Abstract:

Aiming at the increasingly serious safety situation of industrial control system (ICS) in petrochemical industry, a forward neural network (FNNPSOGSA) based on particle swarm optimization (PSO) and universal gravitation search algorithm (GSA) is proposed to solve the problem of intrusion detection. A hybrid algorithm based on PSO and GSA is proposed by using the global optimization ability of GSA and the fast local convergence of PSO. PSOGSA is used as a new training method of FNNs to study the effectiveness of FNNPSOGSA model in practical engineering application scenarios. By comparing the FNNPSOGSA prediction results with the FNNPSO, FNNGSA and the improved FRGNN (K-NN) and FRGNN (Naive Bayes) prediction results in the reference literature, the results show that PSOGSA is feasible in training FNNs and has higher prediction accuracy and more generalization ability. The algorithm is applied to attack prediction in intrusion detection of industrial control system (ICS), and simulation study is carried out using industrial intrusion detection standard data set. The results show that the algorithm can achieve very good results in intrusion detection of industrial control systems.

摘要：

针对日趋严峻的石化行业工业控制系统（ICS）安全形势，提出一种基于粒子群优化（PSO）和万有引力搜索算法（GSA）的前向神经网络（FNNPSOGSA），用于解决其中的入侵检测问题。分别利用GSA的全局寻优能力和PSO快速局部收敛优势，提出了一种基于PSO和GSA的混合算法PSOGSA，并将其用于前向神经网络（FNNs）的训练。通过多组基准测试数据集，将FNNPSOGSA预测结果同FNNPSO、FNNGSA和参考文献中改进的FRGNN（K-NN）和FRGNN（Naive Bayes）预测结果相比较，验证了PSOGSA在训练FNNs中是可行的，并且FNNPSOGSA具有更高的预测准确率和更强的泛化能力。更进一步，对工控入侵检测标准数据集的仿真结果表明其在工控系统入侵检测中的可行性和有效性。