基于改进的SOM神经网络在水质评价分析中的应用

随着人们对水资源环境的日益重视,各种水质评价方法层出不穷.传统的水质评价方法多采用精确的数学模型进行描述,无法很好的反映水环境中存在的复杂非线性关系,从而影响整体评价结果.因此,本文提出采用一种改进的自组织特征映射神经网络(SOM)方法来进行水质评价,利用SOM神经网络能在无监督、无先验知识的状态下对样本进行自组织、自学习,实现对样本的评价与分类这一特点.通过引入主成分分析,解决SOM神经网络处理高维和相关性强的指标时出现的问题,提高网络收敛速度和聚类准确性.仿真结果表明:改进后的SOM神经网络能够直观准确地评价水体质量,反映水质整体状况.

Water quality evaluation analysis based on improved SOM neural network

As more and more attention is put on water environment, various water quality evaluation methods emerge endlessly. Most traditional evaluation methods with exact math model cannot completely reflect the complicated non-linear relationship in water environment, so it degrades the accuracy of evaluation results. This paper presents an improved self-organizing feature map (SOM) neural network to estimate water quality. SOM neural network can achieve sample evaluation and classification under an unsupervised self-organization and self-learning state, with no prior knowledge required. Principal component analysis is introduced to handle high-dimensional and highly related indicator problems, and improve network convergence speed and veracity. Simulation results show that the improved SOM neural network can accurately evaluate water quality and reflect the overall water situation.