Application Research On Pest Prediction Based On Improved Fireworks Algorithm

As a large agricultural country,one of the most important industries in China is agriculture,the development of which is linked to the prosperity of the national economy and the improvement of living standards.The yield and quality of crops are directly related to people’s daily life.Pests and diseases have a direct impact on the growth and development of crops,so the monitoring and control of pests and diseases is extremely important.The use of intelligent methods to monitor and control pests is of great importance in the development of modern Chinese agriculture.BP neural networks have a strong self-learning capability and are well suited to solving non-linear complex problems.The use of BP neural networks to construct prediction models based on pest-related factors opens up new avenues for pest prediction,but some shortcomings have also been identified,such as the tendency to fall into local minima and uncertainty in the structure of the hidden layer.To address these problems,this paper proposes a method to optimize the initial weights and thresholds of the BP neural network model based on the adaptive radius fireworks algorithm with t-distribution perturbation.The fireworks algorithm is a population intelligence algorithm based on the fireworks explosion phenomenon,and is an optimization algorithm with a global search mechanism.However,like other population intelligence algorithms the Fireworks Algorithm(FWA)has the problems of easily falling into local optimum and less population diversity.To address these shortcomings,this paper proposes an adaptive radius Fireworks Algorithm(TOBL-FWA)based on t-distribution perturbation.The algorithm introduces Tent chaotic mapping in the population initialization stage,using the properties of chaotic mapping to make the initialized population random while containing intrinsic laws,improving the diversity of the initial population,using t-distribution variation instead of Gaussian variation to improve the global search capability of the algorithm while ensuring its local search capability,and dynamically adjusting the algorithm based on the sparkle information of t-distribution variation blast radius,improving the local search ability of firework individuals and increasing the diversity of the population,and introducing a reverse elite learning strategy in the selection strategy,which retains more information about elite individuals in the iterative process and improves the global search ability of the algorithm,which in turn improves the algorithm’s optimization performance.Finally,the experimental results of 12 standard test function experiments and six offset test experiments with six offsets show that the TOBL-FWA algorithm has better convergence and stability,proving that the adaptive radius fireworks algorithm based on t-distribution perturbation has a better optimization effect for the fireworks algorithm.The initial weights and thresholds of a BP neural network directly affect the training and construction of the network.The initial weights and thresholds of the BP neural network model were optimized using an improved fireworks algorithm to build a new network model(TOBL-FWABP),which was proven to perform better in the prediction problem through simulation experiments to an effective prediction model.The association between pest and meteorological factors was determined,pest data samples were collected,the TOBL-FWABP network model was trained and tested,and the feasibility of the TOBL-FWABP neural network in the pest prediction problem was demonstrated through the analysis and comparison of experimental results.