AI for multi-energy systems decision-support  

Any multi-energy system has an energy production mix, which could be defined as the combination of the different energy production technologies that are able to satisfy the required energy demand.

Specifically, in thermal multi-energy systems, the control of heat and/or cold production provides a range of possibilities for achieving improvements in the efficiency of this production.

In order to carry out this control, it is necessary to know the different energy production technologies available, the price of electricity and the different fuels used in the system, the capacity and availability of the installed energy storage, the demand predicted and the facility must have a real-time monitoring system that was capable of characterising the installation by using the measured data.

In I-NERGY project, an Artificial Intelligence tool dedicated to decision-making in multi-energy systems is going to be developed, and for this purpose, the Reina Sofia hospital in Cordoba is used as a pilot facility, whose ESCO is Veolia, one of the leading companies in the world of energy systems, involved in the I-NERGY project, as well as in many other research projects at a European scale. Specifically in this project, and for the corresponding use case, number 5, the services associated with the project are:

  • Service 10: Energy load/consumption/demand forecasting
  • Service 12: Optimisation of the production mix

A system is needed that runs autonomously and is capable of collecting and analysing all the information and making decisions at all times, ordering which production and storage technology or technologies should be used and how to distribute the required demand among these production technologies in order to achieve the greatest energy and economic savings, as well as in terms of emissions. All these features can be offered by an AI tool, capable of analysing the data collected, making decisions and carrying out automatic learning to improve decision-making, predicting demand and improving the energy mix of the thermal system.

The pilot used in this project, the Reina Sofía Hospital (HURS), consists of a multi energy system that provides steam, heating, cooling and Domestic Hot Water (DHW) to the different buildings of the hospital. There are a total of 12 substations. The thermal plant of the hospital has 4 heating boilers, 2 steam boilers, 6 chillers and 2 fan coolers. This plant can produce 9MW of heating power, 4 MW of steam power and 11.5 MW of cooling power. Besides the equipment installed in the thermal plant: to produce heat, in 2 substations there are heat pumps and solar thermal; to generate cool, in 2 substations there is an individual chiller; and to supply electricity, there is a PV plant for self-consumption with a power of 1.7 MWp. There are a total 131 variables available to monitor the facility. These variables allow I-NERGY to develop the tool that optimises the production mix.

In order to determine if the pilot achieves the goals, the following user stories of UC5 are considered:

  • Account demand forecasting
    • Economic benefit optimization
    • Energy consumption reduction
    • GHG emissions reduction
  • Energy consumption reduction
    • Cost and GHG emissions reduction
  • Optimization of the production mix: match the demand to the generation
  • Improved operation and maintenance of the network
  • Energy efficiency increases
  • Advanced facility analysis
    • Behavioural usage prediction
    • Error detection of facilities
    • Advanced reporting

To quantify the user stories, the following Key Performance Indicators (KPIs) are defined:

KPI Definition
Demand forecasting accuracy The Mean Absolute Percentage Error (MAPE) between actual and forecasted demand. The evolution of MAPE will be monitored through a moving window, observing how the MAPE value evolves, verifying the solution reduces its forecasting errors as time evolves.
Energy savings The difference between measured and reference consumption data, evaluated within a predefined period of time.
Reduction of operation and maintenance costs The difference between measured and reference data about operation and maintenance costs, evaluated within a predefined period of time.
Reduced difference between generation and demand The difference between measured and reference data about operation and maintenance costs, evaluated within a predefined period of time.
Increase of RES contribution Evaluate the increase in the use of RES for the energy production.