Target transformation by using CEEMDAN - PE

CEEMDAN - Complete Ensemble Empirical Mode Decomposition With Adaptive Noise

• CEEMDAN is a variant of the ensemble empirical mode decomposition (EEMD) algorithm, which provides an accurate reconstruction of the original signal and achieves better mode spectrum separation at a lower computational cost.

PE - Permutation Entropy

• PE algorithm is a dynamic mutation detection method that can easily and accurately locate the time when the mutation occurs and amplifies the small changes of the signal.

A. Generating Process

• Step 1: Obtaining multiple IMFs by CEEMDAN decomposition
• Step 2: Calculate PE according to different IMFs
• Step 3: IMFs with PE differences of 0.1 or less are summed and combined to form multiple (possibly one) new time series
• Step 4: Do one step prediction (horizon = 1) by using different models

B. Fitting models

• ARIMA
• XGBoost

C. Evaluation Criteria

• RMSE
• SMAPE