AN LSTM model-based Prediction of Chaotic System: Analyzing the Impact of Training Dataset Precision on the Performance

Abstract

The chaotic systems are crucial due to their spread applications in different fields. The modeling and prediction of chaotic time series improved significantly with the recent advances in artificial intelligence algorithms, especially deep learning models. In this research, we use a deep learning-based long short-term memory (LSTM) model for the prediction of chaotic time series of the Lorenz attractor. This paper uses three datasets of (25k) samples with three precisions (the step size$\triangle$ t=0.01, 0.005, and 0.001) for training the LSTM model. The mean square error MSE and root mean square error RMSE metrics are used to measure the training performance. The best performance was obtained by increasing the precisions of the training data, where the values of metrics were 5. 3033e - 5 and 0.0073 for MSE and RMSE respectively. It is found that the training performance can be improved by increasing the precision of the training data i.e., reducing the step size. This provides useful knowledge towards reducing the number of data samples and corresponding acquisition time for a prediction.