基于LSTM的连栋温室能耗预测模型

Prediction Model of Energy Consumption of Multi-span Greenhouse Based on LSTM Neural Network

为了解决北方地区玻璃温室作生产中能耗过大、能源浪费严重的问题,提出了基于LSTM神经网络的温室能耗预测模型。首先,使用Python的Keras深度学习框架构建能耗预测模型,然后根据预设的损失函数,优化器等进行训练和验证,最后根据玻璃温室的气象数据和能耗数据对模型进行验证,并与BP神经网络和RNN神经网络能耗预测模型进行了对比分析。结果表明:BP、RNN和LSTM网络模型预测值和真实值的均方差分别为0.054、0.040、0.037,平均绝对误差分别为0.142、0.121、0.114,LSTM模型的MSE和MAE误差均优于BP和RNN网络模型。综上,基于LSTM神经网络构建的温室能耗预测模型对能耗值的预测更加准确,为温室能耗精准管控提供了理论依据。

In view of the problem of excessive energy consumption and serious energy waste in the production of glass greenhouse in the northern region,this paper proposed a greenhouse energy consumption prediction model based on LSTM neural network.First,the energy consumption prediction model was constructed using Python's Keras deep learning framework.Second,the model was trained and verified according to the preset loss function,optimizer,and so on.Finally,the model was verified according to the meteorological data and energy consumption data of the glass greenhouse,and then the comparision was carried out between this model and the other two models i.e.the energy consumption prediction model of BP neural network and RNN neural network.The results showed that the mean square deviation of the predicted value and the true value of BP,RNN and LSTM network models was 0.054,0.040 and 0.037 respectively,and the average absolute error was 0.142,0.121 and 0.114 respectively.The MSE and MAE errors of LSTM model were better than those of BP and RNN network models.In conclusion,the greenhouse energy consumption prediction model based on LSTM neural network is more accurate in predicting the energy consumption value,which provides a theoretical basis for precise control of greenhouse energy consumption.