Host Load Forecasting by Elman Neural Networks

The core issue of load forecast is the mathematical model. Traditional mathematical models lack abilities of self-learning and self adaptation, and guarantee for the robustness of the prediction system. The host load obtains with the utilization of CPU , memory , network bandwidth, which has the characteristics of non-linear, time-varying and uncertainty. In order to improve the forecastable accuracy of host load and enhance the robustness of the host system in a distributed system, especially instantaneous growth of the host load under the service asynchronous scheduling by an effective processing, the paper establishes a model of host load forecasting based on Elman neural network and a gradient-falling learning algorithm. Simulation results indicate that the model has better results in prediction effect, relative to the linear model and BP neural network model with higher precision and better adaptability.