Short-term Demand Forecasting Based Dispatching Optimization For Online Ride-hailing

Online ride-hailing is a non-cruising taxi service that utilizes internet technology to dynamically match passengers and vehicles.It has the characteristics of convenience,speed,and low price,and is loved by many passengers,gradually becoming an important supplement to urban public transportation.In recent years,the online ride-hailing industry has developed rapidly and the market size has gradually expanded.At the same time,the influx of a large number of online ride-hailing services has brought a lot of pressure to the transportation system and caused many environmental pollution issues,and the internal competition in the industry has also been exacerbated.In actual operations,the order matching method of nearby allocation is generally used,but there are some shortcomings in flexibility and efficiency,and there is still a supply-demand contradiction between drivers receiving orders and passengers taking taxis.In this situation,the platform and drivers hope to minimize costs,increase revenue,improve service quality and market competitiveness.To improve the above issues,this article takes the overall benefits of the platform and drivers and the waiting time of passengers into consideration to design a more practical dispatching strategy for online-ride hailing.At the same time,short-term demand prediction is incorporated into system modeling to reduce the ineffective patrols of idle vehicles and improve efficiency.The specific work of this article is as follows:(1)Preprocessing and analysis of trajectory data of online ride-hailing order.In order to eliminate some "problematic" data,data cleaning is performed on the obtained trajectory data of online ride-hailing order.Then the Origin-destination data of online ride-hailing is extracted from the cleaned trajectory data,and the research area is spatially divided.To obtain the spatiotemporal correlation of online ride-hailing demand,the distribution characteristics of online ride-hailing demand are analyzed from the temporal and spatial dimensions respectively.Finally,in order to determine the feature variables that are input into the short-term demand prediction model of online ride-hailing and the experimental data for online ride-hailing dispatching,this paper conducts the correlation analysis of the factors affecting the demand for online ride-hailing.(2)Short-term demand of online ride-hailing based on CNN-LSTM-Attention.The demand data of online ride-hailing has nonlinear characteristics.According to the correlation analysis of influencing factors,it can be known that it will be affected by weather,region,etc.This paper designs a short-term demand forecasting model,which combines the CNN model with the Bi LSTM model,and attention mechanism is also added.The prediction model uses CNN to capture the spatial characteristics of input feature variables,and uses Bi LSTM to capture temporal information from both forward and backward time series.Attention mechanism is also introduced into the prediction model to adjust the impact of the information of different time periods on the prediction period.Through experiments,it can be verified that the fusion model CNN-Bi LSTM-Attention has the ability to extract temporal and spatial features well,and it also has ability to focus on key information.At the same time,the prediction results serve as an important support basis for idle vehicle dispatching during online ride-hailing dispatching.(3)Modeling and algorithm design for optimizing online ride-hailing dispatching based on hybrid particle swarm optimization algorithm.The online ride-hailing dispatching model established in this paper takes into account both the interests of the operator and the quality of system service.Therefore,the optimization dispatching model is established with maximizing the common benefits of online ride-hailing platforms and platform drivers and minimizing the waiting time cost of passengers as the optimization objective,the matching scheme between online ride-hailing and orders as a decision variable,and constraints such as online ride-hailing constraints,order constraints,time window constraints,and dispatching distance constraints as model constraints.At the same time,idle vehicles are matched and dispatched to reduce the idle cost of online ride-hailing based on the short-term demand prediction results for online ride-hailing in various regions,A receding horizon control strategy is employed to achieve multi-cycle dynamic dispatching of online ride-hailing,and a hybrid particle swarm optimization algorithm is designed to optimize the dispatching scheme for each dispatching interval.The hybrid particle swarm optimization algorithm improves the particle swarm algorithm based on the sequential crossover method in genetic algorithm,avoiding the situation where the solution results fall into local optima.The experiment shows that the dispatching algorithm designed in this article can improve the target value of the objective function,reduce the idle cruising cost of online ride-hailing,reduce the waiting cost of passengers,and improve the order completion rate.The research in this paper helps to optimize the dispatching strategy of online ride-hailing platforms and improve the interests and service quality of online ride-hailing operators.It is of great significance to study the dynamic dispatching strategy of online ride-hailing.