Search

U ovom radu opisan je postupak generiranja i optimizacije rasporeda zaposlenika tvrtke korištenjem genetskog algoritma. Općenito, evolucijsko računanje već se dugo koristi za rješavanja ovakvih tipova problema gdje je potrebno pretraživati velika područja svih mogućih rješenja u najkraćem mogućem roku. Budući da su takvi optimizacijski problemi najčešće kompleksnosti "NP-hard", "brute force" pristup nije primjenjiv jer obično košta previše procesorskog vremena. U rješavanju našeg problema koristili smo sve genetske operatore (selekcija, križanja i mutacija), dok je za potrebe implementacije razvijena aplikacija u programskom jeziku Java. Dobiveni rezultati u gotovo svim slučajevima predstavljaju optimalna rješenja (rasporede), a kada nije moguće doći do optimalnog rješenja, naš pristup daje jedno ili više rješenja koja su najbliža traženom optimumu. Pojedini genetski operatori i dobiveni rezultati opisani su u nastavku.

U ovom radu opisan je postupak generiranja i optimizacije rasporeda zaposlenika tvrtke korištenjem genetskog algoritma. Općenito, evolucijsko računanje već se dugo koristi za rješavanja ovakvih tipova problema gdje je potrebno pretraživati velika područja svih mogućih rješenja u najkraćem mogućem roku. Budući da su takvi optimizacijski problemi najčešće kompleksnosti "NP-hard", "brute force" pristup nije primjenjiv jer obično košta previše procesorskog vremena. U rješavanju našeg problema koristili smo sve genetske operatore (selekcija, križanja i mutacija), dok je za potrebe implementacije razvijena aplikacija u programskom jeziku Java. Dobiveni rezultati u gotovo svim slučajevima predstavljaju optimalna rješenja (rasporede), a kada nije moguće doći do optimalnog rješenja, naš pristup daje jedno ili više rješenja koja su najbliža traženom optimumu. Pojedini genetski operatori i dobiveni rezultati opisani su u nastavku., This paper describes the process of generating and optimizing company employee schedules using genetic algorithm. In general, evolutionary computing has long been used to solve these types of problems where it is necessary to search large areas of all possible solutions in the shortest possible time. Since such optimization problems are most often of "NP-hard" complexity, the "brute force" approach is not applicable because it usually costs too much CPU time. In solving our problem, we used all genetic operators (selection, crossover, and mutation), while for the needs of implementation, an application in the Java programming language was developed. The obtained results in almost all cases represent optimal solutions (schedules), and when it is not possible to reach the optimal solution, our approach gives one or more solutions that are closest to the required optimum. The individual genetic operators and the obtained results are described below.

Problem raspoređivanja vozila s vremenskim prozorima (DARP) je važan logistički NP-težak problem koji je začet kao verzija Problema usmjeravanja vozila (VRP), koji je pak podvrsta vrlo poznatog računalnog problema, Problema trgovačkog putnika (TSP). U ovom radu istražuju se stohastične metode optimizacije DARP-a, specifično genetski algoritmi. Genetski algoritmi iz obitelji Evolucijskih algoritama su se kroz nedavnu povijest pokazali kao važan alat moderne računarne znanosti u optimizaciji teških problema. Inspirirani prirodnom selekcijom, na razne načine imitiraju mehanizme preživljavanja najbolje prilagođenih jedinki jedne vrste, i odumiranja onih manje prilagođenih. U ovom radu primijenile su se neke generalne metode genetskih algoritama, ali su se razvile i nove metode specifične ovom problemu kako bi se dodatno proučio i unaprijedio proces optimizacije. Iskušane su različite verzije genetskih algoritama, te su se usporedili i komentirali dobiveni rezultati. Istražene su prednosti i mane ovakve optimizacije, te su postignuti rezultati uspoređeni s postojećim rezultatima drugih metoda. Implementirane su i jednostavne heuristike za optimizaciju, te su uspoređene s radom genetskih algoritama. Komentirana su i područja za poboljšanje metoda korištenih u ovom radu, kao i smjernice za budućnost. Dial-a-ride problem (DARP) is an important logistic NP-hard problem that was concieved as a version of the Vehicle Routing Problem (VRP), which is a subtype of a famous computer science problem, the Traveling Salesman Problem (TSP). In this thesis we explore stohastic methods for optimizing DARP, specifically the Genetic Algorithms. Genetic Algorithms, from the family of Evolutionary Algorithms, have in recent history proved themselves an important tool in modern computer science in optimizing hard problems. Inspired by natural selection, they imitate survival mechanisms of best adjusted individuals of a species in various ways, and also the mechanisms that deny less adjusted individuals reproduction and survival. Some general methods were used in genetic algorithms were applied in this thesis, but also some new methods specific to this problem were developed in order to additionaly explore and improve upon the optimization process. Various versions of genetic algorithms were tried, and the results were compared to the existing results of other methods. Also, simple heuristic methods were implemented, tested, and compared to the performance of genetic algorithms. Areas to improve this implementation were discussed, as well as points for the future.

V diplomskem delu rešujemo problem optimizacije, s katerim se na dnevni ravni srečuje vedno več logističnih podjetij in ostalih prevoznikov. Ustrezna rešitev problema usmerjanja vozil pomeni za podjetje precejšen prihranek na gorivu in s tem znižanje vsakodnevnih stroškov. Reševanje problema smo izvedli s pomočjo naravno-navdihnjenih algoritmov, ki v svojem delovanju opisujejo principe biološke evolucije. Za izvedbo eksperimenta smo razvili sistem, ki omogoča aplikacijo poljubnega evolucijskega algoritma na problem usmerjanja vozil. Eksperiment sestoji iz testiranja uporabe petih evolucijskih tehnik na petih primerih problema. Analiza rezultatov je sestavljena iz pregleda različnih nastavitev, uporabljenih za reševanje in opisa pridobljenih rezultatov, ločenega na primerjavo pridobljenih fitnes vrednosti in časov reševanja algoritmov. In the thesis we are solving optimization problem, which more and more logistics companies and other carriers meet with at a daily level. An appropriate solution for the vehicle routing problem means considerable savings on fuel for the company and thereby reduction of daily expenses. We tackled the problem using nature inspired algorithms, which describe the principles of biological evolution in their operation. In order to execute the experiment, we developed system that allows application of any evolutionary algorithm to the vehicle routing problem. The experiment consists of testing the use of five evolution techniques on five vehicle routing problems. The analysis of results consists of an overview of the various settings that were used to solve the problem and the description of obtained results, compared on fitness values and solving times of algorithms.

Za optimiziranje procesov, ki so v današnji industriji čedalje kompleksnejši in vedno bolj avtomatizirani, se ne moremo več zanašati na preproste izračune in izkušnje. Zato vedno pogosteje uporabljamo diskretne simulacije, ki predstavljajo digitalni dvojček realnega sistema. S pomočjo simulacije lahko preizkušamo različne scenarije, tudi povsem hipotetične, ne da bi motili realni sistem. Problem delavniške proizvodnje (angl. JSSP) predstavlja elementarno težavo planiranja zaporedja proizvodnje različnih naročil. Cilj optimizacije je bil najti zaporedje naročil, v katerem bodo vsi proizvodi končani v cim krajšem času. Naš 10-dimenzionalni problem smo optimizirali naključno in z genetskim algoritmom. Z obema metodama smo pridobili enak najboljši rezultat (2 d 8 h 44 min), kvalitativno pa smo lahko vrednotili le konvergiranega, ki ga je pridobil genetski algoritem. Ta se je izkazal kot dobro orodje simulacijske optimizacije, saj v kratkem času preveri veliko število permutacij, s pomočjo mutacij in križanja pa relativno hitro in z zadostno stopnjo gotovosti najde aproksimacijo optimalne vrednosti. Prav zaradi enostavnosti, hitrosti in relativne natančnosti ostaja med vodilnimi algoritmi optimizacije aplikativnih simulacij. Todays industrys increasingly complex and automated processes cannot be optimised efficiently only using simple equations and experiences. Therefore the use of discrete simulations is on the rise, presenting real systems with digital twins. Using a simulation we can test different scenarios, even hypothetical ones, without disturbing the real system. Job shop scheduling problem represents a basic production planning problem of different sets of products. The goal of optimisation was finding the sequence of orders that equates in the shortest possible production time. Our 10 dimensional problem was optimised using pseudo-random generated sequences and using genetic algorithm. Both methods produced same optimisation result (2d 8h 44min), but we were only able to qualitatively assess the converging result, that of genetic algorithm. The algorithm has proved itself to be very useful solving simulation optimisation problems, with its fast calculations including mutations and cross-overs. It produces approximate results of the optimal solution, which is usually sufficient for real life applications. Its simplicity, speed and relative accuracy place it amongst the most used simulation optimisation algorithms.

Pri izdelavi animacij, ki se uporabljajo predvsem pri izdelavi animiranih filmov, pogosto prihaja do problemov, ko je scena predstavljena s skupino premikajočih se objektov. Animiranje vsakega objekta oz. člana skupine je namreč časovno potratno, zato se v praksi uporabljajo simulacije, ki izračunajo gibanje skupine. Pri animiranju skupin, kjer se člani zavedajo sosedov v skupini in okolice, se za simulacijo lahko uporabi model kolektivnega gibanja. Simulacije prihranijo veliko časa pri delu, a z njihovo uporabo ni natančnega nadzora nad gibanjem posameznih članov. Simulacijo, ki uporablja model kolektivnega gibanja, lahko parametriziramo do te mere, da se animacija skupine izvede po naših željah. V tej nalogi so uporabljeni genetski algoritmi za izbiro optimalnih parametrov modela kolektivnega gibanja. Če poteka gibanje skupine brez trkov med člani ali z okolico, je videti takšno vedenje članov skupine bolj avtentično. Genetski algoritmi so bili uporabljeni tudi zato, da se je zmanjšal čas računanja simulacije. Implementiran je bil program za izvajanje takšnih simulacij, preverjen pa je bil na enostavnem scenariju, kjer se je skupina pojavila na določenem začetnem območju in se je morala mimo ovir premakniti do cilja. Z izdelanim programom so se izvajale simulacije z različnimi nastavitvami, za katere so se beležili trki in meril čas računanja posamezne sličice. Iz rezultatov je razvidno, da se lahko s tem programom izberejo optimalne nastavitve simulacije kolektivnega gibanja za uporabljen scenarij. S programom se lahko prepreči trke v simulacijah skupin z velikostjo do 340 članov. Zaradi uporabe genetskih algoritmov se je zmanjšal tudi čas računanja posamezne sličice simulacije. In the production of animated films, problems frequently arise in scenes with groups of moving objects. Animating each object or member of the group individually can be very time consuming, which is why simulations that calculate all the movements are normally used in practice. If group members need to be aware of other members and their surroundings, the most appropriate model for simulating the movement is a model of flocking behaviour. Such simulations save a lot of time at the expense of precise control over individual group members. To achieve the desired movement of the group, flocking simulation settings are iteratively changed until a satisfactory result is achieved. This thesis tests and evaluates the use of genetic algorithms for selecting the optimal parameters of simulated flocking. The aim was to simulate group movement without collisions between group members or with the surroundings to get an authentic animation of group members' behaviour. A programme for flocking simulations was implemented and tested on a short scenario. A group of objects scattered across a designated starting area had to pass an obstacle course to reach a final destination. The programme was run at different settings, and collisions and frame calculation times were recorded. The results show that the programme was able to set optimal parameters for flocking simulations in the scenario. Measurements show that groups of up to 340 members can be simulated without collisions using the programme. Additionally, the use of a genetic algorithm reduced the time needed to calculate the simulation.

Zbog svog značajnog uticaja na energetske performanse i cenu gradnje, oblik i položaj objekta predstavjlaju važne faktore u projektovanju zelenih zgrada. Međutim, zbog velikog broja promenljivih i složenosti funkcija cilja optimalno projektovanje energetski efikasnih zgrada spada u teške kombinatorne probleme višekriterijumske optimizacije. Stoga je neophodno opisati konstrukciju i njen položaj u okruženju što je moguće preciznije sa što manje promenljivih. Prikazano istraživanje bilo je usmereno ka iznalaženju podesne metodologije za definisanje geometrije i orijentacije date zgrade, kao i njenih elemenata koji su od značaja za analizu energetske efikasnosti. Kako se osnova zgrade predstavlja zatvorenim poligonom, razmatrani su i ocenjivani različiti vidovi opisivanja geometrije poligona u cilju utvrđivanja mogućih problema kao što su epistaza, do koje dolazi kad jedno rešenje „maskira“ druga, i kodni izomorfizam, što je pojava da se isto rešenje pojavljuje više puta opisano na različite načine. Stoga su upoređena dva vida opisivanja geometrije kako bi se utvrdio njihov uticaj na efikasnost i efektivnost pretrage oblasti definisanosti. U optimizacijski model su pored promenljivih vezanih za geometriju uključene i još neke karakteristike fasadnih zidova, kao što su veličina i tip prozora i nadstrešnice. Matematička formulacija problema zasnovana je na dve funkcije cilja, a to su uticaj na okolinu i ukupni troškovi tokom čitavog životnog ciklusa objekta. Mwtodologija je testirana na studiji slučaja, odnosno na primeru tipskog sprata poslovne zgrade petougaone osnove primenom višekriterijumskog genetskog algoritma.

V zaključnem delu je predstavljen začetni del razvoja enofaznega motorja brez utorov, ki predstavlja enega od možnih motorjev, ki bi lahko bili uporabljeni v naslednji generaciji ročnih sesalnikov. Osrednji del naloge je analitični model za izbrano geometrijo in tip motorja. Na podlagi že izpeljanih enačb za izračun polja v zračni reži je bil analitično izračunan električni navor, inducirana napetost ter induktivnost glavnega navitja. Model omogoča tudi izračun izgub v bakru in železu. Preverjen je bil z MKE v programu Flux, ki je pokazal zelo dobro ujemanje. Rezultati elektromagnetnega modela služijo kot vhodi za v analitični termični model, ki je prav tako del naloge. Analitični model je bil uporabljen za optimizacijo motorja pri različnih kriterijih. Razvit je bil tudi tranzientni model, ki služi kot osnova za razvoj krmilnega algoritma in elektronike. V zadnjem poglavju je predstavljen še začetni koncept in prototipno tiskano vezje. The main part of the work is the analytical model of single phase slotless motor for the chosen geometry. Based on already derived equations for calculating the field in the air gap, the electric torque, induced voltage and the inductance of the main winding were analytically calculated. Calculation of copper and iron losses was also done. The model was checked by FEM in the Flux program, which showed a very good match. The results of the electromagnetic model serve as inputs to the analytical thermal model, which is also part of the work. The analytical model was used to optimize the motor for various criteria. A transient model was developed, which serves as a basis for the development of a control algorithm and electronics. In the final chapter, the initial concept of electronics, and PCB is presented.

Background and Purpose: In a complex strictly hierarchical organizational structure, undesired oscillations may occur, which have not yet been adequately addressed. Therefore, parameter values, which define fluctuations and transitions from one state to another, need to be optimized to prevent oscillations and to keep parameter values between lower and upper bounds. The objective was to develop a simulation model of hierarchical organizational structure as a web application to help in solving the aforementioned problem. Design/Methodology/Approach: The hierarchical structure was modeled according to the principles of System Dynamics. The problem of the undesired oscillatory behavior was addressed with deterministic finite automata, while the flow parameter values were optimized with genetic algorithms. These principles were implemented as a web application with JavaScript/ECMAScript. Results: Genetic algorithms were tested against well-known instances of problems for which the optimal analytical values were found. Deterministic finite automata was verified and validated via a three-state hierarchical organizational model, successfully preventing the oscillatory behavior of the structure. Conclusion: The results indicate that the hierarchical organizational model, genetic algorithms and deterministic finite automata have been successfully implemented with JavaScript as a web application that can be used on mobile devices. The objective of the paper was to optimize the flow parameter values in the hierarchical organizational model with genetic algorithms and finite automata. The web application was successfully used on a three-state hierarchical organizational structure, where the optimal flow parameter values were determined and undesired oscillatory behavior was prevented. Therefore, we have provided a decision support system for determination of quality restructuring strategies. Ozadje in namen: V kompleksnih striktnih hierarhičnih organizacijskih strukturah se lahko pojavijo neželene oscilacije stanj, kar do sedaj še ni bilo zadovoljivo naslovljeno. Da preprečimo neželene oscilacije je potrebno optimirati vrednosti parametrov, ki določajo prehode med stanji strukture in fluktuacije iz stanj. Hkrati je potrebno obdržati vrednosti parametrov znotraj določenih meja. Cilj je bil razviti simulacijski model hierarhične organizacijske strukture v obliki spletne aplikacije, ki bi pomagal pri reševanju prej omenjenega problema. Metodologija: Hierarhična struktura je bila modelirana po principih sistemske dinamike. Problem neželenih oscilacij je bil naslovljen z uporabo determinističnega končnega avtomata, medtem ko smo vrednosti parametrov pretoka optimirali z genetskimi algoritmi. Vsi omenjeni pristopi so bili implementirani kot spletna aplikacija z JavaScript/ECMAScript programskim jezikom. Rezultati: Genetski algoritmi so bili testirani na znanih problemskih instancah za katere so znane optimalne analitične rešitve. Deterministični končni avtomat je bil verificiran in validiran na hierarhičnem organizacijskem modelu s tremi stanji, kjer smo z njegovo uporabo uspeli preprečiti oscilacije v organizacijski strukturi. Zaključek: Rezultati nakazujejo, da smo uspešno implementirali hierarhični organizacijski model, genetske algoritme in deterministični končni avtomat z JavaScript programskim jezikom v obliki spletne aplikacije, ki se lahko uporablja na mobilnih napravah. Razvito spletno aplikacijo smo uspešno uporabili na hierarhičnem organizacijski strukturi s tremi stanji, kjer smo določili optimalne parametre pretoka in fluktuacij ter preprečili neželene oscilacije stanj. S tem smo zagotovili sistem za podporo odločanju, ki omogoča določanje kakovostnih strategij za prestrukturiranje organizacij.

In the framework of the doctoral dissertation, it is presented a development of control, modeling and optimization of the casting process, based on application of artificial i.e. computational intelligence methods. Improving of production processes in foundries, in the framework of researches that contains doctoral dissertation consist of modeling of metal melting process, optimization of sand casting system and control of gravity casting process. The first part of research deals with modeling of melting metal in furnaces. It is a complex process due to dynamics of non-linear chemical reactions. Conventional mathematical and thermodynamical models, which are certainly helpful, are not fully reliable due to complex interactions of process variables. Neural networks, as one of the key techniques of computational intelligence, are able to identify internal relationships through training examples. As such, they are used as an effective tool in defining the needed chemical composition of the molten metal by providing alloying process improvement. Proper and complete mold filling is the main objective of the sand casting process, which ensures high quality of the casting part. To this end, the designs of the gating system and feeding system is of major importance, as improper designing of the system results in a number of defects in the casting process, e.g. low quality of the casting part. Тherefore, the gating system aims at providing a smooth, uniform and complete filling of the mold with pure, molten metal. Smooth filling eliminates turbulence, uniform filling ensures continuous mold filling, whereas complete filling allows the molten metal to reach minor, end cavities of the mold. During cooling casting part comes to the volume metal shrinkage, ie. reducing the volume of casting part. The importance of feeders is in their function to compensate the lack of molten metal in the casting part due to volume shrinkage. The second part of research is an optimal construction of gating system and feeders, which satisfy the requirements with high quality casting part and savings of materials. The combination of global optimization method and CAD/CAM software has provided the possibility of effective optimization, design and verification of optimal solutions by numerical simulation. Unlike some other industrial processes, many casting plants have not benefited from the advanced automation therefore the process of filling the mold cavity is still manually performed. Automated systems that minimize the participation of workers and enhance the precision of casting, are also the subject of research in this doctoral dissertation. The part of research, which is dedicated to the development of casting process control, contains a proposal of an automated system in casting plants, development of laboratory simulator of the casting process and the development of possibility of application of intelligent control of mold filling.

U radu je opisan matematički model koji je korišten za potrebe optimiranja parnoturbinskog dijela kombiniranog postrojenja. Generator pare na otpadnu toplinu modeliran je na način da različiti krugovi pare, uključujući i međupregrijanje prvog kruga, čine skup izmjenjivača u kojima se može izmjenjivati toplinski tok između dimnih plinova i radnih. Izmjenjivači mogu biti u međusobnom paralelnom i serijskom razmještaju. Za radni medij prvog kruga odabrana je voda dok je u drugom krugu pare, osim vode, korišten radni organski fluid. Provedena je termodinamička i eksergoekonomska optimizacija rada postrojenja, gdje su optimizacijske varijable bile razmještaj izmjenjivačkih površina unutar generatora pare na otpadnu toplinu, te parametri radnog medija svakog kruga pare i međupregrijanja. Kod termodinamičke optimizacije funkcija cilja je bila maksimiranje termodinamičke iskoristivosti parnoturbinskog postrojenja, dok je kod eksergoekonomske optimizacije funkcija cilja bila minimiziranje eksergijskih gubitaka i minimiziranje investicijskih troškova. Kao optimizacijski alat koristili su se genetski algoritam i gradijentne optimizacijske metode. Matematički model je napravljen u programskom paketu Matlab. Dokazano je da se korištenjem paralelnih i serijskih razmještaja izmjenjivačkih površina, te korištenjem vode u prvom krugu pare i organskih radnih medija u drugom krugu pare, mogu postići više vrijednosti termodinamičke iskoristivosti parnoturbinskog postrojenja. In this work a mathematical model of a steam cycle (bottom cycle, Rankine cycle), which is part of combined cycle power plant, was made. The heat recovery steam generator is modeled in such a way that various pressure levels, including reheating of first pressure level, form a heat exchangers network in which heat can be exchanged between flue gases and working fluids in their parallel and serial configuration. Water is chosen as the working fluid of the first pressure level, while for the other pressure levels an organic fluid is chosen as the working fluid. A thermodynamic and exergoeconomic optimization was conducted. Optimization variables were configuration of heat exchangers inside the heat recovery steam generator and operating parameters of working fluid of each pressure level and reheating. In the case of thermodynamic optimization objective function is represented by the maximum thermodynamic efficiency of the bottom cycle, but in the case of exergoeconomic optimization objective function is represented by minimal exergetic losses and minimal investment costs. The genetic algorithm and gradient optimization methods were used as optimization tools. The mathematical model was made in Matlab software package. In comparison to other available works it is shown that by using parallel and serial configuration of heat exchangers and by using water in the first level pressure and organic working fluids in second pressure level better thermodynamic efficiency of the bottom cycle can be achieved.

Optimizacija je v današnjem konkurenčnem in hitro odzivnem okolju bistvena za doseganje najboljših rezultatov in uspešno poslovanje. V podjetju Štore Steel d.o.o. nenehno stremijo k izboljšavam in povečanju produktivnosti posameznih proizvodnih obratov. V tem magistrskem delu se optimizacija nanaša na valjarno, in sicer na proces valjanja okroglih profilov jeklenih palic. Pri valjanju za preoblikovanje obdelovanca uporabljamo valjarska ogrodja, v katera so vstavljeni valji. Valji imajo po svojem obodu postružene oblike-kalibre, s katerimi z natančnim vodenjem valjanca neposredno v kalibro dajemo valjancu novo obliko in dimenzijo prereza. Vodenje se izvaja s skrinjami, ki so montirane na ogrodjih, pred in za valjem. Pri prehodu valjanja iz ene dimenzije na drugo se v sistemu pojavi prilagoditev posameznih valjev, tako da bomo z novo postavitvijo dosegli želeno dimenzijo. Ob tem se pojavi tudi prestavitev skrinj, tako da vodijo valjanec v zahtevano kalibro. Z analizo valjanja, opreme, s katero ga izvajamo, in planov, ki ji v proizvodnem procesu upoštevamo, smo iskali optimizacijo razmestitve kaliber na valjih ob montaži več skrinj na ogrodje. S tem smo želeli zmanjšati število menjav skrinj in zastoje, ki jih menjava povzroči. Z analizo valjev, kaliber in pripadajočih skrinj smo ugotovili glavne pogoje za optimizacijo. Za iskanje rešitve smo uporabili genetski algoritem. Cilj magistrskega dela je bil zmanjšati število menjav za 20 %, s predstavljeno optimizacijo pa se je število menjav zmanjšalo za 36,3 %. S tem je zastavljeni cilj dosežen. In today's competitive and highly responsive environment optimization is crucial for achieving the best results and successful conduct of business. Štore Steel d.o.o. strives to better and optimize the productivity of each individual production site. The optimization discussed in this master's thesis refers to the rolling mills namely to the process of rolling round profiles of metal rods. When rolling steel for the purpose of transforming a workpiece we use rolling frames with cylinders. The cylinders have sculpted shapes-calibers all over their circumference. Through precise guiding of the rolled product directly to the caliber we give the rolled product a new shape and a new dimension of the cross-section. The guiding is done with the help of the rolling mill guides that are attached to the rolling frames which are located before and after the cylinder. In rolling, when transitioning from one dimension to another, the adjustment of individual cylinders appears in the system in such a way that we will achieve the desired dimension with the new setting. The rolling mill guides move in such a way that they guide the rolled product in to the desired caliber. By analyzing the rolling process, the equipment with which the rolling is done and the plans of the rolling process, we searched for the optimization of the caliber layout on the cylinders with multiple rolling mill guides on the framework. With this we wanted to minimize the number of moving the rolling mill guides by 20 % and the halts such a move causes. By analyzing the cylinders, calibers and the corresponding rolling mill guides we found the main conditions for optimization. When searching for the solutions we modeled a genetic algorithm. The goal of this thesis was to minimize the moving of the rolling mill guides by 20%. With the before mentioned optimization the number of the moves decreased by 36,3%. With this the goal of the thesis is achieved.

V magistrskem delu smo uporabili algoritme po vzorih iz narave za finančna modeliranja. Najprej smo uporabili umetne nevronske mreže za napovedovanje cene delnice, nato pa še genetske algoritme za optimizacijo portfelja delnic, ki smo jih primerjali s kvadratnim programiranjem. V raziskavi se je izkazalo, da lahko s umetnimi nevronskimi mrežami bolje ocenimo variančno-kovariančno matriko, kot če bi uporabili zgodovinske podatke. Pri reševanju problema optimizacije portfelja delnic se je izkazalo, da lahko z genetski algoritmi dobimo rezultate primerljive s kvadratnim programiranjem, saj rezultati med tehnikama, predvsem pri manjšem porteflju, v glavnem niso statistično značilni. In a master work we used nature inspired algorithms for financial modeling. Firstly we use an artificial neural networks to predict stock prices and secondly, we used genetic algorithms for stock portfolio optmization. The results show better assessing covariance matrix with neural networks gives more accurate results in a portfolio optimization than if we are taking historical prices. We can assert that results obtained with a genetic algorithms are in general statistically the same as they are with quadratic programming, especially in cases with less stocks in a portfolio.

Optimizacijom vodovodnih sistema u postupku projektovanja može se ostvariti značajna ušteda u poređenju sa tradicionalnim pristupom Trial and Error. Najčešće se koriste genetski algoritmi kao optimizacioni metod. Razmatra se ekonomski aspekt kao kriterijum za izbor optimalnog rešenja, i primer kako pravilno formulisati kriterijumsku funkciju. Da bi bila ispunjena hidraulička ograničenja, npr u pogledu minimalnih i maksimalnih pritisaka, koristi se kaznena funkcija koja rešenja koja ne ispunjavaju ove uslove udaljava iz skupa povoljnih.

Štore Steel d.o.o. je fleksibilna mini jeklarna, specializirana za dobavo jekel v manjših serijah. V letu 2010 je predviden zagon nove konti proge s tehnično letno kapaciteto 250.000 ton. Nova konti valjarska proga bo poleg povišane produktivnosti omogočala bistveno višji nivo kvalitete valjancev (tolerance, ravnost, površina, trdote), obenem pa s sodobnejšo avtomatizacijo tudi bistveno višji nivo zbiranja in prenosa informacij o izvršeni proizvodnji. Konfiguracija proge bo omogočala temeljitejšo on-line vzorčno kontrolo valjancev in posledično hitrejše ukrepanje, če bi prišlo do neskladnosti. Zgoraj navedene spremembe bodo bistveno vplivale tudi na bodoče izvajanje dodatnih opravil na valjancih (obrat adjustaže), obenem pa bo obstajala možnost premestitve strojev obrata adjustaže na nove lokacije. Cilj naloge je poiskati najprimernejšo razporeditev strojne opreme v obratu adjustaže in posledično transportnih poti glede na obstoječ proizvodni program in prostorske zmožnosti. Za reševanje problema razvrščanja strojev je bil uporabljen genetski algoritem, ki posnema evolucijo živih bitij. Dosežena je bila optimalna razmestitev strojev in s tem krajše transportne poti glede na dosedanje stanje. Štore Steel d.o.o. is a flexible mini steel factory that specializes in supply of steel in smaller series. In 2010 a starting of a new konti line, having a technical capacity of 250.000 tons a year is planned. A new konti fulling mill lane will, beside the enhanced productivity, enable a higher level of cylinder quality (tolerance, evenness, surface, hardness) and with Level 2 automation significantly higher level of gathering and transferring of data on executed production. Configuration of the lane will enable a more thorough on-line cylinder sample control and consequently faster responsiveness in a sense of eliminating inconsistencies. The above mentioned changes will have a significant influence on future execution of additional actions on cylinders (finishing plant), and will also open a possibility for relocation of machines in the finishing plant to a new location. The goal of this research is finding the most appropriate arrangement of the machinery equipment in the finishing plant as well as finding the most appropriate transport route, according to the current production programme and spatial ability. To solve the problem of arranging machines, a genetic algorithm imitating the evolution of living creatures has been used. We reached the optimal layout of machinery and the shorter routes on the existing situation.

Diplomsko delo obravnava določitev strategij za doseganje ciljnih stanj v hierarhičnih sistemih človeških virov. Rešitev je razvita z objektno orientiranim programskim jezikom Java in integriranim razvojnim orodjem Eclipse. Predstavljeni so osnovni koncepti teorije sistemske dinamike in optimizacije z uporabo genetskih algoritmov in njihova implementacija. Model sistemske dinamike obravnava tri stanja z možnostjo razširitve na poljubno število stanj. Za optimizacijo je uporabljena metoda genetskih algoritmov za reševanje enodimenzionalnih in dvodimenzionalnih problemov. Rešitev na testnih primerih nam kot rezultat vrne strategijo, ki nam zagotovi dinamiko sistema skladno s ciljnimi stanji. This thesis shows determination of strategies for attaining target states in hierarhical manpower systems. The solution has been developed with object-oriented programming language Java and integrated development enviroment Eclipse. The basic system dynamics concepts and optimization techniques using genetic alghoritms are introduced, as well as their implementation. The model of system dynamics is dealing with three states with the possibility of expanding it to an arbitrarynumber of states. Optimization is developed for solving one-dimensional and two-dimensional problems. The solution based on used test cases results in a strategy, which provides with the desired dynamics of the system in terms of prescribed target states.