Optical Power Prediction Model for Integrated Learning Based on Dendrite Networks

Optical power is one of the indexes to measure the running state of optical fiber communication network. It is a kind of time series data with the characteristics of complexity, time variability and nonlinearity. To solve the problem of poor prediction accuracy of existing models, a Stacking integrated learning optical power prediction model based on dendrite networks is proposed. Firstly, Classification and Regression Tree (CART), Long Short-Term Memory network (Long short-term Memory, Three kinds of models, LSTM and Support Vector Regression (SVR), are used as learning tools. Dendrite network algorithm is introduced to establish the state prediction model of power optical communication system based on Stacking learning. Secondly, in order to verify the performance of the proposed model, it is compared with three base learners and the prediction models based on dendrite network Bagging ensemble learning and Boosting ensemble learning. The experimental results show that the MSE value of the proposed model is 0.0105, which decreases by 58.1183%, 6.3055% and 93.9329% compared with the three base learners, and decreases by 90.3618% and 42.1601% compared with Bagging-DD and Boosting-DD, respectively. The prediction accuracy and stability of the proposed model on the optical power data set are the highest, and compared with the single prediction model, the model can realize the state prediction of the power optical communication system.