基于变分自编码高斯混合模型的发电企业串谋智能预警

Intelligent Early-warning of Collusion Between Power Generation Enterprises Based on Variational Autoencoding Gaussian Mixture Model

随着市场交易规模越来越大,交易数据量增加,结合数据进行串谋分析成为可能。为此,结合发电企业的串谋预警指标体系和无监督的变分自编码高斯混合模型(VAEGMM),实现了对发电企业串谋的智能预警。首先,提出了完善的串谋预警指标体系和详细的指标计算方法。其次,针对指标集具有高维且正负样本不均衡的数据特点,结合异常检测思想提出了VAEGMM。然后,详细阐述了VAEGMM的网络结构,并且重新构建了联合损失函数,使得该网络能够更好地学习得到原始数据的低维表达,从而有助于进行更准确的密度估计。最后,实例测算表明,与其他传统的无监督学习模型相比较,VAEGMM可以更加高效和准确地预警串谋风险。

As the scale of market transactions becomes larger and the amount of transaction data increases, it becomes possible to conduct collusion analysis with data. Therefore, combining with the collusion early-warning indicator system of the power generation enterprises and the unsupervised variational autoencoding Gaussian mixture model(VAEGMM), the intelligent earlywarning of the collusion between power generation enterprises is realized. Firstly, a complete indicator system for the collusion early-warning and a detailed indicator calculation method are proposed. Secondly, in view of the high-dimensional data characteristics of the index set and the imbalance of positive and negative samples, the VAEGMM is proposed based on the idea of anomaly detection. Then, the network structure of VAEGMM is described in detail, and the joint loss function is reconstructed,making the network better learn the low dimensional expression of the original data. Thus it is helpful for more accurate density estimation. Finally, the actual case study shows that compared with other traditional unsupervised learning models, VAEGMM can warn the risk of collusion more efficiently and accurately.