Supervised Classification of Distributed Data Streams for Smart Grids

"The electricity system inherited from the 19$^{th}$ and 20$^{th}$. centuries has been a reliable but centralized system. With the. spreading of local, distributed and intermittent renewable energy. resources, top-down central control of the grid no longer meets. modern requirements. For these reasons, the power grid has been. equipped with smart meters integrating bi-directional. communications, advanced power measurement and management. capabilities. Smart meters make it possible to remotely turn. power on or off to a customer, read usage information, detect a. service outage and the unauthorized use of electricity. To fully. exploit their capabilities, we foresee the usage of distributed. supervised classification algorithms. By gathering data available. from meters and other sensors, such algorithms can create local. classification models for attack detection, online monitoring,. privacy preservation, workload balancing, prediction of energy. demand and incoming faults. In this paper we present a. decentralized distributed classification algorithm based on. proximal support vector machines. The method uses partial. knowledge, in form of data streams, to build its local model on. each meter. We demonstrate the performance of the proposed scheme. on synthetic datasets.. "