New approaches to the AGC nonconforming load problem

Northern Indiana Public Service Company (NIPSCO) has many nonconforming loads that are a challenge for their automatic generation control (AGC) software. Their area control error (ACE) values often have peak-to-peak swings of 150+ MW in a few minutes, and the ACE signal is noisy due to numerous asynchronous nonconforming industrial loads. These loads include over 20% of the base US steel production. A number of investigations targeted at assessing the actual impact of highly varying industrial loads on system reliability have been undertaken. This paper reports results for one of these investigations. The current AGC program attempts to account for short duration excursions which leads to increased generation over and under shoots and possible increased wear and tear on generating equipment. It has been shown that conventional control strategies of this nature can degrade rather than improve overall control of ACE. As a consequence, an effort has been undertaken to develop algorithms that will discriminate between noncontrollable short-term excursions and controllable long-term excursions. This paper presents two techniques that address the nonconforming load problem for AGC. One technique used a neural network algorithm for pattern recognition of controllable signals, and the other technique is based on the detection of a controllable signal in the presence of a noisy random load using a random signal probability model. Both algorithms were tested with actual field load data via a dispatcher training simulator that utilized a generic system model