Study On Power System Of Parallel Hybrid Electric Bus Based On Neural Network

The powertrain matching technology is one core technology of the Hybrid Electric Vehicles, which are the most promising green cars currently, and it is an important aspect of the automobile enterprise’s abilities of independent development. In this paper, a hybrid power system based on neural network is developed for a traditional bus with an internal combustion engine. The main works are as follows:1. Typical Driving Cycle ConstructionAfter analyzing the real-time acquired driving cycle data of the city bus in Jinan city, based on the two-stage clustering analysis algorithm and the Principal Component Analysis theory, a typical driving cycle was constructed. And another driving cycle was constructed based on the traditional single clustering analysis algorithm within the same data. Then the two typical driving cycles were compared with each other via methods of the scatter distribution, the modeling and simulation in turn. The comparison result showed that the typical driving cycle applying the two-stage clustering method was more close to the actual road condition. Based on the two-stage clustering theory, a software has been built on the Visual Studio platform to construct the driving cycle for the Hybrid Commercial Vehicles. With the massive measured data processed and analyzed, the typical driving cycle for Jinan city buses has been built to provide a research basis of the hybrid powertrain matching and the control strategy.2. Hybrid Powertrain MatchingBased on the ADVISOR software platform, a structure model of the traditional bus with an internal combustion engine was built and also was verified accurate with the experiment results. After studying the requirement of the powertrain in dynamic property and on the emission aspect, the bus was transformed into a uniaxial parallel hybrid electric bus with a new engine and a set of electric motor and batteries. The simulation result comparison between the traditional bus and the hybrid bus, which loaded the Parallel Electric Assist Control Strategy, showed that the hybrid bus improved the ability for cycle tracking and the whole vehicle performance. And the hybrid bus model provided a reliable basis of the follow-up study on control strategy.3. Study of Control Strategy Based on Neural NetworkWith the hybrid bus model built above, the Logic Threshold control strategy which was optimized by the Multi-Level Parametric Sweeps (MLPS) algorithm was simulated off-line. And the sample data were used to train the General Radial Basis Function (GRBF) neural network. Then the hybrid bus real-time control strategy based on GRBF neural network was established and also was redeveloped on ADVISOR software platform. The simulation result comparison between GRBF and MLPS Logic Threshold control strategy in several different driving cycles showed that the GRBF control strategy performed better for the hybrid bus. The GRBF control strategy not only decreased the fuel consumption and reduced the exhaust emission, but also controlled the batteries SOC in good condition and improved the whole vehicle efficiency.