Research On Demand Prediction And Parking Lot Location Of Bike-Sharing

In recent years,various machine learning models and deep learning models have emerged one after another,providing many effective solutions for predicting the demand for bike-sharing.However,these models often only consider the time series characteristic data of the demand for bike-sharing at a site,and do not combine the time series characteristics of the demand for bike-sharing at a certain site with the spatial domain characteristics of different sites,so the effect of these demand forecasts is not very desirable.In view of this,the dissertation builds two kinds of spatiotemporal domain shared bicycle demand prediction models based on attention mechanism.By introducing the attention mechanism,the prediction accuracy can be effectively improved.Nevertheless,in the existing planning of parking spots for bike-sharing,it is generally assumed that users only choose vehicles from the parking spots closest to the demand point,while ignoring the possibility of choosing vehicles from other parking spots within the user’s maximum walking range.In the dissertation,a joint assignment strategy is proposed to solve this problem,and an Improved Non-dominated Sorting Genetic Algorithm-II(INSGA-II)is used to solve it.The dissertation combines the spatial domain features of bike-sharing with time series features,and introduces attention in the process of combining the graph convolutional neural network(GCN)with the long short-term memory recurrent neural network(LSTM),and builds a graph convolution and LSTM model based on attention mechanism.Due to the artificial introduction of the attention mechanism,the model has the bottleneck of improving the accuracy.Therefore,the dissertation introduces the transformer model into the demand forecast model of bike-sharing,and combines it with the graph convolutional neural network model to realize the learning of the temporal and spatial domain features of bike-sharing based on the native attention mechanism of the transformer model.The experimental results show that the native attention mechanism of the transformer model can achieve better prediction results of the demand for bike-sharing than the LSTM model combined with the artificially introduced attention mechanism.Based on the possibility of multiple parking spots within the maximum walking distance,the dissertation proposed a joint allocation strategy to meet the demand of multiple spots,and established a corresponding multi-objective optimization mathematical model.The chromosome coding mode of the original Non-dominated Sorting Genetic AlgorithmII(NSGA-Ⅱ)algorithm is redesigned,and genetic operators such as crossover and mutation are improved.The INSGA-Ⅱ algorithm is then used to determine the parking location and the corresponding number of bicycles.