Situation element acquisition, assessment, and prediction are critical components of situation awareness. Enhancing the accuracy of risk trend extraction in security frameworks is essential. To address the issues of low prediction accuracy and slow convergence in industrial control network situation prediction, this paper proposes SAIPSO-BiLSTM, an adaptive particle swarm optimization-based prediction algorithm. Opposition-based learning is used to initialize the particle swarm to improve population diversity, while Gaussian and Cauchy mutation mechanisms dynamically adjust inertia weights to accelerate convergence. In addition, a Levy flight-based perturbation strategy, integrating similarity and aggregation degree, helps particles escape local optima. Taking into account the temporal dependencies in the situation data, a BiLSTM network is employed, with model optimization performed via the adaptive PSO. The experimental results show that the proposed SAIPSO-BiLSTM improves prediction accuracy and convergence speed, achieving a lowest MAPE of 1.40 and up to 0.16 increase in \(R^2\) compared to baseline models, demonstrating its clear advantage.