Vehicle Driving Cycle Discriminant Based On Multi-Sensor Information Fusion

The energy and pollution problem has been the most important problem since 21 th century as the industrialization and the growth of automobile. How to develop new energy and to use the existing energy effectively is more and more important. HEV(Hybrid Electric Vehicle) is regarded as transition product from gasoline vehicle to the future new energy vehicle and it will play an important role in energy shortage and environmental pollution problem for a long time.Vehicle driving cycle discriminant is a basic technology for HEV development. If a batter vehicle driving cycle discriminant is developed, it is good for developing energy management strategy, good for increasing fuel efficiency, and good for reducing pollutant emission, it can increase vehicle safety and comfort too. In order to develop an more efficient HEV, this paper apply Multi-Sensor Information Fusion method to analyze driving cycle, and divide driving cycle into four categories. then apply BP neural network technology, and SOM self-organizing maps technology to driving cycle discriminant. We can use this method to make vehicle aware of the driving cycle itself be in.The article has analyzed the present situation about driving cycle domestic and abroad.We design the driving cycle test plan in the city and use multi-sensor technology to get data and analysis the data. Base on the driving cycle characteristics, divide driving cycle into four categories:Urban congestion, urban smooth, suburban condition and high speed condition.Then define five parameters(average speed, average positive acceleration, the accelerator pedal opening on average, the average gear, idle time proportion) as the key parameters by the methods of Kruskal-Wallis Single factor variance analysis and correlation analysis.At last, apply BP neural network and SOM self-organizing maps technology to distinguish driving cycle. Firstly, the network model is trained by the all kinds of driving cycles sample data. After that we test the network model by test sample. As a result, the two network models have a good accuracy performance, about 90% accuracy rate.