AI Adoption for Steam Boiler Trip Prevention in Thermal Power Plants

Abstract

This study introduces two advanced artificial intelligence systems designed to model and predict various boiler trips, playing a pivotal role in maintaining boilers' normal and safe functioning. These AI systems have been meticulously developed using MATLAB, thus offering sophisticated tools for diagnosing boiler trip occurrences. Real-world operational data from a coal-fired power plant, encompassing a comprehensive range of thirty-two operational variables tied to seven distinct boiler trips, was harnessed for these innovative systems' training, validation, and analysis. The first intelligent system capitalizes on a pure Artificial Neural Network (ANN) approach, leveraging the insights drawn from plant operators' decision-making processes concerning the key variables influencing each specific boiler trip. On the other hand, the second system takes a hybrid approach, incorporating Genetic Algorithms (GAs) to emulate the decision-making role of plant operators in identifying the most influential variables for each trip. Moreover, different topology combinations were explored to pinpoint the optimal diagnostic structure. The outcomes of our investigation underline the impressive capabilities of the ANN system, successfully detecting all six considered boiler trips either before or concurrently with the detection by the plant's control system. Furthermore, the hybrid system exhibited a marginal improvement of 0.1% in Root Mean Square error compared to the pure ANN system. These findings collectively emphasize the potential of AI-driven methods in enhancing early detection and prevention of boiler trips, thereby contributing to improved operational safety and efficiency.