Incorporating green energy into power systems

Mathematical optimization has long been in the tool kit of energy modelers. With the transition to renewables, this trusted prescriptive analytics technology is rising in popularity, along with newer AI-driven techniques, to assess the incorporation of green energy into power systems. In this article, we explore three use cases.

Modeling energy storage for the U.S. power grid. According to the U.S. Department of Energy, solar installations in the United States have grown 35-fold since 2008 while the average cost of solar photovoltaic (PV) panels has dropped nearly 50% since 2014. This growth in renewables is posing new challenges for the electricity market. Unlike most commodities, electricity must be produced as it’s consumed. As more renewable energy sources are added to the grid, the net load to be met by dispatchable generation becomes more variable. For instance, you may produce surplus solar energy at the height of the day, but how do you then meet load demand during the remaining evening peak hours when solar is no longer available? Storage offers a solution to this problem. How do you assess the cost implications and determine the required storage capacity needed throughout the year to meet demand at peak times? This is the central question behind OnLocation’s REStore model.

OnLocation is an energy consulting company based in the Washington, D.C., area. Their REStore model is a submodule of the National Energy Modeling System (NEMS), which is the most trusted integrated energy model used for projections and policy analysis in the United States. REStore is a linear programming optimization model developed in AIMMS. The model takes into account electricity demands, generation capacity and operating costs, hourly generation profiles for solar and wind capacity, storage capacity and discharge rate, storage efficiencies, maximum ramp up/down rates and other data points. Primary outputs include marginal electricity prices, storage operation hours, utilization hours by plant type, storage arbitrage values and variable renewable energy curtailments.

The output of the model is used to find the minimum cost of dispatch and storage utilization, using both existing capacity and new storage increments. This informs planning and policy decisions surrounding the introduction and optimal use of renewables in the grid. The implications of the model are far-reaching. For instance, as more electric cars enter the market, REStore might also be applied to determine the optimal timing of electric vehicle charging in the future.

Combining machine learning and optimization to determine wind power investments. Optimization is also frequently used in combination with machine learning to support strategic and operational decision-making such as decisions surrounding the optimal configuration and operation of wind farms. “The most interesting challenge when it comes to wind is predicting production,” says Chiara Bordin, associate professor of energy informatics at the Arctic University of Norway (UiT). “There are a multitude of related factors that play a role here, such as historical wind power data and temperature data.”

Machine-learning models are increasingly being developed to improve the prediction tasks for wind power generation. If you can use these to generate wind-related datasets, you can then combine them within mathematical optimization models to address a wide spectrum of optimal investment decision-making and operational management problems. In other words, you can use machine learning to generate the forecast dataset, and optimization to make optimal investment decisions.

Bordin and her colleagues Sambeet Mishra, Kota Taharaguchi and Ivo Palu conducted a performance comparison of five deep learning models, each combined with three types of data preprocessing. The goal was to assess the effectiveness of these models for wind power prediction and offer a reference point to better understand which model to choose given a forecast horizon, as well as what factors are significant. The machine-learning models they analyzed demonstrated very promising results within the field of wind power and temperature forecasts, especially when wavelet and FFT (fast Fourier) transformations were applied to the original time series data. Using the forecast from these types of models, policymakers can make more informed decisions about the amount of contracted wind energy and planners can make wiser operational choices surrounding pricing and dispatch. 

Integrated energy modeling. Another interesting area where optimization is increasingly being used is integrated energy modeling. The models used to understand electricity markets can easily be deployed to other systems, such as food and water distribution systems. With governments under pressure to meet carbon emission targets, an integrated approach can be valuable to understand the interconnectedness of these systems and optimize toward emission reductions.  

Nnamdi Nwulu, professor of electrical engineering at the University of Johannesburg, explains: “The consensus is that it is often better to optimize food, energy and water systems as a whole as opposed to individual optimization. If you have a power system that is based on fossil fuels like in South Africa, those plants will use a lot of water. Suppose you only optimize the amount of electricity to be produced without considering its impact on water flow and water demand. In that case, you will have an optimal system in the energy space, but it will be sub-optimal in the water space.”

Biomass is another interesting example. “Suppose you use food crops to produce energy,” Nwulu adds. “This may have an impact on food security. It reduces the amount of food available to feed your population. So, you have to manage those trade-offs. Do you focus on power output using all the biomass you can find, or do you also consider food security? If an area has more food security issues, you should place more focus on that. In areas where you have severe power shortages, but there is enough food, more weight should be given to power generation. It all depends on the trade-offs.”

The Netherlands Organization for Applied Scientific Research is also conducting research into integrated electricity markets modeling from its Energy Transition unit to model integrated electricity, hydrogen and gas markets. While these models have been traditionally planned and operated independently, their model analyzes the future energy system while considering interactions and linkages between these markets.

These are just a handful of examples of how optimization is being used in the transition to low-carbon energy systems. With more computing power and more advanced models emerging in this space, we expect the use of this prescriptive analytics technique to increase even more.