| Popis: |
Within thesis an environment implementing machine learning techniques for the analysis and study of electrical network contingencies is proposed. A set of contingencies is defined and studied on a filtered set of simulated possible operating points. Machine learning methods based on proposed in bibliography indices are calculated per simulated operating point and are trained per contingency. They are capable to generate a prediction for an unknown network state about the possible outcome of a contingency. In the scope of thesis a novel computing environment has been developed for the analysis and risk assessment of a predefined set of contingencies comprising line outage combinations. In parallel, a general methodology for contingency analysis is proposed that is in particular implemented on the electrical network of the island of Crete. The environment of study is electrical network independent and trains machine learning algorithms based on previous knowledge. It aims at the estimation of the application result of a contingency to a certain network operating state. Previous knowledge is acquired through simulation, that is the application of a contingency to different filtered operating states. Thus, the risk estimation of a contingency is achieved, bypassing the computer intensive procedure of power flow analysis. |
