论文部分内容阅读
根据西安市雁塔区小寨环境空气监测点2011年7月31日起400 d的SO224小时平均浓度监测数据时间序列建立BP人工神经网络(ANN)预测模型,并用接下来100 d的数据对模型的仿真性能进行检验,从而验证了BP人工神经网络模型预测环境空气SO224小时平均浓度的可行性与准确度。经反复调试,最终选用2-3-1的网络结构并以trainbr作为训练算法,经34次迭代网络收敛,耗时7 s,预测结果相对于实际监测数据的平均绝对百分比误差为0.082,模型显示出良好的预测性能。预测结果表明,结构设定合理、训练算法选用适宜的BP人工神经网络模型能较好地反映SO2浓度的动态变化规律,具有可行性。
According to time series of monitoring data of SO224 hourly averaged concentration for 400 days from July 31, 2011 in Zhaotao district of Xi’an, Yanta District, BP artificial neural network (ANN) prediction model was established and compared with the data of the next 100 days The simulation performance of BP neural network model was tested to verify the feasibility and accuracy of BP artificial neural network model in predicting the average hourly concentration of SO224 in ambient air. After repeated debugging, the network structure of 2-3-1 is finally selected and trainbr is used as a training algorithm. After 34 iterations, the network convergence takes 7 seconds and the average absolute percentage error of the prediction result relative to the actual monitoring data is 0.082. The model shows Out of good predictive performance. The results show that the structure is reasonable and the BP artificial neural network model of training algorithm can reflect the dynamic change law of SO2 concentration well, which is feasible.