Smart Flow aimed at applying AI within process industry through using machine learning, simulations, industrial IoT and cloud technology, specifically in the case of district heating and the district heating network (DH-Net). The goal was to make a concept evaluation of AI for decision support in a core industrial process. The project has through a successful concept evaluation shown the technological possibilities and the economic potential within the energy utility application. Furthermore, the project has identified many challenges related to industrial use of AI.

Some of the highlights from the project include: • Successful development of AI algorithms that make detailed predictions of consumer and network behaviour. • An operational IIoT and Cloud pilot in the cloud that manages 15 000 meters daily and more than 3000 meters every 15 minutes. • A physical simulation model showing a potential energy saving in a single DH area of up to 4.2 MW. • Several articles in sector specific media, e.g., Fjärrvärmetidningen and Nordiska Projekt, sharing the ideas of the project. • Many event presentations of project results and experiences, including at Internetdagarna 2017, Microsoft and Dagens Industri AI event 2018, and Science and Innovation day 2019. • A film made by Microsoft that shows the concepts of Smart Flows and its application to utilities.

Mälarenergi has identified that many savings can be possible by optimizing the DH-net, and by getting a more detailed knowledge of how the net behaves one can make large improvements and developments. The keywords for Mälarenergi to continue to be successful is to focus on” the right amount of energy at the right time and place with the right quality”. By creating smartness, and tools for a better understanding, Mälarenergi will reach a higher level of decisions making and support for both the operators and data-analysis group. Instead of working with the production of heat such as the total amount produced from the Combined Heat and Power- Plant (CHP-plant), which is the traditional way to make forecast and heat demand, Smart Flow has focused on working with the consumers data. By trying to sum up the consumption from all the customers one can find another approach to know what to produce from the plant. Smart Flow demonstrated that using machine learning one can estimate how much energy is needed from the customers point of view. Moreover, the results show that it is possible to estimate the heat demand from different zones in the DH-net, using the same type of techniques. It was also found out that the time delays for different areas in Västerås are an important key for a better understanding of how the DH-net behaves. Indeed, knowing the time delays in advance, the operators will know when and how to react, to optimize the delivery of heat. A data-based methodology was developed, which can be used to estimate the time delays for different zones in the DH-net