Answers are grounded in this paper and cite their sources.
Analysis reveals improved forecasting accuracy in photovoltaic power by integrating CNN-LSTM with GMM error correction.
Short-term photovoltaic power forecasting is essential for real-time power system scheduling, yet current forecasting accuracy remains insufficient for practical operation. To address this issue, this study analyzes the statistical characteristics of forecasting errors and proposes an integrated framework comprising primary forecasting, error distribution modeling, error-level classification, and scenario-based weighted correction. Experimental results demonstrate that stratifying forecasting errors by weather type and power level effectively captures distinct residual patterns under different operating scenarios. Based on this, applying error-level classification and scenario-weighted correction significantly improves prediction accuracy. Compared with the uncorrected baseline, the proposed method achieves notable improvements across multiple representative scenarios and test sets. In a three-day independent test, the coefficient of determination (R2) increased from 0.949 to 0.980, while in a one-month continuous test, R2 improved from 0.9755 to 0.9836 (improved by about 0.83%), accompanied by substantial reductions in root mean square error and mean absolute error. These results confirm the method's robustness, generalization capability, and strong potential for practical deployment.
No takes yet. Share an insight, caveat, or question.
Gao et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: