Genetic algorithm-optimized state space neural operator for power plant data platforms

Modern power plant data platforms need to manage large-scale operational data, which are heterogeneous and high-speed and are not stationary. High-computational models like Recurrent and Attention-based often have high computational cost, sensitivity to noise and load variations, and, in the attention-based case, catastrophic forgetting. To overcome these difficulties, this paper introduces a novel framework of GA-SSNOOP (Genetic Algorithm–Optimized State Space Neural Operator) to efficiently model a power plant system in the time domain. The framework combines continuous-time state-space neural operators and multi-objective Genetic Algorithm optimization to enhance the prediction accuracy, stability, and computational efficiency. Approximately 50,000 data points with 15–20 sensor variables were tested under a variety of operating conditions for both the Steam Turbine and the Boiler datasets. GA-SSNOOP unites the advantages of state space learning with the hyperparameter optimization of evolution, enabling it to capture the complex dynamics of the system and automatically determine the optimal model configuration. Results show a better forecasting performance than the baseline models of LSTM and Transformer, with the models converging faster, making lower prediction errors and consuming less memory. These results indicate the potential of GA-SSNOOP to scalable, real-time industrial analytics, intelligent monitoring and decision support in modern power plant environments.