A handful of academic papers have appeared over the last two years applying model predictive control (MPC) to the problem of integrating mining waste heat into greenhouse climate systems. The premise is appealing: use a predictive model of greenhouse thermal dynamics combined with weather forecasts and electricity price signals to optimally schedule mining operation and heat delivery. Run the miner hard when heat is needed and electricity is cheap. Throttle back when the greenhouse is warm enough. Store excess heat when appropriate.

The papers read well. The simulations look promising. And if you are a small grower considering heat reuse, you will probably encounter references to this research at some point. So it is worth understanding what the MPC studies actually show, where their assumptions hold, and where they diverge from what happens in a real greenhouse with real equipment and real weather.

What MPC Means in This Context

Model predictive control is a control strategy that uses a mathematical model of the system to predict future behaviour and optimise control actions over a time horizon. Instead of reacting to current conditions (like a thermostat that turns on when the temperature drops), MPC looks ahead and plans.

For a greenhouse with mining heat, the MPC controller would consider:

  • Current greenhouse temperature and humidity
  • Weather forecast (solar radiation, outside temperature, wind)
  • Electricity price schedule (if on a time-of-use tariff)
  • Mining profitability conditions (Bitcoin price, difficulty)
  • Heat storage state (if a buffer tank is present)
  • Crop temperature requirements

It then calculates the optimal mining schedule and heat delivery plan for the next several hours, minimising some combination of energy cost, temperature deviation from the setpoint, and mining revenue maximisation.

In theory, this can outperform simple thermostatic control by anticipating a cold night before it arrives and pre-charging the buffer tank, or by recognising that tomorrow will be sunny and warm so overnight heating can be minimal.

What the Studies Show

The recent papers, primarily from Northern European university groups working on controlled environment agriculture, demonstrate several consistent findings:

MPC reduces temperature deviation. Compared to simple on/off thermostatic control, MPC maintains tighter temperature bounds around the setpoint. The simulations show 30 to 50 percent reduction in temperature excursion events (moments where the greenhouse temperature deviates more than 2 degrees from target).

MPC improves energy utilisation. By anticipating heat demand and pre-charging storage during off-peak electricity hours, MPC reduces total energy cost by 15 to 25 percent in the simulated scenarios. This assumes a meaningful difference between peak and off-peak electricity rates, which exists in some markets and not others.

MPC can coordinate mining and heating. The most interesting contribution is the dual optimisation: scheduling mining to coincide with both low electricity prices and high heat demand. In the best-case simulations, this alignment improves the effective mining profitability by 8 to 15 percent compared to running the miner at a constant rate.

Buffer tanks are essential for MPC benefit. In every study I have reviewed, the MPC advantage over simple control depends on having thermal storage. Without a buffer tank, the MPC controller has limited ability to decouple mining scheduling from heat delivery, and its advantage shrinks to marginal.

Where the Studies Fall Short

Idealised Greenhouse Models

The thermal models used in most of these papers are simplified. They treat the greenhouse as a single thermal zone with uniform temperature, which is reasonable for simulation but does not match reality. A real greenhouse has temperature gradients, cold spots near the walls, warm spots near the heat source, and stratification from floor to ceiling. These spatial variations matter for plant health and cannot be captured by a single-zone model.

Perfect Weather Forecasts

MPC depends on forecast quality. The simulations typically use historical weather data fed forward in time, which is effectively a perfect forecast. Real-world weather forecasts are imperfect, especially for the solar radiation component that dominates greenhouse thermal dynamics. A cloud passing over at the wrong time can swing greenhouse temperature by several degrees. MPC handles this through re-optimisation at each time step, but the practical benefit degrades as forecast quality drops.

Clean Hardware Behaviour

The studies model miners as a simple wattage input with a fixed relationship between power and heat output. Real miners have startup delays, thermal ramp times, hash rate variation with temperature, and occasional firmware glitches that cause restarts. Throttling a miner up and down on an MPC schedule means more thermal cycling of the chips, which accelerates wear. The papers do not account for hardware degradation from variable operation.

No Maintenance Downtime

Every real mining setup has unplanned downtime. A power supply fails, a fan dies, firmware needs updating, a duct joint comes loose. During any of these events, the MPC controller's plan becomes invalid. The studies run simulations over months or years of continuous, uninterrupted operation. Realistic availability for a small mining setup is 90 to 96 percent, not 100 percent.

Scale Mismatch

Most of the published work models installations with 10 to 50 kW of mining capacity feeding substantial greenhouse operations (500 to 2,000 square metres). The control engineering overhead of implementing MPC is similar whether you have one miner or twenty. For a small grower with one or two miners and 15 to 30 square metres of greenhouse, the engineering cost of a proper MPC system is disproportionate to the benefit.

What a Small Grower Should Take From This

The research confirms what practical experience also shows: thermal storage, combined with basic scheduling intelligence, significantly improves the value of mining heat in a greenhouse. You do not need a full MPC implementation to capture most of that benefit.

Practical takeaways:

  1. Install a buffer tank. This is the single most impactful change identified in the research, and it does not require sophisticated controls. Even with a simple thermostat-based charge/discharge strategy, a buffer tank smooths delivery and reduces waste. See our buffer tank guide for sizing and practical details.

  2. Use time-based scheduling if your tariff supports it. If you are on a time-of-use electricity rate, you can implement a simple version of the MPC scheduling benefit by running miners at full power during cheap-rate hours and throttling during peak-rate hours. No predictive model required. Just a timer or a simple script.

  3. Do not over-engineer the controls for a small setup. A thermostat, a buffer tank, and a bypass damper capture 80 percent of the MPC benefit for 5 percent of the implementation cost. Sophistication pays off at scale. At small scale, reliability and simplicity are worth more.

  4. Weather awareness is useful, not essential. Checking the forecast and manually adjusting your system before a cold snap or a warm spell gives you most of the anticipatory benefit that MPC automates. For a small operation, this is free and effective.

  5. The research validates the concept. If someone questions whether mining heat reuse in greenhouses is theoretically sound, the academic work provides rigorous support. The thermal synergy is real. The debate is about practical implementation, not fundamentals.

Looking Forward

As control hardware becomes cheaper and open-source greenhouse automation matures, MPC-style optimisation may become accessible for small operations. Several open-source projects are already implementing simplified predictive control for greenhouse climate management. Connecting these to mining scheduling is a logical next step, and I expect to see practical, affordable implementations within the next few years.

For now, the message is: the research is encouraging, the fundamentals are sound, but the practical path for small growers is still simpler controls with good infrastructure. Start with the layout guide, size your system with the calculator, and invest in a proper buffer tank before investing in control algorithms.