基于遗传神经网络的工程造价估算方法研究

人工神经网络已经成功的运用到工程造价估算方法研究中,高度的鲁棒性和容错能力使它优于多元线性判别分析（MDA）、逻辑回归等方法（Logistic Regression）,针对传统的BP神经网络在工程造价估算方法中存在收敛速度慢和容易陷入局部最小值等问题,提出遗传神经网络的估算方法。将遗传算法和神经网络结合,充分利用两者的优点,使新算法既有遗传算法的全局随机搜索能力,又有神经网络的学习能力和鲁棒性。利用遗传算法的全局搜索能力,针对传统误差反向传播算法的不足,采用染色体编码对神经网络的权值和阈值等主要参数进行优化,通过仿真试验验证其稳定性和有效性,表明该算法在工程造价估算方法中具备较高的实用性。

Research on Model of Engineering Cost Estimation Based on Genetic Neural Network

Artificial neural network has been successfully applied to project cost estimation methods study. Its high degree of robustness and fault tolerance makes it superior to multiple linear discriminant analysis （MDA）,logistic regression and other methods. This paper,in view of slow convergence and local minimum value of the traditional BP neural network in project cost estimation methods,proposes genetic neural network estimation method. By combing genetic algorithms and neural network and fully utilizing the advantages of both,the new algorithm not only has the random search capability of genetic algorithm,but also has the learning ability and robustness of neural network. With the global search capabilities of genetic algorithm,considering the drawbacks of traditional error back-propagation algorithm,using chromosome coding,weights and thresholds and other key parameters of neural network are optimized. The stability and effectiveness of that algorithm is verified through simulation tests,which shows this method has a high practicality in cost estimation.