A Confidence-Aware Machine Learning Framework for Dynamic Security Assessment
ABSTARCT :
Dynamic Security Assessment for the future power system is expected to be increasingly complicated with the higher level penetration of renewable energy sources and the widespread deployment of power electronic devices, which drive new dynamic phenomena. As a result, the increasing complexity and the severe computational bottleneck in real time operation encourages researchers to exploit machine learning to extract offline security rules for the online assessment. However, traditional machine learning methods lack in providing information on the confidence of their corresponding predictions.Understanding better the confidence of the prediction made is of key importance for Transmission System Operators (TSOs) to use and rely on these machine learning methods. Specifically, from the perspective of topological changes, it is often unclear whether the machine learning model can still be used.Hence, being aware of the confidence of the prediction supports the transition to use machine learning in real time operation.In this paper, we propose a novel Conditional Bayesian Deep Auto-encoder (CBDAC) based security assessment framework to compute a confidence metric of the prediction. This informs not only the operator to judge whether the prediction can be trusted, but it also allows for judging whether the model needs updating.
EXISTING SYSTEM :
? In this study, we start showing how existing approaches fail to detect alarms in some cases when not accounting for class imbalances.
? We proceed by showing that both proposed approaches are able to drive system operation much closer to the global optimum than existing approaches, while also abiding to the user-defined risk tolerance level.
? Moreover, we show that the proposed risk-averse approach is capable of identifying cost-effective control actions under a large range of unseen operation conditions.
DISADVANTAGE :
? Training of a neural network involves solving a nonlinear optimization problem where the goal is to minimize an error function depending on the type of network chosen.
? Different gradient techniques are available online which are chosen during the training phase to reduce the problems that arise due to prsence of local minima.
? In this study, the gradient descent method has been used in order to solve the optimization problem.
PROPOSED SYSTEM :
• This work proposes to use machine learning for probabilistic security assessment including dynamic security.
• A probabilistic framework is proposed that combines the strength of conventional security assessment methods with the strength of machine learning for realtime operation.
• This overcomes the technical challenge of the computational complexity of assessing dynamic security in realtime operation, typically resulting in probabilistic security assessments mainly focusing on static security and not on dynamic security
ADVANTAGE :
• The proposed scheme of using seasonal load has proved that under the same set of contingencies for a different season, the number of violations differ. Therefore, there is a need to include seasonal variability while doing DSA.
• The performance of RF was found to be the best among all algorithms when measurement errors were not considered in the study.
• Substantial degradation in performance of RF was observed when measurement errors were introduced into the system.
• The performance of SVM was better than the other algorithms considered in the study when measurement errors were fed into the testing models.
|