Traffic KPI Correlation Analysis And Short-term Forecasting Using Multi-source Big Data

With the process of urbanization and the development of social economy,urban traffic congestion problem becomes more and more serious.It is urgent to alleviate urban traffic congestion.Through short-term traffic forecasting,we can gain insight into the future traffic situation,adopt appropriate traffic control strategies,and make more efficient use of the existing road network,which is an economic and effective solution.Traffic Key Performance Index(KPI),which is used to reflect road traffic situation is the object of short-term traffic forecasting.It includes traffic flow,speed,density,etc.The difficulty of short-term traffic forecasting lies in the complexity of the traffic system.The traffic operation will be affected by many factors.The key to improve the forecasting accuracy is to introduce strong correlation factors into the forecasting process.Using real word data,we take traffic flow as an example to clarify the correlation between various factors and traffic state change,and extract correlation features.The correlation with itself in time dimension,with other sections on the road network in space dimension,and with weather conditions in environment dimension are analyzed.A two-stage Top-k correlation section extraction algorithm is designed to extract spatial association features.A combined model using Long Short-Term Memory(LSTM)and Artificial Neural Network(ANN)is designed for prediction.LSTM part is used to learn the depth-dependent features of data in time,and ANN part extends the diversity of features,embedding the non-temporal dependent features.Experiment based on real world traffic,road network and meteorological data shows that LSTM-ANN with multi-correlation factors gains mean absolute percent error less than 5%.It's superior in prediction accuracy compared with traditional prediction methods such as moving average and support vector regression.Further analysis shows that the introduction of spatial correlation factors significantly improves the prediction accuracy.The prediction ability is greatly improved in the case of frequent sudden changes in traffic,indicating the efficiency of top-k correlation section selection method.