日光温室颗粒物动态运移机制与预测模型研究

Dynamic Transport Mechanisms and Predictive Modeling of Particulate Matter in Solar Greenhouse

摘要:

精确预测温室内颗粒物浓度变化可为实时环境控制系统提供科学依据。基于中国东北地区典型温室生产场景,本文系统分析了温室内外不同粒径颗粒物（PM2.5和PM10）时序特征及其关联机制,构建了适用于不同通风模式的颗粒物动态运移模型。通过筛选关键环境驱动因子,建立了适用于温室环境的颗粒物浓度机器学习预测模型。研究结果表明：同一粒径范围内室内与室外颗粒物浓度变化相关性最强,室外颗粒物浓度是主导室内颗粒物浓度变化的主要驱动源;相比PM2.5,温室内PM10对室外颗粒物变化的响应较弱。所构建的动态运移模型在不同通风模式下颗粒物浓度计算值与实测值相对误差为0.68%~9.73%,能够较准确地量化温室内外颗粒物运移机制。在预测建模时,筛选室外颗粒物浓度、室内温度、相对湿度与风速等作为输入变量,建立的遗传算法-小波神经网络（GA-WNN）模型表现出优异的预测性能,PM2.5和PM10浓度决定系数分别为0.912~0.938和0.898~0.917,并且超过92.7%和90.3%的实测PM2.5和PM10浓度落在95%预测区间内,验证了模型的准确性与可靠性。提出的基于室外颗粒物浓度与室内环境参数相结合的分析框架,为理解与预测温室内颗粒物浓度提供了有效方法,为温室环境精准调控与作物生产的有效管理提供了重要支撑。

Abstract:

Accurate prediction of temporal variations in particulate matter concentrations inside greenhouses is essential for data-driven,real-time environmental control. Focusing on representative production scenarios in Northeast China,the temporal characteristics and coupling mechanisms of indoor and outdoor particulate matter across different size fractions (PM2.5 and PM10) were systematically analyzed,and a dynamic particulate transport model applicable to multiple ventilation modes was developed. Key environmental drivers were then identified to build machine-learning models for predicting particulate concentrations in greenhouse environments. The results showed that indoor and outdoor concentrations exhibited the strongest correlations within the same particle-size fraction,and that outdoor concentration was the dominant driver of indoor variability. Compared with PM2.5,indoor PM10 responded more weakly to outdoor fluctuations. Across ventilation modes,the dynamic transport model achieved mean deviations of 0.68%~9.73% between simulated and observed values,enabling quantitative characterization of indoor-outdoor particulate exchange. For data-driven prediction,outdoor particulate concentration,indoor temperature,relative humidity,and air velocity were selected as input variables;the proposed GA-WNN model demonstrated excellent performance,with coefficients of determination of 0.912~0.938 and 0.898~0.917 for PM2.5 and PM10,respectively. Moreover,over 92.7% and 90.3% of measured PM2.5 and PM10 concentrations fell within the 95% prediction intervals,confirming the accuracy and reliability of the model. Overall,the proposed framework integrating outdoor pollution levels with indoor microclimate parameters provided an effective approach for interpreting and predicting particulate matter dynamics in greenhouses,supporting precision environmental regulation and effective crop production management.