Development of synthetic power distribution networks and datasets with industrial validation

Abstract

This thesis addresses a key challenge for creating synthetic distribution networks and open-source datasets by combining the public databases and data synthesis algorithms. Novel techniques for the creation of synthetic networks and open-source datasets that enable model validation and demonstration without the need for private data are developed. The developed algorithms are thoroughly benchmarked against existing approaches and validated on industry servers to highlight their usefulness in solving real-world problems.

A review using novel techniques that provides unique insights into the literature is conducted to identify research gaps. Based on this review, three contributions have been made in this thesis. The first contribution is the development of a data protection framework for anonymizing sensitive network data. A novel approach is proposed based on the maximum likelihood estimate for estimating the parameters that represent the actual data. A data anonymization algorithm that uses the estimated parameters to generate realistic anonymized datasets is developed. A Kolmogorov-Smirnov test criteria is used to create realistic anonymized datasets. Validation is carried out by collecting actual network data from an energy company and comparing it to anonymized datasets created using the methods developed in this thesis. The application of this method is shown by performing simulation studies on the IEEE 123-node test feeder.

The second contribution is developing a practical approach for creating synthetic networks and datasets by integrating the open-source data platforms and synthesis methods. New data synthesis algorithms are proposed to obtain the network datasets for electricity systems in a chosen geographical area. The proposed algorithms include a topology for designing power lines from road infrastructure, a method for computing the lengths of power lines, a hub-line algorithm for determining the number of consumers connected to a single transformer, a virtual layer approach based on FromNode and ToNode for establishing electrical connectivity, and a technique for ingesting raw data from the developed network to industrial data platforms. The practical feasibility of the proposed solutions is shown by creating a synthetic test network and datasets for a distribution feeder in the Colac region in Australia. The datasets are then validated by deploying them on industry servers. The results are compared with actual datasets using geo-based visualizations and by including feedback from industry experts familiar with the analysis.

The third contribution of this thesis is to address the problem of electric load profile classification in the context of buildings. This classification is essential to effectively manage energy resources across power distribution networks. Two new methods based on sparse autoencoders (SAEs), and multi-stage transfer learning (MSTL) are proposed for load profile classification. Different from conventional hand-crafted feature representations, SAEs can learn useful features from vast amounts of building data in an unsupervised automatic way. The problems of missing data and class imbalance for building datasets are addressed by proposing a minority over-sampling algorithm that effectively balances missing or unbalanced data by equalizing minority and majority samples for fair comparisons. The practical feasibility of the methodology is shown using two case studies that include both public benchmark and real-world datasets of buildings. An empirical comparison is conducted between the proposed and the state-of-the-art methods in the literature. The results indicate that the proposed method is superior to traditional methods, with a performance improvement from 1 to 10 percent.