Automated Regression Methods for Turbine-Penstock Modeling and Simulation

Abstract

Utilization of turbine-generator system modeling can provide distinct advantages when considering capital improvement projects in hydropower facilities. Such modeling facilitates the design and validation of specific turbine control systems. However, because modeling can often prove complex and time intensive, such techniques are not often utilized as part of the turbine control system development process.

In order to simplify and shorten this modeling process an automated regression methodology can be utilized in conjunction with a series of simple performance tests to provide requisite system characteristics. Such a methodology combines performance test data with gradient-based, multi-variable optimization algorithms to determine a system model through the following steps:

1. Modeling the Turbine-Penstock System
2. Validating the Turbine Model using Empirical Data
3. Optimizing Turbine Control Parameters for a Modernized System
4. Validating the Optimized Turbine Control Parameters on the Modernized System

Utilizing this automated regression method yields results that can help to assess performance improvements for governor modernization proposals, and validate new parameters prior to implementing actual hardware improvements. This paper will outline the process and results of this approach as it applies to the Castaic Pumped Storage automation upgrade project currently underway with the Los Angeles Department of Water and Power.