Keywords

These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

1 Naturally Inspired Methods

Throughout history, nature has always been an inspiration for mankind. It is not an exaggeration to say that almost every human invention, from engineering to social sciences, has been an attempt to replicate nature. In fact, nature continues to play an important roll in different human activities. From a scientific perspective, nature-inspired methods have proven to be an efficient tool for tackling real-life problems that are difficult to solve because of their high complexity or the limitation of resources to analyze them.

Computer science and Natural sciences have evolved considerably during the last years, and their interaction has produced new notions, techniques and methodologies that can be adapted to the complexities of real-life problems [12].

The last years of the 20th century mark the birth of natural computing, defined as a research field that focuses on the computation that takes place in nature. Natural computing works with three methodologies [6, 12]:

  1. 1.

    Human-designed problem-solving techniques inspired by nature.

  2. 2.

    Synthesis of natural phenomena based on computer simulations.

  3. 3.

    Use of nature-inspired materials to perform computations.

The core idea is the fact that several natural phenomena, from simple to complex, always try to optimize certain parameters. For example, ants can find the shortest path to a source food after several iterations, light always follows the path that takes the shortest time to travel, unstable chemical components interact with their environment until they reach stability, the hexagonal honeycomb shape in beehives maximizes the quantity of honey stored, while using the less quantity of wax in its construction. These few examples can provide the reader with a glimpse of why nature has inspired scientists in several research areas.

Throughout this book, several works about the advantages of nature-inspired methods are considered. In particular, this book focuses on methods that can simulate natural phenomena using computers, and the benefits of applying this methodology to the analysis and design of engineering control systems.

2 Types of Naturally Inspired Methods

The list of nature-inspired methods is vast and continues expanding [2, 3, 8, 12]. Several authors have proposed different classifications [2, 3], but in general, nature-inspired methods can come from three different areas of inspirations: biological, chemical, and physical [3]. Following, these naturally inspired methods are described, and Table 1 summarizes them.

Table 1 Comparative chart of different nature-inspired methods
Full size table

2.1 Biological Inspiration

Biological processes have proven to be a great source of inspiration for nature-inspired methods, which can be divided in four categories: evolutionary algorithms, swarm algorithms, ecology algorithms, and reasoning algorithms. Following, these categories are briefly described.

2.1.1 Evolutionary Algorithms

Evolutionary algorithms are inspired by the processes and mechanisms of biological evolution [2, 3, 8]. Some of the algorithms that belong to this category are:

  • Genetic Algorithm (GA): population-based stochastic search algorithms inspired on the micro-level genetic adaptation of organisms.

  • Genetic Programming (GP): similar to genetic algorithms, although the difference resides in the tree-like representation of the solution.

  • Evolution Strategies (ES): optimization algorithms based on the theory of adaptation and evolution by means of natural selection.

  • Differential Evolution (DE): similar to genetic algorithms, but here the mutation operator differs in that it is the result of an arithmetic operation between individuals.

  • Evolutionary Programming (EP): similar to genetic algorithms, although this technique focuses on the macro-level of evolution.

  • Grammatical Evolution (GE): optimization technique inspired on the generation process of a protein from genetic material.

  • Gene Expression Programming (GEP): this algorithm is inspired by the replication and expression process of the DNA molecule.

  • Memetic Algorithms (MA): these algorithms are inspired by the interaction of genetic and cultural evolution.

In general, evolutionary algorithms work well with problems whose search space is huge and there is no mathematical expression that can fully describe them. However, the main disadvantage is that these types of algorithms tend to converge to local optima.

2.1.2 Swarm Algorithms

The inspiration for swarm algorithms comes from the collective intelligence that emerges when a large number of homogeneous agents cooperate for a certain goal in the environment [2]. Some of the algorithms that possess this characteristic are:

  • Particle Swarm Optimization (PSO): population- and trajectory-based stochastic search algorithms inspired by the social behavior of birds looking for food.

  • Ant Colony System (ACS): similar to PSO, this is a population- and trajectory-based stochastic search algorithm that is inspired by the ability of ants to find the shortest path between their nest and the food source.

  • Bees Algorithm (BA): optimization technique inspired by the hierarchical structures within beehives.

  • Bacterial Foraging Optimization (BFO): algorithm inspired by the processes and mechanisms that occur within bacterial populations.

  • Artificial Immune System (AIS): population-based algorithm based on the clonal selection principle.

  • Intelligent Water Drops (IWD): population-based algorithm based on the processes that occur within natural river systems.

The main advantage of swarm algorithms is that they can escape from local optima more easily than evolutionary algorithms. However, the implementation and usage of several agents can make these algorithms memory demanding and computationally expensive.

2.1.3 Ecology Algorithms

Ecology algorithms are inspired by the interaction of living organisms with the environment [2]. Some of the algorithms that belong to this category are:

  • PS2O Algorithm: optimization algorithm that extends the original PSO to consider the notion of symbiotic co-evolution between species.

  • Invasive Weed Optimization (IWO): this is a stochastic search algorithm based on the ecological process of weed colonization and distribution.

  • Biogeography-Based Optimization (BBO): this algorithm is inspired by the study of immigration and emigration of species across time and space.

These algorithms share several characteristics with swarm algorithms. In particular, they have a good balance between exploration and exploitation. However, these types of algorithms can be computationally expensive given the complexity of the different attributes they are simulating.

2.1.4 Reasoning Algorithms

The inspiration for these types of algorithms comes from the ability of life beings to process vague information from the environment and still produce an acceptable response depending on the input [3]. Two of the most popular schemes are:

  • Artificial Neural Networks (ANN): this is a whole research field that mimics the structure and function of neurons in the brain. It focuses on developing computational models with a network-like structure that can be used to solve a great variety of scientific and engineering problems.

  • Fuzzy Logic (FL): a research area inspired by the vagueness of human language and logic. Its application has produced successful results from control theory to artificial intelligence.

The main advantage of artificial neural networks and fuzzy logic is the ability to identify hidden patterns within complex data structures [7]. Once the patterns have been found, predictions can be performed fast and easily. However, the biggest disadvantage is that the training process for the neural network or the fuzzy model is not trivial and there are several parameters that need to be tuned to avoid the overfitting problem. To overcome this, several neuro-fuzzy hybridization techniques have been developed that try to combine the strengths of these two research areas.

2.2 Chemical Inspiration

As the name suggests it, these kinds of nature-inspired methods are based on chemical processes [5, 9]. Some algorithms that have chemical inspiration are:

  • Artificial Organic Networks (AON): artificial intelligence technique inspired by the characteristics of organic compounds in chemistry.

  • Molecular Computing (MC): computational approach that performs computations based on molecules.

  • Chemical Reaction Optimization (CRO): optimization technique inspired by the natural process of converting unstable substances into stable ones.

  • DNA computing: computational technique that makes use of the properties of DNA to make mathematical computations.

These techniques claim that chemical energy can be used as the heuristic guide for modeling and optimization processes. Since, chemical energy tends to a stable state; then, these chemically inspired methods promote minimal resources and finding global optima. In particular to DNA computing, it exploits parallelism to find solutions to hard problems. Disadvantages in these techniques is memory consuming.

2.3 Physical Inspiration

There are several nature-inspired methods that have their foundation in physical processes like metallurgy, music, and complex dynamics systems [3]. Some nature-inspired methods that have physical inspiration are:

  • Simulated Annealing (SA): optimization algorithm based on the annealing process in metallurgy.

  • Extremal Optimization (EO): optimization technique inspired by the field of statistical physics.

  • Harmony Search (HS): this is a stochastic optimization algorithm based on the improvisation skills of Jazz musicians.

  • Cultural Algorithm (CA): optimization technique inspired by the principles of cultural evolution.

  • Quantum Computing (QC): a new computational paradigm inspired by the concepts and properties of quantum mechanics.

The main advantages of these algorithms are that they can deal with a variety of problems, and statistically, they can find global optima. However, most of the algorithms tend to be slow. Furthermore, harmony search and cultural algorithms have been seen as special cases of evolutionary algorithms [1, 11, 14].

3 Unconventional Control Systems

The above description considers Nature as the primary resource for inspiration in different computational methods. Since there are several applications for these techniques, this book will be reserved for those applications in terms of engineering control systems.

In control theory, conventional control systems are techniques employed for designing systems with desired behaviors. It means that any variable of interest in a mechanism, machine or device is maintained or altered with a defined rule [4]. Modern control theory has been applied in real-world problems, using the information of the plant. However, real-world problems are ill of uncertainty and vagueness that alter the behavior of the plant [4, 13]. As a result, conventional control techniques are not able to handle systems properly. In that sense, novel unconventional control systems have been applied [4].

Unconventional control systems are those methods employed to counter real-world problems in control systems. Typical problems identified are [4, 9, 10]: nonlinearities in the plant, impossibility for modeling the plant, uncertainty and noise in the system, vagueness in the behavior of the plant, dynamic operational point, and so forth. In that sense, unconventional control systems compute the control signal using different approaches such as: intelligent control and nature-inspired computing. On the one hand, intelligent control systems have used recently applying techniques like fuzzy logic and artificial neural networks. Complexity of real-world systems forced to mix these methods creating hybrid neuro-fuzzy controllers [7]. On the other hand, naturally inspired methods have been applied for improving the performance of intelligent control systems [4, 13].

In fact, unconventional control systems are designed to minimize the uncertainty and vagueness of engineering systems, by using controllers with high autonomy and robustness. It means that unconventional control systems provide enough flexibility and freedom to control laws to learn, predict and adapt to actual circumstances in systems. This flexibility and autonomy for handling unexpected behaviors bring to unconventional control systems the opportunity to design controllers without having the model of the plant [4].

3.1 Challenges in Unconventional Control Systems

There are still many challenges when using unconventional control systems [4, 13]. For instance, intelligent control and nature-inspired methods require more computational power than conventional control systems. In addition, real-time controllers are also a problem because these methods are highly time consuming. Since, nature-inspired methods need additional information about the performance of the system, they also are memory consuming. These challenges are not easy to tackle, and many efforts are still required.

3.2 Trends in Nature-Inspired Control Systems

Throughout this book, different applications are reviewed in which nature-inspired techniques for control systems have been employed to improve the performance of them. Thus, the book is organized as follows:

Part I begins with general control approaches like the implementation of spiking neural networks in hardware, a type of artificial neural networks, as the biological inspiration in a real-time control for laser spot tracking; and it finishes with the design of a hybrid control system using artificial neural networks and a new proposal so-called grey wolf optimization.

Part II explores some recent works in control tuning and adaptive control systems. For example, evolutionary algorithms were tested to perform modeling and control of industrial plants using auto-tuning of proportional-integral-derivative (PID) controllers; a fuzzy controller in a multiple-input and multiple-output nonlinear systems was implemented in which hybrid elite genetic algorithms and Tabu search methods were employed for optimal tuning; a neural controller was tuned for a special input-output manifolds in constrained linear systems; and an adaptive neuro-fuzzy controller was designed for induction machines.

Lastly, Part III presents some robotics applications such as: the design of a neuro-fuzzy controller for quadrotors using an adaptive-network-based fuzzy inference system (ANFIS); a novel image processing approach for mobile robots using a real-time optical flow technique based on Hermite transform; and the implementation of evolutionary function approximation for gait performance on legged robots.