Resource Optimization in the Princeton University Energy System

Abstract

The Princeton Energy Plant provides Princeton University with electricity, steam and chilled water. Electricity is supplied through a mix of purchase from the grid, import from the recently installed solar collector field in West Windsor, and on-site generation in the University Cogeneration Plant. Steam and chilled water are produced locally in boilers and chillers attached to the cogeneration facility. Demands for these resources, as well as the price of electricity, are volatile making online management of the energy plant necessary. This thesis focuses on optimization of resource purchase and allocation decisions in the Princeton Energy Plant in order to meet loads economically. A mixed-integer linear programming approach is used to solve the optimization problem for twenty four hour periods, which is implemented and updated in five minute increments. The physical behavior of the plant assets is captured through thermodynamic and statistical models and serves as a set of equations that constrain the linear program. The decisions generated by the model based on historical conditions are compared with actual decisions made in corresponding periods. In current tests over ten day periods in each month of 2012, the model achieves average cost savings of 10%. The availability of perfect future information about loads and prices raises savings for particular months by up to 6%. Simulations of scenarios where the size of the University is increased and where taxes on carbon dioxide emissions are imposed reveal that total energy costs rise approximately linearly with these variables, but their effect on individual plant behavior is non-linear.