APPLICATION OF DDDP FOR OPTIMAL DAILY OPERATION OF THE NIAGARA GENERATING SYSTEM
by
H.A. North¹, K. Lacivita¹, and I. Corbu¹

ABSTRACT

This paper describes the application of Discrete Differential Dynamic Programming (DDDP) to determine the optimal operation of Ontario Hydro's Niagara Generating System, on an hourly basis, over a time horizon of one day. The system consists of one partially controlled reservoir called Grass Island Pool (GIP) and one fully controlled Pump Generating Station (PGS) reservoir. There are four conventional hydraulic stations and one Pump Generating Station in this system, having a total installed capacity of approximately 2,100 MW. The objective of this DDDP optimization model is to maximize the worth of energy produced by Ontario Hydro's Niagara Generating System. Opportunities to further increase production by transferring water from some of Ontario Hydro's older, less efficient stations to the New York Power Authority's Robert Moses Generating Station at Lewiston, New York, are also identified. A "trajectory saving" technique is used in the DDDP optimization model to reduce the computational requirements by an expected 35%.

INTRODUCTION

The Niagara Generating System is a very significant hydraulic resource for Ontario Hydro. This system provides an average of 13,800 GWh of energy, which represents 37% of Ontario Hydro's total hydraulic production. The worth of the annual energy produced by the Niagara Generating System is on the order of $300 million per year(Canadian).
In the long term, the Niagara Generating System can be viewed as a run-of-the-river system, but in the short-term it is possible to significantly vary the output of this system on an hourly basis over the time horizon of one day. This flexibility is provided by the partially controlled GIP reservoir above Niagara Falls and the fully controlled PGS reservoir below the Falls at Queenston.
Given the importance and complexity of the Niagara Generating System, a step-by-step approach has been used to develop a set of computer models to improve the operation of this system. These models can be summarized as follows: (1) a simulation model for the operation of GIP; (2) a Discrete Differential Dynamic Programming (DDDP) model to optimize the operation of GIP; and (3) a DDDP optimization model with trajectory savings to optimize the operation of both the GIP and PGS reservoirs (currently under development).
The purpose of this paper is to describe the above models and their application to improving the short-term operation of the Niagara Generating System.