What to do when a sensor cannot be physically implemented?
Mass deployment of sensors is not always possible. In some cases, or remote locations, it is too expensive; in others, there is not even a network. In oil & gas distribution sector, tanks are often on the customer property (or the tanks are owned by the customer), installing a sensor is not even possible. The customer may refuse it or want to remain fully independent of the supplier.
In that case, companies rely on reliable lists. From excel spreadsheets to historic data stored in their ERP, those lists remain only partially valid. Customers may often choose to purchase from a competitor (therefore there is a lack of data), or change their consumption (add a new pool, renovate the house, change the heater, etc.). Experience has shown that a customer is willing to purchase oil when his tank is between 20% and 40% full. Below it, he probably already ordered, and above, he will most probably wait until the tank is lower. Therefore, pinpointing the right time to call a customer becomes very important.
Big Data and Machine Learning in your Oil Tank
Brain-IT developed an algorithm that simulates a sensor. It is, literally, a virtual sensor. Using various data, we can predict the energy consumption of a unit in a confined climatic zone with a very high degree of precision (less than 3% of error on average). This could be for a heating or a cooling unit whether it is gas, oil or any other source of energy.
In the case of AVIA (with oil tank), we have on average an error of less than 300L on a 10’000L Tank. This way customer service agents can predict when a customer will run below a certain % in their Tank and they can call them at the perfect time to motivate the customer to make a purchase.
How it works
Based on the norm SIA 380, our module generates a consumption model of the thermic need of a building without any sensors. We use degree day unified data, meteorological data (wind, temperature, pressure, etc.), GPS coordinates, altitude, and historical data (what the customer has consumed over a certain time). It allows us to visualize the consumption of a building (customer) and predict its future need and evolution of its tank. In the same way as with physical sensors, we can also monitor and alert in case a tank is going to run out of oil. We can also help select which customers should be called based on various criteria.
The algorithm is self-learning. With at least 2 deliveries of history and the information on the real level of the tank (%), the algorithm starts to become effective. The more data and the more regular a customer, the more reliable it will be.
About AVIA – an energy provider
In 1927, independent Swiss importers of petroleum products joined forces and founded the AVIA Federation in Switzerland. 89 years later, its 11 member companies are at the head of a vast network of Swiss service stations, consisting of about 600 stations and more than 100 shops under the AVIA brand.
