Artificial intelligence is key to creating a truly smart building, as explained by professor Michael Krödel, CEO, Institute of Building Technology, Ottobrunn, Germany and professor for Building Automation and Technology, University of Applied Sciences at Rosenheim, as well as Graham Martin, chairman and CEO, EnOcean Alliance.
Today’s building automation systems operate ‘statically’ in response to fixed time programs or simple control parameters. For example, room temperature control is based on a preset temperature that is the same throughout the day. This is not truly ‘intelligent.’
The new dimension that AI can add into the building automation environment is to use autonomous analysis of the data as a basis for optimised operation. Thus the heating and cooling dynamic of rooms, weather forecasts, predicted room occupancy during the course of the day can all be factored into the operation of the heating.
All this – and much more – is possible when data on building system status and conditions is intelligently evaluated, which requires intensive processing of large amounts of data. AI offers many new, tailor-made solutions which are exceptionally suited to efficient building management.
The term AI is increasingly associated with buildings and building automation. The question is: what is it, where do its tangible benefits lie in this field, and how does the building infrastructure need to be adapted to realise those benefits?
The different ‘smart buildings’
Initially, building automation was comparatively un-intelligent. Systems were programmed to follow a set of simple rules, allowing for quick system start-up and subsequent ease of maintenance.
The smart building typically builds on this classic building automation with flexible IT-based management systems. These offer unrestricted programming using modern IT languages and tools, easy integration with other IT systems, such as workspace/room reservation systems or data banks, and data visualisation for facility managers and for ordinary users.
The growing assimilation of sensor-generated data into the IT-based management level opens the way for more advanced data processing solutions to come into play – such as AI tools. This is the pre-condition for the implementation of any prognosis-based form of building management. The sophisticated processing of sensor-generated data makes the smart building into a ‘cognitive building’
AI-learning process
The first step in any AI process is system learning. This can take three forms:
- Unsupervised Learning
- Supervised Learning
- Reinforcement Learning
Unsupervised Learning is used when large quantities of data must be processed and categorised. This grouping enables the recognition of deviations from norms and interdependencies.
Supervised Learning often makes use of neural networks. They consist of entry and exit nodes, as well as further nodes in the intermediate layers. Mathematically weighted relationships exist between the diverse nodes (neurons). In order to optimise these relationships, the neural network is subjected to a training phase with known input and output patterns.
Another form of AI is represented by processes that autonomously determine which actions are appropriate in a given situation. They emulate human behaviour whereby different solutions are tried in order to determine the best way forwards in a hitherto unknown situation, and conclusions drawn retrospectively. The learning task becomes more challenging when feedback is given much later and hinges upon events in the relatively distant past. This is true in a human context, and equally true in computer environments. The best-known example in this category is Reinforcement Learning.
It can be seen that these three approaches are complementary. The learning method should be chosen depending on the task in hand – each has its merits.
Concrete applications
Many diverse AI-based applications are available in the field of building automation. They can be broadly categorised as follows:
- Optimised facility management: needs-based control of heating plants, circulating pumps, lighting etc. (as opposed to control on the basis of simple parameters or by timer).
- Optimised utilisation of spaces and infrastructure: capacity analysis and forecasting, e.g. for
- meeting rooms, canteens, pantries, transit areas, toilets and parking spaces as well as the provision of information in the short term (for building occupants) and in the long term (for facility managers, e.g. in form of advice on building restructuring).
- Load management: forward-looking operation of electrical systems in order to avoid (costly) peak loads.
- Precautionary maintenance and optimised servicing: analysis of failure probability, timely maintenance and consequential avoidance of technical failures.
- Employee-oriented value added services: mobile devices can – for instance – be used to generate space utilisation forecasts, view canteen usage intensity, request parking space availability and preferred workspace location or select individual meals.
- Compensation of skilled-staff shortages: making effective use of facility maintenance staff in managing the building’s technical systems.
- Focus on meaningful sensor data: generate as much data as possible from as few sensors as possible – reducing redundancy, cutting investment and operating costs.
Demands upon system architecture
An AI platform is indispensable for the introduction of intelligent learning processes such as those described above. This can be either cloud-based or server-based. Cloud-based server farms offer more processing power, and cloud-based AI frameworks offer a broader range of features, so this currently represents common practice.
The AI platform is built on a smart building infrastructure, and all technical systems should ideally be connected to a BMS (building management system). The BMS must be able to govern the building facility and room automation systems.
Demands on building infrastructure
The AI platform requires a rich set of data from a variety of sensors around the building to operate effectively. Cognitive buildings store and analyse historical sensor data to make predictions for the future. For this reason, such buildings are even more critically dependent on the data inputs they receive for their success. They need to be equipped throughout with IoT sensor devices that make the algorithms fully aware of every aspect of their operation. The richer the data, the more intelligent the response of the AI.
Wiring sufficient sensors into an established building is hugely expensive – and even if it were done would create an inflexible architecture that couldn’t be adapted. The only effective solution is battery- and maintenance- free energy harvesting sensors that can be fitted in a moment and moved at will.
Conclusions
AI-based processes enable a broad range of applications in the field of building automation. The concrete benefits anticipated from AI-based solutions should be clearly defined before implementation, since this plays a determining role in the choice of learning process and its modelling, as well as in the choice of AI platform and the type, number and location of the energy harvesting sensors needed to supply the data inputs.
This article originally appeared in Electrical Review January/February 2021