Novel Data-Driven Building Electricity Consumption Prediction Method for Public Buildings Based on Historical Data and K-Nearest Neighbor-Matrix Algorithm

The energy crisis is one of the most serious crises in modern society. In this context, building energy prediction is more and more important, as it can guide energy management and efficiency retrofitting, especially for public buildings that often have more energy-using equipment and larger numbers of users. However, existing physical modeling methods and data-driven methods are difficult to predict suitably and accurately. Therefore, the KNN-Matrix algorithm method is developed. It introduces the concept of fuzzy quantification levels for electricity intensity, and uses historical data (after screening and processing) as the prediction data. The validation results on nine buildings with different functions and climate zones show that the average of hourly prediction error is 13.1%. This algorithm relies only on historical data without simulation software and unpredictable parameters as the initial conditions. Moreover, this algorithm can better address noise data, allowing for better accuracy and less data volume.