Research On State Estimation Of Lithium Battery Based On Improved Feedback Neural Network

In order to protect and monitor the battery,the BMS is born.It mainly includes real-time monitoring of state parameters such as SOC and SOE.SOC and SOE mainly reflect the amount of remaining power and remaining energy of lithium-ion battery.SOC and SOE are the key to realize real-time detection in battery management system.In this thesis,the SOC and SOE estimation of lithium battery are studied,and the following work is mainly completed:(1)In order to explore the nonlinear change law of internal parameters of lithium battery under different influencing factors.In this thesis,the relationship between internal parameters and external conditions is established and the internal parameters of lithium battery are modified.In this thesis,open-circuit voltage characteristic experiment,capacity characteristic experiment and impulse response characteristic experiment are carried out.The variation law of internal parameters of lithium battery under different external conditions was summarized by experiments.The relationship between internal parameters and external factors is clarified and the internal parameters of lithium battery are modified.(2)In order to solve the problem of different correlation between different vectors and state parameters.The state parameter characteristic model is constructed with different evaluation indexes.The model is used to realize the strong correlation characteristics of neural network.Firstly,the appropriate vector is selected,and the vector is preliminarily selected.This thesis uses evaluation indicators to explore the influence of different vectors on SOC and SOE.Finally,a strong correlation system of SOC and SOE is constructed,and the vector with strong correlation is used as the training of neural network.(3)In order to reduce the influence of historical state and loss function loss,a neural network model considering historical factors and harmonic loss is constructed to realize the extraction of past state information by neural network.In order to reduce the influence of historical factors,the structure of encoding and decoding is constructed.The intermediate variable C is obtained by weighted sum,and the timing sequence information is extracted.In order to reduce the influence of the loss function,the harmonic loss is used as the historical loss function by adding the basic forgetting curve.In order to overcome the loss function divergence caused by harmonic loss,the weight distribution method is used to improve it.(4)In order to verify the effectiveness of the proposed algorithm,the experimental verification system of SOC and SOE estimation for lithium battery is designed.The SOC and SOE are verified by stress testing conditions and dynamic stress testing conditions of Beijing buses.The maximum RMSE of SOC estimation is 1.49% and the maximum MAE is 1.29%under different working conditions.The maximum RMSE of SOE estimation is 1.30%,and the maximum MAE is 1.09%.In this thesis,the characteristics of capacity and energy under different conditions are obtained by obtaining the working characteristics of lithium-ion batteries.In determining the best input vector,HGRU with harmonic loss is used to accurately estimate SOC and SOE.Through experimental verification and analysis,the proposed method can improve the estimation accuracy of SOC and SOE state parameters,and provide a theoretical basis for the management of BMS.