M. PLOIX Stephane

Teachings

Since 2007, I have been involved in research into the management of energy systems involving human actors. This has naturally guided my teaching, which is all centered around digital technology, energy systems and the consideration of human actors. In particular, this has led to the development of a PARIN architect-engineer training program in which 2 students from ENSE3 and the Grenoble School of Architecture meet for 2 years (M1 and M2), one day a week, to learn how to work together, with a better understanding of each other's profession. I teach heat transfer and bio-climatic design. To achieve this, I developed an open-source project in Python open-source code (over 130,000 lines of code today, see https://gricad-gitlab.univ-grenoble-alpes.fr/ploixs/buildingenergy) to transform my teaching of energy systems. This project includes 11 interactive tutorials in the form of Jupyter spreadsheets: data acquisition, energy modeling (heat transfer, solar radiance model, PV production), parametric estimation on in-situ measurements, calculation of energy strategies, collective self-consumption, the lambda house concept dedicated to student architects to apprehend a place... as well as inspiring Python code examples for the realization of projects related to energy transitions. PARIN students obtain a certificate of competence from the other school. At the same time, I have developed a Specialized Master's degree entitled “Transitions Energétique et Environnementale des Territoires” (Energy and Environmental Transitions for Territories), which is aimed at professionals wishing to reorient themselves towards territorial management: transition scenarios, social and technological obstacles, interaction with decision-makers (politicians) and players in the field, mobility issues, energy flow collection, changing lifestyles (telecommuting, etc.), taking into account the differences between urban, peri-urban and rural areas.

I run 270 hours of teaching a year, including 90 hours of lectures with or without computer studies, 60 hours of consultancy and 118 hours of projects, which are characterized by the following keywords: smart-buildings, computer sciences, collaborative design and innovations, building energy and architectural design, energy management and flexibility, human involvement.

• Head of the TEET Specialized Master's program (Transitions Energétique et Environnementale des Territoires) since 2019
• Co-responsible with Olivier Baverel (ENSAG/ENSPC) for the PARIN Architect Engineer course (M1 and M2 level) since 2015
• Responsible for the 2nd year of the Energy Systems and Markets program at ENSE3 from 2014 to 2021

Research @G-SCOP/OSP/DOME2S team

The diagnostic techniques developed in a very general theoretical context during my initial research were adapted to the building context, where the problem is highly specific: the number of sensors is large, and behavioral models are often imprecise except in certain contexts. Solutions based on the first principles of diagnostics (BRIDGE approach), on the concept of validity and on expert tests whose conclusions are known only by the diagnostic aid system, were proposed (Houda Najeh's thesis).
Since 2007, I have been developing tools to dynamically generate energy management strategies that optimize the comfort (thermal, air quality, etc.) / cost (energy, euros, etc.) trade-offs in residential housing, using a Model Predictive Control approach. My research in this field has now shifted towards finding solutions that put residents at the heart of the development of good energy strategies, rather than in the loop. Indeed, in the context of the Solar Decathlon Europe competition and the implementation of an MPC-type solution, we were obliged to note the limits of all-automation in the context of residential housing. Just as residents sometimes have an erroneous understanding of energy phenomena, so the decision support system may have an erroneous representation of the complex energy phenomena that occur in an occupied home: radiative phenomena, advection, etc., but also unmeasured phenomena and equipment. As part of the ANR INVOLVED project, we explored the possibility of involving occupants by proposing tools for establishing their preferences, as well as interactive tools enabling them to develop step-by-step energy strategies for the day ahead. With the ANR LearningHome project, we are currently working towards an engaging approach in which residents guide the search for solutions by programming experiments based on facts, activities and intentions (Estefania Alvarez Cardoso Del Castillo's thesis). We have shown (Fateh Boulmaiz's thesis) that, unlike the optimization methods we initially proposed, such as the problem generators of Linear Integer Programming including an automatic linearization principle, suggestions for energy strategies can be made without an input-output model using case-based reasoning: the best energy strategies close to the current context are selected to search for the solution best suited to the current context. This requires not only annotation-type interactions, but also information on the quality of services perceived by occupants to synchronize the energy management assistance system, and machine learning classifiers to predict the future context and limit interactions with occupants to what is strictly necessary.

Various classifier/regressors have been adapted to the estimation of occupancy and occupant activity, either by supervised learning or by a knowledge-based approach using Bayesian networks (Manar Amayri's thesis). This work has led to the development of the concept of interactive learning, which aims to determine the best times to ask occupants about their activities. Simulation of occupant behavior has also been improved, on the one hand by multi-agent approaches (Ayesha Kashif's thesis) requiring detailed knowledge of occupants evolving in their environment, and on the other, by less knowledge-intensive approaches based on Bayesian networks (Khadija Tijani's thesis). Better integration of occupants was also developed as part of the INVOLVED project, with the development of generators of differential causal explanations (existence of a reference scenario) or not (Amr Alyafi's thesis). Different types of models have been explored: based on physical knowledge (RC and equational models) (Lisa Scanu's thesis), based on input-output relations or data models constituted by historical data. Tools for the automatic transformation of models based on physical knowledge into PLNE solvers were developed (Yanis Hadj Said's thesis) by studying archetypes of life zone models and the associated parametric estimation method, with in particular the development of a meta-optimization method (Audrey Le Mounier's thesis) enabling SQP optimizations to avoid slipping into physically non-significant trap zones. To validate the results, which point to the high sensitivity of a building system, we developed a hidden Markov chain model (Huynh Phan thesis). Nana Kofi Twum Dua's thesis illustrates the concrete problems of flexibility in vehicle load management. Whereas energy management used to be limited to sobriety issues, today, due to the growing share of flow energies, demand has to be adjusted via signals in order to maintain the production = consumption balance. With Mircea Stefan Simoiu's thesis, we took into account the flexibility (notably indirect) of energy use. With Mircea Stefan Simoiu, with whom I am still working under a subcontract on the SOFTEN regional project, we have developed tools to assess the emergence of virtuous behaviors in response to various types of signals, simple signals or signals with reinforcement, in the context of residential housing and a Bucharest metro station where users participate in the flexibility of uses. The SOFTEN project is in line with energy management issues involving human actors, since it aims to better manage ski resorts by finding the best compromise between reducing ski-lift consumption (by reducing speed) while maintaining skier satisfaction linked to ski-lift waiting time. In addition to adjusting lift speeds according to ridership, we are currently developing a new form of control: after control by involvement and control by commitment, we are going to propose control by influence, where attractive points (organization of events, positive information given to skiers) and repellent points are defined to better distribute the number of skiers on the ski area in question. Case-based control techniques (Fateh Boulmaiz's thesis) and reinforcement learning (Khoder Jneid's thesis) will be tested.