BI and AI help Tornberget cut energy costs

“Tornberget, like many other property owners and facility managers, are doing whatever they can to cut energy consumption as the cost of electricity and energy is becoming dangerously high. Getting detailed and correlated data on historical usage, energy prices and weather data helps you better understand your energy costs and what can be done to reduce them. The Business Intelligence System that we have built for them provides both historical knowledge and future insights, which will help Tornberget understand what drives costs and how they can optimize energy consumption and heating”, says Clinton McCabe, Head of Data and Analytics at Precio Fishbone.

Escalating energy costs requires up-to-date intel at hand

Tornberget Fastighetsförvaltnings AB are owned by Haninge municipality, in Stockholm, Sweden. The company builds and manages a wide range of properties in the municipality such as schools, care homes, leisure parks, libraries, sports facilities, offices and more. Their total portfolio comprises roughly 300,000 square meters spread across about 250 buildings and facilities, most of which are owned by Tornberget.

Previously, Tornberget used a series of Excel sheets with a rather rudimentary analytic model to analyze the properties’ energy consumption. The underlying data were gathered manually from energy providers and Tornberget’s own metering systems. The work was time-consuming and the solution had very limited forecasting capabilities and scalability.

Tornberget needed to get a better grip on their energy consumption and all the interacting parameters that determine the total cost of energy. They wanted a readily available system that can source the data automatically and provide their controllers, property and facility managers and other users with constantly updated charts and actionable insights. Having this knowledge at their fingertips they could then act faster on historical deviations and future scenarios in order to optimize scheduled heating and implement other process and technical improvements designed to cut energy costs.

“We met with Tornberget in May 2022 to discuss their challenges, needs and requirements. After a few very productive workshops with Tornberget’s key users and decision-makers, we outlined a highly scalable, user-friendly, cost-effective and cloud-based analytical platform. The solution we decided to build is primarily based on Microsoft Azure and Power BI”, Clinton explains and adds:

“The choice of technology depends, among other things, on Azure’s outstanding scalability, interoperability, cost-effectiveness and ease of deployment of hybrid cloud solutions. Power BI is one of the most flexible and powerful Business Intelligence solutions and it’s very easy to use. If you’re familiar with Excel, you’ll have no trouble implementing Power BI.”

Automatic retrieval of fresh data

Clinton and his team of developers began building the solution at the beginning of June. The first version of Tornberget’s Energy Consumption BI System was launched by the end of August. The solution sources data from energy utility providers (mainly electricity companies) and Tornberget’s energy consumption meters installed in their properties via the energy data collector and aggregator Metry. Complementary data, such as additional relevant information about each property and facility, are also sourced from weather data suppliers and from Tornberget’s internal databases.

The information that Metry provides includes factual consumption data retrieved from both smart and analogue meters. The data also include historical deviations and alarms, depending on meter type and model, as well as a selection of metadata about the meter and the building it’s installed in, such as building area and the meter’s location. Depending on the smart meters’ functionality and the energy providers’ services most of the consumption data collected are no more than an hour old. Tornberget’s ambition is to have access to all consumption data on an hourly basis.

Additionally, Metry collects utility data from Tornberget’s energy providers, for instance the price of electricity per hour, day or month as determined by Tornberget’s energy subscription plans. In this first version of the BI solution, the information that Metry provides includes electricity and district heating data, but it can just as well be consumption data collected from water, cooling and gas meters. It can also include data gathered from just about any type of sensor installed in a facility, like temperature measurements or sensor data from heat pumps and solar panels.

“Work is underway to include data from other energy sources, like solar panels. Their vision is to have all energy and heating systems accessible in the BI solution”, Clinton underlines.

Deviations call for action

Currently Tornberget can do similar analyses and generate the same fixed reports as they used to in their previous Excel-based solution. However, today it’s done with just a few mouse clicks and the data is always up-to-date. Getting the same knowledge out of the old system took days and much cost-intensive work.

Overview of Tornberget’s energy consumption over the last 12 months compared to previous years.

“The solution’s dashboard provides users with a very clear and straightforward overview of Tornberget’s entire properties portfolio with its energy consumption and savings performance. They can easily check and make comparative analyses of the current or historical consumption for the last 12 months, a specific year, month, week, day or hour. And they can narrow it down to a specific building, facility or meter. With the weather data brought into the system, the so-called Heating Degree Days are also made available, providing Tornberget with a better understanding of consumption deviations”, Clinton concludes.

Tornberget have chosen to set an automatic deviation detector in the system, identifying consumption deviations exceeding 25%. This means that users will be alerted on any consumption that is, during a specified time period, 25% higher than any other comparable period of time. When deviations of this magnitude are identified, investigations are carried out to find the underlying causes.

Predictive modeling leads to cost-optimization

The first version focuses on historical insights, but with the ability to source data down to the hourly level and make use of weather forecasting data, Clinton and his team are looking at enhancing the system’s predictive capabilities. This would, for example, enable Tornberget to program their scheduled heating systems so as to cost-optimize heating based on electricity price and weather projections.

“In future versions we plan to upgrade the system’s forecasting capabilities and implement What if- algorithms. We can make use of a lot more data and Azure’s machine learning functionality to build an advanced predictive model, which will enable Tornberget to better plan ahead by adopting forecasting scenarios”, Clinton concludes.