Research And Application Of Time Series Registration And Missing Value Processing Methods In Meteorological Field

Meteorological data is the main basis for the study of meteorological services and meteorological issues,and its quality fundamentally affects the effectiveness of services and research.The meteorological data collected by the sensor is a kind of time series data,and there are often problems such as inconsistency and loss,so it needs to be effectively preprocessed.In this paper,the problem of registration and missing value processing in time series preprocessing in meteorological field is deeply studied,and the sliding registration method based on sliding window and the missing value processing method based on generative adversarial network are proposed.The main work is as follows:1)Translation registration method based on sliding windowBy analyzing the existing precision of time registration method and the minimum frequency of synchronization frequency,combined with the mining requirements of time series in the context of big data,a new sliding window based registration model is designed and proposed.A sliding registration method based on sliding window.The method adopts the principle of sliding window and neighborhood proximity.By calculating the offset time interval,the low-frequency sampling time data is shifted to the high-frequency sampling time series,and the registration fitting degree can reach 96.7%,which effectively improves the registration accuracy and reaches the time.The registration target of the sequence.2)Missing value processing method based on generative adversarial networkAiming at the problem of neglecting the time dimension of time series in the current missing value processing method and the need to complete training data for constructing the padding model,a time series missing value processing method GAN-TSI based on the generative adversarial network is proposed.The method uses the time series model BiLSTM as the main structure of the generator and discriminator in the model,and combines the proposed adaptive learning strategy to simulate the original data distribution,and then generates the missing value.The minimum square error of the missing value is0.771.Effectively achieving the purpose of filling.3)Demonstration application in meteorologyThrough the deployment of six self-developed air quality and meteorological elements integrated monitoring equipment,a number of data were collected in the field,and the two pre-processing methods proposed in this paper were applied in the data set.Thepre-processed data by the proposed method can effectively improve the accuracy of smog concentration prediction,indicating that the proposed sliding window registration method and the missing value processing method based on the generative adversarial network have important practical applications value.