Loading...

Optimum Power Management Strategy for a Hybrid Vehicle Via Artificial Intelligence

Golmohammadi, Alireza | 2013

844 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: English
  3. Document No: 45702 (58)
  4. University: Sharif University of Technology, International Campus, Kish Island
  5. Department: Science and Engineering
  6. Advisor(s): Salarieh, Hassan; Khayyat, Amir Ali Akbar
  7. Abstract:
  8. In the recent years, increasing concerns about the environmental issues necessitatecientists and industrial people to introduce clean and zero emissions vehicles to the global market. Relying on this acute demand,Hybrid Electric Vehicles (HEVs) are knownas one potential solution to lessen the negative effects of transportation industry on ecological systems.HEVs combine an Internal Combustion Engine (ICE) with an Electric Motor/Generator (EMG) and an Energy Storage System (ESS) to produce a fuel efficient and low or zero pollutant powertrain. Increasing the number of components in the powertrain requires a sophisticated strategy to control all drive-line components suitably. Surely, the best fuel efficiency and emissions reduction of an HEV is realizable only by a competent Power Management Strategy (PMS), which splits the vehicle power demand between different sources of energy (ICE and ESS) to achieve simultaneous objectives such as minimum overall fuel consumption and pollutant emissions, regulation of batteries State Of Charge (SOC), and maximum regeneration of braking energy. Designing a PMS depends on the architecture of an HEV (i.e., parallel, series, parallel-series, or complex), the Degree Of Hybridization (DOH) (i.e., relative driving capacities of the ICE and the ESS), and the available future driving information. The main focus of this thesis is to design a PMS that explores the best fuel saving and SOC regulation capability of a Parallel HEV (PHEV).
    In the present work we have proposed a novel approach using Reinforcement Learning (RL), a powerful technique in the machine learning discipline, which solves the optimal (near-optimal) power split between the ICE and the ESS in a PHEV, taking minimal fuel consumption considerations into account. The power split must perform in a way that in every moment fulfills the demanding power on the final drive by either the internal combustion engine alone, the electricmotor alone or both together, while the batteries SOC maintains in a predefined bound (SOC sustenance ). It is worth to mention that a simple technique for component sizing of the PHEV is applied, and the proposed PMS is implemented on the components sized vehicle.
    In the reinforcement learning technique, the controller called the agent interacts by its environment and achieves a scalar reward signal which represents the desirability of its action (control signal) and state transition. Defining a proper state(s), action(s), and a reward function are three vital tasks of designing a reinforcement learning agent. Considering this very succinct presentation about RL, we designed an intelligent energy management agent based on the RL method (RL-based PMS) (for the first time), which performs suitably. In designing the RL-agent, the most challenging phase was defining anappropriate reward function containing all of our expectations about a proper PMS. To this end,by combining heuristic methods, engineering perception, and trial and errors, we finally designed an efficient and effective reward function.The proposed RL-based PMS outperforms the fuel economy comparing with a wildly used deterministic rule based PMS known as base line strategy, and in some simulations the fuel consumption decreased up to ; besides it is easy and applicable for online implementation, necessarily not reliant on predefined trajectories, adaptive with changing situations and demands, and independent of accurate dynamics of the system
  9. Keywords:
  10. Reinforcement Learning ; Component Sizing ; Hybrid Electric Vehicle (HEV) ; Power Management ; State of Charge ; Reward Function ; Power Split

 Digital Object List

 Bookmark

No TOC