Short-Term Load Forecasting Method of TCN-BiGRU Based on RTH Optimization

Addressing the intricacies and volatility of short-term load series data, traditional forecasting approaches frequently yield substantial inaccuracies. This paper presents a refined TCN-BiGRU methodology for short-term load forecasting, fortified by the RTH optimization technique. Initially, the MIC feature correlation analysis is employed to meticulously quantify the meteorological and other factors' influence on load, pinpointing those with robust correlations. These pertinent features are then fed into both TCN and BiGRU models for intensive training and prediction. Subsequently, the RTH algorithm meticulously tunes the hyperparameters of both models, augmenting prediction accuracy. Ultimately, a novel error reciprocal approach is leveraged to seamlessly blend the TCN and BiGRU predictions, yielding a final load forecast. When applied to real-world power load data from a designated region, the proposed method exhibits a marked reduction in error rates compared to standalone models and other benchmarks, underscoring its proficiency in delivering high-fidelity short-term load forecasting.