Research On Demand Forecasting For Ride-hailing Based On Improved EMD-LSTM Model

The continuous development of cities has brought about economic growth and prosperous living for residents,but it has also brought many challenges to the transportation industry.The increasingly complex urban transport structure leads to traffic paralysis in high-density areas and waste of vehicle resources in low-density areas.How to allocate vehicle resources responsibly for residents’ needs,relieve the imbalance between vehicle supply and demand,and ease corresponding traffic pressure is a difficult problem for transportation departments and related vehicle platforms.Based on the ridehailing order data of Haikou and Beijing,this article analyzes the characteristics of residents’ travel in different areas,explores passenger hotspots,and further predicts the demand for ride-hailing,providing reference data for relevant departments and scholars.The main research work of this article includes:(1)Research on the spatiotemporal characteristics of urban residents’ ride-hailing demand.Based on ride-hailing order data,this article studies the ride-hailing demand characteristics of Haikou and Beijing residents from the perspective of time and space.Specifically,the study focuses on the overall,working day,and non-working day features to understand residents’ ride-hailing demand spatiotemporal characteristics,laying the foundation for building hotspot area mining models and ride-hailing demand prediction models.(2)Research on hotspot area mining based on improved K-Means.Based on the spatiotemporal information of Haikou’s ride-hailing GPS dataset,this article further understands residents’ travel behavior characteristics and corresponding passenger hotspots.Taking the ride-hailing demand dataset during the peak period of working days as an example,the KMeans clustering algorithm is selected to mine passenger hotspots.Considering the influence of the number of clusters K on the model results,this article introduces the genetic algorithm(GA)to adaptively find the optimal cluster center and obtain the best cluster.By comparing the mining effect with the DBSCAN algorithm and the original K-Means algorithm,this study confirms the effectiveness of the model.(3)Research on ride-hailing demand prediction based on improved EMD-LSTM.To further alleviate traffic congestion in hotspot areas,this article proposes an improved EMD-LSTM model to predict ride-hailing demand time series.The ride-hailing time series is decomposed by empirical mode decomposition(EMD)to reduce its instability and obtain multiple stable intrinsic mode functions(IMF).Considering that too many IMFs will lead to excessively high model complexity and weak temporal characteristics between IMF,this article further uses the K-Means algorithm improved by GA to determine the optimal number of clusters and obtain new sub-time series.The new sub-time series is input into the prediction model for prediction,and the results are summed up to obtain the final prediction result.Finally,the model’s prediction accuracy is compared with linear models,nonlinear models,and unimproved models using Haikou and Beijing datasets to confirm its accuracy.