Data-driven equipment control in open ventilated greenhouses through web integration and machine learning microclimate forecasting

Name: Thwin, Kyaw Maung Maung

keyword: Open-ventilated greenhouse

; Impact of equipment operating

; MM-LSTM

; Hybrid model

; Control the equipment

; Web automation

; Temperature and humidity

; Leaf wetness sensor

; Forecasting

Abstract: Demand for vegetables and fruits will increase yearly, according to the statistics of (Linehan, 2012) and (Mason-D'Croz, 2019) by grouping the agri-food categories and predicting 2050. Indoor plantations became more essential for growing many leafy vegetables and fruits. In contrast, outdoor or traditional plantations were less trustworthy regarding dramatic climate-changing effects, more diseases and pests, and extreme weather conditions such as storms, heavy rain, floods, heat waves, etc. Indoor greenhouses come into place as one of the solutions to these issues for growers, but not only do greenhouses have different growing systems, but also the structure of the greenhouse is different, such as fully enclosed greenhouses, semi- enclosed greenhouses, and open ventilation systems in terms of greenhouse protection cover and equipment requirements. Among developing countries and tropical regions in Southeast Asia, most greenhouse structures are constructed as open-ventilated greenhouses due to lower initial investment for construction, energy- savings due to the allowable air flow into the greenhouse, and lower maintenance costs. On the other hand, such greenhouses had various limitations regarding easy-to- influence outdoor environment fluctuation to indoors and challenging plants to favor indoor environments. Moreover, controlling or operating equipment such as FAN, different types of water sprays, and sun shading is essential. Knowing how the control equipment operates effectively in an open-ventilated greenhouse is crucial to optimizing the microclimate indoors. More research needs to be focused on this part. Data-driven equipment operating, machine learning forecasting, and cooperating with real-time sensor data in an open-ventilated greenhouse are some of the most essential and exciting parts of indoor cultivation that led to this research. Experimented greenhouses were equipped with two fans and five fans in some greenhouses, top and side water spray, shading, water irrigation, a top-view camera to monitor plant growth and indoor and outdoor environmental sensors. That greenhouse was open-ventilated and semi-automated ; a web portal can control it ; however, humans must interact most of the time for indoor environmental changes and equipment operation. This research conducted two separate experiments. The first experiment, set up in greenhouses No. 4, 5, and 6, involved different settings for FAN operating time and water spray. Greenhouse No. 5 was significantly reduced. Indoor temperatures did not exceed 34 degrees Celsius during midday compared to outdoor temperatures, around 40 degrees Celsius during equipment operating hours. A dataset was collected from those three greenhouses, and the temperature and humidity of those greenhouses were forecasted. The data included equipment operation during the data collection. In the first experiment, the two different deep learning and machine learning approaches were MM-LSTM (multivariate multistep long short-term memory) and hybrid machine learning models. The preliminary first experiment's environmental parameters are indoor and outdoor temperature, humidity, and LUX (light intensity). From October 18 to November 20, data was collected at 1-minute intervals. The result proved that the MM-LSTM model could train the model for those kinds of data in dynamic and complex microclimates impacted by equipment operating. Hybrid model accuracy performance was better in humidity prediction than LSTM in short-term prediction. However, the first experiment only intended to show the model performance in a two-model comparison on forecasting the performance of equipment operated in microclimates without consideration of equipment controls, real-time feedback, or temperature and humidity patterns. The final experiment was more broadly considered for equipment control in cooperation with MM-LSTM microclimate forecasting and handling the inconsistent sensor data to control the microclimate in an open-ventilated greenhouse to maintain an ideal temperature during the hottest summer in Thailand. In this system, extended development is based on the first experiment result and analyzing the data to do a practical experiment in one greenhouse with complete experiment control via automation of the web portal, model development for multivariate LSTM, and data handling flow for inconsistent data. A system was proposed for the final experiment. This experiment consists of more sensor data than the first experiment, such as a CO2 sensor and a leaf wetness sensor for better accuracy and performance in model forecasting and feedback sensor data to deal with humidity in the greenhouse during operation hours. The final experiment only uses indoor environmental parameters such as indoor temperature, humidity, indoor lamination, CO2, and leaf wetness sensors ; however, only indoor parameters can give good accuracy and performance. Web automation uses Selenium to control the equipment programmatically via the greenhouse portal. Indoor temperature was achieved below the average of 3 to 4°C below outdoor temperature during critical times, preventing overwater vaporization using a leaf wetness sensor at leaf-level precision. The model performed well throughout the experiment since it was evaluated again with the experimented environment dataset and achieved RMSEs of 0.515 and 0.976.

Thammasat University. Thammasat University Library

Address: BANGKOK

Email: preserv@tu.ac.th

Name: Nguyen, Duy Hung

Role: advisor

Name: Teerayut Horanont

Role: co-advisor

Created: 2024

Modified: 2568-05-15

Issued: 2025-04-21

วิทยานิพนธ์/Thesis

application/pdf

DOI: 10.14457/TU.the.2024.95

eng

DegreeName: Master of Engineering

Level: Master's Degree

Descipline: Artificial Intelligence and Internet of Things

Grantor: Thammasat University

©copyrights Thammasat University

ลำดับที่.	ชื่อแฟ้มข้อมูล	ขนาดแฟ้มข้อมูล	จำนวนเข้าถึง	วัน-เวลาเข้าถึงล่าสุด
1	Thwin.pdf	19.45 MB

ใช้เวลา

0.040811 วินาที