Network Security Situation Prediction Model Based on VMD Decomposition and DWOA Optimized BiGRU-ATTN Neural Network

Abstract

The widespread adoption of Internet-of-Things (IoT) devices has resulted in a comprehensive transformation of human life. However, the network security challenges posed by the IoT devices have become increasingly severe, necessitating the implementation of effective security mechanisms. Network security situational awareness enables an effective network state prediction for better formulation of network security defense strategies. Existing network security situational prediction methods are typically constrained by situational sequence data, especially those sequences with a high degree of non-stationarity, leading to unstable predictions and low performance. Moreover, in real-world application scenarios, the network security situational sequences are often highly non-stationary. To address these challenges, we introduce a novel hybrid prediction model named Variational Mode Decomposition (VMD) - Dynamic Whale Optimization Algorithm (DWOA) - Bidirectional Gated Recurrent Unit (BiGRU) - Attention Mechanism (ATTN). The proposed model integrates VMD, BiGRU, ATTN, and DWOA. Initially, network security situational awareness sequences are processed using VMD to decompose them into a series of subsequences, thus reducing the non-stationarity of the original sequences. Subsequently, an enhanced DWOA optimization algorithm is introduced for tuning the hyperparameters of the BiGRU-ATTN network. Ultimately, BiGRU-ATTN is employed to predict each of these subsequences, which are then aggregated to yield the final network security situational prediction value. When compared with several existing methods on public network security datasets, the proposed VMD-DWOA-BiGRU-ATTN method demonstrated an improvement in the R-2 values ranging from 6.34% to 52.61%. These results substantiate that the model significantly enhances predictive performance.