Ultra-short-term solar power forecasting by deep learning and data reconstruction

2025-09-21

AI Summary

To address insufficient spatiotemporal dependency modeling, inadequate meteorological information integration, and substantial long-horizon prediction errors in ultra-short-term solar power forecasting, this paper proposes a CEEMDAN-deep learning collaborative framework. First, historical power output series are decomposed via Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) to isolate high- and low-frequency components, followed by feature reconstruction for enhanced representation. Second, multi-source meteorological data are fused into a dual-path neural network that separately models short-term dynamics and long-term trends. Finally, an interval-based bias suppression mechanism is introduced to improve generalization and prediction stability. Experimental results across diverse weather conditions demonstrate that the proposed method reduces average Mean Absolute Error (MAE) by 12.7%–18.3% compared to state-of-the-art baselines, significantly enhancing forecasting accuracy and reliability—thereby supporting near-real-time dispatch of distributed photovoltaic systems and grid operational stability.

Technology Category

• Machine Learning: Time-Series/Data Streams
• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Web Mining and Content Analysis: Web data integration and cleaning

Abstract

The integration of solar power has been increasing as the green energy transition rolls out. The penetration of solar power challenges the grid stability and energy scheduling, due to its intermittent energy generation. Accurate and near real-time solar power prediction is of critical importance to tolerant and support the permeation of distributed and volatile solar power production in the energy system. In this paper, we propose a deep-learning based ultra-short-term solar power prediction with data reconstruction. We decompose the data for the prediction to facilitate extensive exploration of the spatial and temporal dependencies within the data. Particularly, we reconstruct the data into low- and high-frequency components, using ensemble empirical model decomposition with adaptive noise (CEEMDAN). We integrate meteorological data with those two components, and employ deep-learning models to capture long- and short-term dependencies towards the target prediction period. In this way, we excessively exploit the features in historical data in predicting a ultra-short-term solar power production. Furthermore, as ultra-short-term prediction is vulnerable to local optima, we modify the optimization in our deep-learning training by penalizing long prediction intervals. Numerical experiments with diverse settings demonstrate that, compared to baseline models, the proposed method achieves improved generalization in data reconstruction and higher prediction accuracy for ultra-short-term solar power production.

Problem

• Forecasting ultra-short-term solar power generation using deep learning
• Addressing solar power intermittency challenges for grid stability
• Improving prediction accuracy through data decomposition and reconstruction

Innovation

• Deep learning with data reconstruction for solar forecasting
• Decomposing data into frequency components using CEEMDAN
• Modified optimization penalizing long prediction intervals