Network Downlink PRB Utilization Rate Forecasting and Evaluation Method based on Multi-feature Construction

Long-term forecasting on wireless network downlink PRB (Physical Resource Block) utilization rate is helpful to guide the network planning and construction. To improve the forecast accuracy and robustness, the paper proposes a multi-feature construction method including network population size, coverage area and network quality features, by extracting the real network traffic data to observe the distribution and correlation of 14 key features for training preparation. Then, we build a decision forest model to forecast network downlink PRB utilization rate and evaluate the model's long-time performance using MAE method. The model is applied in 5G network planning and construction system to forecast the next 6 months downlink PRB utilization rate. The test results indicate that the average MAE is 1.98%, and over 90% of the sample cells forecast deviation is within 5%.