基于LSTM-SVM的日光温室环境预测模型

Environmental Prediction Model of Solar Greenhouse Based on LSTM-SVM

摘要:

准确预测日光温室内环境参数可有效降低温室控制的滞后性，是实现日光温室智能化控制的关键，对于实现日光温室生产过程的精细化智能管理具有重要意义。本文通过设置多层LSTM网络模型并结合SVM设计用于预测温室环境参数的混合模型，在模型预测过程中加入未来一段时期的天气信息辅以修正模型，以提高模型预测结果的准确性，实现在复杂天气条件下温室内环境参数预测。实验结果表明，该模型充分利用了LSTM在序列数据长期依赖关系处理上的特长以及SVM在特征提取和分类上的优势，实现了在复杂变化的室外环境条件对日光温室内关键环境参数的精确预测，预测决定系数R2均不小于0.93。实验验证对比结果表明，提出的加入未来一段时间天气信息修正的LSTM-SVM混合预测模型在复杂天气条件下对日光温室环境参数预测准确性和稳定性明显优于单一模型和其他传统方法。

Abstract:

Accurate prediction of environmental parameters in solar greenhouse can effectively reduce the lag of greenhouse control, which is the key to realize the intelligent control of solar greenhouse. It is of great significance to realize the fine intelligent management of solar greenhouse production process. A multi-layer LSTM network model was set up and combined with SVM to design a hybrid model for predicting greenhouse environmental parameters. In the process of model prediction, weather information in the future period was added to supplement the modified model to improve the accuracy of model prediction results and realize the prediction of environmental parameters in greenhouse under complex weather conditions. The experimental results showed that the model made full use of the advantages of LSTM in long-term dependency processing of sequence data and the advantages of SVM in feature extraction and classification. The accurate prediction of the key environmental parameters in the solar greenhouse under the complex outdoor environmental conditions was realized, and the prediction determination coefficient R2 was not less than 0.93. The experimental verification and comparison showed that the proposed LSTM-SVM hybrid prediction model with future weather information correction was superior to the single model and other traditional methods in the prediction accuracy and stability of solar greenhouse environmental parameters under complex weather conditions.