基于机器学习的建筑能耗SVM模型降阶分析与研究

Reducing SVM Model of Building Energy Consumption Based on Machine Learning

在建筑能耗预测模型训练中,选定的特征在某些环境下很难保证预测结果的实效性和准确性.如何科学合理地选择适合建筑本身属性的特征子集用于模型学习,在机器学习研究领域中一直备受研究者的青睐.基于解决使用不同的特征集会改变模型的精度性能和学习速度等问题,本文提出一种“探索式”方法用于特征子集选择,并针对它是如何影响模型的性能进行一系列的实验和系统分析,探索一种足够简单且实用,同时又可以在实践中容易获取和准确记录的特征集.基于选取出的3个数据集,利用径向基函数核和多项式函数核对模型进行训练,通过特征选择前后模型性能的数据比较分析发现所采用的方法对模型的预测精度具有一定的提升作用.

In the model training of building energy consumption prediction,selected features are difficult to ensure the effectiveness and accuracy of the predicted results in some environments.How to scientifically and reasonably select the feature subset which is suitable for the building's own attributes for model learning has always attracted the attention of researchers in the field of machine learning.To solve the problem that using different feature sets can change the accuracy performance and learning speed of the model,this research proposed an exploratory method for feature subset selection,and a series of experiments and system analyses were carried out on how this method would affect the performance of the model.The research also tried to find a feature set which is simple and practical enough,and can be easily acquired and accurately recorded.Based on three selected data sets,the model was trained by using RBF kernel and polynomial function check.Through comparative analyses of the model performance data before and after feature selection,it was found that the method proposed and used in this research can improve the prediction accuracy of the model to a certain extent.