Construction of An Early Warning Model for Market Power Identification in the Electricity Spot Market Based on Improved BP Neural Networks

When using traditional methods to analyse market power, it requires frequent calculation of multiple indicator values and comprehensive assessment of market power through manpower. With the opening of the electricity spot market, the frequency of market transactions has increased, trading strategies have become more complex, market power is exercised by more subjects, channels and opportunities, and the huge amount of market operation data has made the identification and control of market power by market operators complex and lagging. In recent years, machine learning has been excelling in big data analysis and prediction due to its self-evolving capability, and its application prospects are very promising. This paper takes machine learning methods as the basic idea, introduces various traditional machine learning algorithms, and selects back propagation(BP) neural network algorithm for improvement. Based on the improved algorithm, it constructs an early warning model for market power identification in the electricity spot market, evolves the optimal mapping relationship between market power behaviour and market power risk level based on massive historical data, and finally achieves the purpose of automatically predicting market power risk level by inputting market power behaviour.