Load Research and Forecasting

Electric and gas utilities have traditionally developed load forecasting models to predict capital expenditure requirements necessary to meet long-term load requirements. These analytical models are also used to provide weather-normalized energy sales and revenue forecasts for near-term operational planning, budgeting, and ratemaking. The Cadmus Group provides load forecasting services to address key issues facing specific markets and end uses. In addition to understanding the influences of the economy and weather, these analyses address impacts of pricing, competition, demand-side management (DSM), and new products on load projections.

We develop forecasting models using SAS® software. As an SAS® Gold Alliance Partner, Cadmus offers state-of-the-art analytical and software solutions to the toughest forecasting challenges facing the utility industry today. Indeed, the demand for timely, accurate forecasting has never been greater. Today’s utility executives need to know:

• How will new building codes and equipment standards affect load growth?
• What will be the impact of new technologies (such as plug-in electric vehicles) on energy sales and the system load shape?
• Where is load growth expected to occur on the distribution system?
• What kinds of customers are contributing to growth?
• What can be done to meet customer needs and better manage growth?

Utilities’ needs have evolved, and the level of accuracy and depth of information they require has increased. Although it was acceptable to be within +/- 5 percent a generation ago, utility executives now want forecast errors to be no larger than 1 percent. Moreover, the traditional top-down aggregation into residential, commercial, industrial, and governmental customer sectors no longer meets utility business requirements. Better, faster forecasting is now a must. Until recently, getting this type of detailed, disaggregated forecast information required a considerable—and unacceptable—investment in time and human resources.

End-Use Forecasting

Originally developed in 1993, End Use Forecaster is a market segmentation, competitive assessment, and sales projection software application developed to respond to market needs and overcome the limitations of existing demand forecasting and market planning tools. Cadmus’ clients rely on End Use Forecaster for both end-use forecasting and as a DSM planning model to estimate technical, economic, and achievable potential for energy-efficiency measures.

End Use Forecaster is capable of accommodating multiple geographic areas, fuels, sectors, market segments, and end uses. Clients obtain realistic impacts of possible future scenarios, providing the crucial load, cost, and revenue impact information necessary for sound decision making.

We have found that each utility’s market structure and competitive environment is unique, and that a major shortcoming of other tools has been an inability to accurately capture this diversity. End Use Forecaster’s Market Segmentation module provides the ability to update the model to reflect new strategies without writing SAS programming code. Unique market conditions translate into an inherently flexible, dynamic modeling framework which can rapidly adapt to new market conditions.

This flexibility is afforded through a model development approach that separates specific market issues from theoretical modeling constructs. Logic and theory is the portion of the system comprised of the programming code and data structures and is stored and managed in one location. Market data, which are unique for every company and strategy, are stored in a separate location.

This structure makes market segmentation and analyses relatively easy tasks compared to adapting spreadsheet models or rewriting “black box” programming code. For example, in the “DSM planning” and “competitive assessment” market dimensions in the table below, the DSM dimensions show a standard end-use forecast model design for the utility industry, while the competitive assessment dimensions illustrate another way to set up End Use Forecaster to analyze new retail competition if retail choice is present in the jurisdiction.

Effectively incorporating the inherent multidimensionality of most business forecasting issues is a key strength of the End Use Forecaster framework. The critical dimensions of business issues (e.g., geography, customers, products, competitors, equipment lives, etc.) are included in every forecast; along with dimensions users can modify to resolve a variety of business issues. For example, forecasters may be interested in the price elasticity of demand; marketing staff may want to study market shares across various scenarios; and corporate finance may need the bottom line revenue forecast. All these (and more) are immediately available in every forecast due to a concentration of rich and flexible dimensionality.

Seven primary modules form the heart of the End Use Forecaster framework:  Market Segmentation, Data Development, Product Usage, Provider Choice, Intervention Strategies, Forecasting, and Reporting. The figure below depicts the relationships between these modules.

End Use Forecaster Modules and Structure

Multi-Dimensional Forecasting

The Cadmus Group’s innovative Multi-Dimensional Forecasting (MDF) approach solves another long-standing problem facing utilities: obtaining a set of forecasts with enough granularity for system planning, distribution planning, customer marketing, and financial forecasting.

The MDF framework combines traditional top-down econometric load forecasting techniques with the sophisticated forecasting abilities of SAS® Forecast Server to quickly address thousands of related, disaggregated series. The resulting solution produces a series of rich, multidimensional forecasts that simultaneously maintain forecast accuracy requirements. MDF combines the best elements of top-down econometric and bottom-up SAS® Forecast Server capabilities by:

• Adding the top-down model specifications to the SAS® Forecast Server model repository, allowing each node in the hierarchy to choose a structural model dependent on weather and economic activity, in addition to the library of time series algorithms built into the SAS® Forecast Server.
• Fully calibrating the bottom-up forecasts to the top-down forecasts, ensuring the long history of accurate load forecasting is transferred to the disaggregated forecasts.

Multi-Dimensional Forecasting

Cadmus’s MDF process has four essential steps:

1. Conduct Traditional Top-Down Forecasting.  Standard econometric models are developed with a focus on overall energy sales, peak demand, and customer growth. While utilities often produce these forecasts by jurisdiction and revenue class, their distinguishing characteristic is they are highly aggregated to support overall system planning activities.
2. Perform Data Mining and Market Segmentation Analysis. Cadmus employs cluster analysis and related data mining techniques to help utilities analyze the three W’s of load growth: who, where, and why. This analysis takes advantage of existing utility knowledge and micro-level data (customers, distribution points, geographies, load profiles, climates, rates, and other market dimensions of interest) to create information subsets necessary to support forecasting at desired detail levels.
3. Conduct Bottom-Up Forecasting. Using the SAS® Forecast Server, Cadmus combines the best features of structural econometric modeling with time series algorithms to develop highly detailed forecasts. Potentially thousands of separate series—one for each market subset developed in Step 2—are analyzed and forecasted. Batch processing allows this to happen quickly and cost-effectively, with the best forecast for each series chosen using pre-specified decision criteria.
4. Calibrate Forecasts. Once top-down and bottom-up forecasts have been developed, it is crucial they become fully calibrated to one another. This process ensures decision makers across the organization are operating from common, internally consistent forecasts.

Peak Demand Forecasting

Cadmus’s peak demand forecasting services provide energy utilities with the high-quality load projections that are needed for current strategic planning requirements, allowing us to address the most challenging forecasting questions facing utilities:

• How does the magnitude and timing of extreme weather events influence customer demand? Does the traditional concept of “normal weather” accurately reflect the “average” of recent weather events, or are better measures available?
• What does the load forecast distribution look like given historical weather and economic growth? Is the distribution of possible revenue symmetric, or is there greater risk (reward) at one extreme of the possible outcomes?

An example of this approach is shown in the figure below. In this example, weather uncertainty—in this case the uncertainty associated with extremely hot and humid periods—creates an asymmetric risk profile.

Energy Forecasting with Weather Uncertainty