Sarah Amrani
Could you introduce yourself in a few words?
I am Lisa Rivalin, a Research Scientist in Applied Statistics, building energy in Berkeley. I work as a Data Science Project Manager at ENGIE Axima, a leading French Company in Climate Engineering. I created a research partnership with Lawrence Berkeley National Laboratory (LBNL), where I am currently based as an affiliate research scholar working on algorithms for “Smart Buildings.”
In addition to my primary research project, I have two other missions for Engie Axima: to collaborate with customers on applying research results to their use cases and to Identify and evaluate innovative start-up companies in my domain to forge partnerships.
What inspired you to choose this research topic?
I received a Ph.D. in Applied Statistics and Building Energy from MINES ParisTech (PSL Research University). My research focused on statistical methods to predict building energy performance before construction for accurate Energy Performance Contracting. The goals were to design a comprehensive study of potential sources of uncertainty in the building modeling process and to create a statistical tool to predict energy performance and identify parameters to monitor given those uncertainties.
Engie Axima funded this research in a “CIFRE” context: a French Ministry agreement enabling a company and a lab to work closely together. I spent most of my time in the company, working inside the market and close to the ground.
During this time, I noticed that Engie Axima’s automation team was collecting a significant amount of data to monitor clients’ buildings. In this context, I realized that we could use those data in real time to optimize the building controls, make predictions, and enhance the comfort while reducing the energy consumption.
As I enjoyed working both in a lab for academic and rigorous research and in a company to apply it in a real context, I suggested creating a research partnership with LBNL.
Could you describe your current research?
My current research deals with how to set up models for Model Predictive Control (MPC) which is a strategy that utilizes historical measurements of the building to predict building performance and optimize control, given constraints, and objectives. These models are created by combining physical and data-driven modeling and are used to optimize building control to reduce energy consumption and cost while improving occupant service. It also helps the integration into larger energy systems such as electric grids or thermal networks.
One of the challenges to be addressed is the implementation cost, particularly labor time and expertise required for the implementer to set up the model. Another challenge we are facing is the bad quality of the data we collect. To make good predictions, you need good data! This is thrilling because we are working from the sensors to the algorithms and models to make predictions.
What problems did you encounter during your PhD thesis?
As I was both working at a lab and a company, the main challenge was to continually adapt my research to fit both the company’s need for solving industry problems and the academic requirement to produce high-value science.
Moreover, the timing was tight because the statistical process and tools that had to be released at the end of my PhD., were to be used by a whole engineering team to create the energy performance contracts. It is great because now I am sure that the result of this work is still used in a practical business context.
Do you prefer working in the private or public sector?
I don’t have a preference, as long as the research has a practical application and we have enough resources to explore the topic deeply.
As I also have a Philosophy of Science background, I enjoy communicating about my research through publication in peer-reviewed conferences and journals. It is a great way to make new connections and discover tools and methods that can be useful for current or forthcoming studies. Most of the time, it is easier to make some time to participate in conferences and be active in research communities in the public sector. However, some companies, especially in the U.S., have excellent research groups and publish a lot of high-quality content.
Why did you choose to move from France to the USA? What was the process?
I have always wanted to work abroad to have a new experience and discover a new culture. I chose Berkeley, California because LBNL is one of the labs with the most publications in my specific domain.
As I was invited as an affiliate researcher in a national lab and employed by a French company, the lab took care of the visa process, and the company took care of the health insurance and other administrative parts.
What differences is there between working in France and in the US?
I had great experiences in both countries, both in labs and companies but I noticed a few differences. The American work environment is more dynamic and flexible: decisions are taken quicker than in France and the idea can come from any employee whatever his position is. As long as the work is done on time, you can, in most of the cases, work when and where you want. To the contrary, in France, companies are stricter on physical presence at the office at specific times, but it prevents from finishing too late.
Social system and labor laws strongly protect the employees in France, which makes the jobs appear more stable and offer more comfort such as vacations. The Bay Area is a very dynamic and competitive area which is very stimulating but leads to a strong turnover. It is surprising from a French perspective where people tend to stay in the same company as long as possible.
What scientific personalities do you follow?
Etienne Klein, who was one of my teachers in Philosophy of Science is one of the few personalities from whom I buy books as soon as they come out. I was also passionate about Stephen Hawking’s vision of the world. Apart from that, I love listening to Ted Talks, and I follow many different feeds related to mathematics, statistics, and data science such as Data Science Central or Simply Statistics.
What scientific events do you recommend?
In my domain, IBPSA’s Building Simulations, ASHRAE’s Building Performance Analysis Conference and SimBuild conferences are some of the largest conferences on this topic.
I would also recommend Haystack Connect related to the Haystack Project which is a great initiative to have a common semantic for Smart Buildings.
Find more about Lisa and read her publications on her profile!
