A Hybrid Source–Load Power Forecasting Model for Distribution Networks with High Penetration of Renewable Energy

INTRODUCTION: High renewable energy penetration introduces significant uncertainties in distribution networks, posing challenges for source-load power forecasting and voltage management.

OBJECTIVES: This study aims to enhance forecasting accuracy and address voltage control difficulties caused by distributed generation and load fluctuations using a novel integrated framework.

METHODS: A hybrid model combining Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), Temporal Convolutional Network (TCN), and DLinear is proposed. First, CEEMDAN decomposes source-load and meteorological data into stable Intrinsic Mode Functions (IMFs) to reduce non-stationarity. Subsequently, TCN captures short-term dependencies, while DLinear extracts multi-scale features by decomposing IMFs into trend and residual components. The final forecast is derived by aggregating the reconstructed subsequence predictions.

RESULTS: Extensive simulations validate that the proposed method significantly outperforms conventional benchmarks, such as BiGRU and TCN-BiGRU. It achieves higher forecasting precision and effectively mitigates the adverse effects of data uncertainty.

CONCLUSION: The proposed CEEMDAN-TCN-DLinear framework demonstrates consistent superiority in handling complex data patterns, offering a robust solution for distribution network voltage control under high renewable penetration scenarios.