基于机器学习的电力数据回归分析和预测技术研究

Regression Analysis and Prediction Technology Research of Power Data Based on Machine Learning

通过机器学习对海量的电力大数据进行分析和研究,可以探索和挖掘电力数据之间的相关性、数据属性间的相关模型,进而分析现状和预测未来,提升电网安全性、经济性和稳定性.通过支持向量机回归、高斯过程回归、回归树3种回归分析算法对采集到的电力数据进行比较和分析,重点从均方根误差、均方误差、平均绝对误差、拟合系数、运行时间等5个方面进行回归效果的讨论.同时,对3种算法进行超参数优化,获取最优回归模型.最后,通过综合比较5个指标,证明了基于网格树优化器的回归树模型对电力验证数据的拟合度最好.

By analyzing and researching the massive power big data through machine learning algorithm,we can explore and mine the correlation between power data and the correlation model between data attributes,then analyze the current situation and predict the future,so as to improve the security,economy and stability of the power grid.The collected power data are compared and analyzed by three regression analysis algorithms:Support vector machine regression,Gaussian process regression and regression tree.The regression effect is mainly discussed from five aspects:root mean square error,mean square error,mean absolute error,fitting coefficient and running time.At the same time,the hyperparameter optimization of the three algorithms was carried out to obtain the optimal regression model.Finally,it is proved that the regression tree model based on grid tree optimizer has the best fit degree to the power verification data by comprehensively comparison the five indexes.