Affiliations: [a] College of Computer Science, Guangdong Polytechnic Normal University, Guangzhou, Guangdong, China | [b] Department of Computer Science, Hong Kong Baptist University, Hong Kong, China
Correspondence:
[*]
Corresponding author: Zhenghong Xiao, College of Computer Science, Guangdong Polytechnic Normal University, Guangzhou, Guangdong 510665, China. E-mail: [email protected].
Abstract: A combined forecasting model of merchandise sales is proposed on the basis of differential evolution algorithm (DEA). Time series forecasting model and back propagation neural network forecasting value are used to construct the combined forecasting model. Forecasting results obtained by two single forecasting methods are set as the inputs of the DEA, whereas actual historical data values are used as the expected outputs of the network on the basis of the principle of minimum sum of squared errors and determine the weights of various forecasting methods. This method is validated on the basis of the actual sales data collected from Haowanjia online and Rossmann stores in Kaggle. The proposed method performs better in terms of forecasting accuracy than the combined forecasting model based on the weight coefficient of reciprocal variance.
Keywords: Differential evolution algorithm, combined forecasting, time series, BP neural network, merchandise sales
