Advanced Regulatory Control: Cascade, Feedforward, Ratio & More

Single PID loops are not enough for real plants. Learn cascade control, feedforward (static and dynamic), ratio control, split-range, override/selector control, valve-position control, Smith predictor for dead time, and decoupling for interacting loops — with a simulation showing feedforward cancelling a disturbance.

Advanced regulatory control (ARC) combines simple PID blocks into structures that handle disturbances, constraints and interactions. These strategies are configured in every DCS and deliver a large part of the benefit of 'advanced process control' at low cost.

Each strategy answers a specific problem: a fast inner disturbance (cascade), a measurable disturbance (feedforward), a blend (ratio), one controller with two valves (split range), a safety limit (override), a long delay (Smith predictor).

An analogy

Feedback is steering after the car drifts; feedforward is turning the wheel when you see the crosswind sign. Cascade is a manager (outer loop) giving targets to a fast worker (inner loop) who fixes small problems before the manager notices.

Cascade control

Two controllers in series: the outer (primary) controller's output is the setpoint of the inner (secondary) controller. Classic example: reactor temperature (slow) → jacket temperature or steam flow (fast). The inner loop corrects valve non-linearity and supply disturbances before they affect the primary variable. Tune the inner loop first, and put it in cascade (remote setpoint) mode before tuning the outer.

Feedforward and ratio

Feedforward measures a disturbance (feed flow, ambient temperature) and acts before it affects the output. Dynamic compensation (lead-lag + delay) matches the timing. Ratio control keeps one flow proportional to another (fuel/air, reflux/feed, blending); it is a special feedforward. In boiler control, a cross-limited fuel/air ratio ensures air leads on load increase and lags on decrease — always excess air for safety.

Split-range, override and valve-position control

Handling multiple actuators and constraints:

Dead time and interaction

Long dead times force slow PID tuning; a Smith predictor uses a process model to let the controller act on the predicted undelayed output — powerful but sensitive to model error. In multivariable processes (distillation top/bottom compositions) loops interact; the relative gain array (RGA) selects pairings (pair on λ close to 1, avoid negative), and decouplers or MPC handle strong interaction.

Formulas

Worked example: Static feedforward gain

Given: Feed flow +10% lowers outlet T by 4 °C; steam valve +1% raises it by 0.5 °C

Find: Feedforward gain.

Answer: +0.8% steam valve per % feed

Worked example: Ratio control

Given: Air/fuel ratio 17.2 (mass), fuel 0.5 kg/s

Find: Air setpoint.

Answer: 8.6 kg/s (plus trim for excess O₂)

Worked example: RGA pairing

Given: 2×2 RGA λ₁₁ = 0.85

Find: Pairing.

Answer: Diagonal pairing; mild interaction

Feedforward cancels a measured disturbance

# Feedforward: disturbance d enters through Gd = Kd/(tau s+1), same lag as the process
K, Kd, tau, dt = 2.0, 1.5, 10.0, 0.1
def run(use_ff):
    y = 0.0; integ = 0.0; worst = 0.0
    Kp, Ti = 1.0, 10.0
    for n in range(600):
        d = 1.0 if n * dt >= 5 else 0.0          # step disturbance at 5 s
        e = 0.0 - y
        integ += e * dt
        u = Kp * (e + integ / Ti)
        if use_ff:
            u += -(Kd / K) * d                    # static feedforward cancels d
        y += dt * (K * u + Kd * d - y) / tau
        worst = max(worst, abs(y))
    return worst
print(f"feedback only : peak deviation {run(False):.3f}")
print(f"with feedforward: peak deviation {run(True):.3f}")
# feedback only : peak deviation 0.376
# with feedforward: peak deviation 0.000

When the disturbance and process have the same dynamics, a static gain −K_d/K_p cancels it perfectly. In reality the model is imperfect, so feedforward is always combined with feedback, which trims the remaining error.

Common mistakes

Applications

Key takeaways

Going further