Budapest University of Technology and Economics Department of

Budapest University of Technology and Economics Department of Measurement and Information Systems Peter Eredics The Intelligent Greenhouse

Greenhouses are built in various sizes and for many different purposes.

Actuators Common actuators in greenhouses: • • Windows Roof vents Shading curtains Heating Irrigation Misting etc.

Traditional Control Traditional solution: independent, set-point based controllers. Advantages: – Simple and cheap solutions, available off the shelf – Low complexity hardware with high reliability – Unambiguous operation

Traditional Control Traditional solution: independent, set-point based controllers. Drawbacks: – The set-points have to be specified by the operator – The control is strongly reactive – Missing synchronization of the actuators

Computer Control Solutions Computers are only used for: – data recording, – visualization, – remote access.

Optimal Control • The goal is to maintain the physical parameters inside the greenhouse in the optimum range required by the plants with the lowest possible cost. Requirements: • Maximize the comfort of the plants • Minimize costs

The Intelligent Greenhouse • Overtaking the drawbacks of traditional control solutions, with AI methods, to realize optimal control. • Set-points • Reactive control • Missing synchronization Goals Predictive control Planning

Goals of the Control • Traditional control left all responsibility to the operator. • The operator knows the plants, not thermal behavior of the greenhouse. • The intelligent control should accept and work directly based on the goals of the operator.

Goals of the Control

Predictive Control • Traditional control can only react when undesirable situations are reported by its sensors. • Traditional control does not utilize regular external influences (i. e. sunset)

Planning and Actuator Synchronization • The intelligent control creates and evaluates plans shared for all actuators. • A reasonable forward planning time is 4 hours. • The space of potential plans is very large: introduction of constraints or incremental planning is required.

Planning - Example

Modeling the Greenhouse • The model has to predict the future thermal state of the greenhouse • The system has large complexity • The greenhouse is in always in change, thus adaptive modeling is essential

The Experimental Greenhouse

Thermal Zones Shading screen 4 Roof vents 3 Side window 2 1 Heating pipe 0 / / Covered desk Temperature / Light / Humidity measurement point

Topology • Using 7 microcontrollers for measuring quantities on 44 physical 23 locations every 5 minutes

The Experimental Control System The control system records measurements, runs a traditional control application and provides remote control interface for data extraction and intelligent control testing.

Data visualization Temperature data Light levels Actuator states Cloud coverage

Data Cleaning

Data Cleaning – Missing Desk Temperatures

The Proposed Model Decomposition

1. Local External Temperature Forecast

2. Missing Regional Weather Data Restoration L = lokálisan mért hőmérséklet – ismert R = lokálisan mért megvilágítás - ismert, az utolsó néhány mérés átlaga G = regionális hőmérséklet adat – ezt szeretnénk pótolni az előbbiek ismeretében

3. Heating Pipe Model Heating System Model Cooling Pipe Model Heating Control Signal Sequence Predicted Temperature Current Temperatures Warming Pipe Model

4. Global Greenhouse Model (1)

4. Global Greenhouse Model (2)

4. Global Greenhouse Model (3)

4. Global Greenhouse Model (4)

Results • The measurement system is collecting data. • The data cleaning system is operating. • Submodels have been implemented. • The global greenhouse model is under development…

Thank you for your attention!