Short-term Traffic Flow Forecasting Based On Date Mining And Date Fusion

The controlling and management of intectlence traffic is based on the real-time and accurate forecasting of short-term traffic flow. Based on a single time dimension time series forecasting methods are difficult to solve the traffic flow a high degree of complexity, randomness and uncertainties of the predicted effect is not satisfactory. State transition of traffic flow is not only the time dimension of one-dimensional time series, spatial distribution of traffic flow changes also played a considerable impact. Thus, the more accurate short-term traffic flow forecasting, we need a predictive value of space dimension time series dimension of time to amend the forecast results.This departure from the traffic flow characteristics, for a time series of real traffic flow characteristics and spatial correlation, was proposed based on two-dimensional space-time integration of the short-term traffic flow forecasting methods. In the short time dimension of traffic flow on the performance of a strong non-linear, time ariability and uncertainty, the general prediction is very difficult to achieve prediction accuracy requirements, therefore, the proposed traffic flow data will be carried out using avelet transform " frequency "decomposition and single reconstruction, the corresponding high frequency and low frequency components, component by component is relatively simple, the signal change is more stable. The various components of different forecasting methods to predict the high frequency components by the strong non-linear and neural network has a strong nonlinear approximation ability, so BP neural network with high frequency; low-frequency component is rendered more regular function curve, so using exponential smoothing. Prediction vector of each component will be added to the sum, the results of comprehensive prediction can be obtained. Dimensions in space, traffic flow under the transient state between the spatial distribution of the correlation with the instantaneous output to the traffic flow and its adjacent upstream and downstream traffic flow related to the nonlinear relationship for this uncertainty, this paper RBF neural network for spatial dimensions of the short-term traffic flow forecasting. Finally, the use of information fusion technology can reasonably coordinated multi-source data, fully integrated useful information in a short period of time, with little cost, by using a single sensor can not get the advantages of the data features, will be based on time series prediction spatial correlation results and prediction results of the optimal weight integration, get a higher overall prediction accuracy of short-term traffic flow prediction.