As new energy technologies mature and become more widely available, hybrid vehicles are increasingly being adopted by consumers. However, the existence of at least two power modes and energy management methods has had a significant impact on their related operating costs. This paper proposes an energy management algorithm for this type of vehicles based on an improved deep Q-Learning neural network. This way the energy consumption for hybrid vehicles could be reduced, the efficiency of their energy utilisation could be improved, and their related cost advantages could be enhanced in comparison with traditional fuel-powered vehicles. By using a mixture of multiple representative operating conditions for testing purposes, the conducted experiment confirmed that the fuel consumption per 100 kilometers for the algorithm based on double deep Q-networks under random operating conditions 1 and 2 was 4.05L/100km and 3.64L/100km, respectively. Moreover, the average fuel consumption per 100 seconds for this algorithm was 43ml/100s, which was significantly lower than that of the other three employed automotive powertrain energy management algorithms. The obtained experimental results proved that the energy management algorithm for the hybrid electric vehicle powertrain presented in this paper featured excellent energy control and management capabilities. To sum up, this study could have certain reference significance for improving the cost-effectiveness of hybrid electric vehicles in China.
Neural network, Reinforcement learning, Hybrid power, Energy management, Fuel consumption.
Lan BAN, "Optimization of Energy Management Algorithm for Hybrid Power Systems Based on Deep Reinforcement Learning", Studies in Informatics and Control, ISSN 1220-1766, vol. 33(2), pp. 15-25, 2024. https://doi.org/10.24846/v33i2y202402
All articles published by Studies in Informatics and Control (SIC) are published in accordance with the terms and conditions of the Creative Commons Attribution License (CC BY-NC-SA 4.0).