Dissertation
Thermal and AC Power Systems Dispatch Optimization Incorporating Storage and Unit Commitment
Doctor of Philosophy (PhD), Washington State University
01/2019
Handle:
https://hdl.handle.net/2376/17879
Abstract
Reliable and efficient power system dispatch is of paramount importance for the increased incorporation of microgrid infrastructures and renewable and distributed energy resources. Operating costs of microgrids and distributed generation can be reduced when optimizer models include control for unit commitment, meeting thermal demands, energy storage devices, and AC power flow constraints, as opposed to a heuristic dispatch or a dispatch that only considers some of these aspects. Augmentations to conic programming (CP) implementations for power systems dispatch with unit commitment allow for the inclusion of thermal demands, energy storage, and AC power flow constraints. The computational demand for inclusion of all the elements of this problem is shown to be remediated with transfer learning of a neural network (NN) with sigmoid and linear activation functions initially trained from Complementary Quadratic Programming (cQP) real power dispatch solutions then trained from conic AC power solutions. The application of a neural network allows for rapid creation of dispatch solutions which can be used to better evaluate the applicability and cost of dispatch methods with higher computational demand, such as conic programming, across a wide range of scenarios. This dissertation evaluates the computational efficiency, dispatch cost, and reliability of neural networks with varying layers as trained by conic programming and cQP for solving the AC microgrid power flow optimization problem with unit commitment, storage, and combined heat and power. Three test cases are used to benchmark cQP, CP, and NNs: a hypothetical test case, a model of Washington State University’s Pullman campus, and a modified model of WSU’s campus to include higher levels of on-site generation from solar and combined heat and power gas turbines. This work provides the first known example of simultaneous solution of the AC optimal power flow problem in conjunction with the combined cooling heat and power unit commitment optimization with storage. This work also demonstrates the advantage of incorporating a sigmoid function for neural network replication of microgrid dispatches which include unit commitment.
Key Terms: dispatch, optimal power flow, energy storage, microgrid, quadratic programming, conic programming, neural network, decision tree, machine learning
Metrics
1 File views/ downloads
21 Record Views
Details
- Title
- Thermal and AC Power Systems Dispatch Optimization Incorporating Storage and Unit Commitment
- Creators
- Nadia Victoria Panossian
- Contributors
- Dustin F McLarty (Advisor)Noel Schulz (Committee Member)Jacob Leachman (Committee Member)John Swensen (Committee Member)
- Awarding Institution
- Washington State University
- Academic Unit
- School of Mechanical and Materials Engineering
- Theses and Dissertations
- Doctor of Philosophy (PhD), Washington State University
- Number of pages
- 162
- Identifiers
- 99900581705101842
- Language
- English
- Resource Type
- Dissertation