Pattern recognition algorithms for electricity load curve analysis of buildings

• We propose a methodology for the investigation of buildings’ electrical consumption through clustering techniques.
• For this purpose the performance of different clustering algorithms are evaluated.
• Detailed parameter optimization guidelines are proposed for the SOM algorithm.
• We use as a test case a set of buildings of different utilization in a university campus.
• The methodology is a useful tool for the load profiling analysis.

Buildings consume 40% of the total primary energy and 30% of the annual electricity, contributing significantly to greenhouse gas emissions. Naturally, therefore, building energy efficiency and notions like the nearly zero energy buildings are continuously gaining importance and popularity as means to reduce carbon emissions and the strong dependence on fossil fuels. A step towards this direction is the incorporation of smart grid technologies, mainly through the widespread of automatic meter reading and smart meters. This enables automatic collection of in depth information of the customer's behavior along with the building's performance and, thus, introduces new opportunities for energy saving and efficient management. However, the recorded amassing ream of data requires efficient processing and interpretation, so as to provide for meaningful information. In order to tackle this problem, this paper proposes a comprehensive methodology for the investigation of the electricity behavior of buildings, using clustering techniques. Utilizing a university campus as a case study, the proposed methodology is applied to the load curves of different buildings leading to the determination of an optimum clustering procedure. The methodology may be generalized for any type of building.