基于APSO优化的小麦条锈病神经网络预测模型

针对小麦条锈病预测模型中样本数有限、样本分布不均的情况,在BP神经网络中引入改进的APSO优化算法,对神经网络的权值及阈值进行优化,一方面加入惯性权值及约束系数,另一方面在适应度函数中加入权值平方惩罚项来提高泛化能力,同时对“早熟”现象引入变异揉作.通过少样本数据的多角度仿真,结果显示,学习收敛速度及对新样本的泛化能力均得到了明显提高.改进的APSO-BP算法能克服常规BP算法在收敛速度及泛化能力方面的局限性,比常规BP算法及常规PSO -BP算法优越.

An Optimized Wheat Rust Neural Network Prediction Model Based on Adaptive Particle Awarm Optimization

Targeted the issues of limited number and uneven distributed samples in wheat rust prediction model,the advanced adaptive particle swarm optimization algorithm was used in back propagation neural network to improve its weights and threshold.On one hand inertia weights and constraint coefficients were applied in the algorithm.On the other hand the square of weights was added in the fitness function to improve the generalization ability of model.Meanwhile variation operation was introduced aimed at the "Premature" phenomena.The results showed that the convergence speed of the prediction model and generalization ability were both been significantly improved by small sample of the data and multi-angle simulation.The modified APSO-BP prediction model was superior to BP and typical PSO-BP prediction models.