Intelligent Computing Techniques for Clustering and Predicting Energy Consumption in Public Buildings

Because intelligent applications may improve the performance of energy consumption, they have recently played a significant role in the energy management of public buildings. Due to their unexpected energy consumption characteristics and the lack of design criteria for sustainable and energy-efficient solutions, these buildings constitute a significant challenge in terms of energy management. Thus, it becomes imperative to investigate the energy usage patterns in public buildings. This highlights how important it is to comprehend and group these buildings' energy usage habits. To assist decision-makers in determining the energy consumption level of each building, this study aims to identify the most intelligent technique for clustering energy consumption of public buildings into levels (e.g., low, medium, and high) and identify critical factors that influence energy consumption. Lastly, predicting energy consumption levels based on clustering model findings utilizing modern intelligence approaches like deep learning techniques. To achieve the objectives of this study, we proposed three main steps as follows: First, we put forth two fundamental models: text mining and the PRISMA approach. Using the PRISMA approach, we examined 822 publications between 2013 and 2020 and narrowed the analysis to 106 that satisfied specific criteria, such as having experiments and passing the title and abstract screening stages. The most popular terms and their relationships in the energy and intelligent computing domains were discovered using a text-mining process and a bibliometric map tool (VOS viewer). This allowed researchers to identify the most critical factors influencing building energy consumption and the most effective intelligent computing techniques for grouping and forecasting energy consumption of various building types, particularly public buildings. Second, two intelligent models, Self-Organizing Map (SOM) and Batch-SOM based on Principal Component Analysis (PCA), were used to determine the number of clusters of energy consumption patterns. We proposed correlation coefficient analysis as a means of identifying critical factors that influence the energy consumption of public buildings. SOM performs better in terms of quantization error than batch-SOM. SOM and Batch-SOM have quantization errors of 8.97 and 9.24, respectively. Two other methods, the Davis-Bouldin method and the Elbow method, were also utilized to calculate the number of clusters. Each building's cluster labels, or levels, were predicted using a genetic algorithm and K-means analysis. In this part, the optimal centroid points in each cluster were identified using a genetic algorithm. If-Then rules have been retrieved by examining cluster levels, so decisionmakers must locate the buildings that use the most energy. Third, Convolutional neural networks (CNNs) and CNNs paired with a Genetic Algorithm (GA) were two intelligent models we suggested using to estimate energy consumption levels. At this stage, we adjusted a few of CNN's settings using a genetic algorithm. The CNN model is beaten by CNN with a genetic algorithm in terms of accuracy and standard error metrics. With accuracy and error of 0.02 and 0.09, respectively, CNN uses a genetic algorithm to achieve 99.01% accuracy on the training dataset and 97.74% accuracy on the validation dataset. On the training dataset, CNN obtains 98.03% accuracy, with 0.05 standard error; on the validation dataset, it achieves 94.91% accuracy and 0.26 standard error. Finally, this study aids in rationalizing energy usage by building occupants during peak energy consumption periods. It facilitates the replacement of energy suppliers for those buildings by decisionmakers in the energy sector. Lastly, we aim to predict energy consumption levels based on clustering model findings utilizing modern intelligence approaches like deep learning techniques.