Modeling Steady State Power Consumption in Split Air Condition Units Based on Indoor Environmental Condition using Cumulative Sum and Machine Learning Techniques

An energy consumption monitoring system logs, records, and measures building power consumption. Real-time measurement covers room temperature, humidity, CO2 gas sensors, circuit breaker power con-sumption. The data on electricity power usage, indoor temperature, and air quality are collected and systematically analyzed and evaluated. These systems are very necessary for the real-time measurement of power usage, indoor temperature, humidity, and CO2. Remote power consumption sub-metering modules are based on Arduino Mega 2560 coupled with a Raspberry Pi 3B+ across a LAN. Each room is installed with a Mitsubishi split-type inverter ACU, with power rating ranges from 1.5 up to 6.0 horsepower, depending on the room type and size. For 11 weeks, the consumption of ACU power, inside temperature, humidity, and CO2 levels were measured every 5 minutes on months with average outdoor temperature was 29.40°C and relative humidity was 67%. The data used for curve fitting and analysis to connect and model power consumption with indoor temperature and air quality conditions are collected only during classes or office operations. Selected were five rooms of mixed use, from instructional rooms down to the 36-seater lecture hall, laboratory room, computer room, to office spaces with a maximum capacity of fifteen (15) persons, and cubicle room built for solitary usage. The ideal operating time of instructional classrooms should be at least two (2) to eight (8) hours a day, while office areas are usually occupied from eight (8) to ten (10) hours. The objective of the present work focuses on the steady-state power consumption of ACUs. Furthermore, to : (1) compare empirical models with those based on ACU power consumption and factors which influence its steady-state time; (2) demonstrate the ability of AI techniques to classify power consumption; and (3) derivation of potential power savings from changing ACU set points. It is the goal of this study to help with the generation of more general and robust models on optimizing the ACU performance, which in turn raises occupant comfort and energy efficiency. Cumulative sum technique (CUSUM) was used to calculate sum of deviations from a target value to detect significant changes in mean value of power and temperature, adjusted by a standard deviation-based allowance. The resulting values can be found on Table 1 showing that higher ACU setpoints (i.e., warmer temperatures) give lower levels of daily and steady-state power use. The MATLAB Classification Learner application is used to develop predictive models based on supervised learning algorithms. Table 3 summarize the efficacy of different machine learning strategies in predicting energy use by ACUs, offering valuable insight into the choice of the models that one can use, given the complexity and nature of the data.能耗监测系统可对建筑电力消耗进行日志记录、存储与测量。其实时监测参数涵盖室温、湿度、二氧化碳（CO₂）气体传感器数据以及断路器的电力消耗。针对电力使用、室内温度与空气质量的相关数据会被采集，并开展系统化的分析与评估。此类系统对于电力使用、室内温湿度及二氧化碳浓度的实时监测而言至关重要。远程能耗分项计量模块基于Arduino Mega 2560开发，并通过局域网（LAN）与树莓派Raspberry Pi 3B+协同工作。每间房间均安装三菱（Mitsubishi）分体式变频空调机组（Air Conditioning Unit, ACU），其额定功率范围为1.5至6.0马力，具体数值根据房间类型与面积而定。在平均室外温度为29.40℃、相对湿度为67%的月份中，研究人员每5分钟采集一次空调机组的电力消耗、室内温度、湿度及二氧化碳浓度数据，持续时长为11周。用于开展曲线拟合与分析、以构建电力消耗与室内温度及空气质量关联模型的数据，仅在授课或办公时段采集。本次研究选取了5间多用途房间，涵盖普通教学教室、36座报告厅、实验室、计算机机房、最多可容纳15人的办公区域以及用于单人办公的独立隔间。教学教室的理想每日运行时长为2至8小时，而办公区域的日常使用时长通常为8至10小时。本研究的核心目标为聚焦空调机组的稳态功耗。此外，本研究还达成三项具体目标：（1）对比经验模型与基于空调机组功耗及其稳态影响因素构建的模型；（2）验证人工智能（Artificial Intelligence, AI）技术对功耗进行分类的能力；（3）推导通过调整空调机组设定点可实现的潜在节能空间。本研究的最终目标是助力构建更具通用性与鲁棒性的空调机组性能优化模型，进而提升人员舒适度与能源利用效率。本研究采用累积和技术（Cumulative Sum Technique, CUSUM）计算目标值的偏差总和，以检测电力与温度均值的显著变化，该计算通过基于标准差的容差进行调整。计算结果见表1，结果表明：空调机组设定点越高（即温度设置越暖和），每日功耗与稳态功耗均越低。研究人员借助MATLAB分类学习器（Classification Learner）应用，基于监督学习算法开发预测模型。表3汇总了不同机器学习策略在预测空调机组能源消耗方面的效果，可为基于数据复杂性与特性选择合适模型提供有价值的参考依据。