Spatio-temporal information based protection scheme for PV integrated microgrid under solar irradiance intermittency using deep convolutional neural network

PV integrated microgrids are characterized by increased intermittency because of the uncertainty in solar irradiance level. In an islanded mode, the irradiance variation significantly affects the current profile both during the healthy and faulty scenario, thereby increasing the risk of maloperation of conventional relays based on threshold settings. A fault-resilient and reliable microgrid demands the development of a protection scheme, which is robust to varying irradiance levels. In this regard, this paper proposes a protection scheme based on Convolutional neural network (ConvNet) approach of deep learning for fault detection/classification, section identification and location during islanded mode. The use of ConvNet allows identifying discriminatory attributes from complex datasets with reduced computational cost. The time-domain voltage and current signals retrieved from the relaying bus post-fault are fed as input to the respective ConvNets trained to perform the intended tasks of classification and regression. The uncertainty in irradiance has been incorporated by modeling irradiance levels over a period using probability distribution function (PDF). The incorporation of spatio-temporal data-based distribution function within the framework of spatial data based ConvNet allows for improved mapping between the multi-domain data and state of the microgrid. The performance of the proposed technique under varying irradiance and fault scenarios has been analyzed through statistical indices and Monte-Carlo simulations. Further, the appropriateness of proposed technique has also been validated for real-time setting using the OPAL-RT digital simulator.