商业研究
商業研究
상업연구
Commercial Research
2007年
9期
6~10
,共null页
财务危机 预警 神经网络 遗传算法
財務危機 預警 神經網絡 遺傳算法
재무위궤 예경 신경망락 유전산법
financial crisis; alarm; neural network; genetic algorithms
针对基于神经网络的财务危机预警方法模型结构复杂、收敛速度慢且容易陷入局部最小的缺点，提出基于遗传神经网络的财务危机预警方法。首先对神经网络模型的结构和参数进行编码并将其串联，形成一个个体，随机产生N个个体形成初始种群。然后分别进行复制、交叉和变异操作，得到神经网络的结构和初始参数。再通过BP算法对该神经网络进行训练，训练后的神经网络即可实现财务危机预警。测试结果表明．该模型训练速度快、预警精度高。
針對基于神經網絡的財務危機預警方法模型結構複雜、收斂速度慢且容易陷入跼部最小的缺點，提齣基于遺傳神經網絡的財務危機預警方法。首先對神經網絡模型的結構和參數進行編碼併將其串聯，形成一箇箇體，隨機產生N箇箇體形成初始種群。然後分彆進行複製、交扠和變異操作，得到神經網絡的結構和初始參數。再通過BP算法對該神經網絡進行訓練，訓練後的神經網絡即可實現財務危機預警。測試結果錶明．該模型訓練速度快、預警精度高。
침대기우신경망락적재무위궤예경방법모형결구복잡、수렴속도만차용역함입국부최소적결점，제출기우유전신경망락적재무위궤예경방법。수선대신경망락모형적결구화삼수진행편마병장기천련，형성일개개체，수궤산생N개개체형성초시충군。연후분별진행복제、교차화변이조작，득도신경망락적결구화초시삼수。재통과BP산법대해신경망락진행훈련，훈련후적신경망락즉가실현재무위궤예경。측시결과표명．해모형훈련속도쾌、예경정도고。
Considering the disadvantages of the neural network for enterprise financial crisis alarm, such as complex structure1 slow convergence rate and being easy to fall into local minimum polnts, the paper presents the genetle neural network for crisis alarm purpose. The structure and weights of neural network are encoded and connected in series to gain individual entity. The N individuals randomly generated construct a population. The reproduction, crossover and mutation operations are performed to bring about the structure and initial weights of the neural network. The neural network is trained through BP algorithms, by which the neural network trained can realize enterprise financial crisis alarm. The testing results show that the method reaches higher training speed and lower error rate.
