A hybrid clustering approach for electrical load profiles considering weather conditions based on matrix-tensor decomposition

In modern power systems, information from different sources is metered using various monitoring technologies such as energy consumption from Smart Meters (SM). These data are known as internal variables since they are measured directly from the power system. Additionally, some external variables must be monitored because of their strong and direct influence over the internal variables, for instance, weather variables. Thus, the collected data are a heterogeneous, holistic, and rich collection conformed by different types of correlated variables. Therefore, a first stage for understanding the data is structuring them in sort of logical clusters, thereby facilitating the identification of meaningful patterns and trends. However, an initial challenge arises due to the diverse natures and sampling times of the variables, both internal and external, making it challenging to incorporate them into a model that can study the different variables as a global model. Inspired by that, this paper proposed a hybrid clustering load profiles approach, which explores the fusion of internal and external smart grid data sources based on matrix-tensor decomposition. To demonstrate the effectiveness of the proposed approach, a case study is presented using the load profiles as internal variables and the weather variables as external variables. Both variables were taken from the database of the electrical utility ERCOT in the USA. The approach is compared against the traditional intrinsic approach using only internal variables and against the extrinsic approach using external variables. The results demonstrate improved feature extraction, leading to enhanced clustering performance and the ability to identify profile trends more effectively.