SEEKING USER INPUT IN INFLOW FORECASTING
by
T. Tao¹, I. Corbu¹, R. Penn¹, F. Benzaquen¹ and L. Lai¹

ABSTRACT

Traditionally, inflow forecasting models were used like black boxes. Users prepared the inputs at one end and received the outputs at the other end. In such an environment, the inflow forecasting models were aimed at generating inflow forecasts without any user intervention. This paper describes anew, user friendly, approach to inflow forecasting. It allows users to input their preferences, interactively, during the execution of the model and to generate inflow forecasts with which they feel comfortable.


INTRODUCTION

One of the requirements for integrating the water management and operation of Ontario Hydro's hydroelectric facilities into its Grid is to generate daily inflow forecasts for up to two years for those facilities and their control reservoirs. It is recognized that inflow forecasts with longer term could hardly be indicative of what will occur due to the random process upon which they are based. In order to assess the risk associated with the use of inflow forecasts, it is necessary to have a variety of equally likely inflow forecasts. The model described in this paper provides such a capability, allowing users to generate inflow forecasts with any desired exceedance probabilities of volume.
Each generated inflow series has three parts. The first part comprises inflows for the first four days. They are provided interactively by a user-selected inflow forecasting model. The second part includes inflows from day 5 to day N-l. These are heuristically modified historic inflows which precede those in the third part. They ensure a smooth transition from the first part to the third part. The third part contains inflows from day N, which ranges from 15 to 60, to day 732. They are part of historic inflow records. The process of generating such local inflow series is accomplished through three modules. They are called expected forecast, heuristic forecast and probabilistic forecast respectively. Users play an active role in deciding what will be produced in each module (Too 1993). The paper describes how such user interfaces are achieved.


EXPECTED FORECAST

The expected forecast generates the inflows for the first four days of the two-year time horizon. Preliminary comparative studies have been conducted to retain the best of several conceptual and stochastic models (Too and Lai, 1991) for the hydrologic conditions in Ontario, Canada. Three …