This paper proposes a context-aware multi-uncertainty-set distributionally robust chanceconstrained DC optimal power flow model. Meteorological features are projected to partition
the non-convex error support into a context-dependent decomposition of conditional local ambiguity sets, with conditional weights inferred via kernel regression. The minimax problem is
reformulated into a finite-dimensional second-order cone program with proven asymptotic consistency. Out-of-sample tests on the IEEE 30-bus and RBTS systems demonstrate that the
proposed model achieves a high level of empirical feasibility close to the prescribed reliability level and avoids excessive conservativeness induced by global convex approximations.