A bidirectional long short-term memory network for electron density diagnostic with double probe

Abstract

Abstract The double probe method is a plasma in situ diagnostic technology. Compared with Langmuir single probe, it has less influence on the background plasma and can obtain relatively accurate results. However, it can only collect some high-energy electrons in the plasma, and cannot directly measure the electron density (N e). In this paper, a double probe N e diagnosis approach based on Bidirectional Long Short-Term Memory (BLSTM) is proposed. After the training is completed, the accurate prediction of N e can be realized by using the double probe data, which solves the problem that the double probe cannot directly measure N e. In the plasma simulation environment of the laboratory, the plasma source is controlled to generate plasma with different densities, the current–voltage (I–V) characteristic data of the double probe at the same position are used as features, and the N e calculated by the triple probe is used as the label to train the BLSTM model. The mean square error is used as the loss function, the root mean square error (RMSE) and the prediction accuracy (Acc) are used as the evaluation indicators. The BLSTM network is evaluated according to the evaluation indicators and the hyperparameters are adjusted. After about 100 iterations, the RMSE of the BLSTM network to N e can be reduced to about 0.03. The final network is evaluated on a separate test set. The results show that in the range of 2 × 1013m−3–3 × 1014 m−3, the model can predict N e more than 95% accurately. This approach extends the application of the double probe method and is of great significance for improving the accuracy of plasma diagnostic methods. If it is applied to ionospheric plasma diagnosis, it can reduce the amount of data collected by the probe and improve the spatial resolution of ionospheric detection.