Font Size: a A A

Research On Prediction Method Of Soil Moisture And Temperature In Tillage Layer Based On Deep Learning

Posted on:2022-11-10Degree:MasterType:Thesis
Country:ChinaCandidate:H D LiuFull Text:PDF
GTID:2493306779462934Subject:Automation Technology
Abstract/Summary:PDF Full Text Request
Soil tillage layer moisture and temperature are the core parameters of agricultural planting management,and in-depth information on soil tillage layer moisture and temperature can help improve agricultural productivity.Accurate prediction of soil tillage moisture and temperature can effectively regulate and manage planting environment parameters and optimize crop planting quality.The depth of soil tillage layer is shallow,and soil moisture and temperature have a large correlation with ambient air temperature and humidity.The prediction of soil moisture and temperature can be carried out based on deep learning methods using ambient air temperature and humidity and soil moisture and temperature information.According to the design requirements and functional characteristics of different agricultural management systems,the deep learningbased soil tillage layer moisture and temperature prediction function can be deployed in the cloud or locally.However,the characteristics of cloud deployment and local deployment are different,so deep learning models need to be studied for both deployment modes.For the practical needs of soil tillage layer moisture and temperature prediction in terms of training set acquisition and model validation,an Io T data collection system based on embedded system and NB-Io T wireless communication technology is designed to realize the long time collection of environmental data in equal interval time series.The experiments show that the designed data acquisition system has high stability and reliability,which can provide accurate data for the deep learning-based soil moisture and temperature time series prediction work.Cloud computing platforms are usually rich in resources,and achieving stable and highly accurate predictions is a core requirement for models deployed on cloud-based platforms.At present,there are more deep learning models,and different models have different characteristics.In order to improve the upper limit of prediction accuracy of models,a model combination prediction strategy based on deep reinforcement learning DQN algorithm is studied,through which a combination model with strong prediction ability and suitable for cloud deployment is obtained.The DQN-L-G-B combination model is proposed with LSTM,GRU and Bi-LSTM as the base models.The simulation experiments show that the DQN-L-G-B combination model outperforms the LSTM,GRU,Bi-LSTM and L-G-B weighted average combined model in terms of RMSE,MAE,MAPE and R2.In the comparison of the prediction effects of the two soil types,the prediction effect of loam soil is better than that of sandy soil,and the DQN-L-G-B combination model obtained by training is deployed on the cloud platform for actual prediction analysis.In order to ensure that the model has good prediction performance while consuming less system resources for deployment in local devices,a single time series prediction model is investigated.The W-MALSTNet soil moisture and temperature prediction model is proposed to address the shortcomings of traditional machine learning models in predicting soil moisture and temperature.Based on the LSTNet model,a wavelet noise reduction module is added to reduce the negative impact of noise in the dataset on the model training to a certain extent.Secondly,a hybrid attention module is introduced to incorporate a dimensional and time-step hybrid attention mechanism.The simulation experiments show that compared with the LSTNet model,the W-MA-LSTNet model is superior in the prediction indexes such as RMSE,MAE,MAPE,and R2,and the prediction effect in loam soil is better than that in sandy soil.The research in this paper can effectively accomplish the accurate prediction of the moisture and temperature of the soil in the tillage layer at the future moment,which can help realize the fine crop planting management and provide support for modern agricultural planting.
Keywords/Search Tags:soil, moisture and temperature, time-series prediction, deep learning, deep reinforcement learning
PDF Full Text Request
Related items