Machine learning based diagnosis support for ShipBoard Power Systems controls

Abstract

In this paper, a machine learning based decision support system for a naval shipboard power management system is proposed considering contingencies and load priority. A probabilistic model based Bayes' classifier is implemented to classify the current operation state of the ShipBoard Power System (SPS), depending upon the power system readiness for critical contingencies. Real power, reactive power, and generator status are taken as input features for the algorithm. Loss of vital/non-vital load is calculated by solving optimal power flow (OPF) to help build the knowledge base. Training data are updated online to increase the accuracy of the proposed approach. The characterization of the operation states helps the shipboard power management system to take the appropriate control action. Initial results from tests are presented and the outcomes from the particular techniques are discussed. Moreover, we also present RTDS based experimental framework towards the ongoing research on overall management system including the diagnosis support. Naïve Bayes' approach has classified the system states with 97.67% accuracy to new instances. Preliminary results show the computation time of this approach is in the order of 25 ms.