Section: New Results

Modeling activities to promote self-consumption of locally produced energy

Participants : Jean-Marie Bonnin, Alexandre Rio, Yoann Maurel [contact] .

Traditional electricity distribution schemes decouple the production sources from the consumers so that it is necessary to transport energy over long distances. This type of organization is illustrated by the consumption of region such as Brittany, where 91% of the energy consumed is imported. It induces inherent inefficiencies due to the line losses and the transformation steps and therefore induces a high infrastructure and distribution cost. To face these problems and in order to reduce the environmental impacts associated with the use of energy, recent years have seen the development of initiatives to produce energy locally.

The sources of renewable energies are good candidates for this because they are varied and adapt easily to the different geographical situations. The infrastructures necessary for their implementation also impose fewer constraints in terms of installation and safety. One of the main obstacles to the unique use of these technologies comes from their strong dependence on physical and meteorological characteristics, which makes it more difficult to foresee production capacities. These characteristics vary from one facility to another and from one region to another. The combined use of these technologies therefore appears to be necessary to ensure that there will always be available energy at the lowest possible cost. In this context, OKWind proposes to deploy self-production units directly where the consumption is done and has developed expertise in multi-source energy production (see section 8.1).

In 2016, we started to study a solution favoring maximum autonomy of the instrumented sites from the traditional channels energy production by modeling business processes and using learning algorithms to shift demanding activities according to local production capacities. For example, the system should be able to anticipate a potential consumption of hot water (and thus of the energy needed for its production) in order to produce it at the best time when the renewable energy is available. It should also choose the best storage solution for this energy: hot water could be directly stored by the heat pump for instance. The system must implement policies that will intelligently shift demanding activities according to the predictions of energy production. It thus requires:

  • capabilities to predict the production of energy. A lot of theoretical work has been done in the literature to predict the production of renewable sources of energy. In addition, in order to evaluate the production of energy and its consumption over time, OKWind has developed data retrieval mechanisms on each deployed sites. They produce accurate statistics on production and consumption. Both approaches should be used as inputs of our decision processes and model. One of our goals is to evaluate the precision of the theoretical prediction models against these real-world data to determine which are the most relevant for the implementation of our approach.

  • capabilities to model the consumption on energy. Numerous works of the literature are interested in similar problems but focuses mainly on building electricity consumption model of machine tools [10]. We propose to focus instead on activity and business processes. In a related domain, modeling work has been conducted on water consumption of farms [7]. The objective was to predict the water consumption of an operating farm by modeling business processes. Our goal is to propose a similar model for electricity targeting a broader scope of economic sectors.

  • capabilities to schedule activities in order to match production and consumption so as to promote self-consumption. This requires developing control loop that will proactively analyze and predict consumption and take measure to shift demand. This can, in a first approach, be done by assisting the consumers and providing them guidance on when to perform certain tasks. Assisted demand shifting have already been developed for the residential domain [6] but this project focused on uses mainly and little on the modeling of business processes. Ultimately, we would like to develop automated process transparently when possible. The learning algorithms will be developed in collaboration with Ubiant (https://www.ubiant.com/en/about/), a company specialized in artificial intelligence to smart-buildings.

To validate the approach and to understand business processes, we have started a field study targeting two types of activities (e.g. farm or hotel). We also want to develop tools to simulate a site so that we can quickly evaluate our policies over simulated long periods of time.