結合類神經網絡及模型預測控制之智慧節能行車輔助系統

Abstract

本研究結合類神經網絡及模型預測控制，開發智慧節能行車輔助控制架構，並運用車載資通訊及監督式學習，提升車輛之行駛效能，以達到智慧節能行駛之目的。首先，模型預測控制會根據車輛動態、道路參數與前方交通動態資訊等，調整最佳車速及馬達力矩，使動力系統運作於高效區。駕駛者風格則透過監督式學習進行分類，分類器採用深度神經網絡(Deep Neural Networks, DNN)與循環神經網絡(Recurrent Neural Network, RNN)兩種模型，透過先前駕駛者資訊以及車輛動態資訊，進行駕駛者風格分類，模擬結果證明此兩種方法皆具有90%之準確度。最後，車輛控制單元會根據駕駛者風格、最佳扭矩與駕駛者控制訊號，進行車輛節能之介入控制。此外，透過駕駛者於模擬環境中(Driver-in-the-Loop, DiL)之模擬結果顯示，運用此套智慧節能行車輔助架構，在高速公路情境下，可減少8~10%之行車能耗。

This research combines neural network and model predictive control to develop intelligent eco-driving assistance control architecture, Using telematics and supervised learning improve the driving performance of vehicles in order to achieve the goal of intelligent eco-driving. First, the model predictive control adjusts the optimal speed and motor torque according to the vehicle dynamics, road parameters and traffic dynamics information in front, the powertrain system can then be adjusted to the corresponding high efficiency zone. Individual driving style is then classified by supervised learning algorithm; The classifier uses two models: Deep Neural Networks (DNN) and Recurrent Neural Network (RNN). The classifier collects past driver and vehicle dynamic information for driver style classification, with simulation results demonstrating 90% accuracy for both classifiers on driving style differentiation. Finally, the vehicle control unit performs interventional control based on driver style, optimal torque and driver control signals. In addition, through the simulation results of the driver in the simulated environment (driver in the loop, DiL), the intelligent eco-driving assistance control architecture can reduce the energy consumption by 8-10% in the highway situation.