A Distributed Based Network Slicing Fault Detection Method
A fault detection and network slicing technology, applied in transmission systems, electrical components, etc., can solve problems such as data islands and reduce communication overhead, achieve the effects of reducing communication overhead, improving generalization effects, and optimizing federated learning effects
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0050] This embodiment provides a network slicing fault detection architecture based on federated learning, which is specifically as follows:
[0051] see figure 2 , a network slicing fault detection framework based on federated learning consists of two networks, one is the local network at the physical nodes, and the other is the overall collaborative learning network among the physical nodes. Each physical node of federated learning learns a shared learning model through global collaboration, and at the same time saves the training data on each physical node. The uplink from the physical node to the global parameter aggregator is used to transmit parameters related to the local federated model, while the downlink is used to transmit parameters related to the global federated model.
[0052] Loss function: Assume that the VNFs that form a service function chain are deployed in N locations with local data sets D 1 ,D 2 ,...,D i ,...,D N physical server nodes. D. i is t...
Embodiment 2
[0064] This embodiment provides an unsupervised fault detection method based on a CNN-GRU network. The model performs training and detection methods of a fault detection model in network slicing as follows:
[0065] An unsupervised CNN-GRU fault detection model that uses the inherent properties of data instances to detect outliers, including an input layer, a CNN unit, a GRU unit, and an output layer. This method takes the preprocessed VNFs observation data as input, uses CNN to extract features, mines as much effective information hidden in the data as possible, constructs the extracted feature vectors as time series and inputs them into GRU to predict the future working status of the network to confirm Whether the network will fail within a certain period of time in the future.
[0066] 1) First, the data set is divided for training, verification and testing of the model. will divide the normal time series into four sets of time series: normal training set s N , the normal...
Embodiment 3
[0086] This embodiment provides a top-k gradient compression mechanism for model compression method, as follows:
[0087] 1) After the local training of each client is completed, execute the pseudo-gradient Δ i the clipping of (t);
[0088] 2) Determine the threshold Thr according to the gradient of the first k% of each client;
[0089] 3) If the current gradient is greater than Thr, send this gradient to the server;
[0090] 4) Otherwise, send this gradient to the buffer of the working program, and accumulate the gradient on the current client;
[0091] 5) If the cumulative gradient is greater than Thr, upload the gradient to the server and clear the cumulative gradient;
[0092] 6) Perform gradient upload.
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More