Smart Greenhouse Retrofit Notes: A Retrospective After Three Months with 200 Sensors Online
Last year we took on a retrofit project for a planting base: 12 greenhouses and nearly 200 sensor nodes, with the goal of "checking data on a phone, and having fans and cooling pads run automatically." Three months later the system is running stably, but the journey was far bumpier than the slides suggested. These notes record a few episodes worth revisiting.
Selection: Cheap Sensors Are the Most Expensive
The first batch of soil temperature-and-humidity probes cost less than fifty yuan apiece. Two weeks after going online, their readings started drifting across the board — the greenhouse stays above 90% humidity for long stretches, and cheap probes fail quickly in that environment. We eventually switched to models with anti-corrosion coatings and IP67 housings at four times the unit price, but they have run fault-free for six months.
Lesson: in agricultural settings, component cost should not be judged by purchase price, but by "purchase price ÷ actual months of survival."
Power and Communication: Two Underestimated Pitfalls
- Power: There are no outlets in a greenhouse. Battery-powered nodes were rated for a year of runtime, but under high heat and humidity they ran dry in four months. The final solution was solar panels plus lithium iron phosphate batteries, which hold up for two weeks even through overcast, rainy weather.
- Communication: WiFi attenuates badly inside greenhouses with metal frames. After switching to a LoRa mesh network, the coverage problem was solved in one move — a single gateway comfortably handles 20 nodes.
Linkage Logic: Conservative First, Aggressive Later
At first we configured "turn on the fan immediately when temperature > 32°C." But when the afternoon sun slanted in and locally heated a sensor, the fans started and stopped more than forty times a day and nearly burned out a motor. The improved strategy:
- Changed single-point triggering to three-point majority voting (action only when two of three sensors exceed the threshold);
- Added a minimum run/stop interval (15 minutes) to protect the motor;
- Every linkage action must push a notification to a person, with a one-tap manual override kept in place.
Three months later, the fans stabilized at 6–8 start/stop cycles per day, and temperature swings inside the greenhouse narrowed by 40%.
The Payoff Ledger
| Item | Before Retrofit | After Retrofit |
|---|---|---|
| Manual greenhouse rounds | 3 times/day × 2 people | 1 spot check per day |
| Heat damage | About 2 overheating incidents per season | 0 incidents |
| Water and fertilizer usage | Based on experience | Down about 25% |
Final Thoughts
The value of agricultural IoT is not in "how flashy the big dashboard looks," but in turning a veteran grower's experience into reproducible rules. The tech stack is actually the simplest part — what is truly hard is understanding the crops, understanding the seasons, and understanding the on-site details that textbooks never write down.
Related Documentation
- Scene Linkage: The multi-point voting and start/stop intervals described in the "Linkage Logic" section above can be configured in the platform's visual scene linkage.
- Thing Model: Before ingesting sensor data, first define properties, functions, and events for measurement points such as temperature and humidity.
- Device Management: Registration, grouping, and online-status monitoring of the 200 sensor nodes.