Predictive modeling for power system state estimation

Load flow analysis is critical for efficient and reliable operation of power systems and is an essential task for grid operators to keep the power system safe. Classical load flow approach does not take into consideration changes in the power systems that come with varying ambient conditions. If load flow incorporates dynamic electrical characteristics of the conductors based on weather conditions, the results would more accurately reflect the real state of the system. On the other hand, mathematical solvers for this type of calculation would take significant computation time. To improve the computational performance and to predict the future state of the system, this paper proposes artificial neural networks (ANN).

In this paper, a modified load flow algorithm is proposed, incorporating thermal modeling of the conductors, as well as consumption and production dependencies with ambient temperature, wind velocity, conductor surface absorptivity, global solar radiation, and wind direction factor. Results of these calculations are then used to train ANN to predict the future state of the system and make planning easier for system operators.

The proposed algorithm is tested on 28-bus 35 kV distribution grid in Banat region, using hourly profiles for 1 year. ANN is designed to predict the state of the power network and technical losses of the network for the subsequent time periods without the need for detailed thermal calculations, while still taking into consideration ambient conditions.