Data Collaboration Based On Artificial Neural Network And Its Application In Smart Grassland

With the rapid development and extensive application of Internet of Things(IoT),a large amount of multi-source heterogeneous data has been collected by different types of sensors located in the physical-layer of IoT.However,there are major problems on how to solve the difficulties of multi-source heterogeneous data recognition,application and collaboration during the development of IoT.Therefore,a series of new methods based on different types of artificial neural networks(ANNs)are proposed to solve the difficulties of multi-source heterogeneous data recognition,application and collaboration in this thesis.Taking the dynamic grass-and-animal balance system as an example,the thesis predicts the spatio-temporal data of the normalized difference vegetation index(NDVI)in future years by collaborating and analyzing the NDVI data generated by satellite sensors and the precipitation data generated by ground-based meteorological sensors.This thesis uses spatio-temporal NDVI to predict the grass yield and theoretical stocking capacity of the study area.The spatio-temporal prediction of the grass yield and theoretical stocking capacity will realize the dynamic spatio-temporal distribution of livestock species and quantity in future years,consequently realizing the smart grassland which means to realize the function of the dynamic grass and animal balance system based on the IoT.The new methods proposed in this thesis are very important to further enrich the theory of multi-source heterogeneous data collaboration.The main contributions of this thesis are as follows.1.The nonlinear autoregressive with exogenous input(NARX)network is firstly applied to build the prediction model of the temporal precipitation during the grass-growing season in this thesis.The results show that the goodness-of-fit between the observed precipitation and predicted precipitation measured by the determination coefficient is greater than 0.93 from the NARX models for the different weather stations.In addition,not only the trends in the precipitation data but also the characteristics of dynamic relationship among the precipitation in the different years can be accurately captured by the NARX network.The results demonstrate that the proposed method can result in an accurate prediction of the precipitation during the grass-growing season.2.To accurately product the collaboration between the precipitation data and the NDVI data,the NARX network is proposed to represent the collaboration.The results show that the goodness-of-fit between the observed NDVI and predicted NDVI measured by the determination coefficient is greater than 0.94 from the NARX model.In addition,the NARX network has been used to capture time-lag effects between the NDVI and the precipitation variation.The time-lags of NDVI response to both the precipitation and NDVI itself can be accurately captured by the NARX network.3.To accurately product the collaboration between the spatio-temporal NDVI and the precipitation,a hybrid neural network(NARX-BPNN-NARX)is further developed.In this hybrid neural network,the BPNN method is proposed to map the precipitation to the spatio-temporal variables,such as the latitude,longitude,elevation and time.NARX is proposed to map the NDVI to the precipitation.The results demonstrate that the spatio-temporal NDVI can be accurately predicted using the developed hybrid neural network(the determination coefficient being greater than 0.95).Therefore,the dynamic grass-livestock balance system based on ANN techniques can be successfully realized in this thesis.