A Response Surface Based Parameter Calibration Method for Traffic Flow Prediction Model

Abstract

The invention discloses a method for quickly calibrating parameters of a traffic flow prediction model based on a response surface, comprising the following steps: S1, determining the solution space of the parameters of the traffic flow prediction model; S2, calculating the model error under the combination of initialization parameters; S3, generating parameters and the response surface between the model error; S4, convergence test; S5, determine the next parameter combination point; S6, calculate the model error under the parameter combination point. The method of the present invention can quickly calibrate the parameters of the traffic flow prediction model without calculating the model prediction results under all parameter combinations, which greatly saves the time occupied by model trial calculations, and the method of the present invention takes into account both search efficiency and fairness, preventing Calibration results are limited to local optimal solutions. At the same time, the method of the present invention has many application scenarios, simple operation and easy programming.

Description

Technical field [0001] The present invention relates to a transportation engineering, a smart transportation field, in particular to a method of calibration of the traffic flow prediction model based on the response surface. Background technique [0002] Traffic flow prediction is the basis of intelligent transportation system for traffic status, navigation vehicle path selection, intelligent traffic control, etc., accurate traffic flow forecasting model can better serve the operation of intelligent transportation systems. In recent years, with the advancement of the new round of technology revolutions such as traffic data, artificial intelligence, deep learning, traffic flow forecasting model has been greatly developed, derived more complex, more precise traffic flow prediction model, such as depth Learning traffic flow forecasting model. [0003] Although this model can greatly improve the prediction accuracy of traffic flow, it is facing the problem of model parameters (ie, lo...

AI Technical Summary

Problems solved by technology

[0003] Although such models can greatly improve the prediction accuracy of traffic flow, they are faced with the difficulty of model parameter calibration (that is, finding the optimal parameters). There are too many model parameters and hyperparameters, the solution space is too large, and it is difficult to pass enumeration, etc. Solve by direct method

Images