Neural Network based Vehicle Speed Prediction for Specific Urban Driving

摘要

Vehicle speed prediction plays an important role in hybrid and electric vehicle energy management. However, vehicle speed prediction is influenced by various factors such as driver's habits, traffic conditions, weather conditions and route geographical features. Therefore, it's difficult to accurately predict speed profile for a specific route. In this paper, two prediction models based on neural networks are established from two perspectives. One is Time-Based Neural Networks Model (TBNNM) and the other is Distance-Based Neural Networks Model (DBNNM). Both models use the same vehicle speed profile for an individual driver repeated drive cycle. The historical speed and road geographic feature data about a specific urban driving have been recorded. Two models have different inputs, in the TBNNM, road information is not premeditated. Both model outputs are next five steps speed. Through the test data validation, both models can predict vehicle speed, but as the prediction length increases, the RMSE increases.