A WEB BASED SOFTWARE SYSTEM FOR DATABASE GENERATION FOR ONLINE DYNAMIC SECURITY ASSESSMENT STUDIES (ML4DSA) Janath Geeganage, U. D. Annakkage * Department of Electrical and Computer Engineering University of Manitoba B.A. Archer, Tony Weekes Manitoba Hydro ABSTRACT This paper presents a software system that generates a database for power system dynamic security assessment. The gener- ated database is intended to be used in machine learning tech- niques. The development of algorithms to generate data is a very time consuming task. This software tool is aimed at facili- tating faster generation of the appropriate database. Further, the system allows the user to plug-in the case specific limit checks and algorithms for system specific corrective actions depending on the type of study. The proposed system automates the Power System Simu- lator for Engineering (PSSE) which is an industry standard software used in many electrical power utilities. The pro- posed software system, ML4DSA, is based on Python which is available in the public domain with plenty of supporting communities and powerful libraries. These features enable the user to develop algorithms for system specific corrective actions. The web interface facilitates access to the authen- ticated users of PSSE over the web, therefore, requires no additional software installed on the client computer. ML4DSA is successfully tested on the 39 Bus New Eng- land test system and the Midwest Reliability Organization (MRO) system which has over 50,000 buses. Index Terms— Dynamic security assessment, Machine learning, Software tools. 1. INTRODUCTION The accurate identification of transient stability margins of a power system is essential for secure operation of power sys- tems. The said assessment requires solving a large number of differential equations and non-linear algebraic equations for a set of credible contingencies. Computing power avail- able in today’s computers is insufficient for real-time simu- lations of the complete transient stability models in the time domain to compute accurate stability margins. In this situ- ation, the application of Machine Learning techniques have * Authors wish to thank Manitoba Hydro and MITACS accelerate program for funding. shown promising results[1, 2]. The training and test data gen- eration is a main task involved in the application of supervised learning techniques. Due to its vast geographic area and large number of planned and unplanned events, it is unlikely to successfully train a single learned network that is applicable at all times. Train- ing of several learned networks initially or at regular intervals during the operation would be required. The aim of the de- veloped system is to produce a learned network appropriate for a given period of time considering a snapshot of the cur- rent state of the power system and the capability of forecast- ing. This raises the need to have automation algorithms for database generation for machine learning. Development of these algorithms occupy a considerable portion of the entire study. Therefore, having a ready-made and a customizable tool is an advantage. The developed software is a general system that can be used for an existing power system model. However, the uniqueness of the software is to facilitate the customization of the power system specific constraints and corrective actions. Customization is essential for data gener- ation for successful application of machine learning[3]. As explained in section 2.2.3, ML4DSA enables the user to in- corporate the expertise of the operating engineers in terms of rules and parameters. Further, the software system automates th PSSE which is an industry standard power systems analy- sis tool. The availability of PSSE system models enables the easy adoption of the developed software. This paper describes the structure, system design of key features, current status of development, and test results ob- tained using the ML4DSA. 2. SYSTEM DESIGN AND KEY FEATURES 2.1. System Structure Python algorithms of ML4DSA execute on the Web2Py frame- work [4]. As shown in the figure 1, ML4DSA interacts with PSSE using the Psspy module[5] available in PSSE. The au- tomation process creates two sets of files using PSSE: steady state operating points (SAV files) and time series data of dy- namic simulations (OUT files). Time series data is input to ML4DSA using the Dyntools[5] module available in PSSE. 2013 26th IEEE Canadian Conference Of Electrical And Computer Engineering (CCECE) 978-1-4799-0033-6/13/$31.00 ©2013 IEEE