A Data-driven Fast Calculation Method for Security-constrained Economic Dispatch With Small Sample Requirements

Security-constrained economic dispatch (SCED) in power systems needs to consider a large number of N-1 scenarios with massive security constraints. However, the SCED model has few active constraints in the actual power system applications. To reduce the computational burden, the industrial community generally set up the active constraints according to the artificial experience, which is inaccurate and incomplete. Recently, the data-driven method has been implemented to directly establish the mapping relationship between SCED input and output, which is expected to break its computing bottleneck. But the operation mode of the power system is complex and changeable, and there are few historical samples in some operation scenarios. It is necessary to solve the problem of data-driven analysis in small sample scenarios of power systems. On this basis, this paper proposed a fast data-driven security-constrained economic dispatch algorithm with small samples. Firstly, to reduce the learning difficulty of the complex SCED model with small samples, a sample pre-classification method based on marginal unit feature clustering was proposed to split the complex functional relationship between the input and output of SCED. Then, this paper used Gaussian process (GP), which is suitable for small sample learning, to establish SCED surrogate model and identify the active constraint set in advance. A fast computing framework of SCED was proposed based on Gaussian process. Finally, the simulation results performed on IEEE 30-bus and IEEE 118-bus standard systems demonstrate that the proposed method could significantly improve the computational efficiency of SCED with small sample requirements. © 2022 Chin. Soc. for Elec. Eng.