Estimating the Stability of Smart Grids Using Optimised Artificial Neural Network

The smart grid is a revolutionary and upsurging methodology for power supply. Smart grid has many advantages like reduced peak demand, inclusion of different energy sources, increase in the number of power suppliers, increased overall security and real time price prediction thus helping to optimize the power usage. Due to the inclusion of different renewable sources as prosumer (producer and consumer), a centralized system is not sufficient enough to dynamically predict the stability of the smart grid systems. In a centralized system, there is one directional flow of electricity and information. The local nodes are not autonomous and do not have a bi-directional flow of information, hence the prediction of price is time taking, fault detection and correction are also not fast. This paper considers decentralized system to predict the stability of the smart grid power supply which is dependent on the frequency of local nodes. The smart grid is said to be stable if the power generation matches the power demand and also there is a reserve to meet the power outage if it happens at any point of time. The paper considers an ANN model based on deep learning techniques and evaluates various factors to optimize its precision, such as the number of hidden layers, the number of nodes in each hidden layer, the appropriate optimizer and the right activation function. We have concluded the relationship between the predictive features and hidden layers, the use of ‘relu’, ‘sigmoid’ and ADAM as the optimized parameters for the ANN model for smart grid stability predictions.