Driving Cycle Predictive Model With Self-learning Strategy For Bus Fuel Consumption Optimization

The driving cycle has an important impact on the vehicle fuel consumption.The existing control strategy,off-line calibrated based on standard driving cycles,usually leads to higher fuel consumptions in the real-world operation.In order to optimize the fuel consumption of city buses,a predictive model for bus route driving cycle is proposed in this study,based on geographical features and self-learning optimization algorithm.Firstly,based on an in-house developed on-board data collection system,the driving data was systematically collected,for six buses in two lines for over one year.Critical information,such as the road types,seasons,driving periods,driver skills,weather,amongst others,are included in the collect data.A driving cycle similarity model was proposed based on the product of velocity and acceleration rank correlation,in order to evaluate effectiveness of a predicted driving cycle.Based on the correlation analysis between the driving cycle similarity and the fuel consumption,it is confirmed that the proposed driving cycle similarity index can effectively reflect the impact of drive cycle on the fuel consumptions.Secondly,the influence of the station density,redlight density,curve density,average slope,road types and other geographical features on driving cycle was systematically investigated.Additionally the inter-station driving characteristic model was established by the multiple linear regression equation,in order to predict interstation average speed,inter-station maximum speed,stop times per kilometers and driving aggressiveness.The velocity kinematic sequences were classified by the driving cycle characteristics,and the speed acceleration Markov transition matrixes were statistically analyzed under different types.The driving cycle predictive model was then established,based on the Markov line iterative algorithm,the inter-station driving cycle characteristic model and probability transition matrix.The proposed driving cycle predictive model was first identified with the data from line 503,and then used to predict the typical driving cycle of line 516.The results show that the driving cycle similarity is 78.69% between the predicted and the actual driving cycle.Furthermore,in order to improve the predictive accuracy of the driving cycle predictive model,the self-learning strategy is proposed based on online operation data,The regression equation coefficients of the inter-station driving cycle characteristics model are periodically updated by least square method,and probability transfer matrix of weights are dynamically modified by the forgetting factor.The results show that the predictive accuracy of driving cycle predictive model can be increased from 85.6% to 78.69% by the self-learning of 180 trips,which was established by 1400 trips previously.Finally,to address the energy programming problem of plug-in hybrid bus lines,a global program model based on equivalent factor was established based on the predictive driving cycle.The self-learning energy programming strategy based on driving cycle self-learning is proposed.The strategy was verified by 400 laps actual speed curve of liner 516,with simulation platform.The results show that the optimized bus can benefit 1.03% fuel saving and 61.44% power saving,and the saving probability was 63% in the former 300 laps;the optimized bus can benefit 2.25% fuel saving and 54% power saving,and saving probability increases to 73% in the following 100 laps after the activation of self-learning strategies.In summary,The proposed driving cycle predictive model based on geographic features and self-learning strategy can improve the ability and self-tendency of bus fuel saving.