Research On Fuel Consumption Prediction Model And Speed Optimization Method Considering Jerk

The rapid increase of car ownership has facilitated people’s daily work and life,and also caused huge energy pressure.Accurate prediction model of automobile fuel consumption and speed optimization control aiming at energy saving are the important basis for realizing Eco-driving.The current studies do not consider the Jerk,which is the first derivative of acceleration and can better characterize driving behavior,resulting in shortcomings such as weak correlation between fuel consumption model and driving behavior,and low prediction accuracy of the model.On the basis of research at home and abroad,this paper puts forward a prediction model of automobile fuel consumption and a speed optimization method,which integrates vehicle speed,acceleration and Jerk.In this paper,the vehicle driving behavior is selected as the main research object,and the following in-depth studies are performed:the classification of driving behavior,the qualitative and quantitative analysis of the influence of driving behavior on fuel consumption,the fuel consumption prediction model considering driving behavior,and the speed optimization method aiming at energy saving,achieving the following research results and technological innovations:(1)Driving behaviors are divided into 11 categories based on the classification method of driving behaviors considering speed,acceleration and Jerk.On this basis,the influence of each driving behavior on fuel consumption is analyzed by statistical analysis and cluster analysis.This can lay the foundation for establishing the relationship between fuel consumption and driving behavior,and the precise fuel consumption prediction model.(2)A prediction model of vehicle fuel consumption is proposed,which integrates vehicle speed,acceleration and Jerk in non-cruise state.The model is based on physics drive and data drive,and the scatter mapping between vehicle instantaneous fuel consumption and the product of vehicle speed and vehicle acceleration is established.On this basis,using Jerk as classifier and quadratic polynomial as fitting function,the quantitative mapping between vehicle instantaneous fuel consumption and the product of vehicle speed and vehicle acceleration is established.Finally,based on nearly 35000 data sets covering urban roads,expressways and rural roads with a round-trip distance of 160km,the model was calibrated and verified.The experimental results show that:the prediction performance indexes of Mean Absolute Percentage Error(MAPE),Root Mean Square Error(RMSE),Correlation Coefficient(R)and Determinant Coefficient(R~2)of this model are significantly better than those of the classic Vehicle Specific Power(VSP)model and Virginia Tech Microscopic(VT-Micro)model,in which MAPE index is 39.5%and 26.1%lower than those of VSP model and VT-Micro model,and RMSE index is 42.6%and 72.3%lower than those of VSP model and VT-Micro model,and R index is 2.5%and 2.4%higher than those of VSP model and VT-Micro model,respectively.R~2 is close to 95.8%,which is increased by 5.0%and 9.7%,respectively compared with VSP model and VT-Micro model.(3)A general prediction model of vehicle fuel consumption is proposed,which integrates vehicle speed,acceleration and Jerk.Based on the idea of surface fitting,two models of the relationship between instantaneous fuel consumption and speed,acceleration and Jerk are established.On this basis,based on Pearson correlation coefficient method,the correlation coefficient of vehicle fuel consumption with different power products of vehicle speed,acceleration and Jerk is quantitatively calculated,and a general vehicle fuel consumption prediction model is established.Classical VSP model and VT-Micro model are the special forms of this model.Finally,based on nearly 35000 data sets covering urban roads,expressways and rural roads with a round-trip distance of 160km,the model was calibrated and verified.The experimental results show that:the prediction performance indexes of MAPE,RMSE,R and R~2 of this model are significantly better than those of the classic VSP model and VT-Micro model,in which the MAPE index is 48.9%and 9.8%lower than those of VSP model and VT-Micro model,and RMSE index is 69.7%and 18.1%lower than those of VSP model and VT-Micro model,and R index is 4.2%and 1.6%higher than those of VSP model and VT-Micro model,respectively,R~2is close to 91%,which is increased by 23.5%and 4.0%compared with VSP model and VT-Micro model.(4)A deep learning fuel consumption prediction model integrating vehicle speed,acceleration and Jerk is proposed.In this model,4 neural networks,which are Long Short-term Memory(LSTM),Recurrent Neural Network(RNN),Nonlinear Auto-regressive Model with Exogenous Inputs(NARX)and Generalized Regression Neural Network(GRNN),are selected respectively,and 4 input parameter combinations,which are(speed),(speed,acceleration),(speed,acceleration and Jerk)and(rotation speed),are selected respectively,and 3 working scenarios,which are the low-speed(0-40km/h)campus scenario,the medium-low-speed(0-70km/h)urban scenario,and the high-medium-low-speed(0-120km/h)expressway scenario,are selected respectively,with a total of 4×4×3=48 kinds of experiments were carried out accumulatively.The experimental results show that:under any of the above input parameters and any speed condition,the prediction performance of LSTM neural network obviously outperforms other neural network prediction models.After introducing Jerk,compared with the fuel consumption prediction algorithm without considering Jerk,LSTM,RNN,NARX and GRNN neural networks have significantly improved their fuel consumption predictionperformance under three different speed conditions.In high-speed expressway scenario,LSTM algorithm shows the greatest improvement,in which RMSE index decreased by 14.3%,RE index decreased by 28.3%,and R~2 index increased by 9.7%.In low-speed campus scenario,NARX algorithm shows the greatest improvement,in which RMSE index decreased by 34.3%,RE index decreased by 43.0%,and R~2 index increased by 22.9%.(5)Selecting the speed variance to represent the driving fluctuation,it is theoretically proved that when the average speed is constant,the fuel consumption is proportional to the speed fluctuation.The average filter is used to simulate different fluctuating speed trajectories,and the classic VSP fuel consumption model is used to calculate the fuel consumption of each group of speed trajectories.The correctness of theoretical derivation is verified from the perspective of simulation.On this basis,a trajectory generation and optimization algorithm combining speed,acceleration,and Jerk is proposed.The algorithm is based on the optimal control theory,and uses the Hamilton function method to solve the functional minimum,which can reduce the vehicular speed fluctuation,and then achieve the optimal energy consumption.After the algorithm has been transplanted to intelligent vehicles,a comparative experiment on the fuel consumption of the same vehicle in unmanned driving mode and manual driving mode has been carried out in a closed field test base of autonomous driving.The experimental data show that the vehicle fuel consumption increases linearly with the increase of speed fluctuation.Compared with unmanned driving mode,the fuel consumption of manual driving mode increases by 5.6%and 14.7%at the average speeds of 20km/h and 40km/h,when the manual driving behavior is more reckless,the fuel consumption even increases to 60%.