Convolutional neural network with genetic algorithm for predicting energy consumption in public buildings

Due to their capacity to improve energy consumption performance, intelligent applications have recently assumed a pivotal position in the energy management of public buildings. Keeping these buildings ' energy consumption under control is a significant issue because of their irregular energy consumption patterns and the lack of design criteria for energy efficiency and sustainability solutions. As a result, it is essential to analyze public building energy consumption patterns and predict future energy demands. Evidence like this highlights the need to identify and categorize energy use trends in commercial and institutional dwellings. This research aims to identify the most effective intelligent method for categorizing and predicting the energy consumption levels of buildings, with a specific study case of public buildings and, ultimately, to identify the scientific rules (If-Then rules) that will aid decision-makers in establishing the proper energy consumption level in each building. The goals of this research were accomplished by employing two intelligent computing models, the Elbow technique and the Davis and Boulden approach, to count the number of clusters of energy consumption patterns. We addressed clustering with K-means and a genetic algorithm. The genetic algorithm was utilized to find the best centroid points for each cluster, allowing the fitting model to function better. Determining which buildings consumed the most energy has been easier thanks to extracting If-Then rules from cluster analysis. Convolutional neural networks (CNNs) and CNNs combined with a Genetic Algorithm (GA) were also employed as intelligent models for energy consumption prediction. At this point, we utilized a genetic algorithm to fine-tune some of CNN's settings. CNN with genetic algorithm outperforms the CNN model regarding the accuracy and standard error metrics. Using a genetic algorithm, CNN achieves a 99.01% accuracy on the training dataset and a 97.74% accuracy on the validation dataset, with accuracy and an error of 0.02 and 0.09, respectively. CNN achieves a 98.03% accuracy, 0.05 standard error on the training dataset, 94.91% accuracy, and 0.26 standard error on the validation dataset. Our research results are useful for policymakers in the energy sector because they allow them to make informed decisions about energy supply and demand for public buildings.