Atmospheric particulate matter, such as PM2.5, contributes to air pollution negatively affecting human health. Many factors determine the change of PM2.5 concentration levels, which can be very sudden, nonlinear, and uncertain. Hence, traditional methods are not always suitable for predicting the exact amount of PM2.5 in the air. Effective forecasting of PM2.5 levels can tell people the air condition and support country's sustainable development; hence, forecasting PM2.5 values has an important social and long-term economic significance. This study proposes a system for monitoring the amount of PM2.5 and other pollutants in the air using the long-range wireless data communication technology LoRa and a cloud-based system model, which includes a proprietary terminal device. LoRa's excellent characteristics, such as low-power consumption and long range, allow effectively obtaining a history of PM2.5 readings. A PM2.5 prediction model based on a long short-term memory (LSTM) cyclic neural network is utilized to predict the next few hours' PM2.5 values and carry out an air quality index analysis of PM2.5. Experiments using atmospheric pollutant datasets collected specifically for this study from the North China University of Technology in 2019 show that the system can analyze well the PM2.5 datasets and accurately predict the hourly variation trend of PM2.5.