Adaptive Dimming of Highway Lights Using Recurrent Neural Networks

Highway operators are in a constant search of techniques and methodologies that can reduce their energy footprint. In this respect, the installation of dimmable light-emitting diode lights on the open road section of highways appears to be a promising solution, due to the reduced energy consumption (compared to high pressure sodium lamps) and the ability to adjust their brightness at various levels, based on the road’s traffic load. However, setting the desired level of light intensity cannot be performed instantly, due to safety and contractual reasons that a highway operator must follow. For this reason, an adaptive and intelligent system is proposed in this work, that models traffic load and is able to predict its future trend, based on current load and light intensity measurements. In this way, the unnecessary use of the lighting equipment is avoided, as brightness is dropped to a minimum level when traffic load is predicted to be low. The proposed model is based on recurrent neural networks and more specifically on long short-term memory cells that are able to model complex dependencies in data with temporal correlations, like traffic load measurements. The overall approach is evaluated on relevant data provided by Olympia Odos S.A. that operates the Elefsina-Korinthos-Patra highway in Greece, with promising results.