Research On Short-time Traffic Flow Data Repair Method And Prediction Model Based On Machine Learning

The total spatial and temporal resources of urban roads are limited.The continuous growth of motor vehicle ownership and travel demand has added pressure to urban traffic,leading to the increasingly severe scope and severity of urban traffic congestion.Therefore,finding a real-time and accurate traffic flow prediction method which suitable for specific environments and applying this method can greatly promoted traffic flow guidance and traffic control.Most scholars are studying the improvement of generalization ability or prediction accuracy of a single model in the research of shortterm traffic flow prediction technology.However,the traffic flow which is affected by many factors such as the environment and road network conditions has strong randomness and nonlinearity.Moreover,there is a big difference in the traffic flow data between the peak and the peak periods of each day,which means that a single model prediction involves a large prediction range and time interval,and the accuracy of the prediction results may be affected if a single model is used for the prediction of traffic flow under multiple changing patterns every day.Therefore,this paper studied the impact of different characteristics,different times,and different regions on the prediction accuracy of the short-term traffic flow prediction model.Then,the road traffic flow data of mountain cities and the application of machine learning methods were used to established a RF repair model which has higher accuracy for abnormal traffic flow data in mountainous cities and a responsive GA-LSTM short-term traffic flow prediction model that considers multiple influencing factors.First,the traffic flow anomaly data repair method and data pre-processing process were studied,and the vehicle RFID data aggregation method and traffic flow anomaly data RF repair method were proposed.Secondly,the LSTM network which commonly used in the field of short-term traffic flow prediction was carried out to improve the parameter optimization research.GA was used to find the best combination of traffic flow time series sliding time window step size N and LSTM network hyperparameters,then a short-term traffic flow prediction model based on GA-LSTM was established.Finally,for the problem of short-time traffic flow prediction in mountainous cities under multiple influencing factors,a multi-model responsive GA-LSTM short-time traffic flow prediction method was proposed in this paper.The method classified the traffic flow time series according to the influencing factors such as whether it is a weekday,the trend of smooth or drastic traffic flow changes,and the range of the prediction area,and then built the corresponding short-time traffic flow prediction sub-models respectively.In this paper,vehicle RFID detection data of a regional road network in Chongqing was used as experimental data,and MAE,MSE,RMSE and MAPE were used as error evaluation criteria to compare the RF anomaly data restoration method and the mountainous city responsive GA-LSTM short-time traffic flow prediction model proposed in this paper with a variety of other mainstream restoration models and prediction models.Experiments show that the RF restoration model and the mountainous city responsive GA-LSTM short-time traffic flow prediction model proposed in this paper have better restoration and prediction effects than other models mentioned in the paper,and the models are reliable.It is expected that the proposed restoration and model will be applied to the field of traffic flow prediction to improve the existing traffic management measures,help the construction of smart cities,and promote the pace of comprehensive deepening reform.