عنوان مقاله
ارزیابی کیفیت آب های زیرزمینی با مدل جدید PSO و RBF
فهرست مطالب
چکیده
مقدمه
مدل شبکه عصبی تابع پایه شعاعی
مدل هیبریدی یا دورگه Pso-Rbf
نتیجه گیری
بخشی از مقاله
لایه ورودی از گره های سورس سیگنال تشکیل شده است. لایه دوم لایه مخفی است که گره هایش با توابع شعاعی مثل تابع گاوسی ساخته شده اند. تعداد گره های لایه مخفی بر حسب نیازهای مسئله تعیین می گردد. لایه سوم لایه خروجی است که از آن برای پاسخگویی به مد ورودی استفاده می گردد.
کلمات کلیدی:
A New PSO-RBF Model for Groundwater Quality AssessmentZhang Bia, Wang Jiayang and Zhang SulanChengdu University of Information Technology, Chengdu 610041, Chinaawjj@cuit.edu.cnKeywords: radial basis function; particle swarm optimization; evaluation model; underground waterAbstract. There are three adjustable parameters in the radial basis function, the center of the basisfunction cj, the width parameterσ and the output unit weight wj. Through optimization theparameters of the radial basis function by Particle swarm optimization algorithm, a neural networkmodel of underground water is generated, which is used to study the grade of underground water inthe ten monitoring points of the black dragon hole. By applying the PSO-RBF model to undergroundwater assessment in the ten monitoring points of the black dragon hole, the results of this evaluation,which correspond with the real conditions, are basically in accord with those obtained by otherevaluation methods, and also show the practicability to groundwater quality assessment.IntroductionRadial Basis Function (RBF) neural network is a feed-forward artificial neural network[1,2], which isbased on the locality that the human brain neurons response to the outside world. RBF is of highcomputing speed and strong nonlinear mapping capability and can approximate an arbitrary nonlinearfunction with any precision. More importantly, the radial basis function (RBF) network provides anew and more effective method for training. Since it can avoid complicated and prolix calculations,its training speed is much faster than that of the BP network[3]. Thus it has been widely used in manyfields.The RBF network contains three adjustable parameters, the center of the basis function cj, thewidth parameter σ and the output unit weight wj. Each node of the output layer is weight estimation.The value of the output unit weight wj can be optimized and determined by the Particle SwarmOptimization (PSO) algorithm. So the hybrid PSO-RBF model was put forward in the article.RBF model has been widely used in many fields, such as the evaluation of underground waterquality. The main task of groundwater quality evaluation is estimate the comprehensive levels of theunderground water quality through the established mathematical model, based on the evaluationindexes of underground water and the water quality assessment standards. In the recent years, somemethods such as principal component analysis, fuzzy analysis, gray cluster method, matter elementanalysis model and artificial neural networks[4-6] are mainly applied to comprehensively theevaluation of the underground water quality. Each method has its advantages and disadvantages.fuzzy analysis, gray cluster method and matter element analysis model need structure many utilityfunctions, which lack of standardization in the function design, and have larger randomness andsubjectivity. So the article adopted the hybrid PSO-RBF model to evaluate the levels of undergroundwater quality.Radial Basis Function Neural Network ModelThe Structure of RBF Network. Radial basis function neural network is a three-layer feedforwardnetwork consisting of an input layer, a hidden layer and an output layer, as shown in Fig. 1(A). Theinput layer is composed of the signal source nodes. The second layer is the hidden layer, whose nodesare constructed by the radialized functions as the Gaussian function. The number of the hidden layernodes is determined by the needs of the problem. The third layer is the output layer, which is used toresponse to the input mode. The transformation from the input space to hidden layer space isnon-linear, while from the hidden layer space to the output layer space is linear, as shown in Fig. 1(B).Advanced Materials Research Vols. 463-464 (2012) pp 922-925© (2012) Trans Tech Publications, Switzerlanddoi:10.4028/www.scientific.net/AMR.463-464.922All rights reserved. No part of contents of this paper may
