Font Size: a A A

Improved Genetic Algorithm And Neural Network Parameter Optimization

Posted on:2004-10-23Degree:MasterType:Thesis
Country:ChinaCandidate:Y Y LiuFull Text:PDF
GTID:2208360125961244Subject:Power electronics and electric drive
Abstract/Summary:PDF Full Text Request
This thesis is based on the science project "The Research of Integration Intelligent Control in Autopilot Design for Marine ship" ,which applied from Shanghai high Education Scientific and Technology Foundation.The maneuvering of a ship along a proscribed course in restrict waters is important from operational, safety and environmental viewpoints. During the ship maneuvering, there are some characteristics of nonlinear, slow time varying and influence of complex disturbance.The PID autopilot based on linear system math model is hard to control ship maneuver system effectively .For PID control does not has a good active properties and it also deeply depend on precise math model. Fuzzy controller shows promise for these non-linear systems that are not known well or the model is difficult to be identified.Among the learning algorithms of the artificial neural networks, the single genetic algorithm(SGA) and the BP algorithm are always the focus of researching. The SGA is a kind of optimization algorithm with which global, parallel and random searching can be achieved, and its global searching performance is very good, while the BP algorithm does quite in local search.In this thesis, author puts forward a improved genetic algorithm(IGA) by adding BP algorithm into SGA. In the IGA, several important operators are improved. Using the mutation and global optimize, the IGA can find potential extremums; using floating-point coding and elitist model, it can accelerate the speed of converging; using BP operator, it can advance converging efficiency at potential extremums.To test the performance of the IGA, author designs two simulation program by C++Builder5.0. One program is simulate the performance of SGA, and the other is IGA. In the simulation of solving the continue XOR question, author makes comparisons among two algorithms-the SGA and the IGA. The result of simulation show that the new algorithm has the advantage of fast convergence and it will not converge at the local nadir.In the end of this thesis, author designs a PID autopilot by MATLAB/SIMULINK . In the simulation process, author drops example data from simulate data. In this stage, author designs a ANN autopilot. And modify the parameters of ANN controller by using the improved genetic algorithm(IGA). After comparing the response curve, the result shows the efficiency of IGA.As a learning algorithms of the artificial neural networks, the IGA is excellent in the comparing with BP and SGA. Of course there are some aspects of this algorithms needs to be improved in the later study. For example, it is hard to be used in the on-line learning. All these questions determine the further study direction in this field.Written by Yaying Liu directed by Professor Yijian Liu...
Keywords/Search Tags:artificial neural networks, genetic algorithm, autopilot, computer simulation
PDF Full Text Request
Related items