In the 1990s, a new paradigm of science characterized by uncertainty, nonlinearity, and irreversibility and tackling complex problems was generally recognized by the academic community. In this new paradigm, traditional analytical methods are ineffectual, and there is recognition of the need to explore new methods to solve the more flexible, more robust system problems. In 1994 the first Computational Intelligence Conference in Orlando, Florida, US, first combined three different areas, smart neural networks, fuzzy systems and genetic algorithms, not only because the three have many similarities, but also because a properly combined system of the three is more effective than a system generated by one single technical field. Various theories and approaches of computational intelligence including neural computing, fuzzy computing and evolutional computing are comprehensively introduced in this chapter.
TopIntroduction
Computational Intelligence is generally considered to incorporate Neural Networks (NN), Fuzzy Systems (FS) (Dong, 2008) and Evolutionary Computation (EC) (Cacciola, 2008)) as its three main areas. Its positive significance is to promote the comprehensive integration of various kinds of smart theories, models and methods which are based on the combination of calculation and symbols, in order to facilitate the development of intelligent behavior which is more advanced on thinking, and more powerful. Computational intelligence is capable of solving more complex systems problems.
The current key issues of computational intelligence are divided into three main areas:
- 1.
Explain the calculation theory of calculation intelligence. The most typical is evolutionary computation. Researchers have been keen to explain the calculation theory of GA, and to establish the precise mathematical theoretical basis, but this research is still in the exploratory stage.
- 2.
Put forward new calculation models (Christine, 2002). There are many neural network models, but most of them offer a one-sided understanding of the biological nervous system. The establishment of a closer understanding of the human neural network model is expected based on the further exploration of the mysteries of human intelligence. Biological evolution seems simple, but GA is unable to fully understand and imitate it, so the improvement of GA (Cantu, 2000) is needed to promote its performance. Researchers have noted that in the course of evolution DNA fragments are explained through RNA (Ribonucleic Acid), amino acids, and proteins, unlike the direct explanation given by GA. At the same time, there are mechanisms in biological systems such as the ant colony in search of food, or the body's immune system’s ability to resist the effects of viruses, which are used for reference by evolution calculation to establish the structure of a new model .
- 3.
Combine various calculation methods to create a new calculation model (ClercM, 2000). The earliest neural networks can achieve good computing performance by applying the idea of simulated annealing to generate a random neural network model. The optimization of the parameters of the neural network structure is by virtue of GA calculation. GA is easily combined with other methods to form hybrid genetic algorithms (HGA), such as in the choice of operation in the simulated annealing method, or in the decoder algorithm in other local searching operations. The integration of fuzzy systems and artificial neural networks allows wide coverage of the application, so the fuzzy neural network is also a burning issue in computational intelligence.
Computational intelligence is widely used in many areas, covering all areas of AI, such as knowledge acquisition, associative memory, speech recognition, image analysis and so on, as well as penetrating into other areas (Clec, 2002). With the development of complexity of calculating theory, intelligence shows a very strong capacity for solving complex planning problems, for example, the complex combinatorial optimization problem solving (TSP), the multi-function complex calculation of the maximum (Katayama, 2004). Computational intelligence methods and related technologies, such as the Simulated Annealing (SA) algorithm and Tabu Search (TS) constitute the basis of modern optimization technology. The applications in the field of optimization technology in turn promote the progress of computational intelligence technology.
Computational intelligence is widely applied in manufacturing (Antonio, 2000). The essential foundation of intelligent control is fuzzy control and neural network control technology (Chen, 2004). In recent years, with the progress of real-time genetic algorithms, the GA algorithm has been widely used in the field of intelligent control (Ritter, 2007). Researchers are used to applying computational intelligence methods to resolve many complex decision-making problems in manufacturing, such as inventory planning, shop layout, production process scheduling and control (Srivastava, 2008), which are of the Non-deterministic Polynomial Completeness (NPC) . The study of these applications has brought about many valuable algorithm models, which have an important role in the theoretical study of computational intelligence.