State estimation function is essential for effective and timely execution of power system automation and control systems, especially in modern active distribution systems where more intermittent renewable energy systems are integrated into the grid. Distribution system state estimation faces a lot of challenges including lack of monitoring devices and possible incorrect topology information. Developing efficient state estimation for distribution systems is thus of great interest. This paper presents results on utilizing artificial neural networks for this purpose.

Artificial neural networks have been used in power distribution system state estimation. However, there is a lack of systematic analysis and study of which types of ANNs and what structures including parameters are most suitable for state estimation applications. When designing an ANN for a state estimator, trial and error approach has been common and there is no systematic method available to guide the process. The ultimate goal of the research is to examine the performance of various types of ANNs (e.g., Multi-Layer Perceptron (MLPs), Convolutional Neural Networks (CNNs) and Long-Short- Term-Memory Networks (LSTMs)) with different structures and also provide possible guidance on how to choose the different parameters, including model parameters such as number of hidden layers and number neurons in a layer, and algorithm parameters such as adjustable learning rate, for desired performance metrics. The paper presents preliminary results based on MLPs. IEEE standard 34-bus test system is used to illustrate the proposed methods and their effectiveness.

The paper seeks to contribute to a more systematic approach to neural network and deep learning applied to power system state estimation, thus enhancing situational awareness, system resiliency and real-time monitoring and control of power distribution systems. Successful state estimation function will increase the ability of distribution systems to integrate more renewable energy based generations.

Document Type


Publication Date


Notes/Citation Information

This paper was presented at the PAC World 2021 Conference.