异常数据识别与修复机制在区域供水预测方案中的应用

Application of Abnormal Data Recognition and Repair Mechanism in Regional Water Supply Prediction Scheme

为进一步提升区域供水预测模型的准确度与泛化能力,从应用场景分析、数据预处理、特征选取等角度,建立循环神经网络并引入数据异常识别和修复机制,在数据处理过程中对基于K-means聚类和Isolation Forest的异常点检测进行比较,并依据时间序列特性,提出区别近邻均值的时序均值修复方法。深度学习框架下,采用LSTM学习算法,以拟合效果、泛化损失作为评估指标,针对区域供水进行小时级预测。分析结果表明,数据异常识别和修复的预处理机制的引入,使得区域供水预测模型拟合及泛化能力进一步提升,且方法可行、有效。

To further improve the accuracy and generalization of regional water supply prediction model,a recurrent neural network is established and data anomaly recognition and repair mechanism is introduced from the perspectives of application scenario analysis,data preprocessing,and feature selection.In the process of data processing,the outlier detection based on K-means clustering and isolation forest is compared.Based on the characteristics of time series,a time series mean repair method is proposed to distinguish the nearest neighbor mean.Under the framework of deep learning,the LSTM learning algorithm is used to predict the hourly-scale regional water supply with the fitting effect and generalization loss as the evaluation index.The analysis results show that the introduction of the preprocessing mechanism of data anomaly recognition and repair makes the ability of regional water supply prediction model fitting and generalization further improved,and the method is feasible and effective.