Informer-LSTM-EWMA for Panax Notoginseng with Prediction of Early Intervention Time

Panax notoginseng is one of economic crop with significant medicinal value in southwest China. However, traditional extensive fertilization and irrigation method usually damages the soil, making it difficult to achieve sustainable cultivation of Panax notoginseng. In order to assess the growth status of Panax notoginseng and take timely action, this paper proposed an Informer-LSTM-EWMA model to predict the growth trend of Panax notoginseng in cultivation fields. In the experimental greenhouse, we set up two experimental fields A and B to verify the feasibility of the proposed method. Firstly, wavelet denoising is utilized on the date collected by the sensor. Then, Informer and Long short-term Memory (LSTM) are used to train the denoised data to obtain data prediction. Further process the predicted data through Exponential Weighted Moving Average (EWMA) for trend prediction. Finally, the early intervention time is obtained by searching for the maximum value. The results indicate that the model can predict early intervention time and its predictive performance is far better than that of a single model.