Research On Power Battery SOC Algorithm Based On Recurrent Neural Network Structure

With the increasing market share of my country’s new energy vehicles,Lithiumion power batteries have become the power source of choice for power storage devices of new energy vehicles due to their various advantages.However,how to accurately estimate the available power of the power battery through the battery management system at different temperatures is still a key issue that needs to be improved.Therefore,this article selects the 18650 lithium-ion power battery commonly used in battery packs for electric vehicles as the experimental object,and develops a platform for estimating the remaining power of the power battery that adapts to different temperatures.First,test the battery charging and discharging characteristics under different temperatures for HPPC test conditions and different cycle conditions,and then perform parameter identification on the equivalent circuit model of the lithium-ion battery.Based on the variant cyclic neural network structure,the BP neural network is built.Network and long short-term memory(long short-term memory,LSTM)models.The autoencoder structure is introduced to optimize the LSTM model.The designed algorithms are compared and verified.The specific research content is as follows:First,at different temperatures(0℃,10℃,25℃,40℃),the LG18650DBHG2 battery was tested in HPPC test conditions.The tests were carried out in the U.S.Urban Road Cycle Service(UDDS)and the U.S.Highway Cycle Service.Test conditions(HWFET)and Los Angeles 92 test conditions(LA92)carried out charging and discharging experiments for power batteries and processed relevant data.Second,the BP neural network identification algorithm was built to identify the second-order equivalent circuit model of the lithium-ion power battery,and the reliability of the identification algorithm was verified.The results showed that the ohmic internal resistance of the lithium-ion power battery can be more accurately reflected Out of the battery life,the greater the ohmic resistance,the more obvious the battery life attenuation.At the same time,during the battery discharge process,the battery polarization effect will significantly affect the discharge performance of the battery.The electrochemical polarization internal resistance value is higher at the beginning and end of the battery discharge,and lower in the middle of the discharge,and the concentration polarization internal resistance also shows Similar laws.Third,based on the above parameter identification algorithm,part of the identification results of the BP neural network are used as the input of the long and short-term memory network to form a BP-LSTM model to estimate the SOC changes of lithium-ion batteries,introduce the autoencoder structure,and propose a BP-LSTM-SOC estimation algorithm of ED model.Compared with other neural network algorithms,the BP-LSTM-ED model has the highest algorithm accuracy.The SOC estimation results of three different working conditions at room temperature show that the average error is within 1.2%,and the root mean square error is within 1.7%.The error of SOC estimation at different temperatures is controlled within 2%.It is verified that the long and short-term memory network model still has high accuracy and stability under different environmental temperatures.Finally,according to the hardware architecture requirements of the lithium-ion power battery management system,the SOC algorithm based on the BP-LSTM-ED structure is selected to build a hardware test platform,and various functional modules are designed.In the hardware-in-the-loop test of the battery,different temperatures are tested.Under different working conditions,the root mean square error can be controlled within 1.8%,and the maximum absolute error is also controlled within 4%.This verifies the SOC estimation method of lithium-ion battery based on the BP-LSTM-ED structure proposed in this paper.feasibility.The SOC estimation model of lithium-ion power battery based on recurrent neural network studied in this paper has certain practical value,and it has certain reference significance for the prediction of SOC of electric vehicle battery pack module.