Analysis and Prediction of Meteorological Data Based on Edge Computing and Neural Network

Introduction

The meteorological environment is the foundation for the survival and development of human society. However, due to the excessive usage of resources for economic development, the meteorological environment is severely damaged and becomes extremely fragile (Li et al., 2012). Therefore, the analysis and prediction of meteorological data is great significance to environmental governance and improvement. Meteorological observation data are mainly collected through automatic weather stations. Modern automatic weather stations have basic functions, such as data collection, data processing, data storage, data transmission, data quality control, and operation monitoring (Tang & Gao, 2008). The realization of these functions depends on the accurate acquisition of real-time data. Due to the existence of electromagnetic waves and man-made interference sources, and the unpredictability of the environment where the automatic weather station collector is located, the data received during the operation of the automatic weather station is inaccurate (Pepin & Duane, 2010). In addition, predicting and verifying the collected real-time data can improve the stability of the automatic weather station operation and the convenience of manual operation, which make the weather data more predictable, and effectively increase the functions of the control system for managers. This provides a reliable basis for decision-making (Liu et al., 2006).

At present, the analysis and prediction methods and theories of meteorological data have developed rapidly, which provides strong support conditions for the analysis and prediction of large, complex and information-rich data sets (Wang, Zhang, Wang et al, 2020). Recently, many researchers have carried out extensive works on the specific problems of meteorological data analysis and prediction, and proposed a variety of approaches, such as methods based on statistics, methods based on machine learning, and methods based on deep learning (DL) method. Among them, deep learning technology is one of the most popular methods, which deals with complex problems by simulating the learning and training functions of biological neural networks (Rasheed et al., 2020; Wang, Han, Leung et al, 2020). It has strong parallel processing, redundancy and fault tolerance, association, memory, self-organization, and self-adaptation functions. However, a network model of deep learning usually contains a large number of neuron parameters, which causes its computational efficiency to be greatly reduced and restricts its development in the field of meteorological data processing. In order to optimize the data processing capabilities of neural networks, many methods have proposed in decades, and edge computing is a fastest growing one approach (Molina-Masegosa & Gozalvez, 2017). The combination of this technology and deep learning mainly reflects two advantages: firstly, DL can be integrated into the framework of edge computing, which enables adaptive management of the network edge. In addition, edge computing also pushes a large number of model calculations from the cloud down to the edge, and achieves low-latency, high-reliability intelligent services (Liu et al., 2020).