Analysis On Short-time Traffic Flow Forecasting

All parts of the world are continuously increasing their investment in intelligent transportation systems and strategically laying out the system in advance to find new development points and new vitality for the city.The traffic information system provides a quick traffic guidance for the city and can provide the necessary technical support for the city’s public travel.It is also an important branch of the traffic accident handling system.Traffic flow forecasting is a key point to solve the traffic information system,but the traffic in the city has its own characteristics,and the traffic flow has certain complexity.It is very difficult to use algebraic expressions to analyze its changes and laws accurately.Therefore,the study of traffic flow prediction has very important practical significance.In-depth study of XGBoost and light GBM principles,found that light GBM has a good advantage of feature screening;light GBM has not yet been applied in the short-term traffic flow prediction research,whether it is more suitable for short-term traffic flow prediction than XGBoost,the prediction results are more reliable Whether the accuracy is higher and the time spent is shorter has become an issue that needs to be verified.Through the analysis of the applicability of short-term traffic flow typical characteristics of the model,based on the road network,holidays and weather and other spatial and temporal complexity analysis modeling,combined with the measured historical traffic flow data for prediction simulation;select BP nerves Network,random forest,linear regression,XGBoost,and light GBM are the five short-term traffic forecasting methods to build corresponding forecasting models to verify the traffic prediction effect of the light GBM algorithm.Then use light GBM to sort the model parameters that have already been built to make the model parameters more reliable and accurate.The model is constructed by light GBM and other four methods to predict traffic flow.Comparing and analyzing the forecast results,we can see that the light GBM model is much better than the other four methods,regardless of whether or not the features are sorted and screened.The short-term traffic flow forecast is the best.In the five model predictions,after the light GBM feature filtering and sorting process,the prediction results of the model are closer to the real data.Compared with light GBM,XGBoost has a slow training speed and can not better reflect the time-varying characteristics of traffic flow.It is verified that light GBM can greatly improve the prediction accuracy of the model and provide a more accurate and reliable method for solving traffic flow prediction problems.