基于改进灰狼算法优化BP神经网络的住宅工程造价预测研究

在可行性研究阶段,可以明确的建筑工程信息相对较少,为保证投资决策的正确性,需要准确地预测建筑的工程造价。首先通过参考文献初步选取出13项工程造价指标,并结合主成分分析法消除各项指标之间的相关性。最后将主成分得分值作为自变量输入到I-GWO-BP模型中进行训练与仿真。预测结果与传统BP神经网络、粒子群算法和传统灰狼算法模型进行对比分析,结果表明I-GWO-BP神经网络模型更加稳定与准确。

In the feasibility study stage,there is relatively little clear information about construction projects,so it is necessary to accurately predict the construction cost,so as to ensure the correctness of investment decisions.Firstly,13 engineering cost indicators are preliminarily selected through literature for reference,and the correlation between the indicators is eliminated by combining principal component analysis.Finally,the principal component score is input into the I-GWO-BP model as an independent variable for training and simulation.The prediction results were compared with BP,PSO-BP and GWO-BP models,and the results showed that the I-GWO-BP neural network model was more stable and accurate.