Research on Artificial Intelligence-driven Resident Demand Forecasting Model for Digital Community Governance

Abstract

This study starts from two aspects, data-driven and demand insights, to realize accurate forecasting of residential demand. Firstly, the raw data are sculpted and feature extraction of load fluctuation is performed using mutual information and principal component analysis. And an intelligent prediction model Bi-ITCN-Att, a bidirectional improved temporal convolutional network incorporating a self-attention mechanism, is built.Innovatively, the article introduces the Kano model as a demand decoder to analyze the motivation of residents' electricity consumption behavior from the perspective of humanistic demand. The model performs well in predicting load in a neighborhood, with a root mean square error of 0.027 and a MAPE of only 1.86%, and the prediction accuracy is significantly better than that of the original TCN and Bi-ITCN models. The analysis of 369 valid questionnaires reveals that 54.47% of the residents regard circuit fault warning (ESR1) and other as the bottom line demand M that must be guaranteed, while the laws such as AI energy conservation advice (ECF2) are regarded as glamorous attributes A. The quadrant diagram of the demand strategy drawn through the analysis of Better-Worse coefficients further proclaims that the key to improving the satisfaction lies in the implementation of the electricity price services such as reminders, which has a satisfaction coefficient of 0.549, while the defense of emergency security must be maintained. The study provides a set of decision support system for digital community governance that is not only accurate in data, but also handy in human needs, thus promoting the community energy service from passive response to a new stage of proactive and humanized intelligent governance.