GA-BP神经网络模型在地区工业技术创新能力评价中的应用

Application of BP Neural Network and Genetic Algorithm on Technological Innovation Capability Evaluation of Regional Enterprises

针对当前技术创新能力评价方法大多建立在线性模型的基础上，且技术创新能力影响因素较多，可能存在多重共线性的缺陷，本文提出了遗传算法优化的BP神经网络模型。GA－BP神经网络模型在以下几方面做出了改进：①利用了神经网络强大的非线性关系映射能力，避免了传统线性模型的缺陷。②利用遗传算法对评价指标进行了降维，去除了多重共线性。③使用遗传算法从全局搜寻BP神经网络权值和阀值向量，优化了BP神经网络模型，避免了BP神经网络由于使用梯度下降算法，容易陷入局部最优解的缺陷。本文最后选取2008～2013年全国31个省市规模以上工业企业技术创新能力124条数据作为训练样本，31条数据作为测试样本，分别测试遗传算法优化的BP神经网络和未优化的BP神经网络，测试结果显示遗传算法优化的BP神经网络模型预测准确率高于未优化的BP神经网络模型。

At present , most technological innovation ability evaluation methods are established on the basis of the linear model , and the factors that affect the technological innovation capability are many , the multicollinearity may exist among variables . According to the above two reasons , the GA-BP neural network model was proposed in this paper . Genetic algorithm （GA） optimized the BP neural net-work model in the following aspects： ①neural network has the strong ability of dealing with nonlinear system . It avoided the disadvantages of the linear model . ②In order to remove the multicollinearity , the genetic algorithm was used to reduce evaluation index dimension . ③BP neural network used gradient descent algorithm that modified weights and thresholds , and it was easy to fall into local optimal solution . Genetic algorithm was introduced to search the BP neural network weights and thresholds in global scope . Finally , the technical innovation data of industrial enterprises above designated size in the 31 provinces , and cities were selected from year 2008 to 2013 , 124 of them are regard as training samples , others as testing samples . Empirical conclusion shows that forecast accuracy of GA -BP neural network is higher than BP neural network .