基于LSTM神经网络的温室气候环境因子预测

Prediction of Greenhouse Climate Environmental Factors Based on LSTM Neural Network

目前温室气候环境中,环境监测数据只能反映当前环境状况,无法对温室气候环境变化趋势进行预测。为解决温室气候环境控制效果差的问题,采用基于LSTM神经网络的温室气候环境因子预测方法,将温室中采集的湿度、温度、二氧化碳浓度进行标准化处理后作为历史数据,将90%数据作为训练集,10%数据作为测试集;设置初始参数并建立LSTM预测模型,通过寻找不同的模型参数,不断调整模型的训练精度;利用测试集对LSTM预测模型进行测试验证;为了更好地说明LSTM预测模型的优越性,同时建立了BP神经网络模型以及GRU预测模型。结果表明,LSTM预测模型能够较好地预测温室中湿度、温度、二氧化碳含量的变化趋势,在预测精度上,LSTM预测模型比BP神经网络以及GRU预测模型,平均提升了5.80%、3.81%。本研究建立的LSTM预测模型可以实现温室气候环境因子的精确预测,为温室环境调控提供一定的决策支持。

The current greenhouse climate environment,environmental monitoring data can only reflect the current environmental conditions and cannot predict the trend of greenhouse environmental changes.To solve the problem of poor greenhouse climate environment control,a method for predicting greenhouse climate environment factors based on the LSTM neural network was adopted.The humidity,temperature and carbon dioxide concentration collected in the greenhouse were standardized as historical data,and 90%of the data was used as the training set and 10%was used as the test set.The LSTM prediction model was established by setting initial parameters,and the training accuracy of the model was adjusted constantly by finding different model parameters.Finally,the LSTM prediction model was tested and validated by the test set.Both a BP neural network model and a GRU prediction model were established in order to better illustrate the superiority of the LSTM prediction model.The results showed that the LSTM prediction model could effectively predict the trend of changes in humidity,temperature,and carbon dioxide concentration in greenhouse and had an average improvement of 5.80%and 3.81%in the prediction accuracy compared to the BP neural network model and GRU prediction model.The LSTM prediction model established in the paper can achieve accurate prediction of greenhouse climate environmental factors and provide certain decision-making support for greenhouse environmental regulation.