Research Of A Combined Demand Forecasting Model For Apples

As the number one fruit and one of the dominant agricultural product types in China,apples are loved by more and more people,while the demand for apples is also increasing year by year.With the continuous development of agricultural information technology,the analysis of agricultural product demand through massive data is conducive to promoting the balance of apple supply and demand,stabilizing market regulation and inventory information construction,and has good research significance and application value.In response to the problems of low prediction efficiency,poor accuracy and poor stability caused by simple data processing and single prediction model in the prediction of relevant agricultural products in China,this paper proposes a dual-structure combined prediction model based on wavelet decomposition,and the specific research contents are as follows:（1）Data collection and pre-processing.We selected representative Fuji apples as the research object,and selected 25 characteristics of Fuji apples and competitor-related factors,economic factors and consumer factors from 1978 to 2021 through the information released by the National Bureau of Statistics and the data provided by the Ministry of Agriculture and Rural Affairs.The objective and scientific nature of the feature vector selection,the diversity of the data while ensuring low redundancy,improve the anti-interference and prediction stability of the model.（2）Single prediction model construction.Wavelet decomposition is used to decompose the apple demand series features into a one-dimensional linear trend feature structure and a twodimensional nonlinear detail feature structure,and the remaining feature vectors are eliminated by trend volume operation to enhance the variation among the data.On the basis of data preprocessing,the apple demand forecasting models based on SVR,ARIMA and BP neural network are constructed by parameter seeking respectively.（3）Two types of apple combination forecasting models are constructed based on different combination methods.The traditional machine learning model with strong linear fitting ability and the neural network model with strong nonlinear mapping ability are used to predict the linear and nonlinear structures of apple demand respectively,and the integration of the two types of structures is completed by wavelet reconstruction algorithm,so as to build a dualstructure combined prediction model based on wavelet decomposition.The comparison experiments were carried out based on the evaluation indexes of mean absolute value error,root mean square error,mean absolute percentage error and regularized square root mean difference.The experimental results show that the average absolute percentage error and the average absolute error of the dual-structure combined forecasting model based on wavelet decomposition is 0.083 and 81.141,which are both higher than the combined forecasting model based on the weighting method in terms of forecasting accuracy and forecasting efficiency,and can accurately achieve the goal of apple demand forecasting,which is a scientific and effective apple demand forecasting model.