Computing Power Network Resources Prediction Based on GRU-FFT

Predicting computing resources can provide scientific basis for the precise allocation of resources and real-time scheduling of computing tasks, thereby optimizing resource allocation and system performance. This paper proposes a model of Computing Power Network resource prediction based on Gated Recurrent Unit and Fast Fourier Transform to address the heterogeneous ubiquity and dynamic time-varying characteristics of computing power resource data. The model analyzes the time-domain and frequency-domain features of computing power data to predict computing power resources. GRU is responsible for capturing the characteristics of data in the time domain, and then converting the time domain data into the frequency domain through FFT to extract the features of the data in the frequency domain. Based on the Alibaba open-source dataset, experimental analysis was conducted, and the results showed that the proposed computing power resource prediction method reduced the Mean Absolute Error by 49.44%, 53.79%, and 72.11%, compared to GRU, LSTM, and RNN models respectively; the Mean Squared Error decreased by 52.87%, 61.13%, and 74.37%, respectively. the Root Mean Squared Error reduced by 31.35%, 37.66%, and 49.38%, respectively.