Driving Cycle Prediction For Energy Optimization Control Of New Energy City Buses

Driving cycle forecast can help optimize the control and the use of vehicles energy.Based on the characteristics of urban buses,this paper studied the routes of buses,spacial distribution of velocity and prediction methods of driving cycle,and then obtained the off-line forecasting method of the whole trip and on-line forecasting methods of the short range considering both prediction accuracy and computing time.The prediction results can be used for whole-trip energy distribution and real-time energy management of new energy buses,and also were important references for the remaining mileage estimation.Firstly,based on Matlab platform,a driving cycle analysis software with functions of data preprocessing and characteristic parameters calculating was developed.The preprocessing module completed the function of dealing with data drifting and losing problems caused by signal loss and interference.Speed,latitude and longitude were included in the data.The calculating module calculated 18 characteristic values in total,including 13 that characterized the velocity and its distribution and five characterized the acceleration.Secondly,using the mass driving data,the whole-trip prediction method of fixed route buses was constructed,and the accuracy of it was verified.The density-based clustering algorithm(DBSCAN)and distance-based clustering algorithm(K-means)were used to automatically identify parking sites and cluster the velocity fragments based on airspace.According to the method of Markov and the results of clustering analysis,the transfer probability matrixes of the fragment categories was constructed and the off-line prediction of the whole trip was realized.The results showed that the root mean square error of the constructed model was 3.6km/h.Then,considering data quantity requirement,prediction accuracy and calculation time,the on-line prediction methods of short-range driving cycles were studied.The comparison of different methods was completed.The results showed that data demand was different for prediction models.In the prediction interval of 5-100 meters,the RBF neural network had the highest prediction accuracy,followed by the BP neural network and the Markov stochastic process was the worst.In the same prediction interval,the Markov method was the fastest method,ARIMA and BP neural network were faster,and the RBF neural network method was the slowest.Finally,the software and the prediction methods developed in this paper were validated.By using the software,the characteristic parameters of hybrid buses’ and fuel cell buses’ operating data were calculated and compared with the traditional buses.The applicability of the off-line whole-trip and the on-line short-range prediction methods for the buses were verified.