Energy management system optimization based on an LSTM deep learning model using vehicle speed prediction

Abstract

The energy management of a Hybrid Electric Vehicle (HEV) is a global optimization problem, and its optimal solution inevitably entails knowing the entire mission profile. The exploitation of Vehicle-to-Everything (V2X) connectivity can pave the way for reliable short-term vehicle speed predictions. As a result, the capabilities of conventional energy management strategies can be enhanced by integrating the predicted vehicle speed into the powertrain control strategy. Therefore, in this paper, an innovative Adaptation algorithm uses the predicted speed profile for an Equivalent Consumption Minimization Strategy (A-V2X-ECMS). Driving pattern identification is employed to adapt the equivalence factor of the ECMS when a change in the driving patterns occurs, or when the State of Charge (SoC) experiences a high deviation from the target value. A Principal Component Analysis (PCA) was performed on several energetic indices to select the ones that predominate in characterizing the different driving patterns. Long Short-Term Memory (LSTM) deep neural networks were trained to choose the optimal value of the equivalence factor for a specific sequence of data (i.e., speed, acceleration, power, and initial SoC). The potentialities of the innovative A-V2X-ECMS were assessed, through numerical simulation, on a diesel Plug-in Hybrid Electric Vehicle (PHEV) available on the European market. A virtual test rig of the investigated vehicle was built in the GT-SUITE software environment and validated against a wide database of experimental data. The simulations proved that the proposed approach achieves results much closer to optimal than the conventional energy management strategies taken as a reference.