Incremental Auto Regressive Prediction Models with External Variables of Greenhouse Air Temperature for Control Purposes

Zhenfeng Xu, Junjie Chen, Jingxia Zhang, Yuan Liu, Qiaowei ZhangInternational Journal of Smart Home 2016 · DOI

TL;DR

This paper introduces a novel set of "Incremental Auto Regressive Models with External Variables" (IARX) specifically designed to predict greenhouse air temperature. By dividing greenhouse operation into distinct sub-states based on actuator activity and deriving models from a mechanistic energy balance approach, these IARX models achieve higher accuracy with fewer coefficients compared to traditional ARX models, paving the way for more efficient greenhouse control.

What Problem Does This Paper Address?

Modern greenhouses, especially those imported from Western Europe, often struggle with energy efficiency in diverse climates like China's, which experiences significant temperature fluctuations. This leads to manual or inefficient automated control of actuators (like ventilation fans and heaters), resulting in wasted energy and laborious management. Accurate prediction of greenhouse air temperature is crucial for developing advanced control strategies that can optimize actuator usage, reduce switching frequency, and minimize energy consumption. Existing ARX models, while computationally efficient, often overlook the complex impact of actuators and can be cumbersome to identify.

Methods

The core contribution of this paper is the development of a new class of IARX models. The authors first establish a mechanistic model of greenhouse air temperature based on energy balance principles. This model accounts for various heat gains and losses, including solar radiation, ventilation, and heat exchange with the environment. Crucially, they then divide the greenhouse's operating state into five sub-states based on the on-off status of its actuators (passive, natural ventilation, mechanical ventilation, fan-pad cooling, and air-fan heating). For each sub-state, they derive a simplified IARX model. These new models are designed to have fewer coefficients and a more deterministic structure than previous ARX models, making them easier to identify and more robust to changes in environmental conditions. The models are then validated using real-world data from a glass greenhouse.

Datasets

The study utilized environmental data collected from a glass greenhouse located near Nanjing, China. The data included measurements of:

  • Internal air temperature
  • Internal relative humidity
  • External air temperature
  • External relative humidity
  • Solar radiation

This data was collected over specific time intervals corresponding to the passive, mechanical ventilation, and fan-pad cooling sub-states. The paper does not introduce a new benchmark dataset but rather uses collected data for model validation.

Key Insights and Ideas

  • Actuator States Matter: The paper strongly emphasizes that actuator status is not a minor detail but a fundamental factor that dictates the dynamics of greenhouse temperature. Treating these distinct operational states separately leads to more accurate models.
  • Mechanistic Foundation for Simplicity: Deriving IARX models from a physics-based (mechanistic) energy balance model, rather than purely empirical methods, allows for a more principled reduction in model complexity and coefficient count.
  • Fewer Coefficients, Better Performance: The proposed IARX models achieve comparable or superior prediction accuracy to existing ARX models despite having significantly fewer coefficients. This is a critical advantage for real-time control applications where computational resources can be limited.
  • Adaptability to Real-World Conditions: The incremental nature of the models, combined with their simplified structure, makes them amenable to recursive identification in real-time. This allows them to adapt to slow variations in greenhouse parameters and environmental conditions.

Key Results

The IARX models demonstrated superior prediction performance compared to one of the two traditional ARX models tested.

  • Passive Sub-State: Maximum prediction error of 0.1°C and mean square error of 0°C for the IARX model, compared to 0.4°C and 0.2°C for "Compared 1" and 0.1°C and 0°C for "Compared 2".
  • Mechanical Ventilation Sub-State: Maximum prediction error of 0.14°C and mean square error of 0.3°C for the IARX model.
  • Fan-Pad Cooling Sub-State: Maximum prediction error of 0.7°C and mean square error of 0.4°C for the IARX model, compared to 1.8°C and 1.0°C for "Compared 1" and 0.5°C and 0.3°C for "Compared 2".

The paper highlights that the IARX models with fewer coefficients performed as well as the second compared model, whose prediction curves almost overlapped.

Limitations and Future Work

  • Limited Sub-States Tested: While the models were developed for five sub-states, the experimental validation was primarily conducted on three (passive, mechanical ventilation, and fan-pad cooling) due to the specific actuators available in the test greenhouse. The authors plan to validate the models for the other two sub-states in future work.
  • Greenhouse Specificity: The coefficients of the IARX models are dependent on greenhouse parameters (volume, heat transfer coefficients, etc.) and actuator characteristics. While these parameters don't affect the model structure, they need to be identified for each specific greenhouse.
  • Future Research: The authors intend to further research switching control strategies for different operating sub-states based on these new IARX prediction models. They also aim to validate the models for the remaining sub-states and explore their application in real-time control processes.

Summary updated

Abstract

The impact of actuators should be considered in the prediction modeling of greenhouse air temperature. In this paper, the operating state of a greenhouse was divided into five sub-states based on the on-off characteristic of actuators. A group of novel incremental auto regressive models with external variables (IARX models) suitable for the five operat-ing sub-states were deduced from the mechanistic modeling of greenhouse air tempera-ture. The new IARX models have fewer coefficients than other known ARX models. In or-der to validate the IARX prediction models, the related environmental factors of a glass greenhouse were measured. The prediction results of the IARX models were compared with two typical ARX models. The maximum prediction errors and the mean square errors of the IARX models, under the three operating sub-states of passive state(all actuators are not working), mechanical ventilation and fan-pad cooling, are 0.1°C, 0.14°C, 0.7°C, and 0°C, 0.3°C, 0.4°C, respectively. The prediction results are much better than those of one compared model, while similar with the other.
BibTeX
@article{xu2016incremental,
  author = {Zhenfeng Xu and Junjie Chen and Jingxia Zhang and Yuan Liu and Qiaowei Zhang},
  title = {{Incremental Auto Regressive Prediction Models with External Variables of Greenhouse Air Temperature for Control Purposes}},
  journal = {International Journal of Smart Home},
  year = {2016},
  doi = {10.14257/ijsh.2016.10.9.05},
}
Figure