Research And Application Of Machine Learning Method Based On Swarm Intelligence Optimization

At present,machine learning technology has been widely used in industry,agriculture,transportation,environment and other fields.Especially in the field of agricultural production,due to the promotion and implementation of agricultural information technology and precision agriculture,the rapid growth and accumulation of agricultural production data has greatly increased the demand for information technology,especially the demand for machine learning technology is more significant.However,due to the complexity of agricultural production problems,the traditional methods of machine learning are often not satisfactory with the decision-making problem.Therefore,how to construct the optimal agricultural intelligent decision-making methods based on machine learning is the urgent problem to be solved.This paper focuses on the use of swarm intelligent optimization technology to improve the existing machine learning technology to build a new methodology for agricultural intelligence decision problems,and then use these new methods to solve the problem of actual agricultural production.We first discuss the problems in the accuracy,applicability and stability of random forests,support vector machine and kernel extreme learning machine and other common machine learning methods,and propose the three-dimensional chaotic fruit fly optimization technology,improved particle swarm optimization technology,grey wolf optimization technology and multi-group grey wolf optimization technology,and then propose random forests prediction model based on improved three-dimensional chaotic fruit fly optimization technology,improved dynamic multi-objective particle swarm optimization based model,improved grey wolf optimized support vector machine diagnosis model,prediction model of multi-group grey wolf optimization intelligent evolutionary kernel extreme learning machine applied to rice pest prediction,disease diagnosis,deficiency diagnosis and yield prediction respectively.The proposed methods can better solve the problem of rice production decision-making.Details are as follows:(1)Aiming at the problem that the prediction process of random forest model is influenced by its parameters,this paper presents a random forest prediction model based on three-dimensional chaotic fruit fly optimization technology.The original fruit fly optimization algorithm from the two-dimensional search space is extended to three-dimensional space,while the introduction of chaos theory for initialization of population,to avoid falling into local optimum,presents a three-dimensional chaotic fruit fly optimization algorithm.This algorithm is tested on a number of test function,the experimental results show that the proposed method is compared with the original intelligent algorithm of fruit fly optimization and particle swarm optimization,which not only has better quality of solution,but also faster in convergence speed.Then we introduce the algorithm into the random forest model,and use the three-dimensional chaotic fruit fly optimization algorithm to train the random forest to establish the optimal calculation model.Finally,the proposed method is tested on the rice pest dataset and compared with other algorithms.The experimental results show that the proposed method has better prediction accuracy and can be more effective in the prediction of rice insect pests.(2)Aiming at the limitation of the machine learning method in single objective particle swarm optimization,a dynamic multi-objective particle swarm optimization model is proposed.Firstly,improving the original particle swarm algorithm,including environmental change factor,inertia factor and variation factor improvement.Then combined with the method of dynamic multi object technology,selected two clustering methods as objective function,design environmental variables and rules by background difference method,a dynamic multi-objective optimization model for improved particle swarm optimization is established to optimize the image recognition algorithm.Finally,after the feature extraction of the pretreated rice disease image,the model was used to test the disease feature set and compared with other methods.The experimental results show that the proposed model can obtain a large quantity,high quality and uniform distribution of the Pareto solution set,and it has a better accuracy than the single target method.(3)For the problem of selecting support vector machine model,an improved grey optimization algorithm diagnosis model of support vector machine is proposed.In this model,firstly,a new population initialization mechanism is introduced to optimize the population position for the grey wolf,so as to avoid getting into the local optimum and get better solution and improve the convergence speed of the algorithm.In a number of unimodal and multimodal functions were tested,the results show that the proposed algorithm,the improved optimization algorithm is better than the original grey wolf optimization algorithm in quality and convergence speed.Then,we introduce this strategy into the support vector machine,and dynamically select and adjust the penalty factor and kernel width in the model,and get the optimal recognition model.At last,the model was used to diagnose the deficiency of rice.The experimental results show that the diagnosis model can obtain over 95% recognition accuracy,the recognition accuracy of the proposed method is better than the support vector machine based on original grey wolf optimization algorithm,support vector machine based on grid search optimization and neural network method,realized accurate decision-making of rice nutrient deficiency problems.(4)Aiming at the problem that the kernel extreme learning machine is affected by the key parameters in the prediction problem,a prediction model of multi-population grey wolf intelligent evolution kernel extreme learning machine is proposed.Firstly,we take advantage of the multi-population intelligent evolution method to diversify the population of the grey wolf algorithm and the search space,and use the elite mechanism to share information among multiple populations,so as to obtain the global optimal solution.Then we introduce the strategy into the kernel extreme learning machine,and make the dynamic adjustment of the penalty coefficient and the Gaussian kernel width in the model,so as to establish the optimal prediction model.Finally,the rice yield data were tested.The experimental results show that the model not only improves the prediction accuracy of rice yield,but also obtains more stable prediction results,compared with the method based on the original grey wolf optimization based kernel extreme learning machine,support vector machine and neural network.This means that the rice yield prediction model proposed in this paper can predict the grain yield better and can be used as an important assistant decision-making tool in agricultural production.