(347c) Multiperiod Conceptual Design of Integrated Energy Systems with Market Interaction Surrogate Models

Integrated energy systems (IES), which integrate multiple energy sources and conversion technologies, are essential technologies to help decarbonize global energy infrastructures [1, 2]. Compared to conventional technologies, IESs are more flexible, can have higher overall efficiency, and can produce multiple products which mitigate risks in changing market conditions. The economics of IES are often governed by the wholesale energy market. As such, the techno-economic optimization of IESs predominantly uses the so-called “price-taker” approximation to incorporate time-vary market conditions (e.g., wholesale electricity prices). This approximation, however, ignores market interaction by assuming that the IES (i) does not affect the market prices, (ii) does not impact the dispatch profile of other generators, and (iii) the market can accept all the electricity generated by the IES. Recent work [3–5] critiques all three underlying assumptions, often leading to misleading analyses and conclusions.

This work presents a novel multiperiod design and operations co-optimization formulation that explicitly considers IES/market interactions using machine learning surrogate models (e.g., time-series clustering, neural networks). Our framework is organized into four steps

1. Perform Production Cost Model (PCM) simulations to assemble initial training data.

As the first step in the proposed framework, we sample different combinations of design variables from the domain of the IES design space. We fix the IES design to each point in the chosen sample and run an annual simulation of the electricity market using a production cost model (PCM), such as Prescient[6]. PCM solves unit commitment and economic dispatch problems in a rolling horizon fashion, and yields detailed day-ahead and real-time dispatch profiles of all generators and the day-ahead and real-time locational marginal prices (LMPs) at all nodes

2. Train surrogate models to predict market outcomes as a function of IES characteristics.

In the second step, we identify a few representative dispatch profiles for the IES using time-series clustering of daily real-time dispatch profiles corresponding to all annual PCM simulation results. We then train a machine learning-based surrogate model to determine the weight associated with each representative dispatch profile as a function of the design of the IES. Another machine learning-based surrogate model is trained to compute the total electricity revenue (determined using the nodal LMPs) as a function of the design of the IES. The surrogate models capture the effect that the IES has on the overall grid, and how that affects the IES performance.

3. Solve conceptual design optimization problems with embedded market surrogates.

We formulate a conceptual design problem by embedding surrogate models for both revenue and dispatch and solve it to obtain the optimal design of the IES. The surrogates in the optimization problem will enable us to quantify the market-level interactions between the IES and the energy market.

4. Verify results with multiscale simulation[7] including PCM.

By solving the conceptual design, we will get the optimal design of the IES. We will use the multiscale simulation framework[7] to simulate the optimal IES design and compare the results from the conceptual design.

Using the proposed framework, we assess the economic benefits of retrofitting an existing baseload nuclear generator with a low-temperature electrolysis unit. The electrolyzer enables the nuclear generator to ramp down the power output to the electricity market, and participate in the hydrogen market during periods of low electricity demand. Assuming that the hydrogen market can accept all the produced hydrogen at a fixed price, the goal is to determine the optimal size of the electrolyzer that maximizes the net present value of the combined system. Here, we demonstrate that the price-taker approach yields a sub-optimal size of the electrolyzer since it substantially underestimates the revenue generated from the electricity market, and thereby illustrates the need for the proposed workflow.

References:

[1] Arent, D. J.; Bragg-Sitton, S. M.; Miller, D. C.; Tarka, T. J.; Engel-Cox, J. A.; Boardman, R. D.;Balash, P. C.; Ruth, M. F.; Cox, J.; Garfield, D. J. Joule 2021, 5, 47–58.

[2] Fodstad, M.; del Granado, P. C.; Hellemo, L.; Knudsen, B. R.; Pisciella, P.; Silvast, A.; Bordin, C.; Schmidt, S.; Straus, J. Renewable and Sustainable Energy Reviews 2022, 160, 112246.

[3] Emmanuel, M. I.; Denholm, P. Applied Energy 2022, 310, 118250.

[4] Frew, B.; Levie, D.; Richards, J.; Desai, J.; Ruth, M. Applied Energy 2023, 329, 120184.

[5] Martinek, J.; Jorgenson, J.; Mehos, M.; Denholm, P. Applied energy 2018, 231, 854–865.

[6] Watson, J. P.; Knueven, B.; Concepcion, R.; Melander, D.; Short, A.; Zhang, P.; Woodruff, D.; USDOE Prescient, Version 1.0, 2020.

[7] Gao, X.; Knueven, B.; Siirola, J. D.; Miller, D. C.; Dowling, A. W. Applied Energy 2022, 316, 119017.