Research On Estimation Model Of Road Friction Coefficient For Passenger Cars Based On Hybrid Neural Network

With the gradual development and improvement of auto-driving technology,Obtaining vehicle motion state parameters and environmental data has become indispensable information for safe driving.Among the commonly used road friction coefficient acquisition technologies,the sensor-based direct technology has limited use conditions and high maintenance costs.And the indirect technology based on the original car signal is poor in accuracy and real-time application,and it is difficult to form a good estimation effect.Since there is no direct calculation model for road friction coefficient.In order to realize the estimation of road friction coefficient.This article proposed an estimation method based on the neural network model,established a convolutional neural network characteristics of passenger car original data sets processing and a long-short term memory network to estimate the hybrid network of time series data.This article used the neural network model as the road friction coefficient estimation model to estimate the road friction coefficient.At the same time,this article analyzed tow error compensations of the network model to optimize the model.The main work is as follows:(1)Research on correction model of automobile angular velocity based on graph optimization.In order to solve the problem that the traditional adaptive filtering method has a large error in calculating the wheel angular velocity.This article proposed a graphbased optimization algotirhm for real-time noise reduction and optimization of the diagonal velocity.In the algorithm,the wheel angular velocity is calculated by the gear ring,and the model is optimized by combining the measurement data of the inertial unit.This article set the constraints of vertices and edges to correct the wheel angular velocity.The signal after graph optimization can eliminate the peak inherent noise,reduce the data fluctuation,and provide reliable data for the road friction coefficient estimation model.(2)Research on construction and evaluation of hybird neural network model.Since there is no analytic model directly related to the estimation of the road friction coefficient,the road friction coefficient can not be directly calculated from the existing vehicle data.This article established a hybrid network of a convolution neural network model characterized by space and a long short-term memory network model characterized by time.By adjusting the weights of the network and the network structure,the hybrid neural network is used as the model for estimating the road friction coefficient.This article collected the sample set in real time and trained the neural network offline,and finally calculated the road friction coefficient in real time with the trained network model.This article verified the effectiveness of the hybrid network by comparing the results with the current mainstream model framework for road friction coefficient estimation.(3)Research on the model optimization based on batch processing and distanceaverage selection.Aiming at the problem of uneven distribution of the original data set of passenger cars when the moving state of the passenger car changes and the roughness of the road surface itself,the result of the neural network model’s estimation of the road friction coefficient has a large fluctuation problem.This article processed and designed a passenger car data processing method based on batch processing and distance average selection to optimize the hybrid neural network model.The effectiveness of this method is verified by comparing the optimization effects of the hybrid network model before and after data enhancement.