Privacy reinforcement learning for faults detection in the smart grid

Recent anticipated advancements in ad hoc Wireless Mesh Networks (WMN) have made them strong natural candidates for Smart Grid's Neighborhood Area Network (NAN) and the ongoing work on Advanced Metering Infrastructure (AMI). Fault detection in these types of energy systems has recently shown lots of interest in the data science community, where anomalous behavior from energy platforms is identified. This paper develops a new framework based on privacy reinforcement learning to accurately identify anomalous patterns in a distributed and heterogeneous energy environment. The local outlier factor is first performed to derive the local simple anomalous patterns in each site of the distributed energy platform. A reinforcement privacy learning is then established using blockchain technology to merge the local anomalous patterns into global complex anomalous patterns. Besides, different optimization strategies are suggested to improve the whole outlier detection process. To demonstrate the applicability of the proposed framework, intensive experiments have been carried out on well-known CASAS (Center of Advanced Studies in Adaptive Systems) platform. Our results show that our proposed framework outperforms the baseline fault detection solutions. Recent anticipated advancements in ad hoc Wireless Mesh Networks (WMN) have made them strong natural candidates for Smart Grid's Neighborhood Area Network (NAN) and the ongoing work on Advanced Metering Infrastructure (AMI). Fault detection in these types of energy systems has recently shown lots of interest in the data science community, where anomalous behavior from energy platforms is identified. This paper develops a new framework based on privacy reinforcement learning to accurately identify anomalous patterns in a distributed and heterogeneous energy environment. The local outlier factor is first performed to derive the local simple anomalous patterns in each site of the distributed energy platform. A reinforcement privacy learning is then established using blockchain technology to merge the local anomalous patterns into global complex anomalous patterns. Besides, different optimization strategies are suggested to improve the whole outlier detection process. To demonstrate the applicability of the proposed framework, intensive experiments have been carried out on well-known CASAS (Center of Advanced Studies in Adaptive Systems) platform. Our results show that our proposed framework outperforms the baseline fault detection solutions.