Weather forecasting plays an important role in human society. Over time, various weather prediction systems have been developed, evolving from historical methods to the present day. In the past, numerical weather prediction (NWP) was the most accurate approach, relying on solving mathematical equations. Subsequently, machine learning techniques were introduced for this time series task. In the era of deep learning (DL), several DL models have been experimented with for weather forecasting, showing promising results. The weather characteristics in Vietnam vary from region to region. There are 03 main regions: Northern, Central, and Southern. This study explores the weather features for feature reduction and various methods for region-based meteorological forecasting in Vietnam. The methods examined in this research span from statistical models and machine learning to modern deep learning approaches, such as Transformer and Temporal Convolutional Network (TCN). The performance of these models is evaluated using meteorological data from 09 major cities representing 03 main regions (e.g.: Hanoi, Danang, Ho Chi Minh City, etc.), covering the period from January 2020 to August 2024. This study divides a day into four intervals (6-hour periods), and forecasts are made for each time frame. The results show that deep learning models significantly improve the accuracy of weather forecasts. Finally, we analyze the results and discuss several directions for improving weather prediction.