Prediction Of Lithium Battery's Remaining Charging Time Based On IndyLSTM

The remaining charging time is an important lithium battery's parameter of the State of Charge(SOC),it reflects the relationship between the state of charge of the lithium battery and the time.Accurate prediction of the remaining charging time can effectively find and avoid unsafe behaviors,such as overcharging of lithium batteries,and provide guarantee for the safety and stability of lithium batteries.In this paper proposes a method for predicting the remaining time of lithium battery charging based on an independent long-short memory cycle neural network(IndyLSTM),designs and implements a system for predicting the remaining time of lithium battery charging.The main research contents are as follows:First,the overall scheme of the lithium battery charging remaining time prediction system is proposed,the data collection terminal and system software are designed and implemented.The data collection terminal mainly uses the signal sampling unit to collect the battery voltage,battery current,battery temperature and charging voltage generated during the lithium battery charging process,and performs data interaction with the IoT cloud platform through the NB-IoT communication module.The system software mainly includes data terminal collection software and application software.The data collection terminal software is responsible for uniformly scheduling each hardware unit to ensure that the system can work stably and orderly;the application software interacts with the Internet of Things cloud platform through HTTPS protocol and RESTful interface.Visualize the acquired data and put it into the model trained by TensorFlow to get the predicted value of the remaining charging time.Secondly,in order to meet the accuracy and stability requirements of the prediction of lithium battery's remaining time,and in-depth study of the recurrent neural network was established,and the prediction model of the remaining time of the lithium battery charge based on IndyLSTM was established,and the hyperparameters in the IndyLSTM pred iction model were set was studied,the influence of different hidden layer layers,node number,batch size and optimization algorithms on the prediction model was compared and analyzed,also compared with the prediction effect of conventional long-short-term memory network(LSTM)and support vector regression(SVR)By comparison and analysis,the experimental results show that the rms error of the IndyLSTM model compared to the LSTM and SVR model test results is respectively reduced by 40.803%and 46.345%,and the average absolute error percentage is respectively reduced by 7.633 and 5.670.Finally,an experimental platform was set up according to the design scheme,and the signal sampling unit,the IoT cloud platform,and the remaining charge time prediction module tests were completed.The test results prove the feasibility of the design scheme and the effectiveness of the prediction method.The research results of this paper can provide theoretical reference and design guidance for the development and design of the prediction of the remaining time of lithium battery charging and similar projects,and have certain promotion value.