The Research And Implementation Of Multi-expert Cooperation In Integrated Environment Of Intelligent Applications

With the progress of science and technology and the development ofAI technology, the expert system has already made the good result in theapplications of many fields, such as agriculture, medical science, geology,the trouble diagnosing, etc. However, with the expansion of the application,the problem that people want the computer to solve is more and morecomplicated, at the same time, because of the specialization of theknowledge system, the traditional simplex expert system can't meetdemands of people. By this, solving and stepping the complicated problemof the field through cooperating between the intellectual application systemsbecomes inevitable.This paper regards the 863 project of the country "the Intelligentsystem integrated environmental IEIA" as the background, and it is animportant component of this project. This platform includes a group ofmiddleware, software component, developing instrument and basic classlibrary in order to realize the fast development of Web intellectualapplication system, operating high-efficiently and the mutual operationbetween the heterogeneous intellectual application systems. The expertsystem with distributing nature based on web can be developed through thedeveloping platform. However, A new question-cooperation has been putforward because of the distribution of the system which is developed newly.Namely how to dispatch and assign resource, how to resolve task to giveevery expert system make their work in coordination;And because of theparticularity of the expert system, namely every expert system can solve theproblem in the corresponding field alone ,to a lot of different kinds of expertsystems (such as the expert systems based on the rule, the evidence theoryand the bayes etc.), or the expert systems of the same type but havingdifferences on the knowledge base making different conclusions to the samequestion or phenomenon, how to analyze and synthesize its conclusion inorder to get the overall results of this whole system as good as possible.In view of the above question, this paper according to the cooperationways of multi-expert systems divides it into four kinds: dividing type,mutual-helping type, collective type and hybrid type. Among them,according to the characteristics of the first two kinds of cooperation waysand some methods of analogy MAS cooperates, the cooperation way basedIVon contract network is introduced in the cooperation of multi-expert systems,and analyses the necessity and feasibility of introduction. This paper hasanalyzed the course of calling for bid, submitting a tender and opening bid.Then, according to the model of the contract network, and provided itstheory model. Because the classification of the agricultural expert is moreclear (such as, rice expert , the maize expert ,etc.), this paper has imitatedthe general phenomenon that the mankind solves the problem throughlooking for "acquaintance", uses the concept of the network of personalconnections, and does not broadcast the bid in a traditional way but onlysends the bid to the "acquaintance" which in the "acquaintance" database.The system who received the bid may repeat this course if he can't finishthis task, until he found the system that can solve this problem , so, thismethod reduced the systematic communication expenses and improved thespeed of operation at the same time.To the way of assembling type expert cooperates, as we know above,in many expert systems, each expert can solve the problem in thecorresponding field independently. But each expert of multi-expert systemjust like human expert has the difference in " knowledge structure " and "mode of thinking ", so, the conclusion that many experts draw on the samequestion is more unanimous sometimes, differing greatly sometimes, thispaper calls course of analyzing and synthesizing many expert's conclusionsas the high level reasoning.Direct against this kind of question, this paper imitates the course thatthe medical expert holds a consultation, and introduces the vote of theweighting, the purpose of this vote law is to dispel those conclusions whichare seldom supported, its hypothesis is that the conclusion that most expertssupport is more reliable. At the same time, in order to avoid thephenomenon which can't be drew because of the scattering of the ticket, hispaper considers the mechanism that the minority is subordinate to themajority is given consideration to most expert opinions, one ticket passingconsidering the authoritative expert is provided. To the assertion havingbeen voted, expert maybe have a conflict among these assertions, as to this,the paper is divided into two kinds of situations and solved, while only oneconclusion left, this paper will regard weighted average of each expert'sconfidence degree to this conclusion as the last result;In case of exceedingone conclusion, this article imitates the method to discuss in groups inactual life, adopts the strategy of trooping to divide the experts into groups,with the hypothesis that the experts in the whole set have a relatively greatconflict, but the conclusions of some experts are to tend to unanimous onein the set. In each expert group, this paper uses four consistency treatmentmethods to deal with the expert's confidence degree unanimously, these fourmethods are the supreme authority, the most confident degree, adjoint toinfluence and influence each other. Through the above disposal, confidencedegree to the problem conclusion of each expert in the group has beenadjusted, and made its trend unanimous;This paper uses the decisionmethod to deal with the new confidence degree with the unanimous trend ofexperts in the group, the result of decision represents the expert's suggestionin the whole group.After consistency disposal, the expert set is made up of expert delegateon behalf of each group, to the suggestion of these expert delegates, thispaper has analyzed the range of application of each synthetically function,classified the problem solved by the experts, and divided the problem intoordered question and unordered question. This paper employs thecomprehensive method based on protruding function theory to deal with theordered question, and as to unordered question, this paper makes up thestructure to different problem and expert's set, and adopts comprehensivemethod which is on the basis of MYCIN and simplified evidence theoryseparately. The results of comprehensive treatment are the conclusions ofthe whole system.In concrete implement, in line with the open principle, this paperadopts obtaining from the platform directly, users inputing by hand directlyand the network obtaining three ways to gain the expert's conclusion. Andbecause the XML documents has become the main way of data transmissionof Internet now , this paper provides the data form that XML formats, whichhas improved systematic interaction and expansibility.In the end of the article, this paper demonstrated the course of concreteapplication based on reasoning on the high level multi-expert cooperationalgorithm through the agricultural question such as predicting the output ofgoal of the rice, diagnosing the plant scarce, etc.In sum, this paper respectively provides the corresponding theory modelto the classification of the type of multiple experts' cooperation and differentkinds of cooperation ways. And this paper has provided detailed analysis anddesign of the aggregative way of multiple experts' cooperation throughanalogizing the form of the medical expert consultation.