State of Energy Estimation of Electric Vehicle Based on GRU-RNN

Abstract

Based on charging platform’s historical data and real-time charging data, this paper put forward a new kind of power battery state of energy(SOE) estimate method which uses RNN model with Gated Recurrent Unit (GRU-RNN). The innovations of this paper are as follows: 1) through the analysis of SOC-SOE relationship and off-line calculation of a large number of charging packets at present, this paper makes a breakthrough in proposing that charging should be stopped before SOC reaches 95% (SOE reaches 100%) in order to reduce battery loss; 2) The method of using RNN to map the battery pack feature input directly to the output can break through the limitation of a single battery, and avoid complex electrochemical modeling; 3) By comparing different normalization and standardization methods, it is concluded that the Z-score method can decrease the operation complexity and increase the accuracy by 0.729%. 4) The improved Adam adaptive optimizer is used to minimize the loss function of GRU-RNN to rise the estimation accuracy and efficiency.