Advances in Integrated Gas and Power Modeling

With the increased interdependency of the natural gas and power markets there are many of the same and many new challenges being faced by both industries.  These challenges are magnified by the fact that we are under a new Administration, which brings both hope and uncertainty to the market. Under the previous Administration we had rapid growth in natural gas production, major restructuring of gas transmissions systems, increasingly severe regulations of emissions including CO2 and large increases in renewable energy and gas-fired generation with numerous coal and nuclear plant closures. 

Although the new administration will be facing many of the same challenges, there is much indication that their approach will be quite different. This has already been displayed by the complete reversal in policy with regard to the Keystone XL and Dakota Access Pipelines.

Understanding all the challenges and being able to track how they not only change, but affect the markets going forward can be very complicated. In the past when the gas and power markets were largely separated, analysts used individual market model to run what if scenarios in each industry. These powerful highly-granular systems which model each market separately are available, but they do not include important aspects of the other market. Gas market models do not consider power transmission constraints, emissions regulations, government mandates and technological advances in renewables, etc. Power market models do not consider gas pipeline constraints and expansions, storage, LNG imports and exports, etc. A useful and realistic model of both gas and power markets would take into account the constraints of both industries.

However this type of integrated gas and power modeling has its own challenges. One of the major drawbacks is the complexity of such as model. Each model uses its own solution techniques to address unique problems. This makes integration very difficult. Dr. Robert Brooks, Founder of RBAC, Inc. and Dr. Rahul Dhaul, Developer and EPIS, LLC. looked at two different approaches to integrated modeling. The Reduced Form Model and the Iterative Combined Model approach.

In the Reduced Model approach, one would estimate a set of regression equations based on the inputs and outputs from many scenario runs to serve as a substitute for the model. The first challenge had to do with the number of inputs and outputs in such models. Which do you choose and how many scenarios do you run? The biggest drawback was rapid obsolescence of regressions.

The Iterative Combined Model approach would establish an iteration process between a gas model and power model to find common solution for both markets. In this case, RBAC’s GPCM? Natural Gas Market Model and EPIS’s AURORAxmp? model were used by Dr. Brooks and Dr. Dhaul to test the approach.

First, some fundamental requirements for useful tools were established. One would need a realistic representation of market structures, a high degree of infrastructure granularity and detailed models of supply and demand. Modeling the North American markets, the US and Canada are established as competitive markets with Mexico moving towards a more open market. You must have a good model representation of the existing infrastructure, not an abstract aggregation, and provide means for market-driven capacity additions. Finally, when modeling supply and demand, the drivers must include economics, weather, price response, alternative fuels and government mandates regarding renewables and emissions reduction. Without these components, one will not get realistic market results.

RBAC’s GPCM model is a non-linear multi-period partial equilibrium model that computes for market clearing. GPCM has a monthly time frame that is calibrated from Jan-2006 through Dec-2016 and forecasts from Jan-2017 through as far as Dec-2070. Inputs into GPCM include: price sensitive supply and demand functions, detailed pipeline and storage infrastructure model, LNG imports and exports assumptions. Outputs include: basin-level natural gas and NGL production, state and sector level natural gas demand, Henry Hub and other market prices, basis and spreads, detailed pipeline flow and storage activity and pipeline capacity expansion requirements forecast.

EPIS’s AURORAxmp model is a fundamentals-based dispatch model with databases for North America and Europe. It computes chronological solution and simulates unit commitment and dispatch at hourly or sub-hourly level. AURORAxmp has a time frame that supports day-ahead to 40 years out. The inputs include: existing planned and additional required power plants and units, fuel and emissions constraints, RPS and CPP and transmission constraints: zonal or nodal-bus level. Outputs include: location-specific power market prices (zonal or nodal), generator dispatch, fuel burns & emissions, power flows, and capacity expansion requirements.

In order to integrate these two market models, GPCM needs a forecast of gas usage in power markets in order to compute gas prices which balance the market and AURORAxmp needs a forecast of gas prices to compute the optimal fuel mix in power generation. Having a successful integration would require the methodology chosen to address a few key questions. What happens if we set up an iterative loop between these two systems? Will they produce a mutually consistent solution? If so, how many iterations will be required for “convergence”? How can we speed up the convergence? How can we know when we’re “close enough”?

Three methods were tested.

Method 1 was a simple loop where we sent GPCM market prices as-is to AURORAxmp and sent AURORAxmp gas-burns as-is to GPCM. 

After seven iterations, the solutions are seen to be approaching a two iteration limit cycle rather than a convergence. Low prices to AURORAxmp yields high gas-burns, high gas-burns to GPCM yields high prices, high prices to AURORAxmp yields low gas-burns and Low gas-burns to GPCM yields low prices.

Correlations between every other iteration approach 100% for both Henry Hub price and total US gas-burn as shown on the following charts. It was very difficult to visually discern any difference between results from iteration 7 and iteration 5.

Method 2 was an average of the last two runs and sent the straight average of last two sets of market prices from GPCM to AURORAxmp and sent the straight average gas-burns from last two AURORAxmp runs to GPCM. 

After 16 iterations, the solutions are seen to be approaching a four-iteration limit cycle rather than a common solution between AURORAxmp and GPCM. By averaging the two most recent sets of prices from GPCM and sending them to AURORAxmp, its demand response is dampened. Similarly by averaging the two most recent sets of gas-burns from AURORAxmp and sending them to GPCM, its price response is dampened. However, the method does not converge to a single common solution for AURORAxmp and GPCM.

Method 3 took an exponential average of prior runs and sent exponential average of all prior sets of market prices from GPCM to AURORAxmp and of all prior gas-burns from AURORAxmp to GPCM. After 9 iterations, the method has effectively converged to an acceptable solution for both GPCM (prices) and AURORAxmp (gas-burns). This shows that a combined gas-power market model methodology not only works, but will bring about realistic results for examining the market.

During uncertain times, analytical tools and methods are an invaluable asset to any firm. The challenges being faced by the gas and power industries can be greatly reduced with the use of integrated gas and power modeling tools. With a now proven methodology, one can not only reduce risks but also identify opportunities within the market.