Enhanced Power System State Estimation Using Grid Aware Learning Method

The demand for electrical power is significantly increasing due to a growing reliance on technology and automation. Control center operators utilize state estimation procedures to obtain reliable data on the current operating conditions of the power system. Since the state estimation calculation is an iterative process, the choice of the initial guess for the state variable is crucial; the closer the initial guess is to the solution, the fewer iterations are needed for convergence in real-time operations. The traditional approach in state estimation procedures employs a “flat start,” with all voltage angles set to zero and all voltage magnitudes set to 1.0 p.u. This paper introduces a hybrid model that leverages grid-aware learning using a Graph Convolution Neural Network to derive an initial estimate of the power system state from available measurements. This initial estimate is then input into the weighted least squares estimator to reduce the number of iterations required for convergence and enhance the accuracy of state estimation. The proposed model is tested on the IEEE 14-bus system and compared with other initialization strategies, demonstrating fewer iterations required for state estimation calculations and yielding more accurate results.