A Power Prediction Model using Bi LSTM network for a Hybrid Electrical Vehicle

The role of power prediction within the Energy Management System (EMS) of Extended Range Electric Vehicles (EREVs) is of paramount significance. The study on power prediction modeling encompasses four distinct aspects: dependence on Intelligent Transportation System (ITS) infrastructure and high-precision mapping, leveraging location and speed information exchange among networked vehicles, adoption of statistical methods based on Markov chains, and the application of deep learning techniques. Deep learning models, renowned for establishing nonlinear relationships between control parameters and vehicle power, exhibit varying degrees of prediction accuracy and computational efficiency. To enhance predictive model performance, this study introduces a power prediction model grounded in a Bidirectional Long Short-Term Memory (Bi-LSTM) network. The methodology involves formulating a detailed vehicle model with a comprehensive operation mode description, developing a deep learning model tailored for power prediction, and conducting experiments to compare the results with two alternative models. The outcomes manifest across three key dimensions. Firstly, the proposed power prediction model yields a substantial reduction of 68.1% and 71.7% in the root mean square error (RMSE) when contrasted with a Recurrent Neural Network (RNN) and a conventional model, respectively. Secondly, the accuracy of system power prediction experiences a notable enhancement of $\mathbf{1. 5 5 \%}$, accompanied by a more concentrated clustering of hybrid vehicle operation points within the superior performance region. Thirdly, an improvement in overall system efficiency is observed, reflected in a fuel consumption reduction of 8.80% and 9.42%.