What Is Feedforward Control?
Correcting Before the Error Arrives
Feedforward control corrects for a disturbance the moment you measure it, before it ever moves the variable you care about. Where ordinary feedback waits for an error to appear and then reacts, feedforward anticipates. It is one of the most powerful ways to tighten a slow or hard-to-control process, and it appears throughout oil and gas heating, blending, and flow control. This guide explains how feedforward works, why it is almost always paired with feedback, and where it fits.
Feedforward Control in one line: Feedforward control measures a disturbance directly and takes corrective action before that disturbance affects the process variable - rather than waiting for an error to develop the way feedback does. It is a predictive, model-based correction added on top of a normal feedback loop.
Feedforward vs Feedback
A feedback controller only acts after the fact: the process variable must first deviate from setpoint before the controller sees an error and responds. On a slow process with long dead time - say the temperature of oil leaving a heater treater - that delay means the upset has already done its damage before feedback catches up.
Feedforward flips the timing. It measures the disturbance itself - the inlet oil flow rate or the feed temperature - and computes the change in fuel or valve position needed to cancel it out, applying that correction immediately. The result is that the primary variable barely moves. The trade-off is that feedforward relies on a model of how the disturbance affects the process, and it is blind to anything you are not measuring.
Why Feedforward Needs Feedback Too
No feedforward model is perfect. Sensor calibration drifts, the process gain changes with throughput, and there are always unmeasured disturbances. If feedforward ran alone, those errors would accumulate and the process variable would slowly wander off setpoint with nothing to correct it.
For that reason feedforward is almost always combined with feedback in a feedforward-plus-feedback scheme. Feedforward does the heavy lifting - knocking out the large, measurable disturbance fast - while a feedback (usually PID) loop trims the residual error and handles everything the model missed. This pairing gives both the speed of prediction and the accuracy of correction.
Feedforward in Oil and Gas
Classic examples include heater and heater-treater temperature control, where inlet flow rate is fed forward to preset the fuel-gas valve before the extra cold oil can drag the outlet temperature down. In blending and injection, a change in the main stream flow is fed forward to scale the additive or chemical injection rate so the ratio stays constant. Compressor and pump control also use feedforward on suction conditions.
In each case the disturbance is measurable, its effect on the process is reasonably well understood, and it enters faster than a feedback loop could catch it. Those three conditions are the practical test for whether feedforward is worth the added engineering.
Static vs Dynamic Compensation
The simplest feedforward is static: a steady-state gain that says how much the manipulated variable must move per unit of disturbance, applied immediately. Static compensation is adequate when the disturbance and the corrective action reach the process variable through paths of similar speed. When the paths differ - the cold feed reaches the outlet sooner than extra fuel can - a static correction arrives at the wrong time even if its size is perfect, and the loop sees a transient in one direction followed by an overshoot in the other.
Dynamic compensation fixes the timing: a lead-lag element shapes the correction so it develops at the same rate the disturbance propagates, and a dead-time element delays it when the disturbance path is slower than the correction path. Getting these right means knowing both response shapes, which is why feedforward design leans on simple identified models - a first order plus dead time model of each path is usually enough - and on an honest measurement of process dead time. The telltale of bad dynamics is a characteristic double bump on every disturbance: the correction landing early or late and doing its own damage on the way through.
A Worked Symbolic Example: Presetting a Heater's Fuel
Write the heater's steady-state energy balance symbolically: the heat delivered must match the heat absorbed by the feed, so fuel demand is proportional to feed flow times the required temperature rise, u = K x F x (T_set - T_in), where F is the measured feed flow, T_in the measured inlet temperature, and K a gain established from process data. The feedforward controller computes this continuously. When feed flow steps up by dF, the fuel command immediately increases by K x dF x (T_set - T_in) - the burner is already compensating while the additional cold oil is still traveling toward the outlet sensor.
The feedback loop then earns its keep on everything the formula does not know: fouling drifting the true gain away from K, fuel-gas composition changes, ambient losses. Watch the feedback controller's steady-state contribution as a health signal - if its trim keeps growing over weeks, the feedforward model has drifted, and K deserves re-identification from fresh process data rather than another round of feedback tuning applied on top.
Commissioning Feedforward Without Upsetting the Process
Feedforward is added to a working feedback loop, and the commissioning discipline is to prove each piece before trusting the whole.
- Tune and verify the feedback loop alone first; feedforward layered over a poorly tuned loop hides both problems.
- Verify the disturbance measurement itself - a noisy flow signal fed forward becomes a noisy valve, so confirm signal quality and filtering before connecting anything.
- Enable the feedforward path with its gain deliberately reduced below the calculated value and watch how much the manipulated variable moves.
- Compare disturbance responses with feedforward on and off, trending the process variable's excursion each way.
- Raise the gain stepwise toward the calculated value; if the process variable begins deviating in the opposite direction on disturbances, the compensation is overcorrecting.
- Document the final gain and dynamic settings together with the data that justified them.
Frequently Asked Questions
What is the difference between feedforward and feedback control?
Feedback reacts to an error after the process variable has already deviated from setpoint. Feedforward measures a disturbance directly and corrects for it before it affects the process variable. Feedforward is predictive; feedback is corrective. They are usually used together.
Can feedforward control be used alone?
In practice, no. Feedforward relies on a model of the process and only acts on disturbances you measure, so errors and unmeasured upsets accumulate with nothing to correct them. It is almost always paired with a feedback loop that trims the residual error.
Where is feedforward control used in oil and gas?
Common examples are heater and heater-treater temperature control with inlet flow fed forward to the fuel valve, chemical injection and blending where main-stream flow scales the additive rate, and compressor or pump control that anticipates suction conditions.
How is feedforward different from cascade control?
Cascade control is still feedback - an inner loop controls an intermediate variable so disturbances entering there are corrected before the primary variable feels them. Feedforward is not feedback at all: it computes a correction from a measured disturbance with no confirmation from the process. The two are complementary, and a heater with feed flow fed forward onto a fuel-flow cascade uses both at once.
What happens if the feedforward gain is too high?
The correction overshoots the disturbance and drives the process variable in the opposite direction - the loop trades an upset one way for an upset the other way, and the feedback controller has to unwind the excess. If disturbance responses show the process variable moving the wrong way first, reduce the feedforward gain and confirm the improvement against trended data.
Automation services
Need help turning this into a working system?
Merobix integrates SCADA, programs Allen-Bradley and Siemens PLCs, and designs and fabricates industrial control panels.
Meeting requests are reviewed before confirmation.