Study On Real-Time Estimation And Prediction Of Road Grade Using Long Short-Term Memory

Accurate road grade information provides key information for the control of fuel cell vehicles,which affects the acceleration performance,economical performance,safety and intelligence of vehicles.In recent years,various vehicle driving data can be obtained in real time through the vehicle CAN bus,which provides the necessary hardware foundation for road grade perception.However,in actual driving process,there are always uncertain noise and various driving conditions,which make it difficult to calculate the accurate road grade through these driving data.Considering the uncertain noise in real driving process,the complex model in traditional methods and the characteristic that road grade changes with time during car driving,a real-time road slope estimation and prediction method based on long short-term memory network is proposed in this study.It can estimate the instant value and predict the future changes of road grade at the same time through driving data collected from vehicle CAN bus.The main contents are as follows.First,aiming at the real time and accuracy of the road grade perception method,driving dynamic model of the fuel cell vehicle is analyzed.According to the established functional relationship between road grade and vehicle driving data,five parameters are selected as the network input features,including accelerator pedal opening,brake pedal opening,vehicle speed,vehicle acceleration and vehicle DC bus power.Then the relationship between them is analyzed,which shows that the five parameters are indeed helpful for the estimation and prediction of road grade during car driving.Second,considering the influence of uncertain noise during car driving,experiments are conducted.To clearly evaluate the performance of this method when road grade continuously changes,roads with changing grade are selected in the experiments.In addition,the results show that this method can work well when car breaks,which makes up for the shortcomings of existing method and expands the application scope of the road grade perception method.Third,to verify the adaptability of this method,two different routes are selected in experiments.The driving data of first circle on first route are used to train the network,while both the driving data of second circle on first route and the driving data on second road are used to test the network.The results show that the algorithm has good performance on different roads.In summary,this study presents a real-time road slope estimation and prediction method based on long short-term memory network through driving data collected from vehicle CAN bus.It can be applied in real driving environments including braking conditions.The experiments prove the effectiveness and adaptability of this method,which can provide a reference for vehicle control.