Detection of traffic anomaly in highways by using recurrent neural network

Using deep learning in all fields has experienced spectacular dynamics in recent years. Among these fields are road traffic in cities and highways. The sensors placed on these routes generate a mine of data for analysis and exploration. In this article, we propose a model based on recurrent neural networks and Multi-Agent Systems (MAS) to find anomalies in vehicle flows on highways compared to normal flows. This model is designed around the events and contexts of cities. By event, we mean periodic events such as cultural events, and political events, etc., and by context, we mean tourist towns, mountain towns, etc. The vehicle flows during city events and contexts plus daily vehicle flow results in univariate time series, hence the choice of recurrent neural networks “Long Short-Term Memory” LSTM which is useful for learning periodic data and which has the ability to hold data in long-term and short-term memory. The different datasets, representing the different time series, will be generated by agents who receive the daily flow of vehicles from sensors placed on the highways at the entry and exit of cities. Each agent represents an event and/or a city context to generate a single time series. These datasets, considered initially as normal data, will be used for training and testing our LSTM model. By this model, we want to look for two cases of anomalies: a decrease or an excessive increase in the flow of vehicles on the highways compared to the normal value. To test our model, we used the predictive error called the standard error of prediction.