Soil Moisture Prediction Model Based On Neural Network

Soil moisture prediction is a technical direction in the concept of precision agriculture.However,there are many problems in soil moisture prediction,such as the lack of training data,model that easily falls into local minima and the low prediction accuracy.Therefore,this dissertation mainly studied two aspects:optimizing data interpolation algorithm and optimizing soil moisture prediction model.Aiming at the data interpolation problem,this dissertation proposed a density-based sine cosine algorithm and optimized the ordinary Kriging interpolation model by using it,and it was a good data foundation for the establishment of neural network prediction model.The ordinary Kriging method which based on a series of assumptions can still be optimized in its accuracy.In this dissertation,the parameter optimization problem in ordinary Kriging method was abstracted as NP-hard problem and the density-based sine cosine algorithm was used to optimize the ordinary Kriging interpolation model.The performance of the algorithm was tested by using standard functions and the test data was from the experimental field in Changge City,Henan Province.The results showed that the density-based sine cosine algorithm has better local avoidance ability and convergence speed Also faster,and shows better performance in optimizing ordinary kriging interpolation models.In order to optimize the soil temperature and moisture prediction model,a multi-objective particle swarm optimization based on dynamic network was proposed and was applied to the optimization of neural network prediction model.The dynamic multi-objective particle swarm optimization neural network prediction model was finally established.Balance the exploration and exploitation of the solution space,make the algorithm have a good convergence,good local optimal avoidance ability and uniform distribution of solution set are the major challenges for most meta-heuristic multi-objective optimization.The algorithm proposed in this dissertation was introduced a dynamic scale-free network topology in the topology structure,which increased the information correlation between particles.According to the information of particles,the network structure dynamically adjusted to balance the exploration and exploitation of the solution space.The particles combined the quality information of the surrounding particles to update,which makes the distribution of solution set uniform and makes the particles avoid over-concentration search.The performance of the proposed algorithm was tested on several benchmark functions and compared with other multi-objective optimization algorithms such as Non-Dominated Sorted Genetic Algorithm-?(NSGA-?),Strength Pareto Evolutionary Algorithm-?(SPEA2),the sigma method in Multi-objective Particle Swarm Optimization(sMOPSO)and the Multi-objective Particle swarm Optimization Based on Decomposition and Dominance(D2MPSO).Experimental results validate the effectiveness of proposed algorithm.In most test cases,the proposed algorithm was superior to NSGA-?,SPEA2,sMOPSO and D2MPSO in convergence,stability and solution distribution.The algorithm was used to optimize the neural network soil temperature and moisture prediction model.The results showed that the algorithm can enhance the local optimal avoidance ability and improve the predictive accuracy of the neural network model.Finally,in order to achieve real-time collection and accurate transmission of crop canopy micro-meteorological data,analyze,predict and display the data,the crop moisture monitoring and forecasting platform was built.The overall architecture and the software architecture of the crop moisture monitoring and prediction platform were designed.Data processing server and the web server were built.The system was shown in this dissertation.