土壤墒情预测自适应遗传神经网络算法研究

Research of adaptive genetic neural network algorithm in soil moisture prediction

针对传统遗传神经网络算法易出现的早熟收敛及锯齿等现象,提出一种新型算法应用于土壤墒情预测。该算法提出了衡量种群基因多样性的遗传多样性函数的概念,自适应调节交叉和变异策略,在全局范围内寻找最优初始网络权值和阈值,从而降低算法迭代次数,提高神经网络预测的精度和效率。仿真结果表明,与其他遗传神经网络算法相比较,该算法平均绝对误差从2%降低到1%,平均相对误差从5%降低到3%,最大相对误差从15%降低到8%,即新型算法可有效提高墒情的预测质量。

Forecasting soil moisture accurately is very important to monitor plant growing. Researchers are resorting to hybrid intelligence algorithms fusing more effective strategies into prediction process. Combination optimization can overcome the disadvantages of single method and improve predictive quality. This paper advances a novel algorithm to conquer the prematurity and sawtooth of traditional neural network. Firstly, it proposes the conception of genetic diversity function which measures genetic diversity of population. Secondly, it uses adaptive crossover strategy and mutation strategy to obtain the best initial weights and thresholds. Finally, it receives neural network results with better precision and efficiency and less iterations. Simulations reveal that in contrast to other genetic neural network, the quality of the soil moisture forecast has a great improvement in the new algorithm.