Neural network approach for improving AGC control performance

As part of the unified Smart Automatic Generation Control (SAGC) solution framework, the Unit Response and Unit Tuning functional block plays an important role in achieving desired control performance. While the other functional blocks such as Very Short Term Load Prediction (VSTLP), Predictive Economic Dispatch (PED) and Predictive CPS Control, work together to compute the generating unit´s economic basepoints and desired regulation amounts, if the AGC units do not follow respective control commands in a satisfactory manner, the control area´s control performance can not be controlled effectively. In the smart grid environment where an ever increasing amount of intermittent, renewable energy presents a great challenge to the reliable operation of the power system, the regular generating units must be tuned to respond to the unit´s respective desired generation as good as it can be so as to meet the system condition changes include load changes and weather condition changes that significantly affect the total power output of renewable generation resources. In addition, the ACE impact on regulation should be assessed, predicted and incorporated in the PED. This paper proposes several innovative schemes to make use of neural networks to improve the control of generating units for AGC and address the ACE prediction.