Design Solution · Digital & IoT
Design Solution · Dream about it
ML predictive control to sequence chilled water plant efficiently.
A digital model trained on plant data uses machine learning and predictive control to select the most efficient combination of chillers, pumps and towers for each load and weather condition. It continuously optimises sequencing rather than relying on fixed rules.
Training a digital model on historical plant data and using it to sequence chillers, pumps and cooling towers against real-time load and weather is a more sophisticated approach than conventional rule-based BEMS control, and the potential gains are real: chilled water plants routinely run at 30-50% of design load for the majority of annual hours, and fixed setpoint-and-sequencing rules are calibrated for peak conditions rather than those typical part-load hours. ML-driven optimisation that learns the plant's actual performance curves rather than relying on design assumptions can recover meaningful efficiency. The evidence gap is the entire record: no deployment data, no before-and-after kWh comparison, no COP or IPLV improvement figures, and no account of the plant type, age, or building typology to which this was applied. A specifier should treat this as a promising operational technology requiring a focused procurement, not a standard design specification — it needs data continuity (sensors, BMS historian), a commissioning dataset to train on, and a governance model for validating that model recommendations are overrideable when conditions fall outside the training envelope.
#machine learning #predictive control #chillers #optimisation