Real Time Speed Trend Analysis and Hours of Service Forecasting Using LSTM Network

Learning to predict waiting time of a fleet truck is a hard problem, specially for long haul journeys. In the logistic industry, unexpected delay is a big concern since it directly correlates to detention cost. Although it is important to drill down to root causes of delay, it is more desired to predict delays, even before they have occurred. In this paper we propose a new field of study called vehicle speed trend analysis and hours of service forecasting using LSTM network. Over the course of journey of a vehicle, location updates are received in real time. The sequence of location updates are then converted to speed. Using historic data for specific route and a sliding window method over sequence of speed, we train layers LSTM network. In real time tracking, speed sequence for the first 20% of the journey is considered as input to the trained LSTM network. We show that the network is able to forecasts speed trend for the rest of the journey. Finally, we also show that for fleet trucks travelling long distance, the network is able to learn trend in hours of service with mean absolute error less than an hour.