Space-time traffic flow prediction method driven by enhanced hierarchical learning

Abstract

The invention provides a space-time traffic flow prediction method driven by enhanced hierarchical learning. With full utilization of mutual correlation of related road sections in time and space, a nonlinear, high-dimensional and random road traffic flow evolution mode is dynamically simulated through a reinforced hierarchical learning network; road traffic flow feature extraction based on a restricted Boltzmann machine model is designed and realized; and dimensionality reduction is further carried out on road traffic flow data of an input layer and the road traffic flow characteristics afterdimensionality reduction are classified by using an SVM method to obtain a final traffic flow prediction result. The tests and on-site detection show that the accuracy of the prediction result is over 85.6% when the reliability of the sample is 75%; and the accuracy of the prediction result is over 96.3%, when the reliability of the sample is 90%. Therefore, the accuracy and reliability of traffic flow prediction are greatly improved. The traffic flow prediction method having advantages of solid theoretical basis, good traffic flow prediction timeliness and good real-time performance of the prediction result has the wide application space.

Description

technical field [0001] The invention relates to a space-time traffic flow prediction method, in particular to a space-time traffic flow prediction method driven by enhanced hierarchical learning, which belongs to the technical field of traffic flow prediction. Background technique [0002] With the rapid development of urban road traffic, traffic pressure is increasing, and road traffic congestion is becoming more and more serious. Intelligent road traffic system plays an important role in relieving urban road traffic pressure, improving the quality of urban life and work, and reducing environmental pollution. Road traffic conditions, especially traffic flow prediction is a very important part. Due to the characteristics of non-linearity, high dimensionality, and randomness of road traffic flow, a large amount of road traffic flow information contains rich information, which plays a key role in the prediction of traffic flow. The speed and accuracy of traffic flow forecasti...

AI Technical Summary

Problems solved by technology

[0004] On the whole, the traffic flow prediction methods in the prior art mainly have the following defects: First, a large amount of work in the prior art is to use the evolution law of road traffic flow in time series to predict traffic flow, and generally adopt autoregressive moving average Model, the autoregressive moving average model is a random sequence formed by road traffic flow over time. The time variation law of this group of random sequences reflects the continuity of traffic flow data in time, so by mining the time in historical data However, these traffic flow prediction methods based on time series characteristics are all statistical methods that only rely on the regularity of traffic flow data in time, and the accuracy and reliability of traffic flow prediction are very low. Low

The second is that there are still some methods in the prior art that use the spatial correlation in the road traffic network to analyze the evolution trend of road traffic flow, but this method is based on an assumption that the traffic flow density on the road is only the same as that of its adjacent The traffic flow density on the road section is related, so this method only uses the association between the adjacent road and the predicted road to predict the flow of the road to be predicted, and does not fully consider the influence of the far road on the predicted road and the time and space of the road to be predicted. Correlation on

The third is that some existing technologies use reinforcement learning methods to use multi-layer network structures to extract typical features in complex data such as traffic flow, but the existing technologies use reinforcement learning methods, which are all static networks and cannot express roads well. The dynamic mode in the traffic flow network, and all the road traffic flow data needs to be entered, and there is no indicator to measure the time-space correlation between the road sections in the area, so it will cause a lot of redundancy, which will seriously affect the training effect and speed of the network , the timeliness of traffic flow prediction is low, and the prediction results lag seriously

Images