NAČRTOVANJE VODENJA DINAMIČNIH SISTEMOV S POMOČJO METOD EVOLUCIJSKEGA RAČUNANJA

Corn, Marko (author), Atanasijević-Kunc, Maja (mentor)

Pri reševanju problemov vodenja različnih sistemov želimo vedno najti čim boljšo rešitev, kar pogosto enačimo s pojmom optimizacije. Optimizacijski proces zahteva definicijo ustrezne cenilke, ki mora čim bolje odražati želene oz. neželene lastnosti delovanja obravnavanega sistema, nato pa je treba poiskati rešitev, ki zagotovi njeno optimalno vrednost. Tudi evolucija je optimizacijski proces, ki poganja naš ekosistem v ciklu nenehnega razvoja od ene generacije organizmov do druge. Kako je naravi uspelo zgraditi tako kompleksne organizme, ki uporabljajo izredno zanimive in raznolike načine delovanja? Kateri so osnovni gradniki evolucije? Kakšni so mehanizmi delovanja evolucije? Kako uspešno je mogoče te mehanizme posnemati v tehniki? Vsaj delno smo na zastavljena vprašanja skušali odgovoriti v pričujočem delu. Osredotočili smo se predvsem na uporabo metod oz. algoritmov, ki temeljijo na evolucijskih mehanizmih in jim pravimo algoritmi evolucijskega računanja. Njihovo uspešnost pa smo testirali na problemih načrtovanja vodenja dinamičnih sistemov. Primerjava je razkrila njihove relativne prednosti in slabosti, ter nam tako olajšala izbiro pravega pristopa in nas usmerjala pri razvoju novega pristopa oz. metode, ki smo ga predstavili v tem delu. V ta namen smo najprej opisali osnove delovanja evolucijskih algoritmov in pojmov, kot so osebek, okolje in reprodukcija, ter mehanizme delovanja, ki vodijo k nastanku rešitev. Podali smo tudi pregled obstoječih metod evolucijskega računanja, kamor sodijo genetski algoritmi, diferenčna evolucija, evolucijske strategije, evolucijsko programiranje, genetsko programiranje in metoda agentnega modeliranja AMEBA (angl. Agent Modeled Evolutionary Based Algorithm). V nadaljevanju smo podrobneje predstavili razvito metodo AMEBA in jo primerjali z drugimi sorodnimi metodami. Za namene testiranja metode smo razvili tudi programsko orodje v okolju Matlab, ki preko grafičnega vmesnika omogoča uporabniku prijazno uporabo algoritma AMEBA ter tudi primerjavo s številnimi drugimi pristopi, ki so realizirani v istem programskem okolju. Predstavljene evolucijske algoritme in metodo AMEBA smo testirali in primerjali na treh različnih dinamičnih sistemih: sistemu treh povezanih shranjevalnikov, sistemu vodikove gorivne celice in sistemu za razporejanje energije v aktivnih distribucijskih elektroenergetskih omrežjih. V povezavi s sistemom treh povezanih shranjevalnikov smo definirali tri probleme: parametrizacijo, kjer smo iskali karakteristiko ventila, identifikacijo, kjer smo želeli določiti model celotnega sistema, in načrtovanje sistema vodenja. Vsi testirani algoritmi so se izkazali za uporabne pri reševanju problema iskanja karakteristike ventila in pri reševanju problema načrtovanja sistema vodenja. Algoritem AMEBA pa je bil uspešno uporabljen tudi pri identifikaciji, kjer je generiral primerno rešitev. Drugi obravnavani sistem je sistem vodikove gorivne celice, v povezavi s katerim smo reševali dva problema: dopolnjevanje modela in izgradnjo sistema vodenja vodikove gorivne celice. Pri problemu dopolnjevanja modela smo demonstrirali možnost uporabe predhodnega znanja o sistemu, pri čemer smo z algoritmom AMEBA zgradili model masnega pretoka anode, ki predstavlja enega izmed sestavnih delov gorivne celice. Pri problemu načrtovanja vodenja sistema vodikove gorivne celice pa smo uporabili dobljeni model vodikove gorivne celice ter nato z algoritmom AMEBA načrtali sistem vodenja, ki zagotavlja učinkovito proizvodnjo električne energije. V petem poglavju smo obravnavali tretji sistem, kjer smo se osredotočili na reševanje problema razporejanja energije v aktivnih distribucijskih elektroenergetskih omrežjih. Problem razporejanja električne energije smo razširili v izgradnjo celotnega sistema, ki bi omogočal povezavo vseh uporabnikov elektroenergetskega sistema. Po splošni predstavitvi elektroenergetskega sistema smo se osredotočili na možnosti vpeljave obnovljivih virov energije v sistem, ki zaradi svoje stohastične narave proizvodnje električne energije povzročajo probleme v omrežju. Nastale probleme poskušajo reševati tudi aktivna distribucijska elektroenergetska omrežja oz. različni sistemi, ki jih to področje ponuja. Enega izmed takih sistemov predstavljamo v pričujočem delu. Razvili smo sistem za zagotavljanje energijske bilance za bilančne skupine z vključitvijo uporabnikov elektroenergetskega sistema. Sistem zagotavlja energijsko bilanco z razporejanjem uporabnikov. Eden ključnih elementov sistema za zagotavljanje energijske bilance je algoritem razporejanja, zato smo naredili primerjavo učinkovitosti posameznih metod evolucijskega računanja na dveh problemih z različno kompleksnostjo. Metode so učinkovito rešile problem manjše kompleksnosti. Pri tem se je za najbolj učinkovito izkazala metoda diferenčne evolucije. Pri reševanju problema večje kompleksnosti pa sta se najbolje odrezali metoda genetskih algoritmov in metoda AMEBA. Na podlagi opravljene primerjave smo se odločili za uporabo algoritma AMEBA kot algoritma razporejanja v sistemu za zagotavljanje energijske bilance. Sistem za zagotavljanje energijske bilance je pokazal, da je možna vključitev različnih tipov uporabnikov, in sicer tako porabnikov kot proizvajalcev električne energije, na način, ki prinaša ekonomski učinek tako zanje kot za bilančno skupino. Za namene testiranja sistema za zagotavljanje energijske bilance smo razvili tudi simulator v okolju Matlab, ki omogoča simulacijo, testiranje, prikaz in vrednotenje rezultatov, ter uporabniku prijazen grafični vmesnik za vnašanje nastavitev simulacije. Metodo AMEBA smo torej uspešno uporabili pri reševanju različnih problemov, povezanih z načrtovanjem vodenja dinamičnih sistemov. Ob njenem razvoju in testiranju pa so se odprla še številna dodatna vprašanja, ki jih bomo skušali reševati v prihodnosti.

• Language: Slovenian
• Keywords: evolucijsko računanje, genetski algoritmi, genetsko programiranje, diferenčna evolucija, AMEBA, dinamični sistemi, aktivna distribucijska elektroenergetska omrežja, vodikova gorivna celica
• Work type: Doctoral dissertation
• Organization: FE - Faculty of Electrical Engineering
• Year: 2015
• PID: 20.500.12556/RUL-30510
• COBISS.SI-ID: 10921556
• Publication date in RUL: 19.01.2015

Secondary language

• Language: English
• Title: CONTROL DESIGN OF DYNAMIC SYSTEMS USING EVOLUTIONARY COMPUTATION METHODS

Abstract

Solving the control-design problems of dynamic systems involves investigations with the goal being to define as good a solution as possible. Such solution searching is, with regards to design problems, equivalent to an optimization process. It demands a definition of the fitness function that has to reflect the desired or undesired behaviors of the observed system. The optimization process has to generate a solution with the optimal value of the fitness function. Also, the evolution represents an optimization process that drives our ecosystem through repeated cycles of a continuous development of organisms from one generation to another. How has evolution managed to produce such complex organisms that can exhibit very interesting and diverse ways of surviving in the environment? What are the basic building blocks of this evolution? How does evolution work? Is it possible to mimic these mechanisms and use them in a computational environment? In this work we have tried to answer these questions, regarding both the process of evolution and its implementation in the computational environment. We have focused on the use of algorithms that are based on the principles of evolution and are called evolutionary computation algorithms. The efficacy of these methods has been studied for the control design of dynamic systems. A comparison has uncovered the relative advantages and disadvantages of the different approaches, which helped us to choose the most effective approach and also guided us in the development of a new method, which is presented in this work. In the dissertation we first present the basics of evolutionary computation algorithms. The important terms are: the individual, the environment, the reproduction and the mechanism that drives the evolution. We have also presented an overview of the methods of evolutionary computation and described in detail the most popular ones, like genetic algorithms, differential evolution, evolutionary strategies, genetic programming and others. A new approach to evolutionary computation is presented, called the Agent Based Evolutionary Algorithm (AMEBA), together with a comparison involving other methods. We have also developed a toolbox in Matlab that enables the user-friendly application of the algorithm AMEBA through a graphical interface and an efficient comparison with other algorithms that are available in the same program environment. The presented methods of evolutionary computation have been tested regarding three different types of dynamic systems: the system of three coupled tanks, the system of a hydrogen fuel cell and the system of energy scheduling. We have defined three problems regarding the system of three coupled tanks: the static characteristic parametrization of the system’s valve, the identification of the system’s model and a controller design. All the methods have proven to be very efficient in defining the valve characteristics and the controller design, but the AMEBA algorithm has also been successfully used in the identification problem, where it generated a suitable model. For the hydrogen fuel cell (the second system) we have defined two problems: supplementing the hydrogen fuel cell stack model, where we tried to demonstrate the possibility of how known information about the system’s operation can be taken into account during the design. Through this problem solving we have developed a model of the anode mass flow of the fuel cell stack model using the AMEBA algorithm. With the same method the control design problem was also solved by the use of the developed model. The developed controller effectively drives the fuel cell and consequently the production of the electrical energy. The third system on which we focused is described in the fifth chapter. This is the system of energy balancing in smart grids. We have developed a system for ensuring an energy balance by taking into account the customers of the electric power system. After a general presentation of the electric power system we focused on the integration of renewable energy sources into the power system. This integration of the renewable energy sources induces problems because of the stochastic nature of the energy production, which is meteorologically dependent. Smart grids or smart-grid systems are one of the answers to these problems. One such system is presented in this work. It enables the participation of all the users in the energy-balance process of the balance group. The system provides the energy balance of the balance group by scheduling the energy offered by the participating users. One of the most important elements that determine the success of the energy-balance system is a scheduling algorithm. For that we have presented a comparison of different algorithms regarding two problems with different complexities. All of the evolutionary computation methods have successfully solved the less complex problem. Regarding the more complex problem, genetic algorithms and the AMEBA algorithm proved to be the best suited. Based on this comparison we have chosen the AMEBA algorithm to be used in the energy-balancing system. The energy-balancing system proved that it is possible to include the users of the power system to participate in the energy-balancing process with an economic benefit to both the users and the balance group. We have also developed a simulation tool in Matlab that enables the simulation, testing, illustration and verification of the results and a user-friendly graphical interface for defining the simulation settings. The AMEBA algorithm was successfully tested on different types of problems that were addressed in the field of the control design of dynamic systems. During the development phase of the AMEBA algorithm several questions have emerged that will be addressed in the future.

• Keywords: evolutionary computation, genetic algorithms, genetic programming, differential evolution, AMEBA, dynamic systems, smart grids, hydrogen fuel cell