PERSUASIVE E-CONSULTANT FOR BUILDING ENERGY MANAGEMENT – INVOLVED

ELITHIS has designed and is building the ELITHIS Danube Tower, the first in Positive Energy Housing. The aim is to build more positive energy housing towers. The energy pledge to investors and users is very strong and strategic for the company. In an apartment building, the occupants and operators are the main actors involved in reducing energy consumption. Hence for ELITHIS the e-consultant developed by INVOLVED is the means to achieve the ambitious energy targets set for its buildings. It is inscribed within the framework of a series of positive energy housing towers for which energy consumption must be controlled. ELITHIS prepared the integration of the e-consultant and carried out dynamic thermal simulations for partners working on energy management modelling (CEA / INES, G-SCOP, LIG / PRIMA). Vesta-System and ELITHIS have been working on the inclusion of previous results, aware of what the Danube Tower context allows, especially with a less extensive instrumentation than that of modern experimental platforms. An analysis of the Vesta-System HMI was conducted by the EHCI team in LIG and Vesta-System has proposed changes to its HMI. The Vesta-Energy computer engine, a software dedicated to building energy management, was revisited to prepare the integration of new features making the commitment easier: mirror indicators and replay tools. Vesta-System is also working on the automatic generation of models based on architectural data as well as tools allowing effective deployment in a building of this magnitude (16 floors, 65 units, 3-4,000 checkpoints ...).

The PACTE laboratory, focusing on human and social sciences, conducted a literature review on energy consumption practices in housing and an analysis of issues reported in the specification documents of the Elithis Danube Tower (TED) project, supplemented by interviews with key designers of the tower. This generic (literature) and site specific (TED) knowledge was then mobilised in a novel methodology to develop, with Elithis, a range of indicators of community issues at different scales to be taken into account by the e-consultant to encourage involvement. CEA / INES and the G-SCOP laboratory worked on a modular and easily configurable modelling system. For easy configuration, the behavioural model settings used in the resulting energy management system should be easily learned through parameter estimation. Two approaches have been tested: direct exploitation of the physical model using non-linear optimisation techniques with SQP approaches and meta-heuristics, as well as automatic linearization methods leading to linear regressions where convergence is fast and guaranteed. The first approach is broader since it incorporates more knowledge while the second is faster and offers a guarantee of convergence. In collaboration with other partners, the G-SCOP laboratory has developed different learning systems for the estimation of occupancy as well as usages and practices: an interactive approach based on decision trees where the occupants are asked about occupancy and actual activity when the information gain is significant, and a knowledge-based approach using questionnaires formalised through Bayesian networks. The latter does not require occupant returns but remains less accurate.

Following a major bibliographic study, a first differential explanation system was developed. It is based on a comparison between a realised situation and an optimised situation relying on a model. Observed differences are rendered qualitative and a contextual causal explanation is dynamically generated in order to facilitate an understanding of the effects of the occupants’ actions. The representation of the explanations is studied in association with the Engineering Human-Computer Interaction LIG team. Meanwhile, new usage indicators have been developed and are being tested. The EHCI team in LIG is studying the persuasive nature of the artefact in support of human-machine interaction. The aim is to better control the levers that allow users to understand the physical phenomena and act better, that is to virtuously and permanently change their behaviour.

The mirror part of the interface is based on occupancy and living practices estimation tools. An approach based on learning supervision was published in the October 2016 issue of Energy and buildings. The modular modelling process appeared in the International Journal of Advanced Computer Science and Applications in 2016. Regarding the persuasiveness of HMI, a problem space identifying properties to consider in designing persuasive systems was presented at the ACM EICS'16 conference in Belgium. MOST4P, a decompositional approach increasing the probability for the user to change their behaviour, was also designed and shown at the HCI Conference 2016, UK in 2016.

The project focuses on bringing solutions to occupants, building owner and managers for upcoming challenges in energy management in residential buildings. Many technical solutions have been proposed so far, but without considering sufficiently actors as keys. It is generally considered that energy management can be done by measurement and computation means with few contributions of actors. The project explores a new paradigm: a user centric energy management system, where user needs and tacit knowledge drive the search of solutions. These are calculated thanks to a flexible energy model of the living areas. The system is personified by energy consultants with which building actors such as building owners, building managers but also occupants, can interact with in order to define energy strategies, benefiting of both assets: tacit knowledge of human actors, and measurement with computation capabilities of calculators. Putting actors into the loop is the needed step before large deployment of energy management solutions. It is proposed to develop interactive energy consultants for all the actors, which are energy management aided systems embedding models in order to support the decision making processes. MIRROR (interactive monitoring), WHAT-IF (interactive quantitative simulation), EXPLAIN (interactive qualitative simulation), SUGGEST-AND-ADJUST (interactive management) and RECOMMEND (interactive diagnosis) functionalities will be developed.
To construct these interactive energy consultants, researchers and engineers from companies propose a collaborative project. ELITHIS is a building designer in relation with building managers. VESTA-SYSTEM is a company that develops building energy management systems. Researchers from CEA/LITEN will develop behavioral models for living areas targeted by experimentations. PACTE will study actor behaviors, identify their tacit knowledge, needs and assesses different interaction models. LIG/IIHM will investigate the engineering of human computer interaction for supporting the INVOLVED features (i.e., MIRROR, WHAT-IF, EXPLAIN, SUGGEST-AND-ADJUST and RECOMMEND). LIG/PRIMA and G-SCOP will develop algorithms to support interactions with and in between actors for energy management.