Optimization of Waste-to-Resource Plants with ML Prediction- Undertook a comprehensive study of waste-to-resource plant processes for operational data cleaning, identifying key process and quality variables by feature selection and consult chemical engineers with domain, thereby enhancing overall operational understanding.- Examined and preprocessed over a million data points using tools like NumPy, Pandas, and Spark ETL jobs for data imputation, in which I used a special KNN imputation method, reducing data anomalies by 15% and elevating the data quality for subsequent analysis.- Employed advanced machine learning algorithms like Random Forest Regression, RNN, LSTM, and My own Dynamic methods. Evaluation matrices such as MSE, R2, RMSE, and fitted plot were used to enable the product to reach its target goal with an error rate below 5%.Anomaly Detection Strategy for Imbalance Data- Designed a robust process monitoring and diagnosis framework for the imbalance data in Membrane Bioreactor process. - Developed a special classification method for imbalance data. First, two statistical indices were designed to classify normal and abnormal samples where the limits of indices were calculated by normal samples. Then Random Forest Classifier was used to classify abnormal samples.- Evaluated the results by Confusion Matrix, Recall, Precision, F1 score and Accuracy.- Enhanced process oversight and diagnostic speed by 30%.Anomaly Detection by Using ViT- Designed a robust fault diagnosis framework for the Trans - Membrane Pressure (TMP) in Membrane Bioreactor process by using ViT. Multi-Objective Reinforcement Learning- Devised a Reinforcement Learning model for the membrane bioreactor plant, creating an automated operation plan that generated Biogas with the target composition ratio. This innovation resulted in a 5% increase in operational efficiency and optimized decision-making processes.

Communicated clearly to students to assist them in understanding course knowledge and provided guidance on assignments and projects.