Plug-in hybrid electric vehicle energy management method based on deep reinforcement learning

Abstract

The invention puts forward a plug-in hybrid electric vehicle energy management method based on deep reinforcement learning. The method comprises the following steps that: carrying out representation extraction on multisource high-dimension driving working condition information, and processing the high-dimension driving working condition information into a low-dimension representation vector; utilizing a redundant information rejection algorithm to carry out dimensionality reduction and fusion processing on obtained working condition state representation, including a low-dimensional representation vector, the own state information of a vehicle, a gradient and the like, to obtain low-dimensional continuous working condition information; constructing a plug-in hybrid electric vehicle energy management framework based on the deep reinforcement learning, and inputting the low-dimensional continuous working condition information to finish offline training; and utilizing a trained strategy tocontrol energy distribution, providing an approach for comprehensively considering the influence of multisource high-dimensional driving working condition information on a plug-in hybrid electric vehicle energy management effect, and utilizing a reinforcement learning autonomic learning optimal energy distribution scheme to mine the energy saving potential.

Description

technical field [0001] The invention relates to an energy management method of a plug-in hybrid electric vehicle, in particular to an energy management method of a plug-in hybrid electric vehicle based on deep reinforcement learning. Background technique [0002] At present, optimization-based plug-in hybrid vehicle energy management methods mainly include global optimization methods represented by dynamic programming, and real-time optimization methods represented by equivalent fuel consumption minimum strategy and model predictive control. When using the method, most of the state parameters of different working conditions need to be discretized by grid division, and then used as the state input of the control system, and the optimal performance is often found by increasing the grid density or the number of states, but this will make The calculation amount of the optimization algorithm increases exponentially, causing a catastrophe of dimensionality and making it impossible...

AI Technical Summary

Problems solved by technology

[0002] At present, optimization-based plug-in hybrid vehicle energy management methods mainly include global optimization methods represented by dynamic programming, and real-time optimization methods represented by equivalent fuel consumption minimum strategy and model predictive control. When using the method, most of the state parameters of different working conditions need to be discretized by grid division, and then used as the state input of the control system, and the optimal performance is often found by increasing the grid density or the number of states, but this will make The calculation amount of the optimization algorithm increases exponentially, causing a catastrophe of dimensionality and making it impossible to solve it. Therefore, it is difficult to comprehensively consider the actual driving environment and traffic conditions (such as vehicles ahead, pedestrians, traffic lights, traffic flow status near the vehicle, etc.) and the state of the vehicle itself. parameters (such as vehicle speed, acceleration, state of charge of the power battery, etc.), and these working condition information will have an important impact on the energy management effect

On the other hand, the information of these driving conditions has different characteristics in terms of continuity, discreteness, and randomness. The information system formed is a multi-source, high-dimensional information system. Potential relationship between energy management effects

Images