Sigmoid membership function based FALCON and its application

Jie XING,De-yun XIAO
DOI: https://doi.org/10.3969/j.issn.1006-7043.2006.z1.005
2006-01-01
Abstract:As a combination model of basic fuzzy logic and neural network structure, FALCON has both fuzzy logical control and neural learning abilities. Traditional FALCON uses Gaussian function as membership function, resulting in high learning efficiency and accurate approximation but comparatively complicated structure and algorithm. As a modifying, FALCON is adapted using sigmoid function as membership function, so as to get simpler computing and learning algorithm. At first, the basic topological structure of FALCON with sigmoid membership function is described. The operation functions of nodes in all the five layers are illustrated and the learning algorithm is derived. Then, the modified FALCON is compared with the archetypal FALCON. The comparison shows the modification has simpler structure and algorithm, and gets similar capability of the archetype. Eventually, the effect and worthiness of the modification are validated by applying it to probability prediction of anode effect in aluminum electrolysis cell.
What problem does this paper attempt to address?