Construction of a task-oriented database for electricity consumption dynamics forecasting

Abstract

Modern software expert systems have the potential to optimize decision-making proc-esses in a wide array of industries, including energy, healthcare, and manufacturing. Never-theless, many machine learning–oriented approaches underestimate the significance of a ro-bust database architecture specifically tailored to each domain. This study aims to develop and validate an algorithmic approach to building a task-oriented (specialized) database structure for an expert system dedicated to forecasting electric power consumption. Real con-sumption measurements are paired with contextual data such as weather patterns, then trans-formed through systematic validation, cleaning, and encoding. Outlier detection and imputa-tion steps further reduce distortions stemming from incomplete or anomalous inputs, while an iterative feature selection process refines the pool of variables most relevant to predictive models. Unlike generic pipelines, the proposed approach incorporates a schema evolution stage, ensuring that newly identified indicators can be integrated without undermining exist-ing data integrity. Experimental application on a practical dataset combining technical power usage and external factors demonstrates improved resilience of forecasting results. By outlin-ing each stage of data preparation and emphasizing schema adaptability, the proposed meth-od supports high-precision predictions even under dynamically changing conditions. By con-tinually revisiting and adjusting the database, the system can adapt to evolving operational conditions, thus enhancing the accuracy of computational modules responsible for projecting consumption patterns. The findings underscore the importance of a specialized, domain-specific data foundation in realizing the full potential of expert systems.