Whale optimization-based clustering for buildings’ energy consumption profiles

The smart grids require advanced building demand-side management solutions to balance variable renewable production. Energy consumption profile segmentation can increase the efficiency of the processes, enabling grid operators to provide more personalized requests to different customer needs and behaviors increasing engagement and participant satisfaction. However, the complexity of clustering energy profiles is amplified by the diversity of consumer behavior and the need to adapt segmentation over time as consumption patterns evolve. Establishing the number of consumer energy profile clusters in advance is challenging, as most clustering algorithms are sensitive to initial parameter settings, which can affect their performance. In this paper, we propose a new clustering method based on the Whale Optimization Algorithm (WOA), which addresses the problem of dynamicity and variability of energy data. The whale individual is defined as a set of active centroids described by a compressive set of features extracted based on peak and valley periods focused on magnitude, variation, and efficiency. To promote compact and well-separated clusters of energy profiles the Calinski-Harabasz index was used as a fitness function. Population initialization is performed using K-Means+ + algorithm which ensures a diversified initial distribution of solutions in the solution space. The evolution of solutions is ensured through the mechanisms of the WOA algorithm, which allow the gradual updating of candidate solutions to accelerate convergence and produces high-quality solutions. The efficiency of the method was evaluated on a real dataset of daily energy consumption of buildings on a university campus. The experimental results show that the WOA method outperforms the selected state-of-the-art methods for comparison. WOA obtained the lowest Davies–Bouldin index (0.527), the highest Silhouette score (0.484) and the lowest Ball–Hall coefficient (1.851), indicating superior segmentation in terms of both internal cohesion and inter-cluster separability.