Robust Real-Time Load Profile Encoding and Classification Framework for Efficient Power Systems Operation

Neatly represented and properly classified load profiles are fundamental to many control optimization techniques of modern power systems, especially in a distribution area. This paper presents a novel load profile management software framework for boosting the efficiency of power systems operation. The proposed framework encodes and classifies load profiles in real-time. Imperfections as well as time-shifts in the input (measured power consumption levels) are tolerated by the suggested system, thus always providing accurate, fast and reliable output. The framework's fully component based structure allows easy customizations of the encoding as well as the classification engines. The default encoding engine is based on an artificial neural network, a variant known as a deep learning auto-encoder comprised from stacked sparse auto-encoders. The default classifier engine is based on an implementation of a locality sensitive hashing algorithm. The developed methodology was tested on the real case of a set of anonymous customers supplied by a power distribution company. The paper also contains an elaboration about the experiences gained during the design, implementation and testing phase of this system as well as a detailed engineering use case of the framework's applicability.