Integrated Time Series Analysis, Clustering, and Forecasting for Energy Efficiency Optimization and Tariff Management

This manuscript explores an advanced approach to the analysis of power demand and energy consumption, focusing on anomaly detection and the characterization of consumption patterns in different energy consumer units in the state of Ceará, Brazil. The research employs time series techniques and clustering algorithms to identify atypical consumption profiles and regular patterns based on average monthly consumption and contracted power demand. Through the application of specialized algorithms, consumer groups are identified, revealing trends and seasonality, providing insights into consumption behaviors. Additionally, the study incorporates consumption forecasting methods to estimate the behavior of new consumer units or those with limited historical data, optimizing the definition of contracted demand and promoting energy efficiency. Databases with daily measured values of energy consumption and power demand for 185 energy consumer units in the state of Ceará, Brazil, were analyzed. The research demonstrated that energy costs for the analyzed public units could be reduced by approximately 2.7 million reais. To ensure the reliability of the clustering process, different techniques were applied and compared. The K-Means and Mini-Batch K-Means algorithms were used to segment consumer units based on contracted demand and average monthly consumption. Additionally, the BIRCH algorithm was evaluated in combination with dimensionality reduction methods, such as Principal Component Analysis (PCA) and Autoencoders, to enhance clustering performance. These comparative analyses provided a deeper understanding of the clustering effectiveness and the suitability of each approach for identifying patterns in energy consumption. The findings reinforce the importance of using machine learning techniques to improve tariff selection, anomaly detection, and demand forecasting, contributing to better energy management and efficiency.