LSTM Based Time Series Forecasting of Noisy Signals

Time series analysis of signals is important to understand, track the pattern of variation of information and develop a suitable prediction model. The prediction model is used to generate, reproduce and forecast the values of the time series data without the practical experimental set up where the information can be useful for various applications. In this paper, Long Short Term Memory (LSTM) neural network based time series prediction for noisy signal is investigated using MATLAB simulations. The effect of noise on the prediction is studied by varying the level of the additive noise strength on the useful data in terms of standard deviation. From the simulation results, it has been found that LSTM can effectively forecast the time variation of the signal even in the presence of additive noise signal. The result also shows that the Root Mean Square Error (RMSE) prediction parameter is increasing with an increase in the level of the noise standard deviation.