20.500.12556/DKUM-40612 MODELI MEHKE LOGIKE IN NEVRONSKE MREŽE ZA ANALIZO GEOTEHNIČNIH KONSTRUKCIJ FUZZY NEURAL NETWORK MODELS FOR ANALYSIS OF GEOTECHNICAL STRUCTURES V doktorski disertaciji smo razvili nove modele za analizo voziščne konstrukcije, podporne konstrukcije in podzemne konstrukcije. Modele smo izdelali z adaptivnimi nevronskimi mrežami in mehkim identifikacijskim sistemom (adaptive network based fuzzy inference system, ANFIS). ANFIS metoda v splošnem omogoča izdelavo geotehničnih modelov, ki imajo večjo sposobnost napovedi kot konvencionalne analitične metode. ANFIS modele smo izdelali na podlagi geomehanskih računskih modelov in optimizacijskih modelov. Optimizacijske modele smo izdelali z nelinearnim programiranjem (nonlinear programming, NLP). Natančnost napovedi modelov je odvisna od nelinearnosti obravnavanega problema. Ugotovili smo, da je v ANFIS modelih bistvenega pomena razvrstitev nevronov. Za ta namen smo razvili ANFIS modele z različno topologijo nevronov in uporabili tisto, ki je imela najmanjšo odstopanje glede na množico testnih podatkov. V doktorski disertaciji razviti ANFIS modeli voziščne konstrukcije omogočajo napovedovanje horizontalne specifične deformacije na dnu asfaltne plasti in vertikalne specifične deformacije na podlagi. Razviti modeli za podporno konstrukcijo s pasivnimi sidri omogočajo napovedovanje faktorja varnosti in optimalnega naklona pasivnih sider. Dobljeni ANFIS modeli za podzemne konstrukcije omogočajo napovedovanje optimalnih izdelavnih stroškov podzemnega skladišča plina in optimalne zasnove kaverne. In this PhD dissertation, we have developed new models for the analysis of pavement, soil nal structure and underground gas storage. The models were developed by ANFIS method (adaptive network based fuzzy inference system). The ANFIS technique has better prediction capability then conventional analitical methods. The ANFIS models were developed on the basis of geotechnical calculation models and optimization models. The optimization is performed by the non-linear programming (NLP) approach. The accuracy of ANFIS models depends on the nonlinearity of the problem and neural network topologies. Therefore, the ANFIS models with different neural network topology were developed. To test and validate the ANFIS models, a data sets were chosen, which was not used while training the network, was employed. In the PhD dissertation ANFIS models are developed for the prediction of pavement horizontal strain at the bottom of the asphalt layer and vertical strain on the top of the sub-grade. The developed ANFIS models for soil nail wall are used to predict safety factor and optimal inclination of soil nail for any design soil nail wall. The ANFIS models for underground gas storage predict the optimal costs per unit of gas and optimal design of underground gas storage. ANFIS NLP voziščna konstrukcija podporna konstrukcija s pasivnimi sidri podzemno skladišče plina ANFIS NLP pavement structure soil nailing underground gas storage true true false P. Jelušič] Slovenski jezik Angleški jezik Doktorska disertacija 2013-06-03 11:47:36 2013-10-17 11:26:39 2022-05-12 23:49:20 2021-04-23 03:12:19 2013 0 [Maribor 0 0000-00-00 NiDoloceno NiDoloceno NiDoloceno 0000-00-00 0000-00-00 0000-00-00 624.1.041:004.8(043.3) 17239574 URN:SI:UM:DK:KQDNVGBH DR_Jelusic_Primoz_2013.pdf DR_Jelusic_Primoz_2013.pdf 1 F94B178C5E88E0FA0FBD40A7730876E1 53e0f0393d1873dd558712b96f5f2f6e03a10823daa5a79d356dc15d261ad805 7b65dd82-7c0a-11eb-bb7a-00155d0001ca 20.500.12556/dkum/815401d9-1d87-482f-a376-6948ef1a509a https://dk.um.si/Dokument.php?lang=slv&id=60062 Fakulteta za gradbeništvo, prometno inženirstvo in arhitekturo 0 0 0