A Short-term Urban Traffic Flow Prediction Method Based on Spatial-Temporal Similarity of Traffic Flow

Abstract

The present invention relates to a short-term urban traffic flow prediction method based on the spatio-temporal similarity of traffic flow. This method transforms the traditional non-parametric regression method and includes the following steps: S1. Based on the spatio-temporal similarity of traffic flow, define the traffic flow Time state vector, space-time state vector; S2, construct the "current space-time state vector" of the traffic flow in the current period; S3, construct the "historical space-time state vector" of the traffic flow in the same period on different dates in history; S4, use the distance metric function to calculate The "spatial and temporal similarity distance" between the current and each historical spatio-temporal state vector; S5. Select the dates where the k historical state vectors with the smallest spatio-temporal similarity distance are located, and find the traffic flow in the prediction period corresponding to these k historical dates; S6. Based on For the traffic flow in the prediction period corresponding to these k historical dates, use the prediction function to calculate the traffic flow of the target road section in the next period; S7. Based on the prediction results and actual results of the traffic flow, evaluate and analyze the prediction error of the target road section. The purpose is Improve the accuracy of urban short-term traffic flow prediction.

Description

technical field [0001] The invention relates to the technical field of application of big data on trajectory of intelligent traffic floating vehicles, in particular to a method for predicting urban short-term traffic flow based on temporal and spatial similarity of traffic flow. Background technique [0002] Traffic flow forecasting is divided into long-term traffic flow forecasting for traffic planning, medium-term traffic flow forecasting for traffic management, and short-term traffic flow forecasting for real-time traffic control and guidance. The time span of short-term traffic flow prediction for control and guidance services is greatly reduced compared with the previous two, generally no more than 15 minutes. In a short period of time, the periodicity of traffic flow changes decreases, randomness and nonlinearity increase, and the impact of various disturbances is also greater. Therefore, short-term traffic flow prediction is more difficult than medium- and long-term p...

AI Technical Summary

Problems solved by technology

[0004] Traditional short-term traffic prediction methods mainly include historical average model (HA), BP neural network model, original short-term traffic flow prediction Original KNN model, etc. In the prediction results and analysis of these models, the HA model only considers the cycle of traffic status It does not consider its time-varying nature, has poor anti-interference ability, and cannot reflect the dynamics and nonlinearity of traffic flow; the BP-NN model can identify complex nonlinear systems and does not require empirical formulas, but there is no unified method for establishing the number of nodes in the hidden layer. The method can only be tried and tested based on experience, there are local minima, the convergence speed is slow, and it is difficult to realize online adjustment; although the Original KNN model is also a data-driven non-parametric regression method, it only considers the similarity of traffic flow in the time dimension , and only using state vectors based on time similarity cannot fully describe the characteristics of road traffic flow

[0005] With the deepening of research on short-term traffic prediction methods, other scholars have proposed an improved KNN model based on spatio-temporal weights (STW-KNN). Although the STW-KNN model also considers the influence of spatial factors on traffic conditions, it simply adds The upstream and downstream traffic conditions are not considered, and the flow relationship between the surrounding road sections and the target road section is not considered, nor is the influence degree of the surrounding road sections measured.

Images