An improved wavelet transform and multi-block forecast engine based on a novel training mechanism.

Area of Science:

• Renewable Energy Systems
• Electrical Engineering
• Data Science

Background:

• Growing integration of wind energy into power grids necessitates accurate generation forecasting.
• Existing wind-power prediction methods face challenges with accuracy and outlier sensitivity.

Purpose of the Study:

• To propose a novel, accurate wind-power generation prediction technique.
• To improve feature selection for robust data-driven diagnostics.
• To enhance the performance of Elman neural network (ENN)-based forecasting engines.

Main Methods:

• Utilized a dual-tree complex wavelet transform for signal processing.
• Implemented an outlier-insensitive combinatorial feature selection procedure.
• Developed a multi-stage forecast engine with a novel training mechanism based on an Elman neural network (ENN) optimized via stochastic search.

Main Results:

• The proposed model demonstrated effectiveness in real-world engineering data from Alberta, Canada, and Oklahoma, United States.
• The novel feature selection procedure reduced diagnostic efficiency degradation caused by outliers.
• The ENN-based forecast engine achieved high learning capabilities through the optimized training mechanism.

Conclusions:

• The proposed novel prediction technique significantly improves wind-power generation forecasting accuracy.
• The implemented methods offer a robust solution for integrating variable wind energy into power grids.
• The study validates the effectiveness of the advanced approach using practical engineering data.