Catering Delivery Order Prediction Model Based On Deep Learning

As the pace of life accelerates,more and more people choose to order meals online.However,the present problem is not being punctual and the high cost of delivering meals.Solving this problem will benefit restaurants,take-out platforms and customers.The main reason why this problem is difficult to solve is the dynamic and high frequency of orders.Forecasting the information of the orders can solve this problem.The forecast is based on the association of future catering take-out orders with historical catering take-out orders.Since the order information has information of time,customer location,and restaurant location,this association involves not only the spatio-temporal association of customer locations(restaurant locations)in the same area,but also the spatio-temporal association of customer locations(restaurant locations)in different areas.In addition,there is a spatio-temporal association between the location of the customer and the location of the restaurant.The traditional flow prediction problem is often made through macro-prediction methods of time series and cannot meet the forecasting requirements of this paper.Based on this,this paper proposes three predictive models of catering delivery orders based on deep learning.The main research work is as follows:First,customers and restaurants are clustered into customer groups and restaurant groups,and find the spatial structure relationship of orders in the customer group space and the spatial structure relations that appear in the restaurant group space.Then,the simultaneous orders in the restaurant group and the customer group are linked together to form the customer-restaurant group space and find the spatial structure connection in the customer-restaurant group space.Furthermore,in the case of actual meal delivery,time is divided into time slices.Then,the three spaces are linked in time-sequence order to find the connection of the three spaces in time series.Finally,considering factors such as preferential activities,holidays and weather,add them to the model to improve the predictive performance of the model.RMSE,MSE,and MAE are the evaluation criteria for the three models,with RMSE around 0.17,MSE around 0.030,and MAE around 0.013.Almost all areas without orders are eliminated.For regions with more orders,the trend of their orders is also can be predicted.This basically meets the requirements of the forecast and can reflect that the model proposed in this paper is feasible and effective.