A bottom-up framework for analysing city-scale energy data using high dimension reduction techniques. (February 2023)

• Record Type: Journal Article
• Authors: Khan, Waqas; Walker, Shalika; Zeiler, Wim
• Is Part Of: Sustainable cities and society. Volume 89(2023)
• Publication Date: 2023-02
• Subjects: Data mining -- Dimensionality reduction -- Machine learning -- Feature importance -- Data mapping; Sustainable urban development -- Periodicals; Sustainable buildings -- Periodicals; Urban ecology (Sociology) -- Periodicals; 307.76
• DOI: 10.1016/j.scs.2022.104323
• Languages: English
• ISSNs: 2210-6707

Abstract

Highlights: Presentation of a novel framework to analyse neighbourhood scale energy data. Identify suitable dimensionality reduction techniques for nonlinear energy data. Use UMAP to cluster data based on building use characteristics and energy features. Preserve the nonlinear structure within the data. Identify key responsible features for high consumption and generation regions. Abstract: Worldwide cities are becoming more sustainable and are being monitored using data collection techniques at various geographical levels. Given the growing volume of data, there is a need to identify challenges associated with the processing, visualization, and analysis of the generated data from an urban scale. This study proposes a framework to investigate the capabilities of dimensionality reduction techniques (t-SNE, and UMAP) applied to city-scale data to identify key features of high consumption and generation areas based on building characteristics. The analysis is performed on measured data from 2735 postcodes consisting of 72000 households/buildings from a city in the Netherlands. The evaluation results showed that the UMAP's algorithm mean sigma quickly approaches a threshold of 0.6 at n_neighbor values of 50 and the low dimensional shape does not change with increasing values. Whereas the t-SNE's mean sigma value increases continuously with the increasing perplexity value, implying that t-SNE is significantly more sensitive to the perplexity parameter. The UMAP algorithm was used to extract information about the high photovoltaic generation and consumption regions. The proposed framework will assist grid operators and energy planners in extracting information from energy consumption data at the neighbourhood level by utilizing high dimensional reduction techniques.