基于AHP与PSO-BP神经网络的电梯安全评价研究

Elevator Safety Evaluation Based on AHP and PSO-BP Neural Network

在评价电梯安全风险程度中,针对BP神经网络收敛速度慢、易陷入局部最优的缺点,提出了基于层次分析法(AHP)与粒子群优化BP神经网络(PSO-BP)的电梯安全评价方法。首先利用层次分析法建立了电梯系统安全评价体系,确定电梯系统安全评价体系中各子系统及各指标的权重,再结合实际经验,根据安全规范构造各指标的风险值。通过BP神经网络建立回归模型,并采用粒子群算法对模型的权重和阀值进行优化,选取电梯系统的11个权重比较大的影响因素的风险值作为PSO-BP的输入,最终得到电梯系统安全状况的综合得分,进而划分安全评价等级,得到电梯系统安全评价的结论。通过将该模型与标准BP模型进行对比,结果表明PSO-BP模型比标准BP模型的准确率要提高10%,PSO-BP有效克服了BP神经网络的缺点。

In assessing the degree of elerator security risk, for the weakness of BP neural network with slow conver- gence speed and falling into local optimum easily, a method based on Analytic Hierarchy Process（AHP） and Particle Swarm Optimization（PSO-BP） is proposed to evaluate the safety of elevator. First of all, the elevator evaluation system can be built by AHP to determine the subsystem and the weight in the elevator system security evaluation system. Then, according to safety norms, combined with practical experience the risk value of every index can be obtained. PSO optimizes the weights and thresholds of the A linear model of BP neural network. Eleven factors in the elevator system whose weight is heavy se- lected as the input of PSO-BP. Finally, the composite scores which represent the security situation of the elevator system are gotten. According to the score, the safety assessment level and the conclusion are gotten. Through comparing PSO-BP with BP, the result shows that PSO-BP can improve the precise by 10% than BP, and PSO-BP overcomes the shortcomings of BP neural network.