Capacity Index Prediction Based on STL Fusion Attention Mechanism

Accurate prediction of service capacity is very important for network optimization and network resource scheduling. It is difficult to predict the service capacity accurately because of the mobility of the population cycle and the complexity of the environment in the present network. Aiming at this problem, this paper proposes an algorithm for forecasting service capacity index based on attention mechanisms that can effectively mine the periodicity of service capacity index. Based on the STL model, the trend items and seasonal items of the capacity index series are decomposed, and then the improved Transformer attention mechanism model and the LSTM model are constructed. Collect and build the service capacity index database, build data sets for a variety of typical scenarios, and divide the training set and verification set. Based on the proposed algorithm, the uplink traffic indicators were modeled, and the average absolute percentage error (MAPE) of the model was reduced to 15.0%. The model accuracy was improved by 5.7%, and the R2 coefficient indicator was 0.903, which is significantly better than other comparative prediction models.