Power management optimisation for hybrid electric systems using reinforcement learning and adaptive dynamic programming
| Autor: | Andrew R. Mills, Ibrahim Sanusi, Tony J. Dodd, George C. Konstantopoulos |
|---|---|
| Rok vydání: | 2019 |
| Předmět: |
Scheme (programming language)
Power management Artificial neural network Computer science 020209 energy Control engineering 02 engineering and technology Power (physics) Dynamic programming Nonlinear system Hybrid system 0202 electrical engineering electronic engineering information engineering Reinforcement learning 020201 artificial intelligence & image processing computer computer.programming_language |
| Zdroj: | ACC |
| DOI: | 10.23919/acc.2019.8814407 |
| Popis: | This paper presents an online learning scheme based on reinforcement learning and adaptive dynamic programming for the power management of hybrid electric systems. Current methods for power management are conservative and unable to fully account for variations in the system due to changes in the health and operational conditions. These conservative schemes result in less efficient use of available power sources, increasing the overall system costs and heightening the risk of failure due to the variations. The proposed scheme is able to compensate for modelling uncertainties and the gradual system variations by adapting its performance function using the observed system measurements as reinforcement signals. The reinforcement signals are nonlinear and consequently neural networks are employed in the implementation of the scheme. Simulation results for the power management of an autonomous hybrid system show improved system performance using the proposed scheme as compared with a conventional offline dynamic programming approach. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|