Research And Implementation Of Commercial Vehicle Fuel Consumption Prediction Method For Remaining Route Based On Driving Style

The rapid development of industry and commerce has made the demand for road transportation higher and higher.Under this trend,the commercial vehicle industry has developed rapidly.While commercial vehicles play an important role in logistics transportation and engineering construction,they also bring about problems of energy consumption and environmental pollution.How to realize energy saving and emission reduction in the commercial vehicle industry has attracted widespread attention.Predicting the fuel consumption of commercial vehicles on road in future can provide decision support for drivers in route planning and energy regulation,and is of great practical significance.With the continuous upgrading of sensors and communication equipment and the rapid development of Internet of Vehicles technology,a large amount of driving data during the driving of commercial vehicles can be collected and studied.These data are collected by different sensors installed on various components of commercial vehicles at a frequency of seconds,and are transmitted back to the Internet of Vehicles big data platform immediately through communication equipment.These data have the characteristics of large data scale,high collection density,fast growth rate,strong time series relationship,and rich types of attributes,which provide us with a good data basis for in-depth analysis of commercial vehicle fuel consumption characteristics and prediction for commercial vehicle fuel consumption.Fuel consumption prediction research based on Internet of Vehicles big data usually uses machine learning and deep learning methods to analyze the factors affecting fuel consumption,and to explore and establish the mapping relationship between various influencing factors and fuel consumption.However,these studies still have the following problems.First,as an important influencing factor,the driving style has a significant impact on the fuel consumption of the vehicle,but the existing research only considers the driver’s driving behavior when extracting the driving style,ignoring the driving environment such as roads and weather.Second,there are few studies on predicting fuel consumption for remaining route,and the existing research ignores the traveledroute from the starting point to the driver’s position,and this part of the route often provides value for improving the accuracy of fuel consumption prediction for remaining route.In order to solve the above problems,based on the cooperation project with a large domestic commercial vehicle manufacturer,this thesis conducts research on the commercial vehicle fuel consumption prediction method for the remaining route based on driving style,aiming to establish an accurate fuel consumption prediction model for remaining route and realize the fuel consumption prediction system based on driving style.The research faces the following challenges.Firstly,how to represent the driving environment features,and how to fuse the environment features and driving features to extract the driving style.Secondly,how to effectively extract and utilize the short-term driving characteristics reflected by the route traveled to improve the accuracy of predicting the commercial vehicle fuel consumption for the remaining route.Thirdly,how to collect and process input data more quickly from massive multi-source data,realize a commercial vehicle fuel consumption prediction system for remaining route based on driving style.In order to solve the above challenges.this thesis proposes a driving style-based commercial vehicle fuel consumption prediction model for remaining route.DS-RFP(Remaining route Fuel consumption Prediction based on Driving Style),and applies this model to the real production environment of the Internet of Vehicles big data platform to build a commercial vehicle fuel consumption prediction system for remaining route based on driving style.The main work and contributions of this thesis are summarized as follows:1.This thesis proposes a commercial vehicle fuel consumption prediction model for remaining route based on driving style,DS-RFP,which realizes accurate prediction of remaining route fuel consumption.The model consists of a driving style representation module and a remaining route fuel consumption prediction module.The driving style representation module proposes to process driving data and environmental data using segmented time frames and trajectory compression technology based on road segment merging,and to extract driving style using multi-head self-attention mechanism,bidirectional LSTM and convolutional network.The fuel consumption prediction module proposes to use the attention mechanism to learn the driving data that may be generated in each section of the remaining route based on the driving environment and driving data of the vehicle on the route it has traveled,and then use the sequence model to fuse the driving environment of the remaining route,historical driving style and vehicle configuration characteristics,to realize the fuel consumption characteristic modeling of the road segment sequence and the accurate prediction of the fuel consumption for remaining route.2.In this thesis,extensive experiments are conducted on two real large-scale historical driving datasets to verify the effectiveness of the proposed DS-RFP model.First,the comparison experiments with the baseline methods verify that the overall performance of the DS-RFP model on the MAE and RMSE evaluation indicators is better than that of the basel ine methods.Secondly,this thesis compares the influence of different parameter settings on the accuracy of fuel consumption prediction results through parameter setting influence experiments,and analyzes and determines the optimal parameter setting.Finally,through a series of ablation experiments,the effectiveness of different parts of the DS-RFP model in enhancing fuel consumption prediction accuracy is verified.3.Based on the Internet of Vehicles big data platform of a large commercial vehicle manufacturer,this thesis uses the proposed DS-RFP model to design and implement a commercial vehicle fuel consumption prediction system for remaining route based on driving style.The system uses MySQL master-slave synchronization technology and big data components such as Kafka and Flink to collect and process the multi-source data required for model prediction in real time.Then the system uses ClickHouse-based distributed storage and query and data warehouse technology to integrate and calculate massive mu lti-source data,and uses TensorFlow,Docker and TensorFlow Serving to realize the offline training and online deployment of the model.Finally,the system uses SpringBoot to visualize the prediction results and realizes effective application of the DS-RFP model in industrial environments.