[Interview] Where Did Greenhouse Heating Costs Leak Away? One-Minute Measurement… GEMS Developer Researcher Yoon Sung-wook

Even when the same strawberries are grown over the same period, greenhouse energy consumption varies. The energy consumption per unit area at a strawberry farm in Damyang was 12.4 kWh/㎡, while that of a strawberry farm in Yangpyeong was 14.2 kWh/㎡, meaning the Damyang farm used about 13% less than the Yangpyeong farm. Until now, however, all farmers had in hand were monthly electricity bills and receipts for heating oil purchases. Without measurement data showing which equipment used how much and when, there was no way to pinpoint the cause of the difference.

The greenhouse energy monitoring system “GEMS” (Greenhouse Energy Monitoring System), developed by the Department of Agricultural Engineering at the National Institute of Agricultural Sciences under the Rural Development Administration, targets this gap. It measures electricity and heating oil used in greenhouses at one-minute intervals, shows consumption by facility, and even predicts future energy use with artificial intelligence (AI). Development was completed over three years starting in 2022, and field demonstration is now under way in Nonsan, while data analysis is being conducted in parallel at strawberry farms in Damyang and Yangpyeong. We met Researcher Yoon Sung-wook, who led the development.

What kind of system is GEMS?

Farmers can know their total costs each month through electricity charges and heating oil purchase costs, but it is not easy to determine which equipment used how much and when. GEMS is a system designed to measure electricity and heating oil use in real time along with greenhouse environmental information, allowing users to check energy use status at a glance and control it directly. It can show consumption by major facilities such as heaters, ventilation fans, and nutrient solution machines, and also displays internal environmental conditions such as temperature and humidity together with external weather information. It was developed through the “Research on Real-Time Energy Monitoring for Horticultural Facilities” conducted from 2022 to 2024, with a total research budget of 210 million won over three years. Since 2025, field demonstration research has been under way by installing the system on farms to verify its stability and field applicability.

How are electricity and heating oil measured, respectively?

For power, current is measured with a current transformer (CT) and combined with voltage information to calculate electricity consumption. Because measurement devices are installed for each major facility or circuit, loads that require management, such as heating-related equipment, ventilation fans, and nutrient solution machines, can be checked separately. Heating oil is measured by attaching a water-level sensor to the storage tank to detect changes in level, then converting them into actual usage based on information about the tank’s shape and capacity.

A flow meter can directly measure the amount of fuel, but it requires cutting existing fuel pipes or adding connections, which can be burdensome for farms already operating heating facilities. Heating oil is consumed intermittently as heaters repeatedly start and stop depending on outside temperature and the set temperature. Even when using a flow meter, it is necessary to choose a model that operates stably in these low-flow ranges. Estimating consumption by boiler operating time has limitations because actual consumption varies depending on heater load and operating conditions. The level-sensor method also can show a sharp rise when oil is refilled, and values may fluctuate if the liquid surface shakes, so the system distinguishes rises caused by refilling from decreases caused by consumption and analyzes data patterns over a certain period to minimize such effects. The level-sensor method was chosen not because it is the most accurate method, but in consideration of its field applicability: it can continuously track consumption while minimizing changes to piping.

How is it different from existing integrated environmental controllers?

Existing integrated environmental controllers have strengths in measuring environmental information such as temperature, humidity, and carbon dioxide (CO₂) and controlling equipment. GEMS takes a different perspective, collecting data centered on “how energy is actually being consumed in the greenhouse.” Its core purpose is to help users understand consumption characteristics by analyzing electricity and fuel consumption, the operating status of major facilities, internal environmental conditions, and external weather information together. Data are collected and stored at one-minute intervals and can be converted into hourly, daily, or monthly units for analysis depending on the purpose, and electricity consumption is measured with the same equipment as commercial watt-hour meters. Application programming interface (API) integration with the platform of the Korea Agency of Education, Promotion and Information Service in Food, Agriculture, Forestry and Fisheries or with manufacturers’ controllers has not yet reached the standardization stage. Related research is being carried out while reviewing an open linkage structure aligned with standards. 

What was the actual cost-saving effect at demonstration farms?

The field demonstration is being conducted at a strawberry farm in Nonsan, South Chungcheong Province, and data from farms in Damyang and Yangpyeong where GEMS had been installed in a previous project are also being used. Looking at only one farm, the results can be determined by the facility conditions or management habits of that particular greenhouse, so the purpose is to verify whether the system can be used even when region, weather, and operating conditions differ.

At the Damyang strawberry farm, the greenhouse was operated while checking winter energy use status through GEMS information, and according to the farm’s own tally, heating costs in the winter of 2025–2026 were about 10% lower than in the winter of 2024–2025. However, caution is needed before interpreting this directly as a 10% reduction in heating costs due to the installation of GEMS. That is because heating energy consumption is affected not only by weather conditions such as outside temperature and solar radiation, but also by cultivation period, set temperature, insulation status, and equipment operation methods. GEMS is not a device that directly reduces energy use; it is a system that helps farmers adjust their operating methods based on data.

Did measurement reveal any equipment that was using unexpectedly large amounts of energy?

With the current data alone, it is difficult to definitively identify a particular piece of equipment as a “hidden culprit” or to generalize that consumption differs by several times from farm to farm. However, when the Damyang and Yangpyeong strawberry farms were compared over the same period, there was a case in which consumption per unit area was 12.4 kWh/㎡ and 14.2 kWh/㎡, respectively, a difference of about 13%. This was the result of regional weather, greenhouse structure, the performance of thermal insulation curtains, heating settings, and operating methods acting in combination.

By measuring consumption by facility, we confirmed that not only equipment expected to use large amounts of energy, such as heaters, but also auxiliary equipment whose consumption is not usually noticed, such as pumps for using groundwater, consumes a certain amount of electricity. Even if individual consumption appears small, the cumulative amount can grow if equipment operates repeatedly for long periods, so management is necessary.

What level is the AI prediction accuracy, and is it at a level where automatic control is possible?

In the case of predicting 15 minutes ahead at the Nonsan strawberry greenhouse, the system shows prediction accuracy of more than 90% (based on the coefficient of determination), and more than 80% for one hour ahead. Because the model was built using only about one year of cultivation data, accuracy is expected to improve further as data accumulate. The prediction model uses a Transformer-based architecture. This was chosen not for immediate accuracy, but to make use of multivariate time-series data accumulated over long periods. Recurrent neural networks (RNNs) or long short-term memory (LSTM) models can have limitations in learning long-range temporal dependencies as the input window becomes longer. The degree of improvement in accuracy should be presented based on comparisons under the same validation conditions, and it is necessary to distinguish the accuracy of classifying fuel consumption status from the accuracy of predicting fuel usage so the two are not confused.

At the current stage, it is not appropriate to describe it as a system that directly and automatically controls heaters based on prediction results. If prediction reliability and safety are verified and technology for linkage with integrated environmental controllers is secured, there is potential for it to develop into automatic control. However, because proper growing conditions for crops must be maintained, what is needed is not control technology that minimizes energy consumption alone, but control technology that considers crop environment and energy efficiency simultaneously, and the process will include final judgment by the farmer.

What are the installation costs and distribution plan?

The installation cost of the system used in the field demonstration is about 19 million won per farm. However, because this amount includes research sensors and data collection and communication devices, it is difficult to view it as the distribution price for ordinary farms. Return on investment (ROI) also varies greatly depending not only on greenhouse area but also on heating energy use, crop type, facility type, and whether existing measurement equipment is already in place, so it is difficult to present a uniform figure. For distribution, the task is to simplify the system around essential measurement items and lower installation costs. The current transformer and level-sensor methods can be applied to existing greenhouses, but a certain level of communication environment is needed for data transmission, so storage and transmission methods for areas with poor communication conditions must also be considered at the distribution stage.

Based on this year’s demonstration results, we plan to propose it for the “New Technology Pilot Distribution Project,” and if it is selected, the goal is pilot distribution in 2028. The number of distribution sites and target regions will be finalized after selection. A patent application for a “greenhouse energy monitoring device using a prediction model and its operation method” has been filed and is undergoing registration procedures, and the related technology has been transferred to one company.

Is there room for connection with carbon neutrality policy? Also, what are your future goals as a researcher?

It is difficult to say that GEMS is currently a system that certifies reductions for the emissions trading scheme or external projects. That is because institutional requirements such as methodology, baseline setting, and measurement, reporting and verification (MRV) must be satisfied separately. However, the starting point for carbon reduction is ultimately accurately measuring how much energy was actually used. The electricity and fuel consumption data accumulated by GEMS could be used to quantitatively identify energy use and greenhouse gas emissions in the agricultural sector.

Going forward, I want to accumulate field data from multiple regions and crops to compare consumption characteristics by farm, and develop the system so farmers can easily judge “how much energy my farm uses compared with other farms under similar conditions, and where there is room to reduce it.” Energy is invisible, but once it is measured, it begins to become visible, and once it is visible, it can be managed. My goal as a researcher is for GEMS to become not just a measurement device, but a tool that reduces energy costs and supports rational farm management.