Multitask Learning Based on Improved Uncertainty Weighted Loss for Multi-Parameter Meteorological Data Prediction

Abstract

With the exponential growth in the amount of available data, traditional meteorological data processing algorithms have become overwhelmed. The application of artificial intelligence in simultaneous prediction of multi-parameter meteorological data has attracted much attention. However, existing single-task network models are generally limited by the data correlation dependence problem. In this paper, we use a priori knowledge for network design and propose a multitask model based on an asymmetric sharing mechanism, which effectively solves the correlation dependence problem in multi-parameter meteorological data prediction and achieves simultaneous prediction of multiple meteorological parameters with complex correlations for the first time. The performance of the multitask model depends largely on the relative weights among the task losses, and manually adjusting these weights is a difficult and expensive process, which makes it difficult for multitask learning to achieve the expected results in practice. In this paper, we propose an improved multitask loss processing method based on the assumptions of homoscedasticity uncertainty and the Laplace loss distribution and validate it using the German Jena dataset. The results show that the method can automatically balance the losses of each subtask and has better performance and robustness.

Mixed-Input Residual Network for Air Temperature Forecasting by Low-Power Embedded Devices

“…The particular Jena dataset points used in the abovementioned Figure 8 of Wang et al (2022) were determined using a simple algorithm. They turned out to be points in a 24hour segment starting from the 50,095 th hour of the Jena dataset.…”

“…Comparison of MIRNet's single-step prediction on the Jena dataset with previous studies on the same dataset reveals a more objective view of the performance of the model. At first glance, MIRNet outperforms all but one(Wang et al, 2022) existing single-step model. A closer look at the Wang et al model however raises a number of questions.…”

“…It can be seen at a glance that none of the three models in the figure were tracking the true temperature particularly well. In order to investigate this further, the figure fromWang et al, (2022) was digitised using the PlotDigitizer online application available at https://plotdigitizer.com, after which the MSE for just that segment was calculated. Since the statistical properties of the Jena dataset are known, we were able to determine that the best Wang et al model achieved an MSE of 4.3 x 10 -4 for the segment.…”