Scholarlink

<p>Sudden departures from expected urban climate conditions, especially dangerous heat, require controllers that maintain comfort while conserving energy. This study introduces a predictive coding inspired controller that forecasts the next hour and acts only when standardized prediction errors, termed surprise, exceed adaptive quantile thresholds. Surprise is computed from heat index, solar irradiance on a horizontal surface, air temperature, relative humidity, dew point, wind speed, wind direction, air pressure, and precipitation. Using harmonized hourly data for Tehran from the year 2020 to the year 2025 drawn from Meteostat station 40,754 and the National Aeronautics and Space Administration POWER service, the study compares a surprise gated policy with a reactive threshold baseline and simple statistical forecasters. Compared with a reactive baseline of 20,930&#xa0;h of actuation, the surprise gated policy requires 962&#xa0;h, a 95.4 percent reduction, and achieves precision of 0.97 and recall of 0.42, with the harmonic mean of precision and recall equal to 0.58. An explicit energy and information objective that balances actuation effort against residual variance reduction traces a clear frontier in which tighter gates capture more informative deviations at higher energy cost. These findings show that prediction error centric control can materially lower resource use without sacrificing critical interventions, and that the open and lightweight pipeline can be replicated and tuned for city specific operations.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Energy information trade-offs in predictive coding inspired neural architectures for urban climate predictive coding for urban climate

  • 期刊论文

  • 发表日期：

    2025年11月24日

查看全文
  • M. A. Afshar Kazemi,
  • H. Esmaeili,
  • R. Radfar,
  • N. Pilevari

摘要

Sudden departures from expected urban climate conditions, especially dangerous heat, require controllers that maintain comfort while conserving energy. This study introduces a predictive coding inspired controller that forecasts the next hour and acts only when standardized prediction errors, termed surprise, exceed adaptive quantile thresholds. Surprise is computed from heat index, solar irradiance on a horizontal surface, air temperature, relative humidity, dew point, wind speed, wind direction, air pressure, and precipitation. Using harmonized hourly data for Tehran from the year 2020 to the year 2025 drawn from Meteostat station 40,754 and the National Aeronautics and Space Administration POWER service, the study compares a surprise gated policy with a reactive threshold baseline and simple statistical forecasters. Compared with a reactive baseline of 20,930 h of actuation, the surprise gated policy requires 962 h, a 95.4 percent reduction, and achieves precision of 0.97 and recall of 0.42, with the harmonic mean of precision and recall equal to 0.58. An explicit energy and information objective that balances actuation effort against residual variance reduction traces a clear frontier in which tighter gates capture more informative deviations at higher energy cost. These findings show that prediction error centric control can materially lower resource use without sacrificing critical interventions, and that the open and lightweight pipeline can be replicated and tuned for city specific operations.