| As one of the main food crops in China,soybean is an important basic material related to the national economy and people’s livelihood.In order to improve soybean yield,it is often necessary to increase the amount of fertilizer application,which leads to environmental pollution issues.Targeted variable rate fertilization can protect the ecological environment while improving soybean yield.The Green Seeker spectral sensor can be used to collect the Normalized Difference Vegetation Index(NDVI)of crop canopy in real-time,and then the collected NDVI data can be used to realize the division of management zones and targeted variable rate fertilization.The Fuzzy C-means algorithm is the most commonly used algorithm for dividing management zoning based on NDVI data.This paper proposes a model-based Fuzzy C-means algorithm based on the Fuzzy C-means algorithm.This algorithm does not need to iterate all the data every time a group of data is obtained in dividing management zones,which can improve the speed of dividing management zones.In order to eliminate the lag effect of the variable spraying control system,a Long Short-term Memory(LSTM)neural network model was used to predict soybean NDVI data,enabling the controller to adjust and control the fertilizer amount in advance.A Smith-Fuzzy PID control algorithm based on Fluent modelling is proposed.The algorithm’s control simulation and performance test are carried out using the simulation tool Simulink of Matlab,which verifies the algorithm’s feasibility.The main research and conclusions of this paper are as follows:(1)LSTM neural network model,exponential smoothing method and autoregressive difference moving average model were used to predict soybean NDVI data.Through the analysis and comparison of the prediction results of the three models,it was found that the LSTM neural network model had the best prediction effect on the two groups of randomly selected NDVI data segments.The mean absolute percentage error is 7.49% and 6.28%,and the root mean square error is 0.04 and 0.03,respectively.It shows that LSTM neural network model has certain advantages in predicting soybean NDVI data.(2)A model-based Fuzzy C-means algorithm is used to partition the management zoning of soybeans by establishing a real-time crop management zoning system.The effect of dividing the management zones is evaluated using the Square Sum of Error,the Silhouette Coefficient,the Adjusted Rand Index,and the Homogeneity Index.The results show that with the increasing amount of NDVI data obtained,the model-based Fuzzy C-means algorithm can partition the management zones faster,0.02-0.15 seconds faster than the Fuzzy C-means algorithm;When the NDVI data volume is 4000,the model-based Fuzzy C-means algorithm has a poor partition effect.At this time,the Fuzzy C-means algorithm is suitable for dividing the management zones;The Adjusted Rand Index and Homogeneity Index can be maintained at about 0.7 when the NDVI data volume reaches 6000,which increases with the increase of the data volume,indicating that the two algorithms have similar partition effect after the data volume reaches6000,and the model-based Fuzzy C-means algorithm can partition the soybean management zones faster in real-time under the premise of ensuring accuracy.(3)Fluent is used to simulate the variable spraying liquid fertilizer control system,and the dynamic mesh technology in Fluent is used to simulate the opening of the flow control valve.According to the preset pressure control input signal,the pressure change data output by the simulation system is obtained.The variable spraying liquid fertilizer control system was built,the variable spraying test was carried out,and the field test’s input and output data were obtained.(4)Matlab is used to identify the input and output data of variable spraying liquid fertilizer system obtained by Fluent simulation and field test to obtain the system’s transfer function.Using Matlab to identify the system input and output data obtained by Fluent simulation,the fitting accuracy of the transfer function is 90.36%,and the fitting accuracy of the transfer function obtained by identifying the system input and output data obtained by field test is89.91%.Fluent simulation and the model of variable spraying liquid fertilizer system obtained from the field test are similar,which can be used for simulation debugging and verification of the control algorithm before the field test.(5)A Smith-Fuzzy PID control algorithm based on fluent modelling is proposed.This algorithm can achieve remote debugging of control parameters,avoiding the tedious operation of on-site debugging of control parameters for spraying.When the physical spraying system changes,the control program can be timely debugged through fluent simulation.Using Simulink to test the control performance of this algorithm,the results show that compared to the PID control algorithm,this algorithm has faster adjustment time,lower overshoot,better stability,and accuracy control,and the control performance is relatively close to Smith-Fuzzy PID control algorithm.The control performance of the algorithm was verified through the performance test of the variable spraying liquid fertilizer control system. |
PDF Full Text Request