Dual-Driven Electricity Package Recommendation via Cognitive-Noise Filtration and Consensus Evolution

The growing variety of customized tariffs in deregulated electricity retail markets increases customers’ decision burden under information asymmetry. Existing recommendation methods generally characterize either objective load patterns or subjective preferences, but rarely account for load-risk heterogeneity and uncertain evaluations within a unified framework. This study proposes a dual-driven electricity package recommendation framework that combines cognitive-noise filtration with dynamic consensus evolution. A risk-driven two-stage affinity propagation (RD-TAP) algorithm first separates customers according to both load-shape similarity and fluctuation risk. Interval-valued intuitionistic fuzzy evaluations are then calibrated using individual-to-subgroup consensus deviations, after which a multi-threshold weighted-fusion mechanism iteratively improves subgroup consensus while retaining moderate individual differences. In a case study based on real-world load data, RD-TAP maintains a clustering quality index above 0.94 for the tested sample sizes, and the complete recommendation framework yields mean RMSE values ranging from 0.1020 to 0.1346 across the four load patterns. These results indicate that jointly modeling objective load risk and subjective evaluation uncertainty can improve the stability and interpretability of electricity package recommendations.