Fast characterization of nonlinear feasible region based on deep neural network association mining

Abstract

Dispatches of tie-line power between regional grids promote the use of natural resources. Therefore, the exact characterization of nonlinear tie-line feasible region becomes an important guarantee to ensure the power interaction. However, solving nonlinear problems using traditional methods usually requires a solver with powerful computational capabilities. We herein propose a feature association mining for nonlinear constraints and feasible region boundary to directly identify the boundary points with deep neural network (DNN) assisted prediction, which divides the identification of feasible region into two stages. Firstly, the cardinal decision variables are identified using the DNN to alleviate the numerical annihilation problem. Secondly, under the guidance of the characteristics of the description results, the association between the input constraints and the output feasible region is obtained and the block feature library of the sample data is constructed to reduce the learning difficulty. Finally, the block mapping of some key decision variables is completed. In the second stage, some cardinal decision variables are used as indicators to straightly locate the points. Moreover, a round of accuracy rectification is carried out using segment translation method and the results are corrected for ensuring the accuracy. Case studies demonstrate the effectiveness of the proposed methods.