AI- and Security-Empowered End-Edge-Cloud Modular Platform in Complex Industrial Processes: A Case Study on Municipal Solid Waste Incineration.

Achieving long-term stable optimization in complex industrial processes (CIPs) is notoriously challenging due to their unclear physical/chemical reaction mechanisms, fluctuating operating conditions, and stringent regulatory constraints. A significant gap persists between promising artificial intelligence (AI) algorithms developed in academic research and their practical deployment in industrial actual processes. To bridge this gap, this article introduces the AI- and security-empowered end-edge-cloud modular platform (AISE3CMP). It consists of four systems such as whole-process AI modeling, end-side basic loop and AI-assisted decision-making, edge-side security isolation and AI control, and cloud-side security transmission and AI optimization. The data isolation collection module of the platform was deployed at a municipal solid waste incineration (MSWI) power plant in Beijing, where it collected multimodal data from real-world industrial sites. The platform's functionality and effectiveness were validated through the software and hardware developed at the Smart Environmental Protection Beijing Laboratory. The experimental results show efficient and reliable signal transmission between the systems, confirming the platform's ability to meet the computational demands of AI-based optimization and control algorithms. Compared to previous platforms, AISE3CMP features a dual-security transmission mechanism to mitigate data exchange risks and a modular design to enhance integration efficiency. To the best of our knowledge, this platform is the first prototype of a portable, end-to-end cloud platform with a dual-layer security mechanism for CIPs. While the platform effectively addresses data transmission security, further strengthening of cloud-side data protection and ensuring operational safety on the end-side remain significant challenges for the future. Additionally, utilizing this architecture to enable multi-region and multi-plant data sharing, in order to develop industry-specific large language models, represents a key research direction.