Research On Daily Tourism Demand Forecasting Of Tourism Attraction Based On Heuristic Optimization-Based Decomposition-Ensemble Model

Over the past few decades,China’s tourism industry has grown rapidly and become the pillar industry in many regions,and the value added of national tourism and related industries reached 3.96% of the gross domestic product(GDP)in 2021.However,as a micro-individual of the tourism industry,the development of tourist attractions is not always smooth,especially the daily tourism demand of tourist attractions is affected by the impact of internal and external factors and fluctuates significantly,which affects the overall operational efficiency and stability of tourist attractions.At the same time,the serious safety accidents caused by the rapid fluctuation of tourist flow in recent years have made the industry realize the importance of accurate and timely tourism forecasts for tourist attraction management.In this context,providing effective daily tourism demand forecasts for tourist attractions is crucial to the industry’s development.Although previous studies have demonstrated that the decomposition-ensemble model can effectively deal with the daily tourism demand forecasting problem,the model limitations regarding the training and ensemble strategy are still worthy of attention.For the training strategy,previous studies have adopted fixed and independent training strategies to train the sub-models,thus ignoring the intrinsic relationship among the subseries,which leads to the potential loss of useful information during forecasting.For the ensemble strategy,although several studies have recognized the necessity of assigning appropriate ensemble weights to submodels,the limitations such as hyper-parameter dependence and single training target during ensemble likewise limit the model overall forecast performance and generalization ability.Therefore,this research focuses on the problem of daily tourism demand forecasting for tourist attractions,and adopts the decomposition-ensemble model as the main forecasting framework.Meanwhile,a hybrid heuristic optimizer consisting of genetic algorithm and adaptive differential evolution with optional external archive is introduced into models,and the sub-forecasting models and ensemble weights corresponding to each sub-series are incorporated into the constructed optimizer,and a hybrid evolutionary approach is proposed in the optimization process to enhance the search range and optimization efficiency.In addition,during the training process,the model takes both the overall training error and the forecast stability as the training objectives,instead of focusing only on the training error corresponding to each sub-series,in order to balance the overall forecast accuracy and generalization ability of the model.Finally,in order to evaluate the forecasting performance of the proposed model,a case study of three well-known domestic tourist attractions,namely,Jiuzhai Valley,Kulangsu and Siguniang Mountain,is conducted.By evaluating and comparing the overall performance of the model on point and interval forecasting,this study verifies the significant advantages of the proposed model in single-step forecasting(1 day ahead)and multi-step forecasting(5 and 9 days ahead),and demonstrates that the improved training and ensemble strategies in constructed model benefit in help improve the overall forecasting performance.