基于深度学习的SOC预测模型比较研究
Comparative Research on SOC Prediction Models Based on Deep Learning
刘建华 1陈治铭 1陈可纬 1陈林颖1
作者信息
- 1. 福建工程学院计算机科学与数学学院 福州 350118;福建省大数据挖掘与应用技术重点实验室 福州 350118
- 折叠
摘要
锂离子电池的荷电状态(SOC)所包含的物理性质和电化学性质十分复杂,通常很难直接测定其数值,利用大数据的机器学习模型来预测SOC成为一个重要技术方法.近年来,随着神经网络的深度学习算法发展,基于深度学习的SOC估计模型已取得突破成果.论文总结了锂离子电池荷电状态预测方法的深度学习方法,主要分析比较CNN、GRU、LSTM、CNN-LSTM和CNN-GRU的经典模型方法与特点,通过实验数据分析对比其各模型的效果.论文对比实验选取不同室温环境下的不同工况作为测试集,通过预测结果的误差评估发现卷积神经网络对于循环神经网络的预测结果优化有较大的提升,其中CNN-LSTM的效果尤为显著.
Abstract
The physical and electrochemical properties of the state of charge(SOC)of lithium-ion batteries are very com-plex,so it is often difficult to directly measure the value of SOC.The machine learning model of big data has become an important technical method to predict SOC.In recent years,with the development of deep learning algorithms of neural networks,SOC estima-tion models based on deep learning have made breakthroughs.This paper summarizes the deep learning method of lithium-ion bat-tery state prediction method,mainly analyzes and compares the classical model methods and characteristics of CNN,GRU,LSTM,CNN-LSTM and CNN-GRU,and compares the effect of each model through the analysis of experimental data.In this paper,differ-ent working conditions under different room temperature environments are selected as test sets in the comparative experiment.Through the error evaluation of prediction results,it is found that convolutional neural network has greatly improved the optimization of prediction results of cyclic neural network,among which CNN-LSTM has a particularly significant effect.
关键词
SOC/卷积神经网络/循环神经网络/CNN-LSTM/CNN-GRUKey words
SOC/convolutional neural network/recurrent neural network/CNN-LSTM/CNN-GRU引用本文复制引用
基金项目
中央引导地方项目(2020L3024)
福州市科技创新平台项目(2021-P-052)
福建工程学院发展基金项目(GY-Z20046)
出版年
2024
国家科技期刊平台
NSTL
万方数据