Competitive advantage from predicting the who, what, and where of future energy demand
If you were to participate in an energy market, how much money could you save if you were able to predict not only what the future demand of energy would be, but also where it would be demanded? Even better, what if you knew exactly who was going to be demanding it? The answer varies, but for a deregulated competitive energy market, this question may mean the difference between profit and loss. Any ability to improve the understanding of future load behavior in a specific market sector or geography can help mitigate risk, gain competitive advantage, and gather market intelligence before your competitors can.
Load forecasting has been around for decades and is actively researched in electrical engineering and statistics. As a result, today, the industry has many advanced tools and software products at its disposal that have shown to be reasonably accurate. However, these methods have by and large been limited to using historical data in order to fit highly aggregate regional forecasting models. Furthermore, in the context of a competitive energy market, chances are that the most accurate predictive tools and strategies rest with those competitors with the deepest pocket. For those that don’t have in house predictive analytics, bidding strategies depend on forecasts provided by the RTO or ISO.
Regardless of your position along the energy value chain, you are likely to find value in establishing demand forecasts for geographical locations such as load zones, towns, or even specific buildings. In particular, implementing forecasting models for individual buildings has the ability to explicitly account for relevant determining factors (e.g., the heating envelope of a structure) as a means to predict energy consumption. Whether at a structure-specific level or at a geographical level, the use of “spatial” load forecasting is a relatively unexplored application. ISO New England (ISONE), a longtime client of DNV GL, is the regional operator of the grid in New England and the facilitator for its wholesale energy markets. Currently, ISONE records historical load and weather data for each one if it’s eight load zones (see Figure 1). But, for its Day Ahead market, the ISO is only required to provide forecasts of regional load. It uses weighted averages and artificial intelligence to produce regional forecasts that are among the most accurate in the country. However, the temperature, humidity, wind speed, and other weather variables used as inputs for its forecasting models are recorded at weather stations across the varied region. In some instances, the closest weather station may be in a different state. While already known nationally for its highly precise load forecasting, the potential value added from explicitly accounting for differences in industry, demographics, or other zone or state specific factors is unknown for ISONE’s Day Ahead market participants.
The short term load forecasting procedures at ISONE are an example of how prediction accuracy can be improved by considering spatial relationships and differences across such a weather-diverse region. The regional forecasts (“top down”) can be used as a compliment to zonal forecasts (“bottom up”). As a RTO or an ISO, what risk would be mitigated if you knew not only that a shortfall of capacity was going to occur, but precisely which load zone or nodal point would produce that shortfall? As a generator, demand response provider, or similar agent bidding into the market for energy, what if you didn’t have to rely on the aggregate energy demand forecasts provided by your regional system operator, but instead knew exactly how much demand would be needed for your specific geographical location? And finally, if you were a utility, what if you knew what each one of your customers was going to consume in the short, middle, and long term?
As the industry level adoption of smart grids and AMI continues to proliferate, there is no tangible reason to accept forecasting error simply because we are doing things in aggregate. The oncoming flood of high quality, high frequency data facilitates the development of a more granular, spatial predictive modeling approach (See Figure 2). These bottom-up models have value as benchmarks for top down modeling, as well as tools to identify those pockets or segments of the market characterized by specific energy demand behavior.
The powerful new DNV GL (formerly DNV KEMA) has clients found across all levels of the energy value chain as well as having well established relationships with every RTO or ISO in the US. Spatial load forecasting is just one way in which we can offer deeper insights, inferences, and analytics using the rich data that our clients are just beginning to access. Predictive analytics and load forecasting methods being researched and developed at DNV GL can provide immediate value to the wealth of data being generated by smart grids, microgrids, and other advanced metering structures. All corners of the energy realm are looking to see what kind of information they can learn from their data, and DNV GL is listening. Contact us to find out how our clients are benefiting from predictive modeling on a spatial or customer basis particularly as it applies to distributed energy resources.
 In terms of Mean Absolute Percent Error (MAPE).