基于神经网络的“回南天”观测数据质量控制方法初探

Preliminary study on the quality control method for observation data of“Continuous Wet Weather”based on neural network

为判别“回南天”观测设备数据可靠性,基于传统反向传播神经网络(BPNN),结合粒子群优化算法(PSOBPNN),对广西“回南天”观测数据进行质量控制研究。结果表明:(1)在模型估算温度与实测温度对比验证中,与BPNN模型相比,PSO-BPNN模型精度更高,PSO-BPNN模型没有明显高估或低估,而BPNN模型在10℃附近出现较大偏差。(2)在使用测试集数据对模型进行测试中,瓷砖地面和墙面温度在10~30℃范围,模型的适用性更强,PSO-BPNN模型稳定性优于BPNN模型。(3)在随机添加人工误差进行的模型检验中,PSO-BPNN模型瓷砖地面、墙面、水泥地面温度的最佳质量控制参数分别为1.73、1.64、1.68,BPNN模型分别为1.82、1.83、1.78。

In order to determine the reliability of the observation data of“Continuous Wet Weather”,a quality control study on the observation data of“Continuous Wet Weather”in Guangxi was carried out based on the traditional back-propagation neural network(BPNN),combined with particle swarm optimization(PSO)algorithm(i.e.PSO-BPNN).The results show that:(1)compared with the traditional BPNN model,the accuracy of the PSO-BPNN model is higher in comparing the model-estimated temperature with the measured temperature,without any significant overestimation or underestimation in the PSO-BPNN model,while the BPNN model shows a large deviation around 10℃.(2)In the tests of PSO-BPNN and BPNN model,tile floor and wall temperatures in the range of 10~30℃ show greater applicability of the models,and the PSO-BPNN model is more stable than the BPNN model.(3)Randomly adding artificial errors for model validation,the optimal quality control parameters for the temperatures of tile ground,wall,and cement ground in the PSOBPNN model are 1.73,1.64,and 1.68,respectively,and 1.82,1.83,and 1.78 for the BPNN model,respectively.