CAJ | 학술논문

现实生活中绝大数系统都是非线性的,BP神经网络通过训练能否达到局部最优值、能否收敛以及训练的时间长短与初始值和阈值的选取关系密切.为此采用了具有动态惯性权重的粒子群算法对BP神经网络初始值进行优化.实验表明具有动态惯性权重的粒子群算法优化BP神经网络预测误差很小,能够跳出局部极小值,得到更优的结果.
현실생활중절대수계통도시비선성적,BP신경망락통과훈련능부체도국부최우치、능부수렴이급훈련적시간장단여초시치화역치적선취관계밀절.위차채용료구유동태관성권중적입자군산법대BP신경망락초시치진행우화.실험표명구유동태관성권중적입자군산법우화BP신경망락예측오차흔소,능구도출국부겁소치,득도경우적결과.
Most systemms are nonlinear in real life.Whether BP neural network could meet the local optimal value through training,whether it is comvergent and how long the training lasts are closely related to the initial value and the threshold value selected.This paper use a particle swarm algorithm with dynamic inertia weight network to optimize the initial value of the BP neural network.The experiment shows that the prediction error is very small when the particle swarm algorithm is used to optimize the BP neural network. Such an algorithm can jump out of local minimum value and get better results.