Preface to the Special Issue “Complex Process Modeling and Control Based on AI Technology”

In the Industry 4.0 era of digital transformation, complex process modeling and control are critical challenges in various industrial sectors, including steel metallurgy [1,2], chemical manufacturing [3,4], and energy systems [5,6]. The integration of artificial intelligence (AI) technologies has transformed traditional methods by offering unparalleled capabilities in handling nonlinearity, uncertainty, and high-dimensional data.

Traditional modeling and control methods often struggle to address the inherent complexity and dynamics of modern industrial processes [7]. They rely on precise mechanistic models and struggle to handle the uncertainties inherent in complex nonlinear systems, lack adaptability, and result in poor control performance in dynamically changing environments. These methods are computationally complex, making it difficult to find globally optimal solutions. They are also sensitive to changes in system parameters, which prevents them from effectively addressing the large-scale high-dimensional control challenges found in modern industry.

Recent AI advancements, particularly in deep learning, reinforcement learning, and federated learning, have demonstrated extraordinary potential in overcoming these challenges [8]. AI-based methods have robust, data-driven, and self-learning capabilities. These capabilities enable end-to-end intelligent processing, which significantly improves efficiency. They can also automatically identify complex patterns and adapt to dynamic environments without relying on precise mathematical models. AI methods also offer strong generalization and adaptability. They continuously optimize performance through ongoing learning, demonstrating flexibility and accuracy that traditional methods cannot match when addressing high-dimensional, nonlinear, and complex problems.

This Special Issue showcases cutting-edge research at the intersection of AI and complex process modeling and control. We are dedicated to bridging the gap between theoretical AI advancements and practical industrial applications while promoting interdisciplinary collaboration and knowledge exchange.

Contribution 1 proposes a two-stage high-precision PCB component recognition framework integrating CBAM-enhanced YOLOv11 with sub-pixel geometric refinement. The system achieves industrial-grade localization accuracy essential for robotic PCB disassembly under complex visual interference conditions. Contribution 2 proposes an imprecise probability-based load transfer control method integrating IDM with NCC for switch timing prediction. This method maintains disconnection–reconnection actions within 20 ms to prevent circulating currents in distribution networks. Contribution 3 proposes a bidirectional LSTM–Transformer interaction architecture with cross-attention mechanisms for dynamically coupling global market context with local temporal features in gold futures price forecasting. The proposed framework achieves superior prediction accuracy with significantly reduced error volatility during extreme market events. Contribution 4 proposes a three-module D2D mode selection framework integrating GRU-based SINR prediction with statistical error analysis and threshold selection using AR and PCR constraints. The method achieves higher system throughput, longer D2D mode residence time, and lower mode-switching frequency compared to existing approaches in dynamic network environments. Contribution 5 proposes a dual-path neural network architecture integrating spatial feature extraction via Gramian angular field-transformed images with temporal feature analysis through GRU networks, optimized by the Crested Porcupine Optimizer algorithm. This method achieves about 99% recognition accuracy with robust fault detection capability under varying noise conditions and different network topologies in distribution systems. Contribution 6 proposes a flow-path-offset joint scheduling algorithm with multi-index priority scoring for time-sensitive networks, achieving a 39% higher scheduling success ratio than the naive algorithm and demonstrates superior performance in large-scale TSN deployments. Contribution 7 proposes an improved RT-DETR integrating super-resolution convolutional module and channel attention mechanism, achieving 2.8% mAP and 1.7% AR improvements with only 0.4 MB additional parameters for rail fastener defect detection. Contribution 8 proposes an anti-disturbance bumpless transfer control framework integrating a switched equivalent-input-disturbance estimator with a compensator-based continuous controller design under average dwell time switching strategy, achieving asymptotic stability and L2-gain disturbance rejection properties, while ensuring controller signal continuity during subsystem transitions, as verified by simulations on a switching RLC circuit system with sawtooth wave disturbances. Contribution 9 proposes a time-series prediction model for the electricity load of charging piles integrating variational mode decomposition with broad learning system and multi-model fusion optimized by particle swarm optimization. It achieves superior prediction accuracy with an R² of 0.9831 and PMAPE of 2.6468, significantly outperforming LSTM, NARX, and standard BLS and ELM models by 52–89% across all evaluation metrics, while maintaining computational efficiency for optimal electricity-load scheduling decision-making. Contribution 10 proposes a fault distance measurement method for distribution networks integrating a novel phase-mode transformation matrix with Markov Transition Field image conversion and Darknet-19 deep learning architecture. This method achieves high accuracy fault location with an average prediction error of only 330 m under 40 dB noise conditions, demonstrating strong robustness and practical applicability as validated on Raspberry Pi-embedded hardware with error margins below 0.26 km across various fault scenarios. Contribution 11, which is a comprehensive review paper on AI-integrated control engineering, synthesizes recent advancements in artificial intelligence applications for system identification, control, and optimization. The authors categorize and analyze neural network control architectures (including adaptive, sliding-mode, backstepping, and iterative learning variants), model predictive control enhancements through deep learning, and reinforcement learning control paradigms, while identifying critical challenges in interpretability, data requirements, computational complexity, and stability guarantees that shape future research directions for intelligent control systems across industrial applications.

Conclusions

This editorial provides an overview of the progress in data-driven intelligent modeling and control. It emphasizes the cutting-edge methodologies and technological applications that have emerged in this field in recent years. These approaches harness data to perform modeling and control tasks, significantly enhancing the adaptability, reliability, and efficiency of complex systems. Studies show that amid the rapid development of Industry 4.0, data-driven intelligent modeling and control have become essential in ensuring product quality, promoting sustainable development, and enhancing production efficiency.