Research On SOC Estimation And Management System Of Lithium Battery Based On Neural Network

In recent years,many countries are vigorously developing various new energy technologies and products that can replace fossil fuels.Lithium battery has become the representative of new energy products and the main power source in many fields due to its superior performance,especially in electric vehicles.Battery pack is the core component of electric vehicle,and battery management system is the direct guarantee for the normal and stable operation of electric vehicle.During the operation of electric vehicles,monitoring various parameters of battery packs and individual batteries and estimating SOC are the basic and core technologies of battery management system.The high-precision monitoring and estimation can greatly improve the service performance and actual life of battery pack,enhancing the safety and stability performance of electric vehicle.In this thesis,the application of traditional simple neural network and deep neural network in SOC estimation of 32650 lithium iron phosphate battery was studied.In addition,the battery management system was designed in detail using 12 series 32650 lithium iron phosphate batteries.The main contents are as follows:(1)The analysis of battery management system and lithium battery characteristics.By analyzing the core technology and current challenge of the battery management system,the main problems existing in our country and the application prospect of neural network in SOC estimation are summarized.Based on the analysis of the structure and performance parameters of lithium battery,the advantages and drawbacks of the widely used lithium battery at present are compared.Based on above analysis,the research object of this thesis is chosen as lithium iron phosphate 32650 battery.Then,based on the experiment of charging and discharging,the dependence of battery capacity and voltage with different ratios of charging and discharging are obtained,which lays a solid foundation for the subsequent research of SOC estimation and the design of management system.(2)The research on SOC estimation model based on traditional neural network.Based on the analysis of the principle and the existing local optimization of BP neural network,the SOC estimation model was initially established.Then,based on the analysis of the particle swarm algorithm optimization process,an adaptive update strategy based on fitness value was proposed to improve it.Finally,the weights and thresholds of the BP neural network are optimized with an improved algorithm,improving the global optimization performance of the BP neural network.(3)The research on SOC estimation model of battery based on LSTM deep neural network.By comparing with traditional simple neural network and deep neural network RNN,the improved model LSTM is analyzed in detail and the SOC estimation model is established.Then,based on the charging and discharging experimental data in this thesis and NASA lithium battery data,the model was evaluated and verified.At the same time,the comparison is made with that based on the traditional neural network estimation model,verifying the estimation performance and applicability of the model.(4)The design of battery management system.Taking 12 lithium iron phosphate batteries as the management object,a battery management system was designed,in which the raspberry pi 3B+ is the main controller and LTC6811-1 is the battery monitor.The LTC6820 chip converts the isoSPI communication interface built into the LTC6811-1 into an SPI interface,which realizes the isolated communication between the LTC6811-1 and the Raspberry Pi 3B+.The hardware circuit of functions such as single cell voltage,battery temperature,battery current,battery balance control,and the software such as control code,battery SOC estimation,visual interface were completed.Experimental results show that the accuracy of SOC estimation model based on deep neural network LSTM is much higher than that of traditional simple neural network and its improved model.The accuracy within 2%is much smaller the required 5%in national standard.The battery management system designed in this thesis is tested,which can basically complete the preset functions.The research results based on this system can provide an important basis for the development of a perfect battery management system.