The capabilities of simulating the human brain have encouraged scientists to examine the functioning of the brain, which is known as the most complete and best model for diagnosing and organizing surrounding events, and by simulating its patterns, solve many complex problems. to provide Just as the human brain can process and analyze new problems using previous experiences and pre-learned problems, intelligent systems, if trained, are able to provide acceptable answers based on the information they have been trained in exchange for. And indefinitely, succeed in providing answers to information that they have not encountered before.
Preparing a program to provide an answer to a problem with a large number of input or output variables, using today’s common programming methods, is a difficult and time-consuming task, because it is very complicated to consider all the variables and their effects on each other. However, by using methods based on computational intelligence based on available information, it is possible to solve such problems more easily. On the other hand, the structure of current programming methods is such that if there is a mistake in the input information, all calculations may suffer, but in computational intelligence methods, because the training is based on experience, even if there is Mistakes in the input data, the trained system has a significant degree of error tolerance. Computational intelligence has been used in many research subjects and various specializations and has caused the scientific cooperation of various engineering branches such as electricity, civil engineering, mechanics, computer. The synergy of all methods is considered a suitable basis for designing and applying intelligent systems. Because a method alone can be excellent in approximating reasoning and modeling uncertainty, but it cannot necessarily succeed in learning experimental data or adaptability to an unknown environment. Therefore, the combination of computational intelligence techniques can increase the overall performance of the system and reduce computational costs and ease of use.
Models of computational intelligence
A system is called computationally intelligent if it deals with low-level data (such as numerical data), has an identification component, and does not use knowledge in the form of artificial intelligence, and on the other hand, It is able to show computational versatility and error tolerance, and also has the ability to approximate human performance with the desired speed and accuracy. The emergence of computational intelligence was held at the American Institute of Electrical and Electronics Engineers World Congress. From that time until today, several definitions have been presented in relation to computational intelligence. Computational Intelligence Association affiliated with the American Institute of Electrical and Electronics Engineers has defined the branches of computational intelligence as neural networks, fuzzy systems and evolutionary algorithms. Some researchers believe that computational intelligence is a collection of research methods such as crowd intelligence, fractals, chaos theory, safety systems and artificial intelligence.
Although computational intelligence and artificial intelligence both pursue common goals, there is a clear distinction between them. Some scientists believe that computational intelligence is a subset of artificial intelligence. Some other researchers have presented a different view of machine intelligence, based on their theories, there is a clear distinction between hard computing methods based on artificial intelligence and soft computing methods based on computational intelligence. In hard calculations, inaccuracy and uncertainty are undesirable features of a system; While in soft computing, they are considered key features. Soft computing is a set of methods that is the basis of designing intelligent systems. Also, some researchers believe that soft computing is a large subset of artificial intelligence. The remarkable feature of these intelligent systems is their human-like ability to make decisions based on inaccurate and uncertain information.
Versatility is considered as the main feature of intelligence, and neural, fuzzy and evolutionary systems technologies are considered as the main sub-branches of computational intelligence that can be solved without relying on explicit human intervention. address new issues. Adaptability is defined as the ability of a system to change with the evolution of its parameters or structure to achieve the goal. Adaptability and self-organization play an important role in computational intelligence. In fact, adaptability is the central core of computational intelligence and includes practical concepts, patterns and algorithms that facilitate intelligent behavior.
Soft computing components
The focus of computational intelligence is to create a process or model that cannot be accepted using traditional or mathematical modeling. The reason for the model’s inadmissibility can be due to its high complexity or it may be affected by the hard evaluation and cost. Also, the existence of uncertainties, non-linearity and having an incomplete and random nature can also be other reasons. A computational intelligence system has the ability to learn or face new or unknown situations and can predict or make decisions regarding future events. Computational intelligence is a combination of soft calculations and numerical methods. In fact, the field of computational intelligence is interdisciplinary and tries to develop and integrate theories and methods such as modern adaptive control, optimal control, learning theory, fuzzy logic, neural networks, and evolutionary computing.
Each field uses computational intelligence from a different perspective and uses different methods and tools to achieve a predetermined goal. Computational intelligence uses empirical knowledge related to each process and creates a model with the help of input-output behavior. How to model human knowledge and its application and computational efficiency is of special importance.
Researchers believe that measurement, modeling and control cannot be completely accurate for real and complex processes. Also, there are always uncertainties such as the randomness of the data and their incompleteness in the models. For the first time in 1973, the concept of fuzzy logic to model human reasoning using imprecise and incomplete data was proposed by Professor Zadeh with the help of defining vague expressions and rules governing the problem. Fuzzy logic can use human experimental knowledge and by combining it with engineering principles, it can model and control nonlinear systems with complex uncertainty.
Artificial neural network is actually simulated biological neural network. Neural networks are parallel and distributed data processing systems. The features of neural networks can be mentioned as computational power, error tolerance, learning from experimental data and generalizability. Learning in neural networks is done in different ways, the most important of which are supervised learning, unsupervised learning, competitive learning and reinforcement learning. In recent years, interest in researching control systems based on neural networks has gained momentum, because neural networks can perform any nonlinear function defined on the basis of a compact set of data. to estimate with specific accuracy and on the other hand, many control systems have specific types of unknown nonlinearities that suggest neural networks as a suitable control technology.
An introduction to optimization
Optimization can be used anywhere, from engineering design to sales marketing, from daily activities to vacation planning, and from computer science to industrial applications. The goal of optimization is to maximize or minimize. An organization reduces costs and increases performance to increase its profit. When we plan a vacation, the goal is to minimize cost to get the most out of it. In engineering designs, product design efficiency should be maximum and cost should be minimum compared to time. In a general comparison between exact and approximate methods for optimization purposes, it can be stated that exact methods achieve the optimal solution and guarantee optimality, but sometimes the execution time is very high and even non-existent. are available Approximate methods lead to near-optimal solutions in a very favorable time, but they have no guarantee of reaching the exact solution.
For example, in engineering sciences, the importance of designing a structure with minimal weight was first noticed by the aerospace industry, and in this study, the weight of the aircraft was more than the cost of the design criterion. In other industries such as construction, mechanics and automobiles, cost may be of primary importance; Although the weight of the system affects its cost and performance. The increasing attention to the lack of raw materials and the severe shortage of energy resources has caused the desire of engineers for light, efficient and cheap structures, and in such conditions, approximate optimization techniques are widely used.
Evolutionary calculations are derived from the process of natural selection in a search method based on the theory of evolution presented by Charles Darwin. In nature, living organisms have certain characteristics that affect their ability to survive in adverse environments, and these characteristics are passed on to future generations in an evolved form. Genetic information of species can be encoded using chromosomes that represent these characteristics. Species reproduce and give birth to new children with characteristics combined with the ability to fight for survival. The process of natural selection ensures that more competent organisms have more opportunities to mate, and as a result, the resulting offspring is expected to have similar or greater competence. Evolutionary calculations use repetitive processes in the population. Then the desired population is selected with a targeted random search with the help of parallel processing in order to achieve the desired population of answers. Such processes are often inspired by the biological mechanisms of evolution. Today, evolutionary algorithms have wide applications in the fields of science and engineering.
Synergy of computational intelligence methods
Synergy of all methods is considered a suitable basis for designing and applying intelligent systems. A single method can be excellent in approximating reasoning and modeling uncertainty, but it cannot necessarily succeed in learning experimental data or adapting to an unknown environment, so combining computational intelligence techniques It can increase the overall performance of the system and reduce computing costs and ease of use. The result of this combination of methods will be hybrid intelligent systems such as neural-fuzzy models. Fuzzy logic is good at approximating reasoning, but it has no learning ability or adaptive capacity. On the other hand, neural networks have effective mechanisms in learning experimental data.
Naderpour, H., Mirrashid, M., Soft Computing in Civil Engineering (2020), Semnan University Press, Iran (in Persian), 511 pages, ISBN: 978-622-7237-02-3