Traffic Forecasting Based on Multi-scale convolutional and Multi-path residual Triple Direction LSTM

With the booming business scale of domestic telecom industry, its business order of magnitude presents exponential growth. Telecom operators put forward higher traffic forecasting requirements based on the rapid growth of network demand and increasingly complex network environment. The traditional single model network architecture is difficult to adapt to the current flexible network requirements and not able to extract the spatio-temporal correlations. In this paper, by integrating the Multi-scale convolutional and Multi-path residual block, integrating the Triple Direction Long Short Term Memory (MSMR-TriLSTM), This paper extracts the iterative features from the network traffic data set of a branch company of telecom, studies the time series prediction model in depth for the traffic data of telecom equipment and combines the clustering results of time series data, so as to realize the long-term prediction of complex traffic data. The experimental results of the proposed network model and algorithm show that, compared with the single model, the hybrid model has a significant advantage in predicting the long-term trend of server traffic. Analysis and comparison of experimental indicators show that Root Mean Square Error (RMSE), Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE) both decreased and the Accuracy of the model was as high as 88%, which indicates that the model could extract the main features of the server traffic data and had the ability of accurate and stable long-term prediction.