Haze prediction method and device

Abstract

The invention provides a haze prediction method and device and belongs to the meteorological prediction technical field. The method includes the following steps that: monitoring points in a monitoring area are determined; a plurality of effective observation values are selected according to the data information of the monitoring points in the monitoring area so as to be adopted as feature selections for haze prediction, and neural fuzzy models are built for the feature selections, wherein the data information includes first data information and second data information, and the neural fuzzy models include an MLR model, an ANN model and an NF model; subordinating degree functions are selected according to the first data information and the second data information of each monitoring point, and the first data information and the second data information are multiplied, so that third data information can be obtained, normalized confidence can be calculated based on the third data information; and a fuzzy result is calculated according to a fuzzy rule and the normalized confidence result of each monitoring point, and a haze prediction result is obtained according to the fuzzy result. With the haze prediction method and device of the invention adopted, the real-time performance, validity and reliability of haze prediction can be effectively improved.

Description

technical field [0001] The invention relates to the technical field of meteorological forecasting, in particular to a haze forecasting method and device. Background technique [0002] With the gradual acceleration of urbanization, the problem of air pollution has become particularly prominent. For accurate forecasting of air pollution, any predictive model tends to perform better than purely persistent forecasts. [0003] The current forecasting models mainly include time series forecasting model, multiple linear regression model and neural network forecasting model. In the time series forecasting model, the model parameters come from a specific region or location, and its parameters are required to change slowly with time, so it is difficult to determine the observation variables with large influence weights. In the multiple linear regression model, using the minimum mean square error method to estimate the regression parameters, the measurement error is large. The neura...

AI Technical Summary

Problems solved by technology

First, the generalization ability of the network model that has completed learning cannot be fully reflected, and the change of input parameters when predicting the haze concentration has not successfully redistributed the number of neurons; second, the convergence speed is relatively slow. Slow, because the neural network often stays in the flat area of ​​​​the error gradient, the convergence problem has become an inevitable difficulty; third, the data training process will lead to prediction errors, and may fall into the misunderstanding of local minima; fourth , the selection of the number of neuron nodes in the hidden layer of the network needs to be manually defined in advance, and cannot be changed, that is, it lacks a unified and complete self-learning ability; Data retraining, resulting in too long training time period

Images