优化进化神经网络的空气质量预测研究

进化神经网络模型与问题内部机制无关,避免了神经网络收敛到局部,但模型存在参数多而过于复杂的问题。对影响基本进化神经网络模型性能的个体编码方式和适应度函数进行优化,并自适应性定义种群交叉率、变异率。以大气中主要污染物SO2为例,考虑气温、相对湿度、风速等影响因子,实验仿真结果表明优化后的进化神经网络较传统的基本进化神经网络模型进化过程收敛更快,预测效果更佳,为环境保护部门提供可靠的决策依据。

Research of forecasting of air quality based on optimization of Evolutionary Neural Network

Evolutionary Neural Network（ENN） does not concern problems’ internal mechanism，avoids the convergence to local brought by neural network model，but the ENN model is complicated because of too many parameters.This paper optimizes the Simple ENN（SENN） with individual encoding，fitness function，and develops self-adaptive crossing rate and mutation rate of the species，which are main factors that influence the performance of the ENN.Simulation of the concentration that SO2 contained in the air with its impact factors and the historical data shows that Optimized ENN（OENN） converges faster in the evolution process than the SENN，and performs better in the forecasting.The OENN model can be capable to supply valuable prediction data for environmental protection.