Traffic flow prediction method based on deep learning

Abstract

The invention provides a traffic flow prediction method based on deep learning, and the method comprises the steps of collecting traffic flow data information through a detector, inputting a discretefeature vector of the traffic information to an FM part in DeepFM, and obtaining an FM output vector with an implicit relation between discrete features; inputting the current detector section information sequence feature vectors at the previous t moments into a detector clustering label-based LSTM model with a multi-layer LSTM encoder to obtain an LSTM output vector; inputting the information sequence feature vector of each lane of the upstream detector and the LSTM output vector into an attention model at the previous t moments to obtain an attention model output vector with a flow change relationship between the current detector section and each lane of the upstream; and predicting the traffic flow according to the FM output vector and the attention model output vector. According to themethod, the accuracy of short-time traffic flow prediction can be effectively improved.

Description

technical field [0001] The invention relates to the field of traffic control, in particular to a traffic flow prediction method based on deep learning. Background technique [0002] In recent years, with the rapid growth of urban resident population and car ownership. The urban traffic problems are becoming more and more complicated. Accurate traffic flow forecasting can improve the accuracy of traffic guidance and the effectiveness of traffic control. [0003] The methods for short-term traffic forecasting in the prior art can be roughly classified into parametric and non-parametric forecasting. Among them, the parameter method for forecasting includes random method and time series. Typical methods include Kalman filter and autoregressive algorithm. The most classic and commonly used parameter method is ARIMA (Autoregressive Integrated Moving Average Model, autoregressive integer moving average model), and there are also A variety of improved algorithms based on the ARIM...

AI Technical Summary

Problems solved by technology

[0006] Since the traffic system is a complex large system with the characteristics of complex road relations, non-linear flow changes and random changes, the existing traditional methods such as parameter-based short-term flow prediction methods are often aimed at certain detection methods in the road network. However, when dealing with short-term flow forecasting, there are limitations such as strong subjective factors and poor forecasting accuracy.

Images