基于神经网络的"回南天"观测数据质量控制方法初探OA

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 PSO-BPNN model are 1.73,1.64,and 1.68,respectively,and 1.82,1.83,and 1.78 for the BPNN model,respectively.