Forecasting optimizes energy production, purchases for RWE Poland
The core challenge for utility company RWE Poland was balancing its customers' likely next-day demands for electricity with its ability to generate or purchase power on the day-ahead market. RWE Poland was unable to go to intra-day markets for spot purchases, so making good next-day calculations were imperative to its profit and loss operations.
Improving forecasts to reduce "balancing costs" is a complex exercise that's complicated by seemingly infinite variables. How many customers will be using electricity at each moment in time? How will they be using it? How will hot or cold weather affect their demand? How will the economic environment increase or decrease usage? How many unique customers are we serving throughout the day?
RWE Poland was attempting to answer all those questions using the homegrown rule-of-thumb estimations that often reside within each separate department. The challenge was that each small digression from actual facts creates ever-larger matching and coordination problems, especially for the day-ahead contracting this utility needed. In addition, it had some data from smart meters to use for forecasting load, but other data that had to be fit in the forecast to get a holistic picture of energy demand.
Closing the forecast gap
To accomplish this goal, RWE Poland quickly discovered that it also needed to clean, manage and structure its customer data to be useful. By first understanding its customers, the utility could get a better grasp of the variables affecting them and their use of electricity and incorporate that dynamic into its planning. Unfortunately, the data in RWE Poland's customer information systems was in bad shape and continued to degrade in quality as new, large volumes of usage data was incorporated by smart meters and automated metering infrastructure.
Data and forecasting go hand-in-hand
Next came the implementation of SAS Demand-Driven Forecasting. During this phase of the project, RWE Poland used the flexible SAS Data Integration tools to aggregate and integrate different types of data coming from multiple lines of business in order to achieve a more comprehensive vision of overall business management.
SAS Demand-Driven Forecasting provides the utility with consensus forecasting in conjunction with the sales and operations planning processes. The solution worked by providing hierarchical forecasting for hundreds of thousands of data series and also synchronizing and allocating forecasts from any level within the hierarchy.
Of special relevance to every utility's minute-by-minute demand profile, the forecasting method included time series methods such single exponential smoothing, Holt's/Brown's two parameter exponential smoothing, and Winter's three parameter exponential smoothing.
The variables involved in utility forecast are handled by causal methods such as ARIMAX (ARIMA with intervention and causal variables), lagged variables/transfer functions, dynamic multiple regression, and the Unobserved Components Model.
Using these statistical analyses, SAS provided the forecasts that reflected the contingencies of RWE Poland's business, improving its planning accuracy. The solution offered additional value because it went a step further, automatically generating reports that indicated the gaps between the financial plan and all individual, departmental and statistical baseline forecasts, with notes indicating reasons. These reports can be reviewed, changed and written back to the data model and offered a compelling ongoing corrective loop for the ever-changing dynamics involved in electricity demand and supply.
Technical losses occur because of the add-on engineering processes of expanding utility systems, in this case in a fast-growing economy. The savings that could be achieved in modeling and planning to correct technical losses could be used to optimize generation operations and reduce investments needed for additional generation capacities. Non-technical losses occurring due to theft of services could also be detected as a result of the clean-up for the forecasting exercise.
Copyright © SAS Institute Inc. All Rights Reserved.
Need for more dynamic predictive modeling of customer demand for day-ahead portfolio planning and purchases.
Achieved predictive modeling of consumer behavior for better operational efficiency, and reduced balancing costs associated with tomorrow's energy.