Distribution System State Estimation: A Comparison of Data-Driven Neural Networks and Model-Based Weighted Least Squares Method

Abstract:
Due to increasing electrification and the growing integration of decentralized energy generation, distribution networks are approaching their capacity limits. State estimation methods are employed to detect potential grid bottlenecks and initiate appropriate countermeasures. This paper compares the two most prominent approaches to state estimation in distribution networks – the Weighted Least Squares (WLS) method and the Neural Network (NN)-based approach – based on multiple criteria applied to a low-voltage grid. The results show that the NN provides slightly more accurate estimates and does so with significantly lower computation times than the WLS method. However, the model-based WLS approach proves advantageous in dynamically changing network environments, as it remains immediately applicable without requiring a large new dataset for retraining, as is the case with the NN approach. Beyond the comparison of estimation methods, the study also demonstrates that in highly electrified scenarios, the deployment and integration of smart meter measurements is essential to ensure sufficiently accurate state estimation.