Research On Multi-meteorological Phase State Inversion Method Based On Machine Learning

Meteorology is closely related to human clothing,food and shelter.Observing meteorological phases can not only accumulate historical meteorological data but also promote national development.The main method of meteorological element forecasting algorithms is numerical forecasting.The essence is to use high-performance computers to calculate partial differential equations of atmospheric motion.Based on the climate background and weather evolution of a certain area,the area is estimated for a few hours and a few days.Then,even a few weeks later,the situation will be circulated and a qualitative or quantitative forecast will be made.However,there are still some problems in the forecast of meteorological elements at this stage.This paper uses machine learning as the main algorithm,and strives to make a breakthrough in the classification and classification of multiple meteorological phases.First,analyze the characteristics of the meteorological data obtained,understand meteorological knowledge,combine professional knowledge and experience,select the required meteorological element data and process it into an effective data set,and then design a multi-meteorological phase classification model to identify a certain future The forecast result of a certain meteorological element in a period of time.main tasks as follows:1.Collected a set of meteorological phase data sets suitable for machine learning training.First collect the ground map meteorological element data and temperature logarithmic pressure data from 1996 to 2015 in China,and then match them.Then remove the anomalous missing data and the mismatch between the two data sets.Finally,after communicating with meteorological professionals,the meteorological elements that do not contribute to the classification and judgment results are discarded,the data dimension is reduced,and finally 38 meteorological elements are obtained as data sources.2.A new meteorological phase classification model is proposed.This model combines support vector machine algorithm,particle swarm optimization algorithm and gradient equalization mechanism.The support vector machine algorithm is the base classifier of the model.The main function of the particle swarm optimization algorithm is to optimize the selection of the parameters within the SVM algorithm to improve the accuracy of the model classification.The gradient equalization mechanism can handle the categories well by reconstructing the loss function The imbalance problem makes the training more stable.Aiming at the meteorological data set,a single support vector machine is modeled for experimentation,and then combined with particle swarm optimization algorithm,the optimization modeling experiment is carried out in 6 SVM algorithms simultaneously.The SVM algorithm combined with the gradient equalization mechanism modeling experiment,and then combined a strong classifier combined by 6 SVM algorithms.We compare the classification accuracy of the three models,and finally verify the views of this article: the combination of support vector machine + particle swarm optimization algorithm + gradient equalization mechanism for classification accuracy is good,and the evaluation results are good,indicating that this model has a good multi-meteorological phase classification accuracy and high classification stability.Provide a certain reference for weather forecast.