We expect light to replace darkness when we turn on a lamp. It seems simple enough. But prior to light filling the room, years of planning occurred. This type of planning, referred to as load forecasting is a critical planning function performed at electric utilities to ensure ample supply of electricity meets demand.
As renewable energy resources increase, the necessity for a sophisticated, robust and integrated technology platform that can adapt to load forecasting requirements becomes crucial. A platform that enables planning for our lights to turn on when we flip the switch, both in the near-term and long-term.
Integrating numerous energy resources within the distribution system (rooftop solar, photovoltaic, electric vehicles, demand response) require the analytics of load forecasting. These analytics are used to know when, how much and where these resources are contributing on the power grid. All while balancing the amount of traditional generating resources to meet demand.
Forecasting the load on assets along the distribution system provides insights on which components are overloaded beyond their manufactured designed. It also identifies where the grid needs to be hardened, as electric power providers are committed to delivering safe, reliable electricity to you, producing a satisfied customer.
Learn more about SAS® Energy Forecasting
Now you can support your entire planning process using advanced AI and machine learning capabilities with SAS Energy Forecasting. Meet the forecasting challenges of the evolving electric grid. Learn how companies are using SAS Energy Forecasting.
Why load forecasting?
The planning begins by forecasting electricity demand five to 30 years in the future, and then narrowing a near-term view of the next hour’s electricity demand. It uses historical patterns of electricity consumption, or load profiles, weather conditions, regional events and economic indicators to represent when and how much electricity is used. These historical patterns vary by customer type, in both the magnitude of electricity consumed and the pattern of usage.
Aggerating the granular level of 15-minute reads from advanced metering infrastructure (AMI) at the customer level, to defined nodal levels on the system, provides insights of energy usage throughout the power grid. Models developed and applied for each level of the system refines the load forecasting process. Rolling the forecasts up to the system level reveals the peak loads within the month, season or year. The magnitude and timing of the peaks determines the generation type and amount that will be required.
For example, the residential load profile is likely to have higher demand in the early morning hours and evening hours when everyone is home, compared to lower electricity demand during daytime hours when most in the household are at work or school. An industrial facility often has a consistent level of demand for a 24-hour production line, and the industrial demand is less influenced by the weather conditions. Commercial load profiles are similar to residential profiles, but also vary based on the type of business.
Long-term forecasting versus short-term
Long-term load forecasting improvements tend to be undervalued because it is difficult to quantify the impacts of improving forecast accuracy. A forecast with improved accuracy may defer building additional generating units or defer the need for a long-term power contract. From an economics perspective, it is the saving of deferring the marginal unit. The accuracy of a load forecast has budgetary implications for the power supplier. A load forecast that is too low may result in expensive spot market purchases or outages. A load forecast that is too high reduces sales revenue by not selling unused capacity.
Lead time to build the appropriate generation type, for the demand growth in five, 10 or 20 years, can take years to acquire the approvals. Along with the permitting needed at the utility, regulatory commissions and public interest groups. The capital cost of building new generating units is incurred by the utility and eventually passed on to the customers. Conversely, if there is not enough generation to meet demand in the future, the power provider will need to secure long-term power supply contracts for the future.
Near term operational decisions rely on short-term load forecasting to meet customer electricity demand in the next hour to 48-hours. Sudden, unexpected changes in weather conditions, or an unplanned outage at a scheduled generating unit, requires adjustments to supply to ensure grid reliability. In markets with limited availability of additional generation and constraints on the power grid, a miscalculation can result in blackouts and a higher cost to provide or secure supply.
A load forecast that is too low may result in expensive spot market purchases or outages. A load forecast that is too high reduces sales revenue by not selling unused capacity.
Growth: A challenge for load forecasting
A forecasting model is developed and applied to the five, 10, or 20-year future periods to produce the load forecast. A vital aspect of load forecasting involves determining the number of hierarchies (or levels) to represent the business needs. Hierarchies can consist of customer classes (residential, commercial, industrial) and a more detailed hierarchy of a distribution system. A distribution hierarchy can be composed of thousands of assets, requiring scalable load forecasting approaches.
Load forecasters are known for their expertise in these types of load forecasts. The new challenge for utilities is the introduction of more energy sources known as distributed energy resources (DERs), which refers to renewables like solar, wind, micro-grids, electric vehicles (EVs) and even newer battery storage capabilities. All of which is adding more variability and complexity into the forecasting process.
The current annual energy outlook from the U.S. Energy Information Administration (EIA) projects consistent growth in renewables, illustrating the importance of accurately forecasting the output from renewable resources.
Modeling renewable sources, such as solar farms, requires advanced, deep learning algorithms. By accurately forecasting how much and when renewable resources can produce the output expected, we are able to see the value of correctly modeling these renewable resources. Generation supplied from renewable sources reduces the amount of generation needed from traditional resources, typically fossil fuels. This is depicted in what is known as a “duck curve”. The duck curve demonstrates the change in electricity demand and the amount of accessible solar energy throughout the day.
The duck curve shows the growing impact of net load changes as the belly of the curve dips deeper with increasing renewable generation. The challenge for the power supply system lies in the steep changes up and down the curve. For instance, in the afternoon, as less solar generation is available to meet the forecasted demand, the system operator is tasked with ensuring traditional generation resources are available and have time to ramp up to production.
Power generation and distribution must be kept in perfect balance with demand. If it is not balanced continuously, customers will suffer outages or power quality issues that can damage equipment. Analytics can provide insights that improve safety, reliability and ultimately help to reduce costs for both the utility and their customers.
- 4 strategies that will change your approach to fraud detectionAs fraudulent activity grows and fighting fraud becomes more costly, financial institutions are turning to anti-fraud technology to build better arsenals for fraud detection. Discover four ways to improve your organization's risk posture.
- Empowering women in analytics through education How you can elevate your learning with SAS and the Women in Analytics network
- Shut the front door on insurance application fraud!Fraudsters love the ease of plying their trade over digital channels. Smart insurance companies are using data from those channels (device fingerprint, IP address, geolocation, etc.) coupled with analytics and machine learning to detect insurance application fraud perpetrated by agents, customers and fraud rings.