科技通报
科技通報
과기통보
BULLETIN OF SCIENCE AND TECHNOLOGY
2014年
4期
50-52
,共3页
混沌%神经网络%蛙跳群%控制
混沌%神經網絡%蛙跳群%控製
혼돈%신경망락%와도군%공제
chaos%neural network%frogs%control
前馈神经网在智能控制设计训练中收敛速度慢、易陷入局部极值,且对初始权值依赖性。为此提出一种基于超维无限折叠反转迭代混沌的蛙跳群优化算法,训练前馈神经网络参数,进行全局优化智能控制设计。算法在充分利用BP算法的误差反传信息和梯度信息的基础上,引入基于超维无限折叠反转迭代混沌的概念,结合混合蛙跳群仿生算法,将混沌蛙跳群作为全局搜索器,以梯度下降信息作为局部搜索器来调整智能控制网络的权值和阈值,实现全局寻优和空间智能搜索。实验和算法对比结果表明,新的人工神经网络智能控制算法在均方差、泛化均方差等指标上较传统算法具有明显优势。改进算法特别适合人工智能控制模型设计与预测建模,具有较高的预测精度和泛化能力,稳健性好,具有较好的全局优化自适应控制能力。
前饋神經網在智能控製設計訓練中收斂速度慢、易陷入跼部極值,且對初始權值依賴性。為此提齣一種基于超維無限摺疊反轉迭代混沌的蛙跳群優化算法,訓練前饋神經網絡參數,進行全跼優化智能控製設計。算法在充分利用BP算法的誤差反傳信息和梯度信息的基礎上,引入基于超維無限摺疊反轉迭代混沌的概唸,結閤混閤蛙跳群倣生算法,將混沌蛙跳群作為全跼搜索器,以梯度下降信息作為跼部搜索器來調整智能控製網絡的權值和閾值,實現全跼尋優和空間智能搜索。實驗和算法對比結果錶明,新的人工神經網絡智能控製算法在均方差、汎化均方差等指標上較傳統算法具有明顯優勢。改進算法特彆適閤人工智能控製模型設計與預測建模,具有較高的預測精度和汎化能力,穩健性好,具有較好的全跼優化自適應控製能力。
전궤신경망재지능공제설계훈련중수렴속도만、역함입국부겁치,차대초시권치의뢰성。위차제출일충기우초유무한절첩반전질대혼돈적와도군우화산법,훈련전궤신경망락삼수,진행전국우화지능공제설계。산법재충분이용BP산법적오차반전신식화제도신식적기출상,인입기우초유무한절첩반전질대혼돈적개념,결합혼합와도군방생산법,장혼돈와도군작위전국수색기,이제도하강신식작위국부수색기래조정지능공제망락적권치화역치,실현전국심우화공간지능수색。실험화산법대비결과표명,신적인공신경망락지능공제산법재균방차、범화균방차등지표상교전통산법구유명현우세。개진산법특별괄합인공지능공제모형설계여예측건모,구유교고적예측정도화범화능력,은건성호,구유교호적전국우화자괄응공제능력。
According to the feed-forward neural network in the design of the intelligent control, the convergence speed is slow, an improved frog leaping swarm optimization algorithm was proposed for training feed-forward neural networks based on the hyper space dimensional and infinite folding inversion iterative chaos mapping. The global optimal control was de-signed. The error back propagation information and gradient information of BP algorithm were made full use, and the con-cept of hyper space dimensional and infinite folding inversion iterative chaos mapping was presented. The global optimiza-tion and intelligent search were realized. The simulation and comparison results show that the algorithm has various advan-tages in the simulation parameters such as mean square error, MSE of generalization etc. It is very suitable for artificial in-telligence control modeling, the training accuracy and generalization accuracy are perfect, and it has better ability of global optimization adaptive control.