The diversity of the weather and unpredictability of an occupant's use of space make the system less suitable for the direct mathematical modelling required to determine fixed control mappings. Despite this it is broadly agreed that individual comfort and satisfaction can be attained by providing individual control of the local environment.
By allowing occupants to make adjustments they can avoid or alleviate discomfort. However, balancing user-overrides with automatic control is difficult if energy benefits and comfort are to be optimised. What is needed is a system that can learn site characteristics and, through a smart user interface, the nature of occupant preferences. This will enable it to adapt to any energy saving strategy.
While well controlled adaptable shading devices can produce considerable energy savings, field studies show that occupants only manually adjust blinds when threatened by discomfort such as overheating or glare. They rarely readjust the blinds after that threat has passed, leading to the unnecessary use of electric lighting. This has stimulated engineers to develop automatic photoelectric controls for lighting and blind systems.
Recent surveys show that the use of blinds as an energy conscious intelligent facade has largely neglected user needs. The surveys highlight the importance of user satisfaction and need for individual control. Early automated blind systems bypassed user interaction to stop interference with the energy strategy. Various environmental comfort equations, which calculate the comfort of a person doing a specific task within a defined environment, were then used to ensure systems would achieve satisfactory conditions.
Although these formula represent good models for the human physiology, they are unable to simulate an individual's perception of comfort, especially when considering the visual environment. In reality the conditions an automated blind must respond to contain a high degree of uncertainty. A control strategy should attempt to resolve variabilities which include weather, context (site micro-climate) and individual occupant preferences.
The interactions are inherently dynamic and non-linear; regional behaviour can differ due to internal and external conditions. Some environmental factors are unpredictable and interact in a complex manner. Other variables are simply difficult to measure or too expensive to evaluate.
If the majority of these variables must be accounted for, the process becomes less amenable to mathematical modelling based on physical laws. Consequently, researchers now believe that individual comfort and satisfaction can only be attained universally by providing individual local control. The challenge lies in integrating users' priorities within an energy efficient control strategy.
Integrating individual control
The most common method of providing individual control within an automated system is to use the automatic control for most of the time, and allow occupant override in discomfort. This provides individuals with three different forms of control:
Decisional and cognitive control will contribute to user satisfaction. However, behavioural control decreases occupant satisfaction as they can get frustrated with a system they have to constantly override. This is compounded by the time taken between making an adjustment and the automated control system taking back overall control.
One solution is an intelligent control system that gradually learns site characteristics and occupant preferences. It does this by seeing each occupant as a sensor, so it can relate user actions to variations in conditions. As the system learns, user interaction will decrease.
By introducing learning elements into a control system it could become more flexible and able to deal with real-world environments. It could adapt and learn about processes, disturbances and operating conditions, acquire and store knowledge and autonomously improve its performance through experience.
As part of research into adaptive facade devices, the Centre for Window and Cladding Technology (CWCT) constructed an integrated building control test cell at the University of Bath. The cell incorporates automatic window and lighting systems and various internal and external sensors; all integrated by LonWorks technology (see 'The test cell').
The control system currently has three types of user interface: local wall switches, hand-held infra-red remote control, or a purpose-built computer based Visual Basic interface.
Initial research findings
Early results from the research identified key issues associated with adaptive automated blind control strategies.
Fieldbus technologies like Lonworks have potential, but manufacturers are unwilling to develop products due to a lack of international standards and fear of veering from proprietary systems. Greater functionality must be built into the devices to enable integration into different processes.
Second, standard control sensors have limited ability to provide information about internal and external conditions. For example, sun sensors provide little information about current sky conditions. More information could be obtained with better sensors, and research is needed into how much information a system requires to balance learning performance with system cost.
If few parameters are used, the system will be unable to learn, with too many it becomes costly, unmanageable and complicated. However, enough measured parameters must be given to enable the system to assess reasons for occupant adjustments.
The level of control provided by Venetian blind motor mechanisms is currently limited. It is difficult to make slight adjustments in slat angle with subtlety and to a reasonable degree of accuracy. Manufacturers have said that they will find a solution to this problem.
Algorithms that calculate important factors such as solar altitude, azimuth and solar heat gain factors are not included in standard control systems. These could be distributed to individual devices such as solar illuminance sensors and internal temperature sensors to add to the information available.
Due to the issues mentioned, the majority of algorithms used for controlling blinds are basic with regards to solar and daylight control. Although some sophisticated systems have become available, they often have large overheads, in terms of design and commissioning time, and yield unconvincing results.
The user interface will influence the performance of the learning system significantly. The wording of controls (such as 'decrease temperature', 'close blinds' or 'open window') can influence how people use them, and consequently how the system learns.
Systems that are inquisitive in their wording may learn faster, but may also cause greater occupant annoyance. The underlying factor should be to make the control system as clear as possible to limit user frustration. However, this will increase the problem's complexity.
System generalisation could be improved from outside the learning algorithm if the designer anticipates how the system can be misused. With this knowledge it would be possible to create a simple rule-base that filters out the majority of erroneous user overrides before they reach a learning algorithm.
For example, if an occupant enters a daylit room from an artificially lit corridor, any adjustment made to the lighting on entering the room could mislead the system. Therefore, the system should ignore adjustments made within a certain period of an occupant entering. A well designed system could speed up the learning process considerably.
Many occupancy sensors are currently inadequate at providing accurate information about occupancy status. This information could be vital for assessing the application of various system modes and the response required to various user interactions, both necessary for improving system performance.
Further research
The next research stage is to investigate the use of different types of learning algorithms to demonstrate how some of the basic issues identified in the initial studies can be overcome. The performance will be governed by its ability to generalise, ie minimise the network's sensitivity to input errors and ensure that the stored behaviours are consistent so that similar inputs produce a similar response.
These characteristics are dependent on the system architecture and algorithms, and on choosing the right configuration. Learning algorithms being considered include fuzzy logic, knowledge based systems and artificial neural networks.
The CWCT study highlights the needs to provide individual occupant control and make control systems clear. In the 21st Century engineers will engage in the evolution of these technologies by integrating them into rooms to produce transparent systems that are beneficial to humans global and individual needs.
The test cell
The top and bottom vent actuators are worked by a 240 V electric drive. This is the only non-LonWorks device but it is integrated by LonWorks compatible relays. The blind motor controller governs a 24 V dc motor mounted in the window rail. Wind speed is measured at the facade by a basic sensor. A high precision roof-mounted sensor provides more detailed information. Lighting is provided by high frequency ballast fluorescent lamps. An illuminance sensor with built-in functions for lamp life maintenance and daylight dimming was included. A directional light sensor is used to determine illuminance from the window wall and two simple, inexpensive sky illuminance sensors provide the lux level on the vertical plan of the facade. Temperature and humidity are measured inside and out. External sensors were roof-mounted facing north with a rain and radiation shield.Downloads
Source
Building Sustainable Design
Postscript
Mark Skelly is a research engineer at the Centre for Window and Cladding Technology, University of Bath. Martin Wilkinson is a lecturer at the University of Bath's school of architecture and civil engineering. Further reading Newsham G R, 'Manual control of window blinds and electric lighting: Implications for comfort and energy consumption', Indoor environment, volume 3, 1994. Rubin A I, Collins B L, 'Window blinds as a potential energy saver: A case study', NBS Building science series no 112, Washington, 1978. Bordass W T, Bromley A K R, Leaman A, 'Comfort, control and energy efficiency in offices', BRE Information Paper, IP3/95. Veitch J A, Gifford R, 'Choice, perceived control and performance decrements in the physical environment', Journal of environmental psychology, volume 16. Astrom K J & McAvoy T J, 'Intelligent control: An overview and evaluation', Handbook of intelligent control: Neural, fuzzy and adaptive approaches, Van Nostrand Reinhold.
