Real-time very short-term load prediction for power-system automatic generation control

A fundamental objective of a power-system operating and control scheme is t o maintain a match between the system's overall real-power load and generat ion. The automatic generation control (AGC) loop addresses this objective b y using system load and electrical frequency samples to periodically update the set-point power for key "swing" generators with a control sample rate ranging from 1 to 10 min. To improve performance, emerging AGC strategies e mploy a look-ahead control algorithm that requires real-time estimates of t he system's future load out to several samples using a one to ten minute sa mple period (a total typical horizon of 30 to 120 min). We term this very s hort-term load prediction. This paper describes a strategy for developing a very short-term load predictor using slow and fast Kalman estimators and a n hourly forecaster. The Kalman model parameters are determined by matching the frequency response of the estimator to the load residuals, The design strategy is applied to the system operated by the Bonneville Power Administ ration and specific performance and sensitivity studies are presented.