Research On SOC Prediction And Balance Control Of Lithium Battery Packs Based On Machine Learning

In recent years,with the popularity of electric vehicles and the development of artificial intelligence,a high-performance battery management system(BMS)is essential to enhance the driving experience and ensure the safety of the vehicle.Battery state of charge(SOC)estimation and battery balance control are two important functions of BMS,which have been studied extensively by related scholars.At present,the traditional SOC prediction algorithm of Li-ion battery pack is not widely used and the accuracy is not high.In addition,the structure of the battery balance control method is complicated and the operation efficiency is low.For this reason,the following work has been done in this paper to address the above problems.(1)Using a variety of representative single-model machine learning algorithms to estimate the SOC of Li-ion battery packs,and evaluating the effectiveness of the algorithms in terms of evaluation index and model fitting effect,it is found that the Random Forest algorithm based on the bagging ensemble method and the Light GBM based on the boosting ensemble method have lower MAE and RMSE.(2)A battery SOC prediction model based on Stacking fusion and weighting algorithm is proposed.The single model with better effect is selected as the base model for model fusion,and the SOC of Li-ion battery pack is predicted by using linear weighted fusion and Stacking fusion respectively,and it is found that both have better effect than the single model for SOC estimation,which verifies the feasibility and effectiveness of the model fusion algorithm.To further improve the accuracy of estimating the battery SOC,the prediction results of Stacking fusion are used as the base model and other single models with better results are linearly weighted,and the final experimental results show that the MAE and RMSE are reduced by15.08%and 14.13%respectively compared with the Light GBM algorithm,while the R~2 is improved by 0.34%.(3)Propose a battery voltage balance reinforcement learning algorithm based on Proximal Policy Optimization(PPO).In this paper,we first build a 3D model based on ANSYS and simulate the discharge of the battery at different multipliers in Fluent.Since the reinforcement learning environment requires continuous iterable computation,a 3D smooth and uniform surface of(voltage,multiplier,time)is reconstructed for this purpose.The battery voltage is the input state,discharge rate of battery is the action output,and the reconstructed surface is the environment to simulate the battery discharging process.The rate is selected by self-updating through the adjustment parameters of the Actor-Critic network in the PPO algorithm,and the two imbalance cases are experimented separately,and the results show that the voltage difference is reduced from 0.2V to about 0.01V.This reinforcement learning algorithm saves resources by eliminating the need for excess battery power compared to passive equalization,and it also adjusts the output battery ratio according to the reward function and has better results.In this paper,we propose a solution based on artificial intelligence algorithm for two main functions in BMS.In the estimation of battery SOC,the proposed Stacking fusion and weighting algorithm for battery SOC prediction model has smaller MAE and RMSE and higher R~2 than the optimal single model.In the battery balance control study,the PPO reinforcement learning algorithm self-adjusts the battery discharge ratio while adjusting the voltage difference from 0.2V to within 0.011V.