Study On Improved Particle Swarm Optimization And Its Application In Artificial Neural Network

Particle swarm optimization (PSO) algorithm is a classical swarm intelligence algorithm,with the advantages of simple structure, less parameters, easy to describe and implement andwith good global search capability, and so on. which is widely used in many fields such asfunction optimization, multi-objective solving, and pattern recognition. But the standard PSOalgorithm has some shortcomings such as premature convergence and poor local searchcapability. Population may have gathered in a stagnation point and have not the global optimal ifit is applied to the high dimensional optimization problem, it is called premature convergence.The emergence of premature convergence does not ensure the algorithm can converge to theglobal minimum point. Besides, the convergence speed becomes slow when the particles havesearched the area near the global values; it is that poor local search capability is show at the endof iteration. Aiming at the shortcomings of the PSO algorithm, scholars have proposed manyimprovement strategy, although the performance and efficiency are improved, study onintelligent algorithm with wider adaptability, higher precision, better performance, efficiency,and good correlation are still the first-line goal in academia and industrial community.In this thesis, in order to improve the convergence speed and the performance of PSOalgorithm, an improved PSO algorithm is proposed and applied to the ANN training. Finally, themodel is applied to the water quality evaluation. The main contents are as follow:(1) An improved PSO algorithm is proposed, the algorithm is developed based onself-adaptive weight adjustment strategy and Lorenz chaos theory, simply called LSAPSO. In theLSAPSO algorithm, in order to improve the convergence speed, the adaptive weight adjustmentstrategy is employed; in order to improve the premature convergence and balance theexploitation and exploration, the Lorenz chaotic sequence is employed to tune the learning factor.Finally, the LSAPSO algorithm is tested by four multi-objective test functions. Compared withthe classical multi-objective algorithm NSGA II and multi-objective PSO algorithm, the resultsshow that the proposed LSAPSO algorithm has faster convergence speed, higher precision, andgood diversity.(2) A hybrid artificial neural network model (HANN) based on LSAPSO algorithm andRBF ANN is proposed, simply called LSAPSO RBF ANN model. In the LSAPSO RBF ANN,LSAPSO algorithm is employed to tune the function hidden centers and spreads. HANN hascombined the advantages of each algorithm, so as to improve the performance of HANN model.(3) Water quality evaluation model is established based on HANN. Through the water quality evaluation instances, the HANN model is approved that it is feasible and reliable forwater quality evaluation. Compared with the performance of classical RBF ANN and PSO RBFANN, the HANN model shows better accuracy and correlation in water quality evaluation.In this thesis, an improved PSO algorithm and a viable and effective HANN model forwater quality evaluation are developed based on PSO algorithm, Lorenz chaos theory, and ANN.The proposed HANN model may be used for reference in many research fields, it has a goodapplication prospect.