Learning a Robust Fuzzy Cognitive Map Based on Bubble Entropy Fusion With SCAD Regularization for Solar Power Generation

Abstract

Accurate and reliable solar photovoltaic (PV) power forecasting are crucial for cost-effective resource planning and stable operation of smart grids. However, current methods are affected by the intermittent, non-stationary and stochastic nature of solar energy and thus cannot satisfy the requirement of high-precision forecasting. To this end, we propose a fuzzy cognitive map (FCM) forecasting method based on bubble entropy and smoothly clipped absolute deviation (SCAD) regularization, called BesFCM. This method first utilizes bubble entropy to fuse two mode decomposition methods to improve the representation of PV data to capture effective features with significant stability and discriminative ability, then employs a FCM with a combination of fuzzy logic, neural networks, and expert systems to model solar PV power generation, and finally develops a high order FCM learning method based on SCAD regularization to alleviate the overfitting problem, enhancing the robustness and generalization ability of forecasting. Experimental results demonstrate that the BesFCM achieves the best overall performance on PV power datasets from multiple sampling intervals in multiple regions of Belgium compared to multiple state-of-the-art baselines, validating the effectiveness for solar power generation forecasting, providing support and reference for improving the quality of smart grid dispatch and reducing spare capacity reserves.