Advanced Control, MPC, and Virtual Sensing

When to use this

Use this skill when NeqSim's existing dynamic/process model needs a controller-facing model, bounded receding-horizon control, a physics-based soft measurement, or steady-state data reconciliation.

• ProcessLinkedMPC links configured manipulated, controlled, disturbance, and state variables to one existing ProcessSystem; it identifies local dynamics and computes/apply moves.
• ProcessLinearizer, StepResponseGenerator, StateSpaceExporter, IndustrialMPCExporter, and SubrModlExporter create model-identification and integration artifacts.
• ModelPredictiveController is a controller-device implementation with a first-order prediction model, constraints, feed-quality prediction, and optional moving-horizon estimation. It is not a DCS emulator.
• VirtualFlowMeter and SoftSensor estimate quantities/properties from a stream's fluid and supplied operating conditions. They are screening/model estimates, not custody-transfer instruments.
• SteadyStateDetector and DataReconciliationEngine screen input data and reconcile measured values against declared linear balance constraints.

Keep each layer distinct: dynamic simulation models process response (neqsim-dynamic-simulation); controllability/operability maps feasible envelopes (neqsim-controllability-operability); calibration and plant-data workflows select and qualify measurements (neqsim-model-calibration-and-data-reconciliation, neqsim-plant-data). Compressor minimum-speed/recycle coordination belongs to neqsim-compressor-antisurge-recycle.

Class map

Build pattern

The example starts from a converged separator/valve process. The local linear model and MPC operate around this configured point; they do not replace transient simulation or a validated control-system design.

import neqsim.process.mpc.ProcessLinkedMPC;
import neqsim.process.equipment.stream.Stream;
import neqsim.process.equipment.valve.ThrottlingValve;
import neqsim.process.equipment.separator.Separator;
import neqsim.process.processmodel.ProcessSystem;
import neqsim.thermo.system.SystemInterface;
import neqsim.thermo.system.SystemSrkEos;

SystemInterface fluid = new SystemSrkEos(298.15, 50.0);
fluid.addComponent("methane", 0.8);
fluid.addComponent("ethane", 0.1);
fluid.addComponent("propane", 0.05);
fluid.addComponent("n-pentane", 0.05);
fluid.setMixingRule("classic");
Stream feed = new Stream("feed", fluid);
feed.setFlowRate(100.0, "kg/hr");
feed.setTemperature(25.0, "C");
feed.setPressure(50.0, "bara");
ThrottlingValve valve = new ThrottlingValve("inlet_valve", feed);
valve.setOutletPressure(30.0);
Separator separator = new Separator("separator", valve.getOutletStream());
ProcessSystem process = new ProcessSystem();
process.add(feed);
process.add(valve);
process.add(separator);
process.run();

ProcessLinkedMPC mpc = new ProcessLinkedMPC("separator-pressure", process);
mpc.addMV("inlet_valve", "opening", 0.0, 1.0, 0.1);
mpc.addCV("separator", "pressure", 30.0);
mpc.setConstraint("separator", "pressure", 20.0, 40.0);
mpc.identifyModel(60.0);
mpc.setPredictionHorizon(20);
mpc.setControlHorizon(5);
double[] moves = mpc.calculate();
mpc.applyMoves();

Python exposes the same Java class lookup pattern. Use the shared NeqSim Python environment, and check whether the installed bridge requires Java overload disambiguation.

from neqsim import jneqsim

SystemSrkEos = jneqsim.thermo.system.SystemSrkEos
Stream = jneqsim.process.equipment.stream.Stream
ThrottlingValve = jneqsim.process.equipment.valve.ThrottlingValve
Separator = jneqsim.process.equipment.separator.Separator
ProcessSystem = jneqsim.process.processmodel.ProcessSystem
ProcessLinkedMPC = jneqsim.process.mpc.ProcessLinkedMPC

fluid = SystemSrkEos(298.15, 50.0)
fluid.addComponent("methane", 0.8)
fluid.addComponent("ethane", 0.1)
fluid.addComponent("propane", 0.05)
fluid.addComponent("n-pentane", 0.05)
fluid.setMixingRule("classic")
feed = Stream("feed", fluid)
feed.setFlowRate(100.0, "kg/hr")
feed.setTemperature(25.0, "C")
feed.setPressure(50.0, "bara")
valve = ThrottlingValve("inlet_valve", feed)
valve.setOutletPressure(30.0)
separator = Separator("separator", valve.getOutletStream())
process = ProcessSystem()
process.add(feed)
process.add(valve)
process.add(separator)
process.run()

mpc = ProcessLinkedMPC("separator-pressure", process)
mpc.addMV("inlet_valve", "opening", 0.0, 1.0, 0.1)
mpc.addCV("separator", "pressure", 30.0)
mpc.setConstraint("separator", "pressure", 20.0, 40.0)
mpc.identifyModel(60.0)
mpc.setPredictionHorizon(20)
mpc.setControlHorizon(5)
moves = mpc.calculate()
mpc.applyMoves()

Result extraction

Gotchas

• Identify the MPC only after the process has a valid solved operating point. calculate() before identifyModel() throws IllegalStateException; applyMoves() before a calculation also throws.
• ProcessLinearizer perturbs a local model; results can be misleading at hard bounds, discontinuities, phase transitions, or a non-converged base point. Re-check nonlinear behavior over the intended operating envelope.
• MVs/CVs use the property access and unit behavior of the bound equipment variables. Use explicit physical bounds, rate limits, and compatible engineering units; don't assume a universal percent, bara, or temperature conversion.
• ModelPredictiveController uses an internal first-order model and analytical constrained move calculation. It does not emulate a DCS, PLC, SIS, communications stack, alarm rationalization, or vendor controller execution environment.
• State-space, Industrial MPC, and SubrModl exporters produce files/configuration for integration; they do not deploy, authenticate, or validate a target controller.
• VirtualFlowMeter implements a simplified differential-pressure multiplier (sqrt(abs(dP))) with a calibration factor, not a standards-based meter sizing/flow equation. Its output is an estimate; quality HIGH is not a custody-transfer approval.
• VFM returns INVALID when stream/fluid is absent or a flash fails. Phase-absent rates remain zero; distinguish a physical zero from an unavailable phase/model result.
• SoftSensor.estimate() returns NaN when stream/fluid is unavailable, the flash fails, or a selected phase property cannot be evaluated. Test Double.isFinite and preserve the output unit.
• DataReconciliationEngine requires more variables than constraints. Constraints are caller-defined linear rows with coefficients ordered exactly as variables were added; the engine does not derive a process flowsheet balance automatically.
• Reconciliation failure returns a result with isConverged()==false and a message such as No variables added, No constraints added, or Need more variables ...; never report default values as reconciled measurements.
• SteadyStateDetector's R-statistic is a data-window screen. A steady-state verdict does not prove tag independence, sensor health, process stability outside the selected window, or calibration validity.

Validation / benchmarks

• Start from a converged steady state; test local gain signs and magnitudes against independent small positive/negative perturbations and the intended process direction.
• Check response bounds, move limits, control horizon, prediction horizon, sample time, quality constraints, and model-validity region before using moves in a study.
• Compare controller performance with ControlsBenchmarkSuite and report ControllerPerformanceMetrics; retain event history and failed scenarios. Benchmark completion is not DCS or functional-safety qualification.
• Validate exported matrices by reconstructing the operating-point response and checking dimensions, named-variable ordering, and units. Test imports in the actual target platform before claiming interoperability.
• For VFM, compare against independent well tests across multiple operating points; report calibration range, uncertainty basis, and EXTRAPOLATED/INVALID cases.
• Run steady-state detection on selected, quality-screened signals; reconcile only after the feed/product balance equations, sigmas, degrees of freedom, and gross-error flags have been reviewed.
• For plant-data calibration and historian comparison, follow neqsim-model-calibration-and-data-reconciliation and neqsim-plant-data rather than folding data acquisition into this controller layer.

Related skills

• neqsim-dynamic-simulation
• neqsim-compressor-antisurge-recycle
• neqsim-model-calibration-and-data-reconciliation
• neqsim-controllability-operability
• neqsim-optimization-and-doe
• neqsim-plant-data
• neqsim-pid-process-operations
• neqsim-process-safety