Name: Lei Gao
Title: Optimum Design and Operation of Combined Cooling Heating and Power (CCHP) System With Uncertainty
Prof. Reinhard Radermacher (Chair/Advisor)
Prof. Mark Fuge
Prof. Steven Gabriel
Prof. Jelena Srebric
Prof. Peter Sunderland, Dean’s Rep
Prof. Yunho Hwang
Prof. Vikrant Aute
Date: Monday, May 2nd, 2022
Time: 9:00 AM
Location: EGR-4164B (ENGR)
Zoom link: https://umd.zoom.us/j/6085437805?pwd=NlRsYWIwSHczR1k4bkxaUnZENkJmQT09 (ID: 6085437805, passcode: 926173)
Combined cooling, heating, and power (CCHP) systems utilize renewable energy sources, waste heat energy, and thermally driven cooling technology to simultaneously provide energy in three forms. They are reliable by virtue of main grid independence and ultra-efficient because of cascade energy utilization. These merits make CCHP systems potential candidates as energy suppliers for commercial buildings. Due to the complexity of CCHP systems and environmental uncertainty, conventional design and operation strategies that depend on expertise or experience might lose effectiveness and protract the prototyping process. Automation-oriented approaches, including machine learning and optimization, can be utilized at both design and operation stages to accelerate decision-making without losing energy efficiency for CCHP systems.
As the premise of design and operation for the combined system, information about building energy consumption should be determined initially. Therefore, this thesis first constructs deep learning (DL) models to forecast energy demands for a large-scale dataset. The building types and multiple energy demands are embedded in the DL model for the first time to make it versatile for prediction. The long short-term memory (LSTM) model forecasts 50.7% of the tasks with a coefficient of variation of root mean square error (CVRMSE) lower than 20%. Moreover, 60% of the tasks predicted by LSTM satisfy ASHRAE Guideline 14 with a CVRMSE under 30%.
Thermal conversion systems, including power generation subsystems and waste heat recovery units, play a vital role in the overall performance of CCHP systems. Whereas a wide choice of components, nonlinear characteristics of these components challenge the automation process of system design. Therefore, this thesis second designs a configuration optimization framework consisting of thermodynamic cycle representation, evaluation, and optimizer to accelerate the system design process and maximize thermal efficiency. The framework is the first one to implement graphic knowledge and thermodynamic laws to generate new CO2 power generation (S-CO2) system configurations. The framework is then validated by optimizing the S-CO2 system’s configurations under simple and complex component number limitations. The optimized S-CO2 system reaches 49.8% thermal efficiency. This efficiency is 2.3% higher than the state of the art.
Third, operation strategy with uncertainty for CCHP systems is proposed in this thesis for a hospital with a floor area of 22,422 m2 at College Park, Maryland. The hospital energy demands are forecasted from the DL model. And the S-CO2 power subsystem is implemented in CCHP after optimizing from the configuration optimizer. A stochastic approximation is combined with an autoregression model to extract uncertain energy demands for the hospital. Load-following strategies, stochastic dynamic programming (SDP), and approximation approaches are implemented for CCHP system operation without and with uncertainties. As a case study, the optimization-based operation overperforms the best load-following strategy by 14% of the annual cost. Approximation-based operation strategy highly improves the computational efficiency of SDP. The daily operating cost with uncertain cooling, heating, and electricity demands is about 0.061 $/m2, and a potential annual cost is about 22.33 $/m2.
This thesis fills the gap in multiple energy types forecast for multiple building types via DL models, prompts the design automation of S-CO2 systems by configuration optimization, and accelerates operation optimization of a CCHP system with uncertainty by an approximation approach. In-depth data-driven methods and diversified optimization techniques should be investigated further to boost the system efficiency and advance the automation process of the CCHP system.