Research And Applications Of GA-FLANN Based On Entropy-AHP

Actual industrial processes are complex and redundant, but neural network can build a good model for production process utilizing the input and output data without knowing the mechanism process of objects because of its "black box" feature. In recent years, a variety of neutral networks have been proposed at home and abroad, the function link neural network (FLANN) is one of them. Industrial production data are high-dimension and they are severely polluted so they contain many noises, building a model for industry data using the FLANN model directly cannot obtain good training and generalization effect. Therefore this paper proposes an improved FLANN network model and a dimensional reduction method. The proposed methods are tested in standard data set and practical industrial application, the results show that the methods are effective and useful. The main content as follows:(1)The traditional FLANN network training algorithm based on gradient descent algorithm and easy to fall into local optimum in the process of training. This paper proposes a FLANN network based on genetic algorithm (GA FLANN). In order to avoid falling into local optimum, before training the FLANN network the initial weight and threshold are optimized and assigned instead of set manually. Finally, the GA-FLANN is tested in standard data set, the results show that it is more stable and useful.(2)For industrial data contain noise and a lot of redundant information, this paper proposed analytic hierarchy process based on data-driven to extract the nature of data and remove redundant information, it can reduce the input of the GA-FLANN. The final experiment shows the method's effectiveness.(3)In order to verify the practicability of the GA-FLANN, we apply it into the prediction of ethylene industrial production and the testing of food data and build corresponding models. Besides, we build a prototype system based on food safety risk early warning model.