CN-122008950-A - Electric automobile energy management method and system based on lithium-sodium hybrid battery system

CN122008950ACN 122008950 ACN122008950 ACN 122008950ACN-122008950-A

Abstract

The invention provides an electric vehicle energy management method and system based on a lithium-sodium hybrid battery system, and relates to the technical field of energy storage systems, wherein the method comprises the steps of obtaining state transition data of a lithium-sodium hybrid battery sample, training a preset depth Q network by using the state transition data, and generating an initial optimal control strategy; the method comprises the steps of carrying out light weight processing on a lightweight time sequence model through knowledge distillation, structural pruning and linear quantization to obtain a target TTM prediction model, inputting a power characteristic sequence of a lithium-sodium hybrid battery sample into the target TTM prediction model to obtain future power prediction information, taking minimized initial output power and initial load power as objective functions, embedding power balance constraint into the objective functions to construct an extended Lagrangian function, inputting the future power prediction information into the extended Lagrangian function, and alternately optimizing the initial output power and the initial load power in an initial optimal control strategy to obtain the target optimal control strategy.

Inventors

ZHOU ZE
ZHOU ZHEXIN
XIA YING
ZHANG LIYAN
CHEN QIHONG
ZHAO DONGQI
LI ZHOUBO
DENG MINGMING

Assignees

武汉理工大学

Dates

Publication Date: 20260512
Application Date: 20260326

Claims (10)

1. An electric vehicle energy management method based on a lithium-sodium hybrid battery system is characterized by comprising the following steps: Acquiring state transition data of a lithium-sodium hybrid battery sample, training a preset depth Q network by using the state transition data, and generating an initial optimal control strategy, wherein the state transition data comprises a current state, an execution action, an obtained reward and a new state after transition; Carrying out light weight treatment on a lightweight time series model through knowledge distillation, structural pruning and linear quantization to obtain a target TTM prediction model, and inputting a power characteristic sequence of the lithium-sodium hybrid battery sample into the target TTM prediction model to obtain future power prediction information; and taking the minimized initial output power and the initial load power as objective functions, embedding power balance constraint into the objective functions, constructing an augmented Lagrangian function, inputting the future power prediction information into the augmented Lagrangian function, and alternately optimizing the initial output power and the initial load power in the initial optimal control strategy to obtain the target optimal control strategy.
2. The method for managing energy of an electric vehicle based on a lithium-sodium hybrid battery system of claim 1, wherein the preset depth Q network comprises a target network and a prediction network, wherein training the preset depth Q network by using the state transition data to generate an initial optimal control strategy comprises: The state transition data are respectively input into the target network and the prediction network to obtain a corresponding target Q value and a current Q value; Calculating a mean square error loss function based on the target Q value and the current Q value, and updating the weight of the target network by utilizing the back propagation iteration of the mean square error loss function until the mean square error loss function converges to obtain a target depth Q network; And traversing the state transition data by utilizing the target depth Q network, screening out a target execution action corresponding to the maximum target Q value, and determining initial output power and initial load power corresponding to the target execution action as an initial optimal control strategy.
3. The method for managing energy of an electric vehicle based on a lithium-sodium hybrid battery system according to claim 2, further comprising, before inputting the state transition data to the target network and the prediction network, respectively, obtaining a corresponding target Q value and a current Q value: determining the sampling priority of each group of state transition data based on the time sequence difference error of each group of state transition data, wherein the sampling priority is higher for the state transition data with larger time sequence difference error; And combining the sampling priority to sample the state transition data to obtain training sample data, and respectively inputting the training sample data into the target network and the prediction network to obtain a corresponding target Q value and a current Q value.
4. The method for managing energy of an electric vehicle based on a lithium-sodium hybrid battery system according to claim 1, wherein the performing light-weight processing on the light-weight time-series model through knowledge distillation, structural pruning and linear quantization to obtain the target TTM prediction model comprises: Training the lightweight time series model by using the large model as a teacher model, and minimizing output difference to obtain a first TTM prediction model; performing structural pruning on a channel with contribution lower than a preset threshold value in the first TTM prediction model based on a BN layer scaling factor to obtain a second TTM prediction model; And linearly mapping the weight of the second TTM prediction model to an 8-bit integer interval by adopting linear quantization to obtain a target TTM prediction model.
5. The lithium-sodium hybrid battery system-based electric vehicle energy management method of claim 1, wherein the constructing an augmented lagrangian function with the objective function of minimizing the initial output power and the initial load power and embedding a power balance constraint in the objective function comprises: Constructing a double-packet collaborative optimization problem in a rolling time domain, taking the initial output power and the initial load power as target functions, embedding power balance constraint into the target functions, and constructing an augmented Lagrangian function, wherein the power balance constraint comprises power balance, power constraint and SOC constraint for the initial output power and the initial load power.
6. The method for managing energy of an electric vehicle based on a lithium-sodium hybrid battery system according to claim 1, wherein inputting the future power prediction information into the augmented lagrangian function, and performing an alternate optimization on the initial output power and the initial load power in the initial optimal control strategy to obtain a target optimal control strategy, comprises: under the condition of keeping the Lagrangian multiplier unchanged, fixing the initial output power, optimizing the initial load power, or fixing the initial load power, and optimizing the initial output power; and sequentially and iteratively updating by adopting a reverse loop coordinate descent method to obtain a target optimal control strategy.
7. The method for managing energy of an electric vehicle based on a lithium-sodium hybrid battery system according to claim 6, wherein the method for sequentially and iteratively updating by a reverse cyclic coordinate descent method to obtain a target optimal control strategy further comprises: And after the power variation of the initial output power and the initial load power is smaller than a preset power threshold, updating Lagrange multipliers and penalty factors in the augmented Lagrange function, and balancing constraint errors and convergence speed.
8. The electric automobile energy management system based on the lithium-sodium hybrid battery system is characterized by comprising an initial strategy generation module, a future power prediction module and a target strategy generation module, wherein, The initial strategy generation module is configured to acquire state transition data of a lithium-sodium hybrid battery sample, train a preset depth Q network by utilizing the state transition data to generate an initial optimal control strategy, wherein the state transition data comprises a current state, an execution action, an obtained reward and a new state after transition; the future power prediction module is configured to perform light weight processing on the lightweight time series model through knowledge distillation, structural pruning and linear quantization to obtain a target TTM prediction model, and input a power characteristic sequence of the lithium-sodium hybrid battery sample into the target TTM prediction model to obtain future power prediction information; The target strategy generation module is configured to take the minimized initial output power and the initial load power as target functions, embed power balance constraint into the target functions, construct an augmented Lagrangian function, input the future power prediction information into the augmented Lagrangian function, and alternately optimize the initial output power and the initial load power in the initial optimal control strategy to obtain a target optimal control strategy.
9. An electronic device comprising a processor and a memory, the memory storing a computer program, wherein the computer program when executed by the processor implements the lithium-sodium hybrid battery system-based electric vehicle energy management method of any one of claims 1-7.
10. A non-transitory computer readable storage medium, having stored thereon a computer program, wherein the computer program when executed by a processor implements the lithium-sodium hybrid battery system-based electric vehicle energy management method of any one of claims 1 to 7.

Description

Electric automobile energy management method and system based on lithium-sodium hybrid battery system Technical Field The invention relates to the technical field of energy storage systems, in particular to an electric automobile energy management method and system based on a lithium-sodium hybrid battery system. Background With the increasing importance of global environmental protection, electric vehicles are receiving more and more attention due to their zero emission and low noise operation. However, in a low temperature environment, the endurance mileage of the electric vehicle is significantly attenuated, mainly due to the decrease in battery performance, the increase in thermal management energy consumption, and the decrease in system efficiency. Lithium batteries have increased internal resistance and greatly reduced usable capacity in low temperature environments, while sodium batteries have excellent low temperature performance but lower energy density. The lithium-sodium hybrid battery is used as a novel energy storage technology combining the advantages of the lithium ion battery and the sodium ion battery, and has been remarkably developed in the aspects of technology cooperation, application expansion and industrialization promotion in recent years. However, the existing energy management strategies often cannot effectively coordinate the power distribution and the thermal management of the lithium-sodium hybrid battery, so that the energy utilization efficiency is low, and the endurance mileage cannot be maximized. Disclosure of Invention In view of the above, the invention provides an electric vehicle energy management method and system based on a lithium-sodium hybrid battery system. The technical scheme of the invention is realized in such a way that the first aspect of the invention provides an electric automobile energy management method based on a lithium-sodium hybrid battery system, which comprises the following steps: Acquiring state transition data of a lithium-sodium hybrid battery sample, training a preset depth Q network by using the state transition data, and generating an initial optimal control strategy, wherein the state transition data comprises a current state, an execution action, an obtained reward and a new state after transition; Carrying out light weight treatment on a lightweight time series model through knowledge distillation, structural pruning and linear quantization to obtain a target TTM prediction model, and inputting a power characteristic sequence of the lithium-sodium hybrid battery sample into the target TTM prediction model to obtain future power prediction information; and taking the minimized initial output power and the initial load power as objective functions, embedding power balance constraint into the objective functions, constructing an augmented Lagrangian function, inputting the future power prediction information into the augmented Lagrangian function, and alternately optimizing the initial output power and the initial load power in the initial optimal control strategy to obtain the target optimal control strategy. On the basis of the technical scheme, preferably, the preset depth Q network comprises a target network and a prediction network, the training of the preset depth Q network by using the state transition data is performed to generate an initial optimal control strategy, and the method comprises the following steps: The state transition data are respectively input into the target network and the prediction network to obtain a corresponding target Q value and a current Q value; Calculating a mean square error loss function based on the target Q value and the current Q value, and updating the weight of the target network by utilizing the back propagation iteration of the mean square error loss function until the mean square error loss function converges to obtain a target depth Q network; And traversing the state transition data by utilizing the target depth Q network, screening out a target execution action corresponding to the maximum target Q value, and determining initial output power and initial load power corresponding to the target execution action as an initial optimal control strategy. On the basis of the above technical solution, preferably, before the state transition data is input to the target network and the prediction network respectively to obtain the corresponding target Q value and the current Q value, the method further includes: determining the sampling priority of each group of state transition data based on the time sequence difference error of each group of state transition data, wherein the sampling priority is higher for the state transition data with larger time sequence difference error; And combining the sampling priority to sample the state transition data to obtain training sample data, and respectively inputting the training sample data into the target network and the prediction network to obtain a correspon