Számítási intelligencia algoritmusok, rendszerek és modellek = Algorithms, systems and models in computational intelligence

Korabbi eredmenyeinkre epitve javasoltuk egy evolucios (pl. bakterialis, reszecskeraj) memetikus algoritmuscsaladot, az LM, maxi gradiens, es kombinacios eljarasokat alkalmaztunk lokalis keresesre. Az uj modszerek jobb konvergenciasebesseggel es –pontossaggal rendelkeznek, kulonosen a fuzzy modellek konstrukciojaban. Javaslatot tettunk multipopulacios, tobbszalas es hibrid evolucios, iterativ moho es utemezett vegyes evolucios es memetikus eljarasokra. Szabvanyos adathalmazokon e modszerekkel az eddig publikalt eredmenyeknel jobbat ertunk el. Vizsgaltuk a fuzzy neuralis halozatokat, uj strukturakat, műveleteket bevezetve megkezdtuk a hardver implementaciot; fuzzy kognitiv terkepeket vizsgaltunk. Javaslatot tettunk a fuzzy szignaturak geometriailag strukturalt altalanositasara, valtozo finomsagu szituacios terkepek leirasara. Javasoltuk a fuzzy 2 dimenzios raszterek alkalmazasat a kepreprezentacioban. Az uj komplex fuzzy - evolucios/moho/gradiens alapu optimalizacios - neuralis halozat eszkozkeszletet a műszaki es alkalmazott problemak szeles koreben hasznaltuk fel, igy a tavkozlesi, a szallitasi es logisztikai halozatok optimalizaciojara, hibadetektalasra; intelligens mobil robotok iranyitasara, kommunikaciojara es autonom egyuttműkodesere; ellatasi lancok es gyartasi folyamatok optimalizalasara; erőforrasallokaciora es –utemezesre; karakterfelismeresre es az epitő- es kornyezetmernoki dontestamogatasra. | Based on our earlier research results we proposed a family of enhanced bacterial and evolutionary other memetic algorithms (e.g. Partical Swarm Optimization), with Levenberg-Marquard and Steepest Descent, viz. combinatorial methods for local search. The new methods have better convergence speed and accuracy, especially in fuzzy rule based model construction. We proposed multipopulation, multithread and hybrid evolutionary, iterative greedy and alternatingly scheduled mixed evolutionary and memetic approaches. We have achieved better results for standard benchmark data sets than any other authors. We studied neural networks based on fuzzy operations, proposing new structures, new operation families and starting hardware implementation; and we simulated fuzzy cognitive maps. We proposed extended classes of fuzzy signatures with geometric structure, modeling situational maps with flexible depth and fineness. We proposed fuzzy 2D grids for image representation. The new complex fuzzy - evolutionary/greedy/gradient optimization - neural network tool kit thus developed was deployed for a wide variety of engineering and applied problems, telecommunication, transport and logistic network optimization and failure detection, intelligent and mobile robot control, communication and co-ordination of autonomous collaboration; optimization of supply chains and production, resource allocation and scheduling, character recognition, and decision support in civil and environmental engineering.