Development of Low-Cost IoT Units for Thermal Comfort Measurement and AC Energy Consumption Prediction System

Yutong Chen,Daisuke Sumiyoshi,Riki Sakai,Takahiro Yamamoto,Takahiro Ueno,Jewon Oh

2024-11-29

Abstract:In response to the substantial energy consumption in buildings, the Japanese government initiated the BI-Tech (Behavioral Insights X Technology) project in 2019, aimed at promoting voluntary energy-saving behaviors through the utilization of AI and IoT technologies. Our study aimed at small and medium-sized office buildings introduces a cost-effective IoT-based BI-Tech system, utilizing the Raspberry Pi 4B+ platform for real-time monitoring of indoor thermal conditions and air conditioner (AC) set-point temperature. Employing machine learning and image recognition, the system analyzes data to calculate the PMV index and predict energy consumption changes due to temperature adjustments. The integration of mobile and desktop applications conveys this information to users, encouraging energy-efficient behavior modifications. The machine learning model achieved with an R2 value of 97%, demonstrating the system's efficiency in promoting energy-saving habits among users.

Machine Learning

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: in small and medium - sized office buildings, how to use low - cost Internet of Things (IoT) devices to monitor indoor thermal comfort in real - time and predict changes in air - conditioning (AC) energy consumption, thereby promoting users to adopt energy - saving behaviors. Specifically, the research aims to:

1. **Develop low - cost IoT devices**: for real - time monitoring of indoor thermal comfort conditions (such as temperature, humidity, air flow velocity, etc.), and identification of air - conditioning set temperatures.
2. **Combine machine learning and image recognition technologies**: analyze data to calculate the PMV (Predicted Mean Vote) index and predict energy consumption changes due to temperature adjustments.
3. **Provide feedback to users through mobile and desktop applications**: encourage users to adjust their behaviors according to the feedback information to achieve more efficient energy use.

### Main problem background

- **Building energy consumption problems**: Buildings, especially office buildings, consume a large amount of energy and account for an important part of CO₂ emissions. Small and medium - sized office buildings often lack effective energy - saving strategies due to financial and technological limitations.
- **BI - Tech project**: The Ministry of the Environment in Japan launched the BI - Tech (Behavioral Insights X Technology) project in 2019, aiming to promote voluntary energy - saving behaviors through AI and IoT technologies.

### Solution overview

This research proposes a low - cost IoT system based on the Raspberry Pi 4B+ platform, combined with machine learning models (such as SVM - RF), achieving the following functions:

1. **Real - time monitoring and calculation**:
- Use sensor modules to collect indoor thermal environment data (such as air temperature \( t_a \), globe temperature \( t_g \), relative humidity \( R_h \), air flow velocity \( V_{air} \), etc.), and calculate the PMV index through formulas.
- Use image recognition technology to obtain the set temperature on the air - conditioning panel.

2. **Energy consumption prediction**:
- Predict the energy consumption change for each 1°C adjustment of the air - conditioning set temperature through machine learning models.
- After being trained, the model can accurately predict energy consumption changes, with an R² value reaching 97%.

3. **User feedback and interaction**:
- Develop iOS and Windows desktop applications to provide users with real - time thermal comfort and energy consumption prediction information.
- Through the notification system, suggest that users make small - scale temperature adjustments to achieve energy - saving goals while maintaining thermal comfort.

### Key technologies and methods

- **PMV index calculation**: Use Sympy and Math libraries to handle complex non - linear equations to ensure the accuracy of PMV values.
- **Machine learning model**: Adopt the SVM - RF model for energy consumption prediction, and optimize model parameters through cross - validation.
- **Data transmission and storage**: Use the MQTT protocol for data transmission, the MySQL database for data storage, and REST API to achieve front - end and back - end interactions.

### Conclusion

This research provides an effective and economical solution for small and medium - sized office buildings. Through real - time monitoring and prediction, it helps users better understand and adjust air - conditioning settings, thereby achieving energy - saving goals. Future research will focus on improving the design of IoT devices and enhancing the monitoring of clothing thermal insulation performance, etc.