Off-campus UMass Amherst users: To download campus access dissertations, please use the following link to log into our proxy server with your UMass Amherst user name and password.

Non-UMass Amherst users: Please talk to your librarian about requesting this dissertation through interlibrary loan.

Dissertations that have an embargo placed on them will not be available to anyone until the embargo expires.

Author ORCID Identifier


Open Access Dissertation

Document Type


Degree Name

Doctor of Philosophy (PhD)

Degree Program

Civil Engineering

Year Degree Awarded


Month Degree Awarded


First Advisor

Konstantinos M. Andreadis

Second Advisor

Song Gao

Third Advisor

Hari Balasubramanian


Reservoir systems operations are challenging given that they must function to meet conflicting goals. Using mathematical programming and deep learning techniques, this dissertation presents innovative methodologies to address some of the challenges. The first chapter focuses on development of a mathematical programming framework for assessing sub-daily hydropower hydropeaking operation and flow regime outcomes of a system of five large sequential hydropower facilities on the mainstem Connecticut River under various operation scenarios. A formulation for the pumped-storage Northfield reservoir is presented that uses binary decision variables to properly model the reservoir operations. The results closely match annual historical power values that indicates the model can replicate the operations. The second chapter presents a novel multiple objective optimization methodology for trade-off analysis of river basins. The novelties include a weighting scheme that normalize different objectives having different range of variabilities and formulations for quantification of ecological and flood control objectives as frequencies of meeting desirable conditions. The methodology is applied to the Connecticut River basin. In this chapter, formulations are developed that use binary decision variables to quantify ecological and flood control objectives along with other operational goals. The key trade-offs of the system objectives are identified. The results indicate hydropower revenue objective highly conflict with any other objective than flood control. Moreover, it is concluded that a balanced operation that equally weight different objectives has the potential to improve all the objectives. The third chapter presents a methodology for designing reservoir operation policy using optimization and deep learning. This chapter addresses the challenge of designing of an operation policy for a reservoir with conflicting objectives under uncertainty of hydrological and energy prices data. A deep neural network is developed to infer near-optimal operation policies under different foresight scenarios using the optimization modeling results. The methodology is applied to the Wilder reservoir on the mainstem Connecticut River. A base method is also developed that uses linear regression and is applied to the problem and the associated results are used as a comparison basis. Results indicate that the designed policies using neural networks perform better than the base method used while having foresight for a longer time improves the performance.