Agricultural Product Price Forecasting Based On Empirical Mode Decomposition

The large price volatility of agricultural products not only increase the economic risks faced by agricultural participants,but also adversely affect the daily life of people,thus including unnecessary social panic.An accurate price prediction of agricultural products is conducive to the timely response of agricultural practitioners to the risks existing in the agricultural product market,and helps relevant institutions and government departments to release effective policies for macro-control of agricultural products market.In order to effectively improve the prediction accuracy of agricultural product price,this paper investigates the integrated prediction model based on empirical mode decomposition method and improves the current method based on the characteristics of agricultural price forecasting,so as to improve the prediction accuracy.First,due to problems of the End Effect and Mode Mixing in the empirical mode decomposition algorithm,this paper proposes a combined expansion method based on characteristics of similar triangle waveforms,and combines the representative complementary noise method to improve the empirical mode decomposition algorithm for suppressing end effect and mode mixing.Simulation experiments show that the improved empirical mode decomposition algorithm can effectively suppress the endpoint effect and mode mixing,thus verifying the effectiveness and feasibility of the algorithm.Second,due to the characteristics of non-stationarity and small sample in agricultural product price time series,this paper proposes an improved GA-SVR prediction model to solve the parameter selection and over-fitting problems of SVR.Through the random analysis of GA algorithm selection operator,a differential selection operator is proposed.Compared with other selection operators,the difference selection operator is more effective in the search for optimal solutions in complex problems.Based on the improved GA algorithm,the SVR prediction model is constructed for agricultural product price.Empirical result shows that GA-SVR has higher predictive performance on agricultural product price prediction than RBF neural network.Finally,it is difficult to solve the problem of predictive modeling of complex time series for single prediction model,and then the complementary ensemble empirical mode decomposition method is used to decompose it into multiple simple and relatively stable Intrinsic Mode Functions(IMF)so as to improve the prediction accuracy.At the same time,considering the mode mixing problem that may occur in empirical mode decomposition,the Permutation Entropy(PE)is used to analyze the complexity of the IMF sequence for further decomposing the high complexity sequence to ensure smoother IMF sequence.Further,the gray correlation degree(GCD)is used to analyze the correlation of multiple sub-sequences,and the sub-sequences with high correlation are combined to reduce the prediction modeling time.The empirical results show that combined with GA-SVR,the complementary ensemble empirical mode decomposition integrated prediction model(CEEMD-PE-GCD)proposed in this paper can effectively improve the prediction accuracy and reduce the computation time.