基于改进型PSO-BP神经网络算法的水环境质量评价

发布时间：2018-11-09 11:15

【摘要】：针对水环境污染的治理与保护,需要科学的水环境评价方法对水环境进行分类。我国目前采用单因子评价法,评价原则为“一票否决制”。这种评价方法具有简单直观,易于操作的优势。但是,单因子评价法在水质监测数据上存在数据利用不从分,评价结果过于悲观的缺点。本文基于云南省洱海流域信息化项目,针对洱源县永安江的水质监测数据采用主成分分析法、BP神经网络、PSO-BP算法进行水质评价研究。研究过程中发现主成分分析法所构成的评价函数物理意义不明确,并且评价过程中不能对污染物中影响程度较大的指标进行重点评价。针对主成分分析法以上不足,改用人工神经网络评价法对水质评价工作进行建模,采用BP神经网络算法对水环境质量进行综合评价。BP神经网络算法具有良好的非线性映射和自学习能力,对具有非线性复杂关系的水环境质量评价工作,结果更有针对性、物理意义更明确。但是,BP神经网络算法有易陷入局部极值点、收敛速度慢、泛化能力弱、对网络初始化参数敏感的缺陷。针对BP神经网络水质评价模型的缺陷,本文考虑使用粒子群算法对BP神经网络的网络参数进行优化。由于粒子群算法具有全局搜索能力强的优点,通过粒子群算法对神经网络的连接参数进行优化,弥补神经网络算法对网络参数初始化设置敏感以及容易陷入局部极小值点的不足。同时,粒子群算法易于实现、结构简单,与其他算法容易结合;粒子群算法采用并行运算,运算速度快,资源利用率高。将两种算法结合以后,提高了神经网络算法的收敛精度和泛化能力。但是,在对BP神经网络进行优化的过程中引入了新的变量和迭代过程,也增加了算法的运行时间。最后,对粒子群算法中惯性权重衰减函数的改进,在保证评价算法收敛精度的条件下,提高了算法的收敛速度,减少算法运行时间。通过实验仿真,验证了改进的评价算法在收敛精度保持一定的条件下,减少了算法运行时间。
[Abstract]:For the treatment and protection of water environment pollution, it is necessary to classify the water environment by scientific water environment evaluation method. At present, the single factor evaluation method is adopted in our country, and the principle of evaluation is "one vote veto system". This evaluation method has the advantages of simple and intuitive, easy to operate. However, the single factor evaluation method has the disadvantage of not using the water quality monitoring data, and the evaluation result is too pessimistic. Based on the information project of Erhai River Basin in Yunnan Province, the water quality monitoring data of Yongan River in Eryuan County were evaluated by principal component analysis, BP neural network and PSO-BP algorithm. It was found that the physical meaning of the evaluation function constituted by principal component analysis was not clear, and the evaluation process could not focus on the indexes with greater influence on pollutants. In view of the deficiency of principal component analysis method, artificial neural network evaluation method is used to model water quality evaluation. The BP neural network algorithm is used to evaluate the water environment quality synthetically. The BP neural network algorithm has good nonlinear mapping and self-learning ability, and the result is more pertinence to the water environment quality evaluation work with nonlinear and complex relationship. The physical meaning is clearer. However, the algorithm of BP neural network is easy to fall into local extremum, slow convergence speed, weak generalization ability and sensitive to network initialization parameters. Aiming at the defects of the water quality evaluation model of BP neural network, particle swarm optimization (PSO) algorithm is considered to optimize the network parameters of BP neural network. Because particle swarm optimization (PSO) has the advantage of global searching ability, the connection parameters of neural network are optimized by PSO. The neural network algorithm is sensitive to the initialization of network parameters and is prone to fall into local minima. At the same time, particle swarm optimization algorithm is easy to realize, simple structure, easy to combine with other algorithms, particle swarm optimization algorithm uses parallel operation, fast operation speed, high resource utilization. After combining the two algorithms, the convergence accuracy and generalization ability of the neural network algorithm are improved. However, in the process of optimization of BP neural networks, new variables and iterative processes are introduced, and the running time of the algorithm is also increased. Finally, the improvement of inertia weight attenuation function in PSO can improve the convergence speed of the algorithm and reduce the running time of the algorithm under the condition of ensuring the convergence accuracy of the evaluation algorithm. The experimental results show that the improved evaluation algorithm can reduce the running time of the algorithm under the condition of keeping the convergence accuracy.
【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：X824;TP183

【相似文献】