Our products find applications in the field of energy.
These are few applications:
1) Assessing thermal performances of buildings
This work is carried out in collaboration with ACTIS – EMM (European Multifoil Manufacturer). It breaks with the state of the art in deriving thermal performances from in situ testing thanks to a breakthrough technology. This work has lead to the participation in the European CEN/TC89/WG13 standardization committee and in the international ISO/TC163/SC1/WG16-ISO9869-1 standardization committee.
On this example, the difference between internal and external temperatures is simulated from the power consumption and some meteorological information.
One can observe a good agreement between the measured data and simulated temperature difference.
This is the basis of the Adagos patent (WO2014147148A2) for thermal building assessement.
2) Wind farm long term power prediction
in order to predict the wind speed in a local point, we use DeepROM, our neural network generator, to build a model that compresses the global weather (area of 300 Km²). Once we have an accurate estimation of the wind speed, we use DeepROM once again, to build a second model capable off reproducing the non linear function that relates the wind speed to the power generated by the wind turbine.
Thanks to this method, we are able to predict the power for a period of 42 hours in advance. The following charts show our results compared to three other providers.
The first chart represents the sum of all the power generated by all the wind turbines in the wind farm. the second one represents the absolute deviation from the reference.
3) Luminous intensity forecast
In order to predict the production of solar power plants, it is important to be able to predict the light intensity at a given point with a good accuracy.
In this context, we used the data from the RAPACE Receiver (Automatic Receiver for Whole Sky Acquisition) deployed by the Atmospheric Research Center in Lannemezan (65). This receiver takes pictures of the whole sky with a time step of 5 min.
Using deepROM, we created a model to determine the evolution of luminous intensity at a point in the center of the image. After learning over a period of 3 months of observation, the validation of the model was carried out over a month.
The model can now be used, on demand, to accurately predict luminous intensity over a 1-hour period from previous 2-hour data.
4) Local rainfall forecast
In order to predict the rainfall intensity in a local point, we use deepROM to build a model that compresses weather radar data: area of 100 x 100 km, centered around the location of the forecast.
Using deepROM again, we created a model to determine the local rainfall intensity based on these compressed data. After learning over a period of 4 months of observation, the validation of the model was carried out over a month.
The model can now be used, on demand, to accurately predict rain intensity over a 2-hour period from previous 4-hour data