTM processing module is used to input the trend item and the residual item into two LSTM networks respectively to obtain the trend prediction feature and the residual prediction feature; The temperature prediction module is used to add the trend prediction feature and the residual prediction feature to obtain the predicted temperature of the battery system.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

10. A multi-step temperature prediction device for a battery system based on a DLSTM model, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

Images