Knowledge Discovery Applied to Agriculture Economy Planning

Transcription

1 Knowledge Discovery Applied to Agriculture Econoy Planning Bing-ru Yang and Shao-un Huang Inforation Engineering School University of Science and Technology, Beiing, China, Eail: Abstract: The paper presents double-base cooperating echanis by studying the knowledge discovery based on database (KDD which changes the structure, running process and the echanis of KDD. Then a new Knowledge discovery based on database is established as KDD*. Applied to agriculture econoy planning, the KDD* provides scientific decision for instructing agricultural production. Key Words: Knowledge Discovery, Agriculture Econoy, Decision. 1. Introduction In the agriculture research, anageent and its basic level departent, a large aount of data, exaples, knowledge and experiences have been accuulated. In the field of agricultural crop the data are not ade full use. The accuulated data on seedling, soil, fertilization, water, harful insect of all kinds of crops as well as weather and calaities are saved as archives. That is to say, the phenoenon of plentiful data and poor knowledge is ore serious in agriculture than other. So the deands for knowledge discovery are ore eager. If soe new rules which are produced by dynaic changed factors can be found through finding interrelations of the factors fro the plentiful data, exaples, coon experiences and knowledge, the econoical and social benefits will be very great. The agriculture is a large and coplex syste. The types of soil in the world are enorous. The kinds of crops are coplex. The calaities of harful insects appears frequently and their sypto changes constantly. The interrelations and its effects aong fertilizer, water, density and weather haven t been recognized. This is also the sae with in the livestock, birds, fish and forestry. The relative database and knowledge base are characterized as large, ulti-diension, dynaic, incoplete and uncertain. In recent years the arket inforation didn t flow soothly in any places, especially the crop production planning isn t instructed by the large dynaic arket inforation. It causes blindness in the production planning and great fluctuation in the price which greatly affect agricultural arket econoy. How to collect the inforation in realization and find valuable and regular knowledge so as to effectively forecast and take easures in tie will play an iportant role to the agricultural production. Knowledge acquisition is always regarded as a bottleneck in the realization of intellectual syste. Knowledge discovery partly solved the proble of knowledge acquisition. At present the developent in knowledge discovery is ainly the traditional knowledge discovery based on database (KDD. Soe intellectual ethods, such as fuzzy logic, neural network, rough set and chaotic theory, are used in the KDD. But the KDD lacks eans used by existing knowledge which helps to focus. The hypotheses and rules produced by KDD are directly evaluated. They are set into the knowledge base if passing the evaluation. Then the following defects are fored: first, any eaningless hypotheses are produced. It increases the burdens of evaluation and check on consistency and redundancy. It is close in the process of knowledge discovery; second, the KDD ines according to the need and interest of person, which lacks creative thought of coputer itself to ine heuristically and directly. third, at present there are any experiental verification and original syste but few practical syste and tools. In accordance with the above question, we first present the double base cooperating echanis which is used to ake basic knowledge base liit and drive KDD. This will lead to an open syste of KDD: KDD* which is based on double base cooperating echanis. KDD* breaks through the closeness of KDD. It akes database cooperate with knowledge base through interruptive and heuristic coordinator to find new knowledge. 2. The Introduction of KDD* subsidized by the ephasis ite of National Natural Science Fund ( General Frae of KDD*

2 Set the acquired knowledge into ining KB, check if there are redundant and contradictive Interruptive Coordinator (Direct Mining Acquire hypotheses Evaluate Direct Mining Derived KB According to users need And interested knowledge Focus Heuristic Coordinator (Direct Mining Produce data sub-class, construct ining database according to sub database Search interrelation in knowledge nodes in ining KB, find knowledge shortage decide priority Divide knowledge node, produce ining KB according to attribute Differentiate sub database Differentiate sub knowledge base Preprocess Real KB Basic KB KB Knowledge base This figure shows the logic structure of the syste and the relations between all the parts. Fro this figure we can see that the odules can be divided into the following parts: Pre-processing: To process the original data by purifying the data, specific changing, etc. and create the DMDB which is used in the process of data ining and knowledge discovery. Hypothesis rules: It is the core process of KDD. It uncoonly abstracts the hidden, unknown and potential valuable inforation in database which has the character of large aount of data, incopleteness, uncertainty, structure and causality qualitative reasoning. The forer ethod will be discussed in 2.2. Double base cooperating echanis: to process the acquired rules by using interruptive coordinator and heuristic coordinator, and to exciting the data focusing for data ining by using relative strength. This will be discussed in 3.2. Fig.1 General Frae of KDD* Focusing: naely to chose data fro data ining. The ain ethod in focusing is clustering analysis and detecting analysis. The ethod to direct the focusing are: (i the expert, through an-achine interaction, inputs the knowledge in which he is interested and direct the direction of the data ining. (ii Data directional ining by using heuristic coordinator. sparseness. In the syste the abstracted inforation is causality relation rule. Thus the basic knowledge base will be further iproved. The ining ethods that are used are statistics induction reasoning and Evaluation: this process is ainly used to evaluate the acquired rules in order to decide whether they will be stored into the derived knowledge base. The ain ethods are: (i relative strength sets up a threshold value and be realized by coputer; (ii experts evaluate through an-achine interaction interface and also evaluates all kinds of figures and analysis aterials provided by visual tools. Experts

3 evaluate ainly by using experiences and the relative strength of acquired rules. The rules are stored as new knowledge into the derivative knowledge base after passing evaluation. 2.2 The Reasoning Algorith of Causality Statistics Induction This algorith uses incoplete induction approxiate inference in statistics and credibility theory in uncertainty theory, by counting the exaples in database and using the property with a large aount of exaples as odule, and gets a set of rules by credibility theory. Possessed conditions: data focusing has been copleted i.e. is ready to ine the two language variable A, B (e.g. the kinds of crop and its production. The ining process is as follows: The Coputer Decide the Relativity of the Corresponding Language Value through Statistic Analysis. Divide A, B as A(A 1 A 2. A B (B 1 B 2. B n according to their language values. If A and B are both single variable then we have A(A 1 A 2 A 3 A 4 A 5, B(B 1 B 2 B 3 B 4 B 5. Given A is the intersection of l variables, =5 1. Given B is the intersection of nl variables, n=5 n1. Thus there are altogether n kinds of cobination A i B i=1, 2. =1,2,,n. To the possibility factor Pk=Cnk/N k=1,2,, n corresponding to each coputation, P=0.5 is the highest possibility. If Pk>0.5, A i B is selected, otherwise it is eliinated and these two are considered to have no relativity Analyze A and B through Visual Tools Experts can use visual tools, such as a distribution figure to decide the cobination of the selected or eliinated areas. The areas here have one to one apng relation with the language value entioned above, i.e. the language value and the corresponding radius equals the corresponding area. The acquired area cobination ust be changed into corresponding language cobination which is to be used in the later coputation. Get the two highly relative properties e.g. A i and B, and draw the corresponding values e.g. statistic value N, statistic value Cn A i B appearing both in Ai and B, statistic value Cn A i appearing in A i, and statistic value Cn B appearing in B to decide which variable have causal relation Get Weight of the Preise in the Hypothesis Rule (A i B Given Ai is single preise, its weight is 1; given A i is the interaction of any preises, i.e. rule R: A i B is: R (P 1 p 1 P p P p Then the corresponding ri in P i p i (Pi, and can be gotten fro the following forula. The weight in its rule can be gotten according ri. r = i q ( q ( q( = 1 = 1 = ( q ( q ( q = 1 = 1 = 1 = 1 q Causality statistics induction reasoning algorith flow is shown as following: Input language variable A.B Differentiate A i.b Acquire P k P k >0.5 Y Keep A i B Visual analysis Fig.2 Causality Statistics Induction Reasoning Algorith Flow 3. Double Bases Cooperating Mechanis: 3.1 Basic Theory The technological realization of double-base cooperating echanis is to construct interruptive and heuristic coordinators. To realize the there are soe requireents: The large (basic knowledge base is divided into several correlative subknowledge bases according to each doain; Meanwhile, the real database is divided into N Get weight of the preise in the hypothesis rule (A i B Discard A i B 2 q

4 correlative sub-databases according to each doain. Thus the layers between knowledge nodes in ining knowledge base and data sub-class (structure in ining database ake a one to one apng. The basic theory which is proposed by us is panhootopy conception and the following structure apng theore: (Details can be found in reference [1][2] Theore (Structure Mapng Theore: Aiing at X, in the sub-database corresponding to subknowledge nodes, <E F > of knowledge nodes and <F D> of data sub-class (structure are identical panhootoc type spaces. This theore presents the apng of layers between knowledge nodes in the sub-knowledge base and data sub-class in corresponding sub-database, shown in fig.3. Sub-knowledge base(corresponding to doain X data sub-base(corresponding to doain X knowledge prie node Pi Correspond to soe layers of Si u l -n l -r l u 2 -n 2 -r 2 M u v -n v -r v S i Prie node P Correspond to soe layers of Si Knowledge prie node Decopound M S Decopound into knowledge node first Prie node P Correspond to soe layers of Si M S k Fig.3 corresponding construct graph On the basis of the research above, we can see that in the knowledge discovery syste atheatical structure of database and knowledge base can be essentially coe down to panhootopy category. Naely database is panhootopy category cobined with data sub-type (structure set and ining path, which is called data ining category; and knowledge base is panhootopy category cobined with knowledge nodes set and reasoning arc, which is called knowledge reasoning category. Moreover soe results about the isoorphy and restricting echanis of knowledge reasoning category C R E in <E F >and data ining category C D F in <F, D > are got, and directional searching and directional ining process are solved. 3.2 The Technological Realization Interruptive Coordinator The ain function of the interruptive coordinator is, when the rules (knowledge have been created fro the focusing of the data in the real base, to interrupt the process of the KDD and to search whether there is a repetition of the created rule in the corresponding position of the knowledge base. If so, cancel this created rule and return to the beginning of the KDD. There need soe special technology and ethods to process contradiction. If not, continue the process of the KDD i.e. evaluate and store the result. Because the interruptive coordinator is introduced into KDD, the inconsistent and redundant knowledge can be canceled earlier. Only those who are possibly accepted as new knowledge are evaluated and the evaluation work is greatly reduced. At the sae tie redundancy is processed in real tie. This avoids coplication of proble accuulated in a long tie. In practical expert syste, the aount of rules which finally becoe new knowledge are rather sall copared with the original knowledge (it is difficult to find new knowledge, and a great nuber of rules are repetitive and redundant, so the introduction of interruptive type coordinator into KDD enhance the efficiency Heuristic Coordinator The function of heuristic coordinator is to search irrelevant state of knowledge nodes in knowledge base under the principle of property Knowledge shortage is found. Data sub-class on which knowledge base is established. corresponding to real database uses heuristics and is

5 activated to produce directional ining process. To find the knowledge shortage in knowledge base especially in rule base, one of the ethods is to copute the causality rule strength in each possible knowledge node in the whole causality network. The causality rule strength consists of a group of three factors which can be expressed as π (H E=< α,β,γ > α=cf(e*p(e β=cf(h E γ=cf(h*p(h Aong it CF(E is the reliability of preise, P(Eis pre-probability, CF(H E is the reliability of rule, CF(His the reliability of conclusion, P(H is preprobability. CF(E and CF(H Eare known. It consists of the whole rando and fuzzy uncertain inforation of the rule. According to the causality rule strength the priority of directional ining can be deterined and those can not be ined will be excluded. 4. Properties of KDD* Copared with KDD, KDD* is a new structure of knowledge discovery which blends KDD and double base cooperating echanis. It has the following characters: 1 KDD* organically ake new knowledge found by KDD* counicate and erge with the knowledge in knowledge base and becoe one organis. 2 In the process of knowledge discovery, KDD* processes those redundant, repetitive and incopatible inforation in real tie. This effectively decreases the coplication of proble caused by a accuulated process. At the sae tie the preconditions are given for the erge and fusion of new and old knowledge. 3 KDD* changes and optiizes the process, structure and running echanis of knowledge discovery. 4 Fro cognition KDD* strengthens and provides intellectual degree of knowledge discovery and enhances the cognition of coputer itself. This is the direction for a long tea. 5 Double base cooperating echanis, the core technology of KDD*, shows the apng between sub-knowledge base and data sub-class under a certain principle of establishing base. It provides a valid technology to decrease search space and iprove ining efficiency. 5. Knowledge Discovery in Agricultural Econoy Planning In agriculture syste there are abundant data which for all kinds of database such as relation database, tie-spatial database, obect-oriented database and ultiedia database. But the data in these database are not ade full use and hold plenty of storing space. Therefore it is necessary to ine. In order to find knowledge fro a database, it is necessary to process the database and establish corresponding basic knowledge base. Then All kinds of ethods are used to ine the data in database. For exaple, Seleniu (siplified as Se is a necessary icroeleent for huan and anial. It has any biological functions. Lack of Se is the ain reason for any diseases, such as cataract, astitis, cancer, large bone disease and so on. Rice is one of the ain foods in the world. The content of Se is related to nutrition of Se in the huan body. But ost rice production areas are short of Se or have low content of Se. Therefore if we can find the dynaic changing rules under which rice sorbs Se, it will play an iportant role to instruct agriculture production and iprove huan health. Now there are soe processed agricultural data which are shown in the following tables. KDD* is applied to analyze the data in the table and finds that the accuulation of dry aterial isn t at the sae speed with that of Se in the rice. The peak of forer is in the iddle of growing, the latter in the late. This is a rule that will be stored in Knowledge base. According to the rule we should fertilize Se again before the of filling starch in rice. On the other hand rice has certain ability to sorb Se. So fertilizing Se in those areas that lack Se or have low content of Se can greatly enhance the content of Se in rice and iprove its nutrition quality. Doing so on one hand can instruct us to fertilize reasonably, on the other hand can instruct anufacturer of fertilizer to add different icroeleent in different stage so as to eet the deand of agriculture production. Other data of agricultural crop can be treated so.

6 Bearing Seedling Sc Filling starch Ripe Growing tie (d Table 1 Dry aterial accuulation of rice in the whole bearing Accuulate speed coparative accuulation ( pot -1 d -1 accuulation ( pot -1 accuulation ( pot -1 (% coparative accuulation (% Bearing Seedlin g Sc Filling starch Ripe Growing tie (d Table 2 Se accuulation of rice in the whole bearing coparative Accuulate speed coparative accuulation ( pot -1 d -1 accuulation ( pot -1 accuulation ( pot -1 accuulation (% (% Conclusion Agriculture production is an iportant thing to a country and its people. Reasonable planning for agriculture production will take great effect on a country. The article provides a new ethod of scientific decision for agriculture econoy planning. It decreases the loss caused by planless production and will be instructive to the developent of agriculture. On the basis of KDD, double base cooperating echanis can be applied to ine knowledge autoatically and directionally. It can also process repetitive, contradiction and redundant rules. This will greatly iprove the ining effeciency. The two kinds of coordinator can be independent syste and install any existing KDD software to counicate with original knowledge base. It expands the function of original KDD greatly. Reference [1] Bingru Yang, KD(D&K and Double-Bases Cooperating Mechanis, Journal of Syste Engineering and Electronics, Vol.10, No.1, [2] Bingru Yang, Double-Base Cooperating Mechanis in KDD, International Syposiu on Coputer, (1998. [3] FIM and CASE for Evaluation of Hazavd level Based on Fuzzy language Field, Fuzzy Sets and Syste, North Holland, Vol.95, No.2, (1997. [4] S.S. Anand,D.A.Bell,J.G.Hughs,EDM:A General Fraework for Data Mining Based on Evidence Theory,Data &Knowledge Eng., 18, (1996. [5] R.H.Lchiang,etal.,A Fraework for the Design and Evaluation of Reverse Engineering Methods for Relational Databases,Data &Knowledge Engineering,21,57-77 (1997. [6] H.-L.ong and H.-Y.Lee,A New Visualization Technique for Knowledge Discovery in OLAP,Proceedings of the First Pacific-Asia Conference on Knowledge Discovery and Data Mining, (1997. [7] C.Li and G.Biswas,Unsupervised Clustering With Mixed Nueric and Noinal Data-A New Siilarity Based Aggloerative Syste,Proceedings of the First Pacific-Asia Conference on Knowledge Discovery and Data Mining,35-48 (1997.