⚡ electrical engineering

A PSO Approach for Optimum Design of Multivariable PID Controller for nonlinear systems

This paper proposes a Particle Swarm Optimization (PSO) algorithm to tune a multivariable PID controller for nonlinear MIMO Takagi-Sugeno fuzzy systems, demonstrating superior performance over conventional methods like Ziegler-Nichols by significantly reducing overshoot.

Original authors: Adel Taeib, Ali Ltaeif, Abdelkader Chaari

Published 2026-06-04

📖 4 min read☕ Coffee break read

CC BY 3.0

Original authors: Adel Taeib, Ali Ltaeif, Abdelkader Chaari

Original paper licensed under CC BY 3.0 (http://creativecommons.org/licenses/by/3.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to steer a very complicated, wobbly boat (a nonlinear MIMO system) through a storm. This boat has two steering wheels and two rudders that all affect each other in confusing ways. If you turn one wheel, the other rudder might jerk unexpectedly. Your goal is to keep the boat on a straight path to a specific destination.

To do this, you need a PID Controller. Think of the PID controller as the "autopilot" or the "smart hand" that constantly adjusts the steering wheels. It has three settings (gains) it can tweak:

Proportional (P): How hard to push right now based on how far off course you are.
Integral (I): How hard to push based on how long you've been off course.
Derivative (D): How hard to push based on how fast you are drifting off course.

The Problem: The "Old School" Way

Traditionally, engineers tune these settings using a rulebook called Ziegler-Nichols (Z-N). The paper compares this to using a generic, one-size-fits-all map. The authors found that while this map gets you moving, it often causes the boat to swing wildly (high overshoot) before it finally settles down. It's like a driver who slams the brakes too hard, skids, and then corrects too much, making the ride bumpy and unsafe.

The Solution: The "Smart Swarm" (PSO)

Instead of using a static rulebook, the authors used a method called Particle Swarm Optimization (PSO).

The Analogy:
Imagine a flock of birds looking for the best place to land in a field.

Each bird is a "particle" representing a different set of PID settings (a different way to steer the boat).
The birds fly around randomly at first.
Every time a bird lands, it checks how smooth the landing was (this is the fitness function).
If a bird finds a spot with a smooth landing, it remembers it.
The birds also talk to each other. If one bird finds a perfect spot, the whole flock adjusts its flight path to move toward that spot.
Over time, the whole flock converges on the single best landing spot.

In this paper, the "birds" are computer simulations testing thousands of different PID settings. They are looking for the combination that minimizes errors (like overshooting the target or taking too long to settle).

How They Did It

Modeling the Boat: First, they used a "fuzzy logic" system to create a digital twin of the wobbly boat. This allowed them to predict how the boat would react to steering inputs without actually crashing a real boat.
The Hunt: They let the "swarm" of virtual birds search for the perfect PID numbers. They tested three different ways to measure "smoothness":
- IAE: Total distance off course.
- ISE: Squaring the errors (punishing big mistakes more).
- ITSE: Squaring errors but also punishing them if they happen late in the flight.
The Result: The swarm found a set of settings that was much better than the old "rulebook" (Z-N) method.

The Outcome

When they tested the new "Swarm-Tuned" autopilot against the old "Rulebook" autopilot:

The Old Way: The boat swung wildly, overshooting the target by 55% to 65% before settling. It took a long time to stabilize.
The New Way: The boat was much steadier. The overshoot dropped to around 40% or less, and it settled into the correct path much faster (in less than half the time).

The Bottom Line

The paper claims that by using this "flock of birds" algorithm (PSO) to tune the autopilot, they can make complex, wobbly machines run much smoother and safer than using traditional, manual tuning methods. The "smart search" found a better balance of settings that the old rules missed.

Technical Summary: A PSO Approach for Optimum Design of Multivariable PID Controller for Nonlinear Systems

Problem Statement
The paper addresses the challenge of controlling Multiple-Input Multiple-Output (MIMO) nonlinear systems, which are prevalent in industrial processes but inherently more difficult to manage than Single-Input Single-Output (SISO) systems. While Proportional-Integral-Derivative (PID) controllers are widely used in industry, conventional tuning methods, such as the Ziegler-Nichols (ZN) technique, often result in significant overshoot and suboptimal performance when applied to complex, nonlinear MIMO dynamics. The authors posit that modern heuristic approaches are necessary to enhance the capabilities of traditional tuning techniques.

Methodology
The proposed solution integrates three primary components: a Takagi-Sugeno (T-S) fuzzy model for system identification, Recursive Weighted Least-Squares (RWLS) for parameter estimation, and Particle Swarm Optimization (PSO) for PID gain tuning.

System Modeling: The MIMO nonlinear process is represented using a T-S fuzzy model. The system is decomposed into Multiple-Input Single-Output (MISO) NARX (Non-linear Auto Regressive with eXogenous input) models. The identification of these models relies on input-output data.
Parameter Estimation: To handle the nonlinearity and allow for online adaptation, the authors employ the Recursive Weighted Least-Squares (RWLS) algorithm. This method estimates the system parameter vectors ( $\theta$ ) by treating the system as linear with respect to its unknown parameters, utilizing a covariance matrix and estimator gain vector to update parameters iteratively.
Control Structure: A multivariable PID control structure is defined where the controller matrix $C(z)$ interacts with the process matrix $H(z)$ . The controller gains ( $k_p, k_i, k_d$ ) are the variables to be optimized.
Optimization Strategy (PSO): Instead of using ZN tuning, the paper utilizes the Particle Swarm Optimization (PSO) algorithm. In this framework:
- Each particle represents a candidate set of PID parameters ( $k_p, k_i, k_d$ ) for the MIMO system.
- The search space is defined by the dimensions of the PID gains.
- The objective function is minimized to find the optimal gains. The paper evaluates three specific performance criteria: Integral of Absolute Error (IAE), Integral of Square Error (ISE), and Integral of Time-weighted Square Error (ITSE).
- The PSO algorithm updates particle velocities and positions based on individual best ($pbest$) and global best ($gbest$) values, utilizing an inertia weight ( $\omega$ ) that decreases linearly over iterations to balance exploration and exploitation.

Key Contributions

Integration of T-S Fuzzy and PSO: The paper proposes a specific architecture where a T-S fuzzy model identifies the nonlinear MIMO dynamics, and PSO tunes the PID controller based on this model.
Comparative Tuning Analysis: The study systematically compares the performance of PID controllers tuned via the traditional Ziegler-Nichols method against those tuned via PSO using different objective functions (IAE, ISE, ITSE).
Algorithmic Implementation: The paper details the step-by-step implementation of the PSO-PID controller, including the initialization of particles, the update rules for velocity and position, and the specific stopping criteria used in the simulation.

Simulation Results
The authors validate the proposed method using a simulated MIMO nonlinear system defined by specific difference equations with two inputs and two outputs.

Parameter Optimization: Tables III and IV demonstrate that PSO yields different sets of PID parameters compared to the ZN method, with significant variations in $k_p, k_i,$ and $k_d$ depending on the chosen objective function (ISE, IAE, or ITSE).
Performance Metrics: Tables V and VI compare the step response characteristics. The results indicate that the ZN-tuned controller produces high overshoot (55.3% for $y_1$ and 64.8% for $y_2$ ) and longer settling times. In contrast, the PSO-tuned controllers significantly reduce overshoot (ranging from ~40% to 48%) and drastically improve rise and settling times. For instance, the settling time for $y_2$ improved from 3.29s (ZN) to as low as 0.52s (PSO-ITSE).
Visual Confirmation: Figures 2 and 3 illustrate the system responses, visually confirming that the PSO-based approach provides a smoother response with less oscillation compared to the ZN approach.

Significance and Claims
The paper concludes that the conventional ZN tuning method is insufficient for the specific nonlinear MIMO system tested, as it leads to large overshoots. The authors claim that the PSO-based optimization approach offers a distinct advantage by effectively tuning PID parameters to minimize performance indices. The study asserts that the proposed PSO-PID controller enhances the capability of traditional techniques, providing a more robust solution for controlling nonlinear multivariable processes. The work positions PSO as a viable, computationally efficient alternative to genetic algorithms for this specific control application, citing its simplicity and deep intelligent background.

The Problem: The "Old School" Way

The Solution: The "Smart Swarm" (PSO)

How They Did It

The Outcome

The Bottom Line

More like this