Method to Discriminate Collusive Bidding among Power Producers based on Deep Self-Coding Gaussian Mixture Model
Abstract
With the increasing transaction scale of the power market and the corresponding increase in transaction data, it is urgent to analyze power companies' collusive bidding behavior based on big data. Therefore, this paper combines an indexing system and unsupervised deep-autoencoder Gaussian mixture model to detect collusive bidding. First, this paper introduces an indexing system that includes structure, behavior, and influence indexes. Second, the Gaussian mixture model of the deep autoencoder is proposed according to the characteristics of high-dimensional index data. Then, using the compression network of the deep autoencoder, the hidden representation and reconstruction errors of the unit index data are obtained, and the loss function is constructed based on this data. The estimation network is then used to estimate the density, and the abnormal energy function of the unit is constructed to determine whether collusive bidding behavior occurs. Finally, a numerical example shows that compared with other traditional unsupervised study models, the deep-autoencoder Gaussian mixture model is more efficient and accurate in discriminating the bidding behavior of power producers.