基于GASA混合策略的BP网络在基准地价测算中的应用

Application of BP Networks Based on GASA Hybrid Strategy to Measuring and Calculating Base Land Prices

结合遗传算法的并行搜索结构和模拟退火的概率突跳特性 ,提出了一种用于BP网络权值学习的GASA混合策略。以江西省赣县县城商业用地为例 ,应用基于GASA混合策略的BP网络对其基准地价进行了测算 ,并与回归模型方法作了比较。结果表明 ,混合策略能有效地避免BP算法陷入局部极小和网络单目标学习易产生的过拟合现象。将神经网络用于基准地价的测算 ,精度优于常规的回归模型方法。

Combining the parallel searching structure of genetic algorithms with the probabilistic jumping property of simulated annealing, a GASA hybrid strategy is proposed for the weights learning of BP networks to speed up training process, improve starting solutions robustness and generalization ability and overcome local minimum. In this new algorithm, the GA population is initialized with an improved BP algorithm.the comprehensive error index of the training error and the testing error is adopted to be the optimization objective function of BP networks,and the strategy of alternately learning between the training sample sets and the testing sample sets in GA is introduced. The application of BP networks based on the GASA hybrid strategy to measuring and calculating base land price is studied, and the results are compared with the conventional method(regression model). The results show that the GASA hybrid strategy can efficiently avoid BP algorithm converging to local optima and the network simple objective learning easily leading to the phenomenon of over fitting.