Development of Hybrid real-time capable Vehicle Simulation Platform for Energy Management Strategy Development and Virtual Validation

Abstract

Model-based calibration enables a shift in development tasks from the real world to the virtual world, allowing for increased system robustness while reducing development costs and time. These benefits are especially pronounced in the case of complex powertrain systems such as Hybrid Electric Vehicles (HEV). In (P)-HEV, Energy Management Strategy (EMS) constitutes the core for fuel consumption and emissions reduction of hybrid electric vehicles. This paper presents an assessment of the feasibility of carrying out automated model calibration within a virtual powertrain test environment for EMS validation. It highlights the accuracy of the engine simulation model under steady-state and transient operating conditions. It also demonstrates the integration of the model with data based pollutant emission models and the after-treatment system. The Internal Combustion Engine (ICE) is modeled by means of the so called Mean Value Engine Model approach (MVEM) with limited amount of input signals. Electrical components are modeled following a classical map-based approach. The novelty of this methodology resides in the pollutant emissions models. Engine out emissions (CO2, CO, HC and NO) are modeled through convolutional neural networks (CNN) giving extremely accurate results, both in instantaneous and cumulative prediction. This allows to calibrate a Three-Way Catalyst model (TWC) enabling the characterization of a complete powertrain system model with fuel consumption, emissions (CO2, CO, HC and NOx) and the electrification part. In this framework EMS can be developed for minimizing not only fuel consumption but also pollutant emissions. As final step, advanced EMS based on reinforcement learning is proposed. The methodology is developed in a co-simulation framework between MATLAB-Simulink and AMESIM. The resulting model runs between 2-3 times faster than real time in an off-the-shelf laptop. This enables the methodology for developing models suitable for HIL (hardware-in-the-loop) and SIL (software-in-the-loop) applications. The final error in predicted pollutant emissions in the studied cycles remains below 2.5% in the case of CO2 emissions, 3.5% in the case of NOx emissions, and below 8.5% when speaking about CO and HC emissions.