基于LC-GAN的电力碳排放数据异常检测方法

Detection Method for Abnormal Carbon Emission Data of Electric Power Based on LC-GAN

针对目前电力碳排放数据存在的坏数据虚假注入问题,提出一种基于生成对抗网络的电力碳排放数据异常检测方法。首先构建面向时序数据的改进生成对抗网络,引入双层长短期记忆网络,深入挖掘时序数据的内在联系;再通过生成对抗网络改进模型的生成器和判别器的共同打分,判别电力碳排放数据异常值;最后在华东某省碳排放监测平台进行电力碳排放数据测试。结果表明,基于生成对抗网络改进模型的电力碳排放数据异常检测方法对抗训练稳定、损失函数收敛速度快,检出率为87.5%,针对电力碳排放时序异常数据检测的准确度较高。

Considering the problem of false injection of bad data in current electricity carbon emission data,a detection method for electricity carbon emission data anomalies based on generative adversarial networks was proposed.Firstly,the improved generation adversarial network for time series data was constructed,and the two-layer long short-term memory network was introduced to deeply mine the internal relationship of time series data;then,the generator and discriminator of the improved model of the generative adversarial network were scored together to identify the anomalies of the electric power carbon emission data;finally,carbon emission data testing for electricity was conducted on a carbon emission monitoring platform in a certain province in East China.The results show that the method of abnormal detection of power carbon emissions data based on the improved generative adversarial network model has stable confrontation training,fast convergence speed of loss function,and a detection rate of 87.5%.The accuracy of abnormal data detection for electric power carbon emissions time series is high.