CO Emission Prediction Method of MSWI Process Based on FHDDM Sliding Window Drift Detection

Municipal solid waste incineration (MSWI) technology plays a pivotal role in addressing solid waste disposal challenges, particularly in densely populated urban areas. This paper introduces a method for predicting carbon monoxide (CO) emissions during the MSWI process utilizing the fast Hoeffding drift detection method (FHDDM) within a sliding window drift detection framework. The methodology involves the development of a long short-term memory (LSTM) neural network model and an FHDDM drift index calculation model based on historical data sets. Each online sample undergoes recursive standardization, and predictions are generated using the historical LSTM model. Subsequently, the prediction errors and historical drift indicators are analyzed to identify any detected drift. If no drift is identified, the historical model is employed for prediction purposes. However, in the presence of drift, the LSTM model is updated by integrating both historical and drift data. Real-time assessments and updates are conducted to enhance prediction accuracy. The efficacy of this approach is verified through actual industrial data simulations from an MSWI facility in Beijing, affirming its rationale and effectiveness.