What Are Fuzzy Neural Networks?
Fuzzy neural network is the product of the combination of fuzzy theory and neural network. It combines the advantages of neural network and fuzzy theory, and integrates learning, association, recognition, and information processing.
- There is a sharp contradiction between the complexity of the system and the required accuracy. For this reason, by simulating human learning and adaptive capabilities, people have proposed the idea of intelligent control. Control theory expert Austrom (1991) pointed out at the IFAC conference: fuzzy logic control, neural network and expert control are three typical intelligent control methods. Expert systems are usually based on expert experience, not on operational data generated by industrial processes, and the inaccuracy and uncertainty of general complex systems are difficult to grasp even by domain experts, which makes it very difficult to establish expert systems. . Fuzzy logic and neural network, as two typical intelligent control methods, have their own advantages and disadvantages. The fusion of fuzzy logic and neural network --- Fuzzy Neural Network because it absorbs the advantages of fuzzy logic and neural network, part Avoiding the shortcomings of the two has become one of the hotspots in intelligent control research. [1]
- In the design of fuzzy neural network, the establishment of fuzzy rules is the bottleneck of system design. Therefore, most of the researches on the combination of neural networks and fuzzy systems focus on the modeling of fuzzy neural networks. The research on the structure and algorithms of fuzzy neural networks is at home and abroad. Scholars' research hotspots, new fuzzy neural network models and learning algorithms continue to emerge. [2]
- The diversity of neural fuzzy network structure leads to the diversity of its learning algorithms. The learning of fuzzy neural network mainly includes structure learning and
- blurry
- blurry
- Fuzzy neural network combines the advantages of neural network system and fuzzy system. It has great advantages in dealing with non-linearity, ambiguity, and other issues, and has great potential in intelligent information processing. It makes more and more experts and scholars. Invested in this field and made fruitful research results. However, most of the researches on fuzzy neural networks are based on the innovation, improvement and improvement of algorithms. Few reviews have summarized it, so that people who are new to this field are often confused, and it is difficult to understand fuzzy in a short time. The concept of neural networks is also difficult to practically apply. On the basis of reading a lot of literature, the author organizes, summarizes, and researches various related theoretical knowledge in order to make a systematic overview and preliminary exploration of fuzzy neural networks. This article actually consists of two parts: the first part is an overview of the fuzzy neural network; the second part is the proposal of an algorithm and its implementation process. Fuzzy neural network is a relatively new concept. The article from the historical discussion of neural network systems and fuzzy systems to its origin and development, demonstrates the possibility and necessity of its generation, and briefly introduces domestic and foreign fuzzy logic neural network software hardware. Based on the concept of fuzzy neurons, the fuzzy neural network is defined. From the perspective of function mapping, the function approximation capabilities of the neural network system and the fuzzy system are discussed. Both can approximate any continuous real function with arbitrary precision. The theory is mature The algorithms and models are briefly introduced. Aiming at the specific problems of the fuzzy neural network terminal (network) reality observation, the learning capabilities, capacity, and structural distribution of the network terminal (network) were retrogradely discussed. This paper proposes a two-step hybrid algorithm of fuzzy neural network. In the first step, the fuzzy inference system and genetic algorithm are used to determine the parameters of the membership function according to the training samples. The genetic algorithm is used to search the optimal solution of the parameters in the domain. The ST model is selected as the reasoning model in the fuzzy inference system. The second step is to determine the network structure. Use the BP algorithm to train the network according to the training samples, adjust the network weights and deviations. In order to avoid local minimum observations and accelerate the network convergence speed, an improved BP algorithm with a momentum factor and a variable learning rate is selected as the training algorithm . In order to apply the genetic algorithm more widely, this paper uses C ++ to implement a generic genetic algorithm class library. In the process of actual observation, it uses a combination of class templates and abstract classes. This library supports optimization of one-dimensional and multi-dimensional functions. For multi-dimensional functions, genes of uniform length or with different lengths in each dimension can be used; fixed and variable mutation rates are supported, fixed iteration algebra is completed, and iterations are met under certain conditions. The author uses MATLAB's Fuzzy Toolbox and NNetToolbox to implement the algorithm. Simulation results show that the algorithm has high efficiency, fast convergence speed, and high model accuracy.