Research On Shared Bicycle Traffic Flow Prediction Based On Hierarchical Clustering And LSTM Model

With the acceleration of industrialization and modernization,environmental pollution and congestion caused by the rapid increase of urban traffic flow have become increasingly serious.With the continuous development of green,sharing,and sustainable development concepts,shared bicycle travel modes have emerged.While sharing bicycles to ease urban road congestion,it also facilitates people's 'last mile' travel.However,due to the mismatch between the understanding of the operating rules of shared bicycles and the speed of shared bicycle development,a more comprehensive management model for shared bicycles has not yet been formed.Problems such as disorderly parking and crowding of roads,and uneven distribution of vehicles have become increasingly serious.Shared bicycles are becoming residents.Life brings convenience and burdens to city management.Therefore,accurately predicting the user's travel needs and arranging a reasonable vehicle scheduling scheme are effective ways to improve the operating efficiency of the shared bicycle travel mode.This article takes a sample of 2.5 million shared bicycle trips in Yangpu District,Shanghai in the 14 th day of September 2019 as a sample to analyze the behavior of shared bicycle trips,and uses the relevant features to predict the 7-bit Geohash area(hereinafter referred to as " Station ")to provide solutions for intelligent operation management of shared bicycles.First,the original trajectory data set was used to analyze the characteristics of shared bicycle trip behavior.An overview of the 14-day cycling behavior in Yangpu District was obtained.The study focused on the time and climate characteristics of the cycling: discovering the cycling users during the working day And non-working days with different fluctuating riding demands,and extracted the riding tide phenomenon during the day;explored the correlation between the amount of trips and temperature,wind speed,weather,etc.The operations such as Geohash inverse transform,sample pre-filtering,custom rule filtering,and standardization are used to complete the cleaning of the original data,reducing the noise and redundancy of the data.Secondly,according to the temporal and spatial characteristics of the traffic flow problem scenario,k-means clustering based on different clustering dimensions was used to complete the classification of the riding geohash sites in Yangpu District: for the spatial nature of geohash site distribution The k-means clustering using absolute distance calculation method based on the geographic location of the site completes the first classification of the sites;for the mobility of shared bicycles between Geohash sites,each Geohash is generated based on the first classification The traffic conversion vector of the site,based on the traffic conversion relationship between the sites,uses kmeans clustering of the Euclidean distance calculation method to complete the second classification.After two classifications,the Geohash site of bike sharing in Yangpu District was divided into eleven internal bike sharing areas with similar usage rules,close connections,and relatively independent travel areas between clusters.Finally,for different riding area clusters,considering the self-mobility and correlation of bicycle traffic between travel sites within the area,and combining user riding behavior characteristics,an LSTM-based riding demand forecasting model was established for each cluster.The experimental verification of shared bicycle track data in Yangpu District shows that the model has achieved good prediction results on denser clusters and sparse clusters on the riding Geohash site.The predicted RMSE values are 0.076 and 0.051,and the results are fit and shared.Actual situation of bicycle dispatch operation.The comparison of the prediction effect with ARIMA model,support vector machine(regression),and LSTM model without cluster clustering validates the effectiveness of the research method in this paper.The main characteristics of this article are reflected in two aspects:(1)Fully consider the spatio-temporal characteristics that affect the number of shared bicycle trips,and use the two-stage clustering method to divide the shared bicycle riding Geohash station in Yangpu District into multiple riding area clusters In the following,the self-fluidity and correlation of bicycle flow are considered to construct a prediction model for each riding area,so that the model is more targeted to the data in the same area.(2)Aiming at the characteristics of large-scale data and complex features,the LSTM neural network model,which has advantages in solving gradient explosions and extracting long-term sequence features,is used as a bicycle demand forecasting algorithm.