A deep reinforced learning spatiotemporal energy demand estimation system using deep learning and electricity demand monitoring data. (15th October 2022)

• Record Type: Journal Article
• Authors: Maki, Seiya; Fujii, Minoru; Fujita, Tsuyoshi; Shiraishi, Yasushi; Ashina, Shuichi; Gomi, Kei; Sun, Lu; Budi Nugroho, Sudarmanto; Nakano, Ryoko; Osawa, Takahiro; Immanuel, Gito; Boer, Rizaldi
• Is Part Of: Applied energy. Volume 324(2022)
• Publication Date: 2022-10-15
• Subjects: Energy monitoring -- Deep Learning -- Data Fusion -- Deep Reinforcement Learning -- Bayesian Kriging method -- Indonesia; Power (Mechanics) -- Periodicals; Energy conservation -- Periodicals; Energy conversion -- Periodicals; 621.042
• DOI: 10.1016/j.apenergy.2022.119652
• Languages: English
• ISSNs: 0306-2619

Abstract

Highlights: Multiple energy demand monitoring systems were installed in residences in Bogor. Electricity consumption time-series and equipment ratio in the city were estimated. Reinforcement learning system for electricity demand estimation systems was created. Hourly energy demand prediction system with scope for automatically improvement was developed. Abstract: Tracking decarbonization effects requires a model for the identification of spatial energy demands on city facilities. However, most developing countries lack detailed discrete time and device-specific energy demand data. In this study, we installed multiple energy demand monitoring systems that could observe electricity demands at the device level in some residences in Bogor, Indonesia. The study aimed to estimate the time-series and equipment ratio of electricity consumption by households in the entire city based on the monitoring data. However, the number of households monitored was small, and therefore unlikely to be regarded as having a representative system. Therefore, we used questionnaire data to create monitored mimic data and increased the number of samples to estimate the energy demand characteristics of the entire city via spatial interpolation. In addition, we developed a reinforcement learning system for discrete time–electricity demand estimation systems for unmonitored households using a 5-step procedure; 1) Analyzing energy demand and its patterns from monitoring data, 2) Questionnaire-based surveying of households, 3) Estimation of energy demand and its patterns based on questionnaire responses in monitored households, 4) Development of a deep learning model that extends the results from (3) to unmonitored households using data fusion, and 5) Spatial interpolation of energy demand characteristics for all households in Bogor using a spatial statistics method. The spatial electricity demand of households was interpolated from GIS and high-resolution satellite data matching procedures. Based on this analysis, we developed an hourly energy demand prediction system that could be automatically improved by adding new data from the reinforced learning framework.