SOC Estimation For Lithium Power Batteries Based On CNN-LSTM Network

With the continuous promotion of ecological environment construction and the key support of the state,low energy consumption and pollution-free electric vehicles are developing rapidly.The main technical bottleneck of electric vehicles lies in battery technology,predicting the state of battery is the main difficulty in battery management.The accurate estimation of battery State of Charge(SOC)is one of the most important State parameters,which plays a very important role in ensuring the safe and effective operation of battery.However,the estimation method of SOC has always had various limitations,so how to improve the accuracy and versatility of the estimation methods is of great significance.Lithium batteries are widely used in electric vehicles due to their superior characteristics.This article selects lithium batteries as the research object,uses data-driven methods,analyzes data characteristics,and establishes a deep learning battery SOC prediction model.The main work of this article is as follows:Firstly,the research status of SOC estimation methods is recommend,and outlined the relative merits of different methods.On this basis,the research direction of this article is determined.Secondly,this article analyzes the working principle and performance characteristics of lithium batteries,and builds an experimental platform to conduct experimental analysis on the possible influencing factors of SOC,such as temperature,discharge rate,internal resistance.After comprehensive consideration and based on the realistic states,three parameters,voltage,current and temperature,are selected as data parameters.The LSTM network is used for modeling and forecasting,analyzing the influencing factors of network parameters,designing the network structure and processing data.Combining with the previous experimental analysis of SOC influencing factors,in consideration of the actual working conditions,the charging and discharging data of different working conditions and different temperatures are used for training and testing.The results show that the LSTM network is suitable for SOC estimation tasks and has a good prediction effect.By analyzing the data and the limitations of the pure LSTM network,a CNN-LSTM network combined with attention mechanism is proposed.The time series relationship of battery data and the spatial characteristics relationship between dimensions can be extracted through the CNN-LSTM network.The attention mechanism to distribute more weight to key time steps and the interference of secondary information is reduced to achieve a better fitting effect.The results indicated the prediction results of the network under different temperatures and working conditions are better than the pure LSTM network,and the robustness of the results can be maintained in different periods of use of the battery and the initial state of unknown power.In addition,the improvement effect of different data volume on the network is explored,which proved that in the case of more data volume,the prediction performance of the network is greatly improved.Compared with other methods on the same data set,the proposed network has higher estimation accuracy at different temperatures and better performance at high temperatures.In summary,the network proposed in this paper fully mines the characteristics of data,effectively improves the prediction accuracy,and has good prediction performance at different temperatures.And the network is less affected by the initial state and has good stability and robustness.When the trained model is used for prediction,the prediction time of each data point is less than0.5ms,so it is suitable for on-board real-time estimation.