Identification of market power abuse in Chinese electricity market based on an improved cost-sensitive support vector machine

Accurate and real-time identification of market power abuse is a key task in the management of electricity market violations. However, there are few effective monitoring methods for extremely imbalanced datasets and progressively increasing amounts of data in actual market transactions. To address the aforementioned problems, this paper proposes an improved support vector machine by considering the index system, which can not only realize identification automatically but also minimize the credit risk of power market transactions. Firstly, the dataset is composed of an indicator system for measuring market power abuses. Secondly, a comprehensive algorithm for identifying offending data is proposed, which combines the K-Nearest Bound Neighbor and the distance between the means of two classes methods to overcome the shortcomings of traditional support vector machines with long training time due to the high dimensionality and progressively increasing amounts of data in actual market transactions, and the Cost-sensitive Support Vector Machine to tackle the problem of inefficient identification due to few tags in transaction data. Finally, five different features of constructed datasets and a power market synthetic dataset are tested, and results indicate that the proposed method can ensure high classification accuracy while significantly improving recognition speed and recall for violation data, which is more suitable for Chinese electricity market data and provide a dynamic detection method to identify market power abuse precisely and quickly.