Smart Grid Data Classification with Deep Networks

Abstract:
In Smart Metering Networks it is an important task to process the measured data with high precision and efficiency. The analysis of the recorded measurements helps to improve the service provided by the networks as well as establishes a firm basis for optimization of the operation. The automated classification of different types of consumers in Smart Grids is one of the critical tasks. Power consuming users, appliances, offices having different power consumption patterns should be treated different ways regarding the capacity planning and distribution, as well as various conditions, also have to be provided to them. The application of deep neural networks achieved remarkable results in case of two-dimensional images and image sequences as well. However, the capabilities of the deep networks in case of other types of data or time series are not investigated extensively. In this paper, we present new results on the classification of power consumers based on their consumption measurements using deep neural networks, with different architectures and data. We demonstrate that the deep networks are capable of solving the classification problem of power consumer at a high level of performance. Current results are obtained on a publicly available measurement database, using several weeks long time series.