| As one of the world’s largest agricultural countries,China has a large population,sparse arable land,and a shortage of resources.How to improve agricultural production has become one of the important and pressing tasks.The environmental factors can affect the photosynthetic intensity of crops in real time,and become the most direct influencing factor during the process of crops growth.Therefore,in this thesis,we make short-time prediction of environmental factors during the maize growth period,and inversely,use the prediction results to predict photosynthetic intensity in order to provide real-time technical support for agricultural production.The research work done in this dissertation is as follows.In terms of data collection and pre-processing,there are two sets of original data sets used in this thesis:Data set A,the time series data of environmental factors of maize growth obtained by this research group using the field information collection system developed by ourselves;Data set B,the time series data of environmental factors of maize growth and its and photosynthetic intensity around a city in the northern China.The two data sets are subjected to the preprocessing procedure with some transformation operations to provide reliable and high-quality data sources for subsequent prediction studies.For the short-time prediction of environmental factors,in this thesis,an improved WPD-CEEMD-GA-Elman neural network model is proposed based on the existing research work in our group.The pre-processed data of each environmental factor are grouped into 24 groups of data by hour,and each group is recorded as hour_i;each hour_iis decomposed by Wavelet Packet Decomposition(WPD),then,decomposed again by Complete Ensemble Empirical Mode Decomposition(CEEMD).The final decomposed signals are fed into the GA-Elman neural network model for the prediction of each hour_i,then,all hour_i are further integrated to obtain the final prediction results.The experimental results show that the mean value of MAE(Mean Absolute Error)evaluation index of the improved WPD-CEEMD-GA-Elman neural network model is2.89 over dataset A,being lower 40.5%,61.18%,78%,and 89%than those of the WPD-CEEMD-Elman model,WPD-Elman model,GA-Elman model and standard Elman,respectively;the mean value of MAE evaluation index over dataset B is 1.99,being lower 51.9%,76.49%,88.4%,and 93.36%than those of WPD-CEEMD-Elman,WPD-Elman,GA-Elman,and standard Elman neural network models,respectively.It can be seen that the improved WPD-CEEMD-GA-Elman neural network model proposed in this thesis has high prediction accuracy rate compared with other network models.In the application of environmental factors,this thesis mainly uses environmental factors to predict the photosynthetic intensity of crop.Firstly,the correlation analysis between environmental factors and photosynthesis intensity is conducted,and the corresponding results show that the correlation between solar radiation intensity and photosynthesis intensity is the most significant.Then,the solar radiation intensity is used to fit the photosynthesis intensity and thereby obtain a fitting equation of solar radiation intensity-photosynthesis intensity.Finally,the predicted environmental data are input into this fitting equation to obtain the prediction results of photosynthesis intensity.The predicted MSE(Mean Squared Error)evaluation index value of photosynthesis intensity is 4.02,and MAE evaluation index value is 2.02,showing that the model proposed in this dissertation had good prediction effect,and the prediction of photosynthesis using environmental factors not only reduces the cost of relative devices,but also provides effective technical support for real-time monitoring of photosynthetic intensity during the growth of maize to serve digital agriculture. |
