A deep reinforcement learning-based autonomous ventilation control system for smart indoor air quality management in a subway station

A review on available energy saving strategies for heating, ventilation and air conditioning in underground metro stations

“…Other control strategies, such as adaptive control and reinforcement learning control, [82,83] have also been applied in existing studies. Wang et al [84] retrofitted HVAC control systems in UMSs based on distributed control architecture with centralized management.…”

“…Both adaptive control and schedule-based policies were developed according to the components of HVAC systems, and helped to reduce the energy consumption of stations between 20-38%. Heo et al [82] have proposed a data-driven approach for a ventilation control system based on a deep reinforcement learning algorithm. The algorithm agent was trained in a virtual environment developed based on a gray-box model, and their results showed that the control system could reduce the energy consumption by up to 14.4% for the validation dataset.…”

A deep learning approach towards the detection and recognition of opening of windows for effective management of building ventilation heat losses and reducing space heating demand

“…The use of video or vision-based methods to detect occupancy behaviour within a building space is promising [40]. Compared to other shallow learning methods, the use of deep learning techniques can lead to a better detection and recognition performance [51].…”

“…The majority of the HVAC system deep learning methods are for mechanical systems. Such application includes [40], where a data-driven ventilation control system based on a deep reinforcement learning (DeepDL) algorithm was developed and [41] where techniques were designed for mechanical ventilation systems. However, deep learning could also be a viable technique to enhance building natural ventilation strategies, but limited solutions are currently available in the literature.…”

Managdment of a Local Urban Heat Supply System Based on Neural Network Modeling Taking into Account the Weather Forecast

“…Например, в работе [8] для создания безмодельной (model-free) оптимальной балансировки HVAC-здания, которое было кондиционировано четырьмя кондиционерами, двумя электрическими чиллерами, градирней и двумя насосами, использовался современный метод обучения с подкреплением DQN, который позволил снизить общее потребление энергии на 15,7 % по сравнению с базовым режимом, сохранив концентрацию CO 2 в помещении ниже установленного ограничения. В работе [9] для повышения качества работы системы вентиляции метрополитена и снижения ее энергопотребления использовалась интеллектуальная система управления вентиляцией, основанная на алгоритме глубокого обучения с подкреплением. Построенная нейросеть позволила снизить потребление энергии до 14,4 % и повысить качество воздуха.…”