Buildings account for 39% of the total primary energy consumption, representing both a significant strain on power systems and a large source of human-induced climate change. While the rise of solar power and other renewable energies seek to address the supply side of the equation, there are still plenty of efficiencies to be extracted on the demand front.
A recent Irish study has taken on demand response algorithms, using a new evaluation approach they demonstrate machine learning’s superiority in reducing electricity use, utility generation cost, and carbon emissions of residential buildings.
Over the last decade, a wave of policy, legislation, and initiatives have sought to improve the energy efficiency of buildings. In the EU, for example, new efficiency measures, such as increasing thermal envelope insulation, improving heating and cooling systems, and the penetration of distributed renewable energy, are all contributing to the reduction of the overall grid energy consumption. However, the intermittent and distributed nature of renewable energy is increasing the supply/balance variability in the power system creating challenges that could cause congestion and atypical power flows. Thereby straining the underlying electricity transmission and distribution network.
Enter Demand Response (DR) a not-so-new technological evolution of the Demand Side Management (DSM) measures that were promoted in the UK and other countries since the 1970’s. The concept was originally brought about to reduce high winter energy spikes as well as avoiding associated grid upgrade costs. More recently, interest in DSM reignited as DR, with a view to increasing the percentage of renewable energies in the system by utilizing new cyber-physical systems rich with data and intelligence.
Hans-Christian Gils, Energy Systems expert at the department of the German Aerospace Center defined DR as ”changes in electricity use by demand-side resources from their normal consumption patterns in response to changes in the price of electricity or to incentive payments designed to induce lower electricity use at times of high wholesale market prices or when system reliability is jeopardized.”
DR brings about consumption efficiency and smoothes grid energy spikes to balance the power system. However, DR measures require building automation systems which have not yet been widely adopted in residential buildings. The proposals for DR in new buildings envision a system, equipped with connected sensors and energy management systems, which can control HVAC systems and appliances. While being responsive to grid signals these “smart controllers” can be embedded into smart home networks and become an integrated part of each dwelling.
On a wider scale, connecting these new residential buildings would make it possible to actuate balancing strategies across neighborhoods and districts, all triggered by DR signals, to create energy efficient cities made up of a network of smart buildings. However, this aging picture of the built environment has not been realized. This is in large part because advanced smart grid features require rapid prototyping and validation of EMS control algorithms. Until equipment and algorithms can be analyzed, it is difficult for utilities and regulators to install, operate and exploit these new resources.
The answer comes from building simulation software, which can be utilized to save considerable resources compared to experimental analysis when assessing the value and the risks associated with the adoption of DR technologies. If properly calibrated, building simulation software also offers the opportunity to perform energy assessments at a range of time intervals from annual to sub-hourly, without the need for hardware and communication network prototypes.
For the majority of building simulation software, DR control algorithms have been developed and trained on simulated data, according to the recent study in Ireland. “This is because the evaluation of the effectiveness of DR control algorithms in real buildings often requires considerable periods of analysis before a consistent validation can be arrived at, and test conditions are not easy to reproduce because of the unpredictable nature of human behavior and weather conditions,” reads the accompanying paper entitled: Demand response algorithms for smart-grid ready residential buildings using machine learning models.
“Moreover, when the evaluation involves critical infrastructure such as a power grid or advanced heating and cooling equipment, trial-and-error approaches can compromise the integrity of the test bed and the related systems,” explain Fabiano Pallonetto et al. “Therefore, the ability to develop control algorithms using measured data, and assess their performance within a co-simulation environment which exploits a calibrated building energy simulation model is a novelty element and provides a step forward to narrow the gap between research, development, and implementation.”
During the study, a typical house, representing the most common building category in Ireland, was fully instrumented as a test-bed for the project. A calibrated building simulation model was then developed and used to assess the effectiveness of demand response strategies under different time-of-use electricity tariffs in conjunction with zone thermal control.
Two demand response algorithms, one based on a rule-based approach, the other based on a predictive-based (machine learning) approach, were deployed for control of an integrated heat pump and thermal storage system. The two algorithms were then evaluated using a common demand response price scheme. “This hybrid methodology can reduce the test cycle time, reduce the hardware infrastructure and also allows for the replication of the experimental conditions,” stated the study.
Compared to a baseline reference scenario, electricity end-use expenditure showed 20.5% using the rule-based algorithm and 41.8% using the predictive algorithm. A similar ratio was seen for utility generation cost, where the rule-based algorithm conceived savings of 18.8%, while the predictive algorithm reduced cost by 39%. Carbon emissions tests saw 20.8% reduction for rule-based and 37.9% for the predictive algorithm.
“In conclusion, advanced demand response control techniques in all-electric residential buildings together with the increased adoption of time of use tariffs in many European countries, can positively contribute to the development of flexible power system frameworks aimed at reducing the carbon footprint of building stocks and supporting the shift towards more sustainable power generation mix,” says the paper.