The Principle Of Information Diffusion And Thought Computation And Their Applications In Earthquake Engineering

This paper systematically discusses the character of fuzzy information, and shows that the fuzzy information comes from not only the measure with fuzziness, but also non—completeness of sample knowledge. When the knowledge which is provided by sample W doesn' t sufficiently and completely describs the rule which the sample would be kept to, the sample W is called as non—completeness sample knowledge. For example, if we take the samples which are the records of earthquake taken place in China since 1900,they can not exactly explain the relation between epicentral intensity and magnitude. So, the sample's knowledge which is provided by the samples is the non — completeness. Based on the recognition of the relation between the fuzzy properties and non — completeness, this paper indicates that fuzzy information analysis would not be limited in objects which is relative to man—made factor, and it would extend to deal with the knowledge which is relative to non — completeness. So far the paper wholly describes the concept of INFORMATION DISTRIBUTION which was suggested by Pro. Liu Zhengrong and author in our early years and interpretes its theoretical cornerstone efficiently applied in earthquake engineering. At the same time author indicates the various disadvantage of the model .Based on the above, this paper developes the concept of information distribution using the principle of INFORMATION DIFFUSION.The concept of the information diffusions comes from non — completeness sample W in which each sample point w_j is regarded a representative of it "around". This implies w_j to provid infomation not only to the point where the obsevered value was obtaind, but also to the other points around that one. Let Vis the universe of discourse of W , v_j is the observed value of w_j , it is called that W provides the information as measure value in 1 to v_j through w_j. If the infomation in v_j provided by w_j is shared by the points around v_j , it is called information diffusing. The way of the information shared on v_j is called information diffusion way.The papes utilizer Parzen's kernel estimate theory which was used in estimating the function of probability intensity and Chen Xiru's results in the field to strictly prove the following fact. Whenusing non—completeness knowledge sample W to find the law which W would be kept to, the from:is closer to real p(v) than the estimate of rectangular diagram. Where u is non—ordinary diffusion way that makes v obtain the information from v, , m is sample size and Am is called window width. The non—ordinary diffusion is defined that , there is v V , v =￡ Vj, but u(u ,Vj) 0 . Due to the exist of non — ordinary diffusion way, and its estimate is better than the ordinary diffusion way (rectangular diagram method corresponding to undiffusion way),we infer the following fact.The principle of the information diffusion: assuming that W = {W\, w2, ... , wm} is the knowledg sample, V is its universe of discourse, let vj is the observed value of Wj and x = (v — vf) , then , when W is non —completeness, existing function fi(z) , it make that the information obtained on the Vj as measure value 1 can diffuse to v in the value which is /ii.x) . And the primary information distribution Q(v) which is obtained by diffusion method;can be better show the law which W would be kept to.Theoretically, we can strictly prove that the optimizatial diffusion of the W is quardric functionwhere x =-------'-. Due to the optimal selecting of window width Am must be used unknow in-tensity function p(v) the better z). has to be rsearched by self—adapting process. It is not easy to do this work. So, with the help of the molecular diffusion theory in Physics, through researching the similarities of the information diffusion and molecular diffusion and sovling the partial differential e-quation, the paper takes normal diffusion way which is more effective ;where x = v — Vj , h is the normal diffusion coefficient. They can be obtaine from m which is the size of W and maximin an minimin of observed data of W...