TISM analysis of smart grid consumption data using ensemble learning approach

Abstract Total Interpretive Structural Modeling (TISM) proves highly effective in analyzing the persistence of smart grid consumption patterns over time. This analysis is crucial for various smart grid applications, including demand-supply management, tariff adjustments, grid stability, and more. However, smart grid consumption data typically gathered from customers’ smart meters is often highly dimensional and poses various challenges for traditional clustering algorithms. This paper presents a novel framework that identifies clusters or consumption patterns from the data in each consecutive window, detects changes in these clusters across windows, and subsequently conducts TISM analysis based on the observed changes. Ensemble clustering has two phases. First, we apply the k -means algorithm several times to data within a time window, using different random starting points to get different clustering results. Then, we combine these results into one using a new ensemble method. We also introduce a change detection technique to spot shifts in patterns between consecutive windows. Based on this detection, we use Total Interpretive Structural Modeling (TISM) to analyze the patterns, which helps in addressing smart grid applications. Experiments using synthetic data showed that the proposed algorithm outperformed multiple well-known state-of-the-art methods. Furthermore, it was tested on real-world power consumption data from various factories in Guangdong province, China. The findings support utilities in optimizing operations, improving grid responsiveness, efficiency, and resilience to fluctuations in demand and supply.