DATA-DRIVEN POWER SYSTEM OPTIMIZATION

Abstract

Power system operators are increasingly utilizing AC Optimal Power Flow (OPF) algorithms for a variety of applications, including planning and real-time operations, due to a variety of causes. However, because it is non-linear and non-convex, the AC Optimal Power Flow issue is typically difficult to solve in its original form. Aside from a wide variety of approximations and relaxations, recent studies have concentrated on Machine Learning methods, particularly neural networks. In this thesis, we are using the Data-driven approach to predict the power system variables to assess the system security of the power network. This mainly comprises three parts. In the 1st part, we are going to generate the data with a uniform distribution and in the 2nd part, we are going to predict the OPF variables using a deep learning tool. In the second part, we have formed a different neural network with a different number of stages and different inputs to predict the security variables. In the 3rd part, we are going to assess and compare the accuracy of the prediction made with different networks formed in the 2nd part. The data-driven approaches will reduce the complexity and time required for determining the system security variables. We have analyzed the different neural networks with different inputs and stages on IEEE 14 bus system. In the later stages of this project, we have tried the federated model algorithm on the IEEE case 14 bus system to accommodate the distributed learning algorithm. We have added the federated model results in the numerical setup and their comparison with all the other algorithms. With the advancement of the computing power of the measuring devices, we exploit the distributed learning mechanism also.