A Study Of Fuzzy Neural Network Prediction Method Based On Rough Sets And Condition Number

Based on 48h numerical forecast productions of Japan Fine Grid Precipitation Model and T213 Model on May & June in 2002-2005 and so on the basis of numerical data products. Fuzzy neural network itself is not a predictor of how to choose the appropriate questions, Rough Set attributes by using simple terms and conditions for the screening and analysis of fuzzy neural predictors Network precipitation forecasting model southwestern Guangxi before the 2006 season (May-June), the average daily precipitation forecast modeling.At first, numerical weather prediction model and the Japanese model of the T213 NWP products related to the census. 29 forecast to be the final factor (28 T213 factor, one factor Japan lattice precipitation forecast) 29 to establish this as a predictor of the basic factors of the forecasting model group. Rough Set Attribute Reduction Act used to forecast the results of the 29 factors, factor analysis of the importance of attributes, remove unnecessary factors in the outcome of the policy change, and elect nine predictors 10 predictors. Two sets of data and use this as a fuzzy neural network model of the input matrix for the two-day pre-average precipitation modeling experiments reported. from two months to 56 days (5 days for lack of numerical data, data products factor), the independent samples of precipitation forecast average absolute error poor were 5.81 mm (nine factor) and 5.37 mm (10 factor). Use the same method to calculate the condition number 29 predictors of multi-collinearity, elect nine predictors and 10 predictors. Fuzzy neural network model as input matrix daily average precipitation in the forecast model test from two independent samples on the 56-day precipitation forecast average absolute error of 5.73 mm (nine factor), and 5.85 mm (10 factor).T213 numerical forecasting model using input from the Japanese model and an average absolute error of 8.74 mm, 5.75 mm. From the result, four of the fuzzy neural network model precipitation forecast better results in the T213 model output. A noticeable increase in the overall forecast accuracy than the T213, with the current operational forecasting reputation in Japan is quite good. certain prediction of the actual value. In addition, The paper also widely used at home and abroad to use traditional methods for forecasting the stepwise regression modeling methods for the prediction of clashes A post-mortem test as a predictor of the above two methods based on the primary predictor of group exactly the same. To ensure a reasonable comparison, the pilot of the F-control, a stepwise regression different predictor of the forecast equation. Comparative tests have been forecast by the independent forecasting and modeling samples under identical samples. Stepwise regression method of forecasting precipitation 56 days to two months of independent samples of 7.21 mm forecast average absolute error (nine factor) and 6.09 mm (10 factor). In contrast than using Attributes and conditions established by the method chosen predictors fuzzy neural network model prediction error. Stepwise regression methods are forecast accuracy is not high when only one hand was mainly due to the modeling consider the nature of a significant regression model The impact factor, but not the factor correlation between forecast and other information thus affecting the forecast model prediction capabilities; Owing to the heat of atmospheric precipitation forecast by atmospheric movement, power and the impact of factors such as moisture conditions have significantly changed the characteristics of the nonlinear, Stepwise regression forecasting factor and does not reflect the non-linear relationship between the amount forecast. Eventually lead to better prediction fuzzy neural network model.Another paper in precipitation forecasting fuzzy neural network model is used in feed-forward neural network fuzzy. BP and using feed-forward neural network algorithm to adjust the parameters. Forward-FNN using the multilayer feed-forward neural network, fuzzy relationship mapping.Comprehensive analysis of the results that the proposed use of attribute reduction rough calculation and analysis of two kinds of conditions, several small number of predictors extracted to construct fuzzy neural network input matrix. Fuzzy neural network model greatly reduce input nodes, the nodes and reduce duplication of information and the noise input. Reduction deposing played a significant role. In the actual hydrology, traffic, geological disasters, and many other areas of economic forecasting and modeling, encounter many facing primary predictors, and how to select a more reasonable predictor of the portfolio. Therefore, the result of research in the field of research and the application of predictive modeling is very useful for the promotion prospects.