Dr. Amy Shaw gives an In-House Presentation on Optimizing Operating Schemes for Hydropower Reservoirs

Amy Shaw, Ph.D., P.E. of AquAeTer’s Brentwood, TN office kicked off our 2019 in-house seminar series with her presentation entitled “Determination of Optimal Operating Schemes for a Hydropower Reservoir Under Environmental Constraints”. Dr. Shaw’s research is described in more detail below:

For hydropower systems under regulatory requirements, high-fidelity water quality models are useful tools for advance prediction of water quality outcomes subject to operating schemes; however, when considered alongside additional reservoir demands and constraints, an optimization approach becomes necessary for determining compliant operations. Hydropower operations optimization subject to environmental constraints is limited by challenges associated with dimensionality and spatial and temporal resolution.

Dr. Shaw’s dissertation describes an advancement for computing hourly power generation schemes for a hydropower reservoir using high-fidelity models, surrogate modeling techniques, and optimization methods. An artificial neural network successfully emulates the predictive power of the high-fidelity hydrodynamic and water quality model CE-QUAL-W2.

A genetic algorithm (GA) optimization approach was applied to maximize hydropower generation subject to constraints on dam operations and water quality (as predicted by the surrogate model) to a multipurpose reservoir near Nashville, Tennessee. The surrogate model successfully reproduced high-fidelity reservoir information while enabling increases in hydropower production value relative to actual operations for dissolved oxygen (DO) limits of 5 and 6 mg/L, respectively, while witnessing an expected decrease in power generation at more restrictive DO constraints. Exploration of simultaneous temperature and DO constraints revealed capability to address multiple water quality constraints.

The optimizer’s solution quality depends upon surrogate model training data selection, and offline training alone (i.e., prior to optimization) yields poor results. We therefore demonstrated a method for adaptively updating the surrogate model with newly supplied training data, influenced by the convergence path of the GA. A random immigrants replacement approach was also explored. Combining both approaches improved solution fitness values compared to those found by optimizing with neither feature. Looking toward expanding this work to a system of reservoirs, we assessed the sensitivity of release water quality in response to an upstream reservoir’s releases. Prediction errors caused by differences in the upstream boundary condition indicate minimal impact on potential solutions to the optimization problem, in which these predictions define water quality constraints. Determining independence between reservoirs could enable efficient expansion of the optimization methodology to reservoir systems.