Improving Performance of Fitting Energy-Saving Measures of Buildings Using Artificial Neural Networks

Abstract

The article is devoted to increasing energy saving and energy efficiency of public sector institutions by automation of the process of energy-saving measures fitting. A classifier based on an artificial neural network is proposed as a technology for selecting measures. The base numeric dataset was supplemented by categorical data on buildings and structures. The urgency of the work is justified by the need to develop solutions aimed at introducing measures to ensure energy saving and energy efficiency in the public sector. The structure and operating principle of a module of auto-selecting energy saving measures as a part of the Energy Resources Management System (SUER) are described. The work studies the quality of fitting energy-saving measures with or without using data on categorical features. An analysis of the selected categorical features, as well as a comparative analysis of their encoding methods, is carried out. An accuracy evaluation method of the classifier in the context of the work is proposed. A series of experiments were conducted by enumerating combinations of numeric and categorical data converting methods, in order to determine the significance of categorical features in general, as well as to determine the most effective combination of encoding methods. A comparative analysis of the results of the experiments was performed and the most successful model was determined, with an additionnal assessment of the quality of the model was made based on the metrics of precision, recall and F1-score. Conclusions were made on the advisability of supplementing the original dataset with categorical data to improve the performance of the system.