Pattern recognition for one shot control in power systems

Indiana University, Bloomington IN

Investigators

Abstract

This work will develop new ways of using pattern recognition for one shot controls in power systems. One shot controls include disconnection of generation and load and fast power changes on HVDC transmission lines. The control actions used in this work generally reduce inter-area power flows which tends to improve transient and steady state stability. The reduction of power flows is also associated with the prevention of relays from tripping on overload. Many blackouts result from cascading outages involving generators and transmission lines tripping off-line due to protective relaying. The proposed controls promise to make these failure modes less likely to occur. In addition to one shot control at the transmission level, the investigators will also develop algorithms to enable consumer appliances such as AC units to detect the loss of a large generator nearby and turn off long enough for the Independent System Operator (ISO) to take corrective action at the transmission level. The proposed methods are inexpensive and easy to incorporate into smart devices because only local measurements are used. No remote communication is required. Utilities could provide rebates or rate discounts to motivate customers to purchase appliances that use the proposed control schemes. All the methods developed in this work from one shot controls at the transmission level to smart appliances that operate autonomously will contribute to a more resilient and self-healing electric grid. Many state variables in turbine governors, voltage regulators and other digital controllers are subject to maximum and minimum limits. The differential algebraic equations used for analytical solution of power system dynamics are only valid if controller variables do not hit their limiting values. This is one reason why accurate prediction of power system behavior requires detailed time domain simulation. Deriving control strategies from a large number of simulations is a big data problem that requires pattern recognition. The proposed controls are response based which means, for example, they do not receive information about the status of particular breakers. Current industry practices focus on event based risk assessment (e.g. Special Protection Scheme) assuming the operators are fully aware of what has happened in the grid. It is challenging to develop response based controls that are actuated quickly enough to stabilize some events. This research will use event detection and pattern recognition prediction to actuate response based control earlier so that the stabilizing effect is greater. Another area of this research will contribute to self healing for a smarter grid. The underlying scientific principle is that local modes of oscillation have higher frequencies than inter-area modes. The derivative engineering contribution of the work will show how a Fourier calculation can help detect nearby loss of generation events using only local frequency measurements. The proposed methods are novel compared to existing proposals in the literature for under frequency load shedding. Utilities are not currently benefitting from autonomous load shedding at the residential level as proposed here.