Early-warning of generator collusion in Chinese electricity market based on Information Deep Autoencoding Gaussian Mixture Model

The collusion of Generating Companies (GenCos) in the domestic power spot market not only damage the fairness of the market, but also seriously affect the healthy development of GenCos. Aiming at the problem of low real-time performance of the current identification methods of collusion behaviors of GenCos, this paper com-bines the collusion index system and the unsupervised Information Deep Autoencoding Gaussian Mixture Model (Info-DAGMM) to realize intelligent real-time early-warning of collusion of GenCos in the market. Firstly, a comprehensive collusion indicator system for GenCos is constructed and the Info-DAGMM is proposed based on the data characteristics of imbalanced positive and negative samples. Secondly, the mutual information of original input and latent variable is added to the loss function, so that the expression network can obtain a more discernible low-dimensional space, which is conducive to the work of the estimator GMM. Finally, a new evaluation index, Collusion Suspicion Index (CSI), is proposed to identify GenCos in which conspiracy occurs. Case study shows that, compared with traditional anomaly detection methods, the Info-DAGMM network pro-posed in this paper is more suitable for electricity market data and can provide early-warning of collusion for GenCos precisely and quickly, with an accuracy rate of more than 80%.