An efficient stacked recurrent broad learning scheme for PV cluster power forecasting.

Area of Science:

• Renewable Energy Systems
• Artificial Intelligence in Energy
• Machine Learning for Power Grids

Background:

• Accurate photovoltaic (PV) power prediction is crucial for efficient grid management and resource optimization.
• Existing methods often struggle with the complexity of PV cluster data and resource limitations of training devices.
• Predicting power for PV clusters offers advantages over single-site predictions for grid integration.

Purpose of the Study:

• To propose a novel Stacked Recurrent Broad Learning System (Stack-RBLS) for precise PV cluster power prediction.
• To develop a method that reduces training time and improves prediction accuracy, suitable for resource-constrained environments.
• To ensure reliable forecasting through an improved data cleaning technique.

Main Methods:

• Implementation of a data cleaning method using a sliding time window (STW)-based isolation forest algorithm and k-nearest neighbors (IFA-KNN) for linear time complexity.
• Utilizing a lightweight broad learning model with Recurrent Broad Learning System (RBLS) for temporal feature learning and Broad Learning Systems (BLSs) for residual fitting.
• Employing a residual-stacked design for enhanced time series fitting performance and Sequential Least Squares Programming (SLSQP) optimization with a validation set to eliminate manual parameter tuning.

Main Results:

• The proposed Stack-RBLS method demonstrates significant improvements in prediction accuracy for PV cluster power.
• The system achieves reduced training times, making it suitable for resource-constrained devices.
• Validation using real-world PV cluster datasets confirms the effectiveness and superiority of the Stack-RBLS approach compared to existing methods.

Conclusions:

• The Stack-RBLS offers an effective and efficient solution for PV cluster power prediction.
• The integrated data cleaning and residual-fitting approach enhances forecasting reliability.
• This method provides a valuable tool for optimizing resource allocation and scheduling in power grids with significant PV penetration.