In the capacity prediction process of telecom operation and maintenance, there are problems of too many capacity indicators and deployed business classes. Most of the existing researches do not consider the difference of indicator data types, and use the same prediction method for all types of data, which results in both good and bad prediction effects. In order to improve the efficiency of indicator prediction, a classification method of data type was proposed, and the data types were divided into trend type, periodic type and irregular type. Aiming at the prediction of periodical data, a periodic capacity indicator prediction model based on Bi-directional Recurrent Neural Network (BiRNN), called BiRNN-BiLSTM-BI, was proposed. Firstly, In order to analyze the periodic characteristics of capacity data, a busy and idle distribution analysis algorithm was proposed. Secondly, a Recurrent Neural Network (RNN) model was built, which included a layer of BiRNN and a layer of Bi-directional Long Short-Term Memory network (BiLSTM). Finally, the output of BiRNN was optimized by the system's busy and idle distribution information. Experimental results compared with the best one among Holt-Winters, AutoRregressive Integrated Moving Average (ARIMA) model and Back Propagation (BP) neural network model show that, the proposed BiRNN-BiLSTM-BI model has the Mean Square Error (MSE) reduced by 15.16% and 45.67% on the unified log dataset and the distributed cache service dataset respectively, showing that the prediction accuracy is greatly improved.

To handle the issues of small sample size and multi-target detection in the hierarchical detection of diabetic retinopathy, a weakly supervised target detection network based on improved GoogLeNet was proposed. Firstly, the GoogLeNet network was improved, the last fully-connected layer of the network was removed and the position information of the detection target was retained. A global max pooling layer was added, and the sigmoid cross entropy was used as the objective function of training to obtain the feature map with multiple feature position information. Secondly, based on the weak supervision method, only the category label was used to train the network. Thirdly, a connected region algorithm was designed to calculate the boundary coordinate set of feature connected regions. Finally, the boundary box was used to locate the lesion in the image to be tested. Experimental results show that under the small sample condition, the accuracy of the improved model reaches 94%, which is improved by 10% compared with SSD (Single Shot mltibox Detector) algorithm. The improved model realizes end-to-end lesion recognition under small sample condition, and the high accuracy of the model ensures its application value in fundus screening.