Accurate forecasting of building energy consumption is as valuable as it is complicated. The ability to reliably predict how much energy a building will use in the next hour, week, or year would bring about a new level of flexibility within and between our smart buildings but a range of internal and external variables have presented a barrier to such development. Recently, however, an international research collaboration has published the results of a new approach that appears to trigger an accuracy-evolution in building energy management.
The paper, published in IEEE journal, proposes a “novel hybrid deep learning prediction approach” that utilizes long short-term memory as an encoder and gated recurrent unit as a decoder in conjunction with the Algebra of Communicating Processes or “ACP theory” — an algebraic approach to reasoning about concurrent systems. The new approach has been tested and validated against real-world datasets and consistently outperformed traditional predictive models, promising a new era for the prediction of energy use in buildings.
"Generally, it is challenging to predict building energy consumption precisely due to many influential environmental factors correlated to energy-consuming such as outdoor temperature, humidity, the day of the week, and special events," said Abdulaziz Almalaq, an author of the paper and assistant professor in the Department of Electrical Engineering in University of Hail's Engineering College in Saudi Arabia.
“While environmental parameters are useful resources for energy consumption prediction, prediction using a large number of a building’s operational parameters, such as room temperature, major appliances and heating, ventilation, and air-conditioning (HVAC) system parameters, is a quite complicated problem, compared with prediction using only historical data,” Almalaq continued.
Traditionally, historical energy consumption datasets are investigated for trends that might suggest what to expect in the coming days, weeks, or years. However, this method is limited by a number of highly unpredictable variables it depends upon, not least weather and human behavior.
These chaotic aspects of daily building life have prevented us from creating the reliability required to further enhance the distribution and management of electricity across facilities, cities, and entire grids. These fields of sociology and meteorology may never offer the quantifiable accuracy demanded by our digital energy infrastructure.
The world’s most advanced seven-day weather forecast is said to be about 80% accurate, while the five-day forecast claims to accurately predict the weather approximately 90% of the time, however, there is no standard measure of a "correct" weather forecast. Claimed accuracy certainly drops quickly beyond a week, with a 10-day forecast said to only be right about half the time, for example. Furthermore, weather forecasts are not localized to the building or even neighborhood scale and still provide intentionally vague guidance, like “partly cloudy”, to represent the inaccuracy.
“The inability to quantify or predict all variables influencing the [atmospheric] boundary layer is a large outcome of what is known as The Butterfly Effect [—the idea that a single flap of a butterfly’s wing could trigger a huricaine],” states an expert article on the Penn State blog: Musings of a Meteorologist. “Because we lack sufficient data and because the state of energy balance in the boundary layer is so chaotic, forecasting meteorologists find it impossible to predict weather phenomenon more than a few days out. While we may be able to add a few more days to the weekly planner in the future, there will come a time when we cannot forecast any further out.”
The article emphasizes the point that “we would need a lot of data to notice all of the slight nuances of thermal energy movement in the boundary layer,” likely prompting a pause-for-thought from readers across the AI and machine learning spectrum. Those of us engaged in AI, smart buildings and cities, or the IoT and big data analytics, are all chasing a dream that with enough data anything is predictable. And while real-time data collection from atmospheric boundary layers is still logistically out-of-reach, gathering masses of data from in and around the building is more feasible than ever.
“These days' smart buildings have high intensive information and massive operational parameters, not only extensive power consumption,” states the new research paper. “With the development of computation capability and future 5G, the ACP theory (i.e., artificial systems, computational experiments, and parallel computing) will play a much more crucial role in modeling and control of complex systems like commercial and academic buildings.”
In essence, better, cheaper sensors and connectivity leaps like 5G enable greater real-time data collection, which then feeds AI systems with the information they need to make sense of the “chaotic” complexity — and this new predictive method uses artificial systems, computational experiments, and parallel computing to bring us closer to that reality. One where we can depend on the accuracy of building energy consumption to such a degree that we could plan our energy infrastructure with a much lower risk of deviation.
This simple improvement in accuracy raises efficiency to a new level, reducing costs, balancing load, and supporting the ultimate goal of reducing the overall consumption of fossil fuels. Better accuracy also increases the flexibility of our energy systems, which reduces costs in conjunction with time-variant electricity pricing, improving our ability to load balance. Hardware and software design, purchasing decisions, service offerings, and potential market growth can all be disrupted with a sufficient boost in the level of accuracy in our predictions.
Whether this new method can provide enough of a boost is yet to be seen but the researchers remain confident about its potential application. "The analysis performed in this paper showed that the hybrid deep learning model is a powerful artificial intelligence tool for modeling multivariable complex systems,” Almalaq concluded. “It has the potential to be applied in different areas, such as the smart office, the smart home, and the smart city."