Autonomous Control

An intelligent sensing and control system, integrated within the physical architecture, allows for real time monitoring and reconfiguring of the states of key architectural elements, which hence become responsive. This capacity revolutionizes architecture, where usually tectonic elements are passive, or require actuation by the users.

A typical window, lighting or heating system requires direct control (switch), or low level control (thermostat), or timed programming. The control system of the prototype developed by MIT Model-based Embedded and Robotic Systems group (MERS manages the operation of the variable transmittance windows of the dynamic facade, the high thermal mass building envelope, and the HVAC and artificial lighting system, in response to the changing interior activities and the conditions of the exterior environment without requiring direct input by the users, the aim being to reduce energy consumption and to enhance living comfort.

To address the technical challenges of the control system Artificial Inteligence methods to building control were used in implementing an on-line risk-sensitive planner called p-Sulu (probabilistic Sulu). The control algorithm balances performance versus risk: energy efficiency versus thermal comfort.

P-Sulu leverages an Iterative Risk Allocation (IRA) algorithm to provide robust planning in the context of building management by taking into account a stochastic plan model, which specifies probabilistic state transitions in a continuous domain.

The controller fulfills three missions:

First, it adapts the management of energy to the activities of the residents by allowing the residents to input a detailed schedule and desired ranges of comfort and planning over these. This capability is called goal-directed planning with continuous effect. The language used to represent a detailed schedule with user constraints in AI terminology is called a CCQSP (chance-constrained qualitative state plan).

Second, it manages the HVAC system, the dynamic windows, and the thermal conservation capabilities of the envelope to limit the consumption of electricity. To make the goal of distributed control possible the control receives data capturing the dynamics of the house environment, and calculates a control plan that minimizes the use of all systems consuming electricity. In technical terms the controller achieves optimal planning of these systems.

And third, it provides a probabilistic guarantee that the resident’s comfort constraints will not be violated, by explicitly acknowledging the sources of uncertainty (such as the weather conditions and the human impovisation) and planning accordingly.

A wireless network of sensors serves the need for distributed monitoring of the interior and exterior environment in real time. The sensors support the decision-making layer and assist the management of the interior. Using this sensory data in combination with historical weather data as well as any user defined preferences, the optimization core manages the house devices, as an “ideal” house resident. A tablet-based user interface allows users to monitor performance and to override the automated controls if desired.

In parallel to optimizing sun heat, light, ventilation, and air conditioning at the house interior – by determining the optimum window configurations for solar radiation, ventilation and interior illuminance – the the control system applies generative rules of a formal grammar to specify a large variety of visual patterns on the electrochromic facade, based on both performance and visual symmetry properties.

Ono, M, Graybill, W, Williams, BC, 2012, “Risk-sensitive plan execution for connected sustainable home” , Proceedings of the Fourth ACM Workshop on Embedded Sensing Systems for Energy-Efficiency in Buildings , New York, USA, pp. 45-52.

Control Diagram 2_small