Research on Grain Yield Prediction Model Based on Contribution Multiplier and Bidirectional LSTM Neural Network

It is extremely difficult to predict grain yield because of several influencing factors and the uncertainty and nonlinearity among them. In order to improve the prediction accuracy of grain yield, one new prediction model is proposed based on bidirectional LSTM Neural Network. Firstly, the correlation coefficients between the grain yield and each influencing factor are computed and sorted, thereby the main influencing factors are chosen; then one contribution multiplier is defined and weighted with the corresponding influence factor by the correlation coefficient; finally, the weighted factors and the historical grain yield will be imputed to bidirectional LSTM neural network and the future grain yield can be obtained. The simulation analysis has shown the contribution multiplier can change the difference among the main influence factors, and compared with the traditional prediction methods as ARIMA, SVR, RF, LSTM, the proposed prediction model can realize the medium and short-term prediction of grain yield with the higher accuracy.