USE OF AN ARMAX MODEL TO PREDICT DISCHARGES AT THE CARILLON POWER-PLANT

This paper discusses the use of an ARMAX model to predict the total discharge at the Carillon power plant on the Ottawa River in the province of Quebec. This prediction is used to manage the power plant's headrace canal level during spring floods to reduce flooding in the area between Carillon and Pointe-Gatineau. The Kalman filter recursive algorithm is adapted to the problem discussed and operates in real time to generate 24-h optimal predictions. Two versions of the forecasting model were considered and classified according to complexity and performance. The first forecasting model is an autoregressive model applied to intermediate inputs and assumes that the input of the three main tributaries is known. It does not account for flood routing or transportation delay between the system's input and output. The Nash global factor is 0.91 and 0.86 for flood peaks only. The ARMAX model takes into account flood routing and transportation time that is assumed to be variable. The Nash global factor is 0.97 and 0.95 for flood peaks only. This latter model was therefore selected to manage and predict the Carillon headrace canal level.