Energy Load Forecasting Overview
Natural gas transportation (pipeline) and local/natural gas distribution companies (LDCs or NGDCs) must meet the demands (contractual or other) of their customers. Pipeline content cannot be changed rapidly, and to operate most effectively and to keep lines in balance, these companies must be able to prepare lines in advance to meet expected demand. Traditionally, companies have relied on the knowledge of controllers who, through years of experience, have learned to anticipate and adjust pipelines accordingly. This experience, while invaluable, cannot quickly be transferred to new controllers. As such, a natural gas load forecasting system is required to estimate demand based on available historical data.
The fundamental challenge of any natural gas load forecasting system is to identify valid causal relationships between environmental, economic, or social factors and their effect on customer demand for natural gas. Once these relationships are identified, sufficient historical data must exist to establish the nature of the causal relationships, as well as adequate forecasted information so that future effects can be estimated.
Natural gas that is delivered from a pipeline to the customer passes through a flow measurement device, called a "meter," which, among other things, records the accumulation of gas volume that has flowed through the pipe at the specified receipt or delivery location. The "meter" is the fundamental entity in the forecasting system, since it is at the meter where actual values are recorded and, therefore, to which estimates will be associated.
It is widely accepted in the industry, and supported by historical data, that weather conditions and the calendar (time of day, day of week/month/year) have strong influences on customer demand for natural gas. Intuitively, this makes sense from a simple perspective. When it gets cold, people turn on their heaters and the demand for natural gas rises. When it gets hot, people turn on their air conditioners, increasing the demand for electrical energy, which in turn, increases the demand for natural gas, if that is the energy source used by power companies to generate electricity. Similarly, the calendar plays an important role in determining a customer’s demand for natural gas. People tend to spend more time at home on weekends or holidays and their demand for electrical energy increases on those days.
Unfortunately, the true causal relationships are much more complex than these simple examples. While the two primary factors affecting demand that must be considered are weather and the calendar, other factors, such as market price, must also be considered if forecast data is available. For instance, when the temperature rises, causing the demand for electricity to increase, the market price of natural gas could be relatively high at the time. Hence, the demand for natural gas might actually decrease, if coal or oil is a more economical option to satisfy the increased demand for electrical energy.
What is needed to solve this problem is a technique to model the very complex relationships between known causes such as weather and the date and their effect on customer demand for natural gas. Fortunately, the field of Artificial Intelligence (AI) provides a very effective approach to this specific forecasting challenge. This forecasting technique is known as "feed-forward, back-propagation neural networks" or "neural networks" for short. CygNet has developed a solution that uses neural networks to generate estimated customer demand for a natural gas pipeline.
CygNet’s ELF system can help companies determine the impacts these known factors have on the customer demand for gas delivery so that contractual obligations (demand capacity) can be met, as well as to improve the estimate for future capacity availability. While the forecast produced by the system may not always be 100% accurate, it is expected that the forecast will produce an acceptable user-defined variance from what proves to be actual demand.
More:
