Models for Predicting RWIS Sensor Misalignments and Their Causes

Abstract

The Minnesota Department of Transportation uses the Road Weather Information System (RWIS) for monitoring the current weather and surface conditions of its highways. The real-time data received from these sensors reduce the need for road patrolling in specific locations by providing information to those responsible for directing winter maintenance operations. Since most road maintenance decisions and weather forecasts are explicitly dependent on the reliability and accuracy of the RWIS sensor data, it is important for one to be able to determine the reliability of the sensor data, that is, to determine whether a sensor is malfunctioning. In a previous project we investigated the use of machine learning techniques to predict sensor malfunctions and thereby improve accuracy in forecasting weather-related conditions. In this project, we used our findings to automate the process of identifying malfunctioning weather sensors in real time. We analyze the weather data reported by various sensors to detect possible anomalies. Our interface system allows users to define decision- making rules based on their real-world experience in identifying malfunctions. Since decision rule parameters set by the user may result in a false indication of a sensor malfunction, the system analyzes all proposed rules based on historical data and recommends optimal rule parameters. If the user follows these automated suggestions, the accuracy of the software to detect a malfunctioning sensor increases significantly. This report provides an overview of the software tool developed to support detection of sensor malfunctions.