Delay-Compensated Control¶
Feedforward controller — instantiates Stock–Flow Accumulation Control
Controls a stock whose response lags the lever, acting on where the level is headed rather than where it is now.
Delay-Compensated Control governs a stock whose response to any lever arrives late. Its defining move is to steer against the stock's predicted future level, not its present reading — because between pulling a lever and seeing the stock move there is a transport or processing lag, and a controller that reacts to the current level always over- or under-shoots by exactly the amount of change already in the pipeline. So it keeps an explicit register of the system's delays, projects the stock forward by accounting for the "in-flight" effect of levers already pulled, and acts on that projection. It is not about which lever to pull or how fast the stock cycles; it is about when — timing the action so that when its effect finally lands, the stock arrives on target instead of sailing past it.
Example¶
A large office building heats its zones with hot-water radiators fed from a central boiler. The controlled stock is the accumulated heat in a zone — its temperature. The trouble is thermal lag: when the controller opens a valve, the water must heat the radiator, the radiator must warm the air, and the air must warm the massive concrete floor slab before the zone temperature actually rises. That chain can take, illustratively, forty minutes. Delay-Compensated Control is the thermostat logic that respects this lag.
A naïve thermostat waits until the room reads below setpoint, then heats at full tilt until it reads above — and because forty minutes of heat is still marching through the slab when the sensor finally reads "warm enough," the room overshoots for another half hour, then overcools the same way, cycling the building between too cold and too hot all day. The delay-compensated controller instead registers the lag and projects forward: it estimates the heat already in transit from valve moves made half an hour ago and asks where the temperature is headed, cutting the valve back while the room still reads cool because it knows the pipeline will carry it the rest of the way. The zone settles smoothly at setpoint because the controller acted on the future, not the present.
How it works¶
- Register the delays. Maintain an explicit estimate of the lag between each lever move and its effect on the stock — the dead time and the settling time — as a first-class input, not an afterthought.
- Define the stock so it can be projected. Pin down the stock and its unit precisely enough that a forward estimate of its trajectory is meaningful.
- Project the level forward. Roll the stock ahead by the delay horizon, adding in the not-yet-realized effect of levers already actuated, to estimate where it will be when a new action would land.
- Act on the projection. Size and time the control move against the predicted level, easing off before the present reading reaches target because the in-flight effect will finish the job.
Tuning parameters¶
- Delay estimate — the assumed lag between action and effect. Too short and the controller behaves almost naïvely and overshoots; too long and it acts prematurely on effects that arrive sooner than expected, undershooting.
- Prediction horizon — how far ahead the projection reaches. A longer horizon smooths a long-lagged stock but compounds model error; a shorter one is robust but reverts toward reacting to the present.
- Model confidence — how strongly the controller trusts its forward projection versus the live measurement. High trust gives smooth, anticipatory control but is fragile if the delay model is wrong; low trust is safe but sacrifices the compensation.
- Correction damping — how gently the controller nudges when projection and measurement disagree. Gentle correction avoids reintroducing oscillation; aggressive correction chases the discrepancy and can undo the anticipation.
When it helps, and when it misleads¶
Its strength is that it tames the single hardest case in stock control — a slow stock and a fast twitchy hand — by converting lag from an ambush into a modeled quantity. Where a present-reading controller is doomed to hunt between the walls, an anticipatory one settles, because it never fights change that is already on its way. This is the logic of the Smith predictor[n1] and every dead-time compensator: subtract out the pipeline you already committed to before deciding to add more.
Its failure mode is that it is only as good as its delay model. If the lag is mis-estimated or drifts — a fouled radiator, a changed load — the controller confidently steers against a wrong prediction and can overshoot worse than the naïve version it replaced, now with false certainty. The classic misuse is the opposite error: ignoring the lag and cranking control gain to force a faster response, which drives the growing, delayed oscillations of the beer-game bullwhip — each over-correction arriving just in time to amplify the next. The guarding discipline is to measure and periodically re-fit the delay register against observed response, and to keep enough weight on live measurement that a stale model is corrected before it does harm.
How it implements the components¶
Delay-Compensated Control fills the archetype's timing-and-lag turf:
delay_and_lag_register— its core asset: an explicit, maintained estimate of the dead time and settling lag between each lever and the stock's response.stock_definition_and_unit— it defines the stock precisely enough that a forward projection of its level is a well-posed quantity to control against.stock_boundary_and_owner— it fixes the boundary across which the lagged transport occurs and the owner accountable for the controller's anticipatory setpoint.
It does not enumerate the intervention_lever_map of which flows to move (that's Net-Flow Lever Adjustment), nor reshape the clearance_and_turnover_model that sets cycling speed (that's Clearance–Turnover Tuning); its concern is purely the timing of action against a delayed response.
Related¶
- Instantiates: Stock–Flow Accumulation Control — it is the controller that keeps a lagged stock on target by acting on its predicted trajectory.
- Consumes: Net-Flow Lever Adjustment supplies the lever choices whose timing this controller corrects for the delay.
- Sibling mechanisms: Stock–Flow Balance Reconciliation · Net-Flow Lever Adjustment · Accumulation Threshold Alert · Clearance–Turnover Tuning · Stock-Level Buffering · Hidden Accumulation Probe
Editorial Notes¶
Form Classification¶
Form family: Control, Automation & Runtime
Rationale: Delay-Compensated Control operates as a live operational control that automatically routes, enforces, adapts, or responds during execution because it controls a stock whose response lags the lever, acting on where the level is headed rather than where it is now.
Independent corroboration: The frozen evidence defines Delay-Compensated Control as 'Controls a stock whose response lags the lever, acting on where the level is headed rather than where it is now', so its operative form is Control, Automation & Runtime.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Systems Thinking & Cybernetics
Origin pattern: Single lineage
Present-day reach: Multi-domain
Rationale: Control theory is primary because the mechanism is dead-time compensation, canonically the Smith predictor: use a model-predicted undelayed state rather than react directly to stale process output. Process engineering supplied its formative plant context and deployment practice.
Related originating lineages:
- Engineering & Design — Process and chemical engineering supplied delayed plants and the operational controller designs in which dead-time compensation became standard.
Review resolution: Control theory is primary because the mechanism is dead-time compensation, canonically the Smith predictor: use a model-predicted undelayed state rather than react directly to stale process output. Process engineering supplied its formative plant context and deployment practice.
Review outcome: Researched adjudication after independent review; high confidence.
Sources consulted:
- MathWorks: Smith Predictor Controller
- Frontiers in Control Engineering: Smith predictor history and review
Notes¶
[n1] The Smith predictor — a classic control structure for processes with significant dead time, which uses an internal model of the plant and its delay to predict the response to control actions already taken, so the controller can act as if the delay were removed. It is the archetypal delay compensator this mechanism generalizes to stock control. ↩