论文研究-A Fault Diagnosis Modeling Method Combined RBF Neural Network with Rough Set Theory.pdf

所需积分/C币:6 2019-08-20 20:13:51 194KB .PDF

基于神经网络和粗糙集的故障诊断模型,周柳阳,石玉文,为了提高诊断精度和减少错误的诊断,提出了一种基于粗糙集理论和RBF神经网络的故障诊断模型。首先,对数据进行预处理,形成故障诊
血国利技论文在线 http://www.paper.edu.cn stored by fixcd connection structure, so it cannot casy be modified or supplemented. neural network cant process semantic input. It can realize unsupervised clustering learning but can be uncertain of which knowledge is redundant or usefulness RBF neural network is made up of input layer, hidden layer, and output layer. Its structure can be illustrated in Figure 1. The net mapping from input to output is shown in the figure x…,xn)∈R" Is Input vector;W=(w1,wWn)∈ R is output weight matrix;f output vector, is radial basis function; C; is the i-th clustering center of radial basis function. Figure. rbf neural nctwork architectur 0(,) X-C (2) The key problem of RBF neural network is to confirm the number of hidden nodes and the locations and width of corresponding center nodes. When they are confirmed, RBF neural network from input to output will become linear equations. At the same time, output weight vector can be gained by the least square method Knowledge acquisition of substation fault diagnosis based on rough sets theory regards substation fault diagnosis as pattern classification. And then use rough sets theory to process knowledge mining so that it can rapidly judge the fault area and fault component when fault appears. Because of the difference of substation's structure, different substation needs different knowledge acquisition. Based on the scale and complexity of substation, we take two-layer structure to establish substation fault diagnosis model. That is to say, firstly judge the fault area based on fault information, and secondly use fault area and fault information to find the fault component That is shown in Figure 2 国利技论文在线 http://www.paper.edu.cn Obtaining diagnostic original data Establishing the decision table Discretizing the decision table Reduciny the decision lable Producing the decision rule Model (rule set Figurc 2. Fault diagnosis information acquisition modcl The fault diagnosis model based on rough sets theory starts with the feather of rBF neural network fault diagnosis model Its model is illustrated in Figure 3 Discreti zator Deeision table timur conditior: attribute ard corresponding lcar ng sarple Sy ItiptuIIt ckxl processing Test sampl network system Figure 3. Structure of fault diagnosis model The basic idea of fault diagnosis model: analyzing initial sample data, establishing decision table based on history faull information, completing or deleting omission or inefficacy information in decision table by data pretreatment, discretizing continuous attributes in decision table. The reduction proccsscs of dccision table includc attribute reduction and attribute valuc reduction And it processes incompatible rules in decision table, then deletes repeated samples and gets corresponding learning samples set which could cover original data feather and bears minimum condition attribute. Training neural network could gain RBF neural network model of fault diagnosis. Inputting discrete fault symptom and being through rules explanation gets the final diagnosis results. Discretization of the dccision tablc is to bc completion and continuous attribute. We need technical treatment of missing data if it exists when establishing a decision table. rough Sets Theory only processes discretized data. So the continuous data should be processed to be discretized data at first and then taking rough Set Theory. There are many discretization methods at present but different methods have different results The formation of initial decision table--Some parameters in sample data collected initially are repeated or deficient. When comes into being integrated information table, it must delete those repeated parameters and supplement essential characteristic parameters The discretization of continuous attributes--When used Rough Sets Theory deals with the decision table, values in decision table are described by discretized data. We set some divide-points(breakpoints in specific continuous attribute range and divide attribute range to be some discrete Intervals and then use different symbols or integers to represent each interval 4 血国利技论文在线 http://www.paper.edu.cn attribute values. This paper uses a kind of hybrid clustering algorithm a hybrid hierarchical k-means clustering algorithm to discretize the continuous data of decision table. This algorithm combines with advantages of hierarchical clustering method and k-means clustering method and it overcomes their disadvantages. The thought way is to carry on hierarchical clustering first and then get some initial information. Using k-means clustering method refines it. High quality clustering results are gotten in the end The formation of decision table--To adopt those attributes or attribute values which has been discrete forms two-dimensional table. Each line describes an object and each column stands for an attribute The attribute reduction of decision table--Reducing decision table attribute is deleting the unnecessary condition attribute. Thus after analyzing we get condition attribute in reduction for decision rules of decision attribute. And people hope to get fewer conditions of reduction results if possible or get fewer decision rules. There is a kind of attribute significance reduction algorithm based on Rough Sets Theory. It is illustrated as follow: a Discretizing attribute values of the decision table in awaiting analysis b. Calculating the relative core core(C, D) c. Calculating the dependent degree y(,) e. or any attribute∈ red C.Calculating the significance sG(, red, SGF h. Calculating the dependent degree y( red, i If y( red, )=Y(,) g Return red; jump out of cycle; or switch to according to the reduction that we have calculated the results with same attribute values are combined RBF neural network model and learning of test sample--According to the input of neural network model, selecting corresponding training data and attributes from initial continuous attribute decision table trains network and uses corresponding test samples to test Diagnosis results--RBF neural network model will be tested by test samples. If it couldnt achieve pre-requirement, repeat B-E until it output satisfied results Our work demonstrated that neural network can be used to develop the complex mappings uired in a high dimensional data. The be effectively used to preprocess Lh training dataset of the neural network. The training samples of the neural network are dramatically reduced by rough set, and the neural network structure can be simplified so that the training time was decreased. It can effectively eliminate false and omit alarm of fault diagnosis and offer fault diagnosis technology a new kind of thought and method [1] Pawlak Z. Rough sets[J]. International Journal of Computer and Information Science. 1982, 11: 341-356 [2] Masahiro Inuiguchi. Attribute Reduction In Variable Precision Rough Set Model. International Journal of Uncertainty, Fuzziness and Knowledge-Based Systems, 2006, 14(4): 461-479 [3] Lee H.J., Ahn B.S., Park Y M. A Fault Diagnosis Expert System for Distribution Substations[]. IEEE Trans on Power Delivery, 2000, 15(1):92-97 [4] Lee H J, Ahn B S, Park Y M(2000).A FaultDiagnosis Expert System for Distribution Substations, IEEE Trans on Power Delivery Vol. 15, No 1, 92-97 血国利技论文在线 http://www.paper.edu.cn [5] SUN YS, YUAN F.Y., YU ZE, etc. Extraction of Decision Rules Based on Rough Set and Evidence Theory Journal of Jilin University, 2007, Vol 45, No 4: 577-581 [6] Ryszard n. evaluation of vibroacoustic diagnostic symptoms by means of the rough sets theory[J]. Computers n Industry,1992,20(2):141-152 [7 H. Goto, Y. Hascgawa, and M. Tanaka, "Efficient Scheduling Focusing on the Duality of mPl Representatives, Proc. IEEE Symp. Computational Intelligence in Scheduling(SCIS 07), IEEE Press, Dec 2007,Pp.57-64,doi:10.109SC2007.357670 [8] Wang B. Wei W.J., Zhang B etc. Support vector clustering based on rough set Journal of Jilin University:,2007,37(4):851-853 [9 Zhang W.x., Qiu G.f. Uncertain decision making based on rough sets. Qinghua University Press, 2006 [10] HAO L Na, WANG W, WU G.Y., etc. Research on Rough Set-Neural Network Fault Diagnosis Mcthod. Journal of northeastern Univcrsity, 2003. VoL24, No.3: 252-255 [11] Swiniarski, R. W,& skowron. A.(2003). Rough setmethods in feature selection and recognition PatternRecognition Letters. 24.833-849 [12] Pawlak, Z,& Skowron, A(2007a). Rough sets and Boolean reasoning. Information Sciences, 177, 41-73 Pattaraintakorn, P, Cerconeb, N,& Naruedomkul, K (2006). Rule learning: Ordinal prediction based on rough sets and soft-computing Applied MathematicsLetters, 19, 1300-1307. [13] He, Y, Hu, S.(2004). A decision analysis method based on rough-fuzzy sets integration model. Controland Decision, 19(3). 315-3

...展开详情
img
  • 至尊王者

    成功上传501个资源即可获取

关注 私信 TA的资源

上传资源赚积分,得勋章
    最新推荐