Short-term prediction network for short-wave MUF based on model-data dual-driven

Abstract

Predicting the maximum available frequency of short-wave communication presents the challenges of low prediction accuracy of classical prediction model methods and difficulty in obtaining training set data for machine learning prediction methods.To address this issue, a model-data dual-driven bidirectional gated recurrent unit (BiGRU) network for short-term prediction of MUF was proposed.On the model-driven, a large-scale dataset generated by the classical MUF prediction model was used as the model-driven training set, and a preliminary network was obtained after joint learning of the 2D CNN and the BiGRU network.On the data-driven, the preliminary network was trained twice using a small-scale measured dataset to obtain the final network CNN-BiGRU-NN.The simulation results show that the proposed network has reduced average root mean squared error (RMSE) at both daily and momentary scales compared with the GRU network, LSTM network and VOACAP model.

Keywords

• short-wave communication;maximum usable frequency;short-term prediction;model-data dual-driven;CNN-BiGRU-NN