Abstract:

This paper proposes a novel algorithm for detecting trade activities suspected to be illegitimate across day-ahead and real-time energy markets. Illegitimate trades bring severe risks to the financial health of energy markets and the operation stability of power grids. It is of critical importance for the operators to identify such activities in a sound and timely manner. The proposed algorithm firstly generates a set of legitimate trade feature samples using historical trade profiles of energy markets based on the comparison of day-ahead and real-time trade profiles and related environmental impacts. Then a set of illegitimate trade feature samples are created using a Genetic-algorithms based negative selection procedure based on those legitimate trade feature samples. The created illegitimate samples are further labeled with specific illegitimate trade types by comparing with pre-defined typical trade feature samples for each illegitimate trade type. After that, deep learning is employed to learn the relationship between the trade features and associated legitimacy labels from the sets of legitimate and illegitimate trade feature samples, and predict the legitimacy status for incoming trade profiles according to corresponding trade features. The effectiveness of the proposed algorithm has been demonstrated using sample trade profiles obtained from New England ISO