[System Design and Implementation] Flink-based distracted driving prediction and data analysis system

基于flinkForecast of distracted driving and data analysis system(GitHubWarehouse click here)

项目简介

通过flinkReal-time identification and random forest model driver whether the user is in a state of alert,And further data statistics and analysis,To save the results of the analysis inhbase,并通过WebsocketTo provide customers with real-time analysis results. (The system try to give a few years ago,Model the truth quite pull)

项目技术栈

后端技术栈：Flink,Flume,kafka,HBase,Mysql,Springboot

前后端通信协议：websocket（stomp）,http（前端axios）

前端技术栈：Vue,Echart,ElementUI

机器学习：sklearn,pmml

项目组成

The project has created three module：

• StayAlertWeb模块：系统的Web客户端,Used to display real-time statistics and query users of real-time statistics.

• StayAlertAPI模块：该模块是一个Springboot程序,Used to provide for persistent data and real-time statisticsapi接口.Simulation characteristic data is through the moduleapi转发到flume和kafka中间件,And then to predict analysis program for processing.

• StayAlertFlink模块：This module is responsible for according to the characteristics of simulation data to predict the user's alert,Then combined with the user's personal information to users' personal data analysis and data analysis of global.该模块 Send the results to the message middleware and persisting data tohbase.StayAlertAPI模块可以从KafkaThe message queue to obtain real-time data,Also can be directlyhbaseObtain the data persistence.

• 数据库：本次项目使用MysqlKeep the user's personal information,并使用HBaseThe statistical results of data persistence.

业务流程：

1. Simulation program toSpringboot编写的apiCharacteristics of user application sends data.
2. springbootCharacteristics of the data byFlume和KafkaForward the message middleware toFlink预测分析程序,Formation of the flow.
3. 通过connectOperation incorporating prediction results and user information data.
   1. Flow load training in goodPMMLRandom forests model to predict.
   2. 程序从MysqlRegularly read user information data in,Through radio sends data to the trunk calculation flow in the form of
4. Program will be persisted to the statistical resultsHbase中,然后发送到Flume和KafkaMessage middleware in waiting forSpringbootProgram gets pushed to users and.
   1. 根据用户id分组,Users' personal data
   2. Using the window operation and data aggregation function global statistics
5. SpringbootTo the user the results
   1. From the message middleware to obtain real-time data
   2. 从Hbase和MysqlThe database access to persistent data
6. Web客户端先向SpringbootTo the initial page request a persistent data show,然后创建Websocket连接,To obtain the statistical results of real time

界面

相关下载

原始数据集

数据集来源

Distractions classification model

Model after downloading on/StayAlertSystem/StayAlertFlink/src/main/resources目录下.

名称	说明	准确率	模型大小
StayAlertRFC.pmml	使用88An estimator	87.56%	200MB
LightStayAlertRFC.pmml	使用20An estimator	84.46%	45MB

百度云 提取码：qz4x

操作文档

项目报告

数据集说明

本数据集来自KaggleA competition,Data set provided by the ford,The data shows that many groups“试验”的结果,Each representative experiment lasted about2Minutes of driving data,The data on the road or in a driving simulator per100Ms record at a time,The characteristics of the data fields including the driver's physiological indexes in the process of the driving,Some environmental indicators and vehicle,May for some reasons,These indicators of the actual name and unit of measure in this task does not disclose.

数据集来源,同上

字段名称	字段解释
TrialID	大约2Minutes of continuous data of each cycle has a unique testID.例如,前1210Observation on behalf of every100msThe continuous observation,Therefore has the same testID
ObsNum	This is a testIDIn increasing Numbers in sequence
IsAlert	If the driver is alarm for1,否则为0
P1-P8	Physiological data
E1-E11	On behalf of the environmental data
V1-V11	According to vehicle data