Three questions at the start of the operating day
Anyone running a cogeneration plant that serves a distribution or district heating network faces three questions every morning, and each needs an answer within a few hours. What will heat demand be over the coming days? Which units should stay in service, and at what load, hour by hour? How much electricity should go to market, and at what price?
In many organisations, those answers take shape in three separate places. The forecast lives in a spreadsheet built up over the years by someone who knows the building stock being served and how it responds to the weather. The production plan comes from the shift supervisor’s experience and the previous season’s history. The market bid is put together by the trader, who adds a safety margin to cover uncertainty across the rest of the chain.
Taken one at a time, the three answers are almost always defensible. Their combination is not, because none of the three functions sees it in full.
How the three decisions are linked
The decisions are linked because the physical system they act on is linked. Heat produced through cogeneration comes with electricity, and the price in the hour that electricity reaches the market determines whether that heat was expensive or cheap. In turn, the volume available to bid depends on forecast heat demand.
A forecasting error also spreads beyond the forecast. It turns into fuel burned for a load that never materialised, or into an operating set-up tuned to price conditions that have since changed. And since a CHP unit has a minimum stable load, start-up times and ramp rates to respect, adjusting the set-up once the day is under way carries a cost that no spreadsheet reveals.
These constraints are physical and non-negotiable: minimum loads, ramp rates, start-up times, storage capacity, supply and return temperatures. Decision-makers who don’t consider them all at once leave margin on the table. That margin never appears in any report, because it is the gap between the solution adopted and a better one that nobody ever calculated.
Calculating it means answering one precise question every day: of the units available right now, which combination minimises cost over the next eight hours while meeting network constraints and service levels? Without an optimisation tool, the answer rarely arrives in time, because it requires forecasts, technical constraints and prices to be handled in the same calculation. That is the job of a digital twin of the system.
The digital twin of the generation system
The Digital Twin is a structured representation of the generation system. It covers the units with their technical constraints and efficiency curves, the energy flows between plant nodes, the incoming fuels, the connection to the power grid and the markets, and the incentives and tax items that make one set-up more cost-effective than another. Where there are multiple sites, the model includes them all, along with the constraints that link them. Continuously changing data flows through this structure: geo-referenced weather forecasts, prices and field measurements.
The operational advantage is that the same model answers questions across very different time horizons, from the setpoints for the next four hours to next year’s budget, right up to replacing a boiler over several seasons. The same applies to configurations that don’t yet exist, when two plant design alternatives need comparing before an investment is approved.
Demand forecasting
Heat demand forecasting is a data problem before it is a question of experience. It depends on temperature, solar irradiance, wind, time of day, user type and day of the week, in combinations that vary from network to network.
OptiEPTM includes weather-driven load forecasting models with a multi-day horizon and automatic updates, backed by continuous performance monitoring through statistical KPIs and dedicated reporting. Errors are measured as they emerge, rather than discovered at the end of the season.
Where an organisation already has an established forecasting system, the platform integrates it and works from it. The value lies in connecting the forecast to the decision that follows.
Dispatch optimisation
The optimisation engine runs on the model. It calculates setpoints for each unit at a granularity of fifteen minutes to one hour, respects the fleet’s technical constraints and minimises operating costs over the selected horizon. The plan is delivered in two forms: one readable by operating staff, and one as files ready for the PLCs, so the automation system can take it on board with no manual steps.
Control stays with the plant operator. In copilot mode, the system proposes and the operator checks. When conditions change, re-optimisation returns an updated plan, while what-if functions make it possible to simulate a scenario before applying it, with technical and economic KPIs side by side to weigh up the trade-offs.
A feature currently in beta uses GenAI and LLMs to explain the optimisation engine’s choices and to query plant data and documentation in natural language. It proves especially useful in the first months of operation, when the control room keeps asking why the system recommends one set-up over another.
Putting flexibility to work on the markets
A district heating network has flexibility that is often underused. Thermal storage and the thermal inertia of the buildings served allow production to be shifted by a few hours without users noticing any change in service.
That flexibility has value on the electricity markets, and the trading module turns it into bids. It starts from the optimal production plan, calculates the margin available within service constraints and builds bids for the day-ahead and intraday markets. OptiEP™ runs these calculations automatically, within technical constraints, and aligns trading decisions with plant operations.
Budget, emissions and investment planning
The annual energy budget is usually a separate exercise, built on its own assumptions and revised once a year. With the digital twin already in place, it no longer has to be: multi-year simulation runs on the same constraints that govern day-to-day operations, and CO₂ emissions are monitored and projected within the same framework.
When a company falls within the consumption thresholds of the Energy Efficiency Directive (EED), or has to meet national renewable-energy requirements, the mandatory audit calls for savings opportunities to be identified. But the audit stops at the list. Quantifying those opportunities, ranking them by impact and testing them against several price scenarios requires a model of the system, and the digital twin already is one.
When the decision becomes which plants to build and in what sequence, the system design optimisation module comes in. Working on the same model, it returns the optimal long-term asset portfolio, the investment plan set against the emissions trajectory, the decommissioning sequence for existing units and the assessment of any network expansion. Operating decisions and capital decisions converge on a single model, while at many utilities they still run on separate tracks.
How the project is set up: from impact assessment to operation
The journey starts with an impact assessment, before anything is installed: six to eight weeks working on historical data, during which we build the digital twin, rerun the past year of operations with optimised dispatch and measure the difference. At the end, you have a quantified economic result, an emissions analysis, a forecast accuracy analysis and a roadmap with the IT requirements for going live.
If that result is convincing, the second phase takes the solution into operation in another six to eight weeks, with timing that depends on data availability and the type of integration required: model calibration, connection to field and monitoring systems, data exchange, automation, reporting and go-live.
One final point concerns the project teams, and it explains why these initiatives reach production instead of stopping at a report. The optimisation model is built by people who know the application domain. The software is developed by the people who built the model. And consultants, operations research specialists and developers all work in the same team at Optit.
Two applications, one method
At Malpensa, moving from experience-based operation to optimised dispatch improved the energy management margin by more than 20%. In New York, for Con Edison, the same approach produced a digital twin of the city’s steam network, used to assess decarbonisation pathways over decades-long horizons. The next step is to bring it into day-to-day operations.
Daily operations in one case, long-term planning in the other: the method is the same, keeping forecasting, production planning and market bidding together. That is what we mean by Decision Intelligence.
Want to know how much margin your plant is leaving on the table? Get in touch: we’ll start from your historical data.
