建设用地需求预测方法的比较研究——以山东省为例

Comparison of Construction Land Demand Forecasting Methods-A Case Study of Shandong Province

为进一步探索不同建设用地需求量预测方法的预测能力与精度,为建设用地需求预测提供理论技术支持,以山东省为例,利用城镇建设用地和社会经济统计等数据,在部分原有模型的基础上进行了一定的模型改进,运用三类灰色预测模型、Lasso回归与OLS耦合模型、PCA-BP神经网络模型以及系统动力学模型等4种模型,对模型拟合程度进行了比较,运用平均绝对误差、均方根误差以及平均预测精度作为判断标准进行了精度评价,筛选出PCA-BP神经网络模型为山东省面板数据条件下的最优预测模型,进一步预测出山东省2028年建设用地需求量的情况,并从模型优化情况、数据利用形式、数据获取难易度、预测过程难易度、模型精度评价和模型适用条件等5个方面对建设用地需求预测模型进行了综合比较,为科学选择建设用地需求预测模型选择提供了合理思路,也为经济可持续发展和土地总体规划提供了科学依据。

In order to improve the forecasting accuracy of construction land demand and compare forecasting methods to provide theoretical technical support,4 types of models are selected for comparison,including three kinds of GM(1,1)models,a Lasso regression coupled OLS model,a PCA-BP neural network model,and a system dynamics model.Predictions based on the improvement of the original models using data such as urban construction land and socio-economic statistics in Shandong Province are made.The fit of the models is compared.The average absolute error,the root mean square error,and the average prediction accuracy are used as the judgment criteria for accuracy evaluation.Finally,under the data conditions of Shandong Province,the PCA-BP neural network model is the best prediction model.Construction land demand prediction methods are compared from the aspects of optimization of the model,the forms of the data,the difficulty in data collection and etc.In addition,it could provide reasonable ideas for the scientific selection of construction land demand forecasting models,and also provide a scientific basis for sustainable economic development and overall land planning.