A Selection Method of Agricultural Product Price Forecasting Model based on Time Series Image Encoding 

Abstract

The forecasting and early warning of agricultural product price time series is an important task in the field of stream data event analysis and agricultural data mining. The existing forecasting methods of agricultural product price time series have the problems of low precision and low efficiency. To solve these problems, we propose a forecasting model selection method based on time-series image encoding technology. Specifically, we use Gramian Angle fields (GAFs), Markov transition fields (MTF), and Recurrence Plots (RP) to encode time series to images and retain all information about the event. Then, we propose a information fusion feature augmentation method (IFFA) to combine time series images. The time series combined images(TSCI) are input into the CNN forecasting model selection classifier. Finally, we introduce the idea of transfer learning to optimize the selection method of agricultural product price time series forecasting model. This idea can effectively reduce the overfitting phenomenon caused by insufficient data or unbalanced samples in real data set. Experimental results show that, compared with the existing methods, our IFFA-TSCI-CNN time series classification method has great advantages in efficiency and accuracy.