參加競賽名稱：2025 International Conference on Smart Devices and Sustainable Energy ( SDSE 2025 ) 論文競賽

獲得獎項：Honorable Mentions Award

題目：Improving Wastewater Treatment Capacity Forecasts Using PID-Corrected Transformer Model

同學姓名：王羽珊、戴佑恩

指導教授：劉弘一

作品/競賽簡介：
以Transformer 深度學習模型結合PID控制原理，改善污水處理廠出水預測的時間延遲與準確度。研究中以實際處理廠高頻 COD 資料進行特徵工程與時序建模，並利用PID 修正層動態效正 Transformer 的延遲預測行為。結果顯示 RMSE降低33.87%、R²提升至0.96，顯著提升模型穩定性與實務應用價值，並榮獲Honorable Mentions Award的肯定。
This study proposes a novel approach to enhancing predictive performance by incorporating a Proportional-Integral-Derivative (PID) controller-based error correction mechanism into a prediction model. Building upon a wastewater treatment capacity forecasting model developed using the Transformer architecture—renowned for its effectiveness in handling sequential data—the study addresses a common yet critical issue in autoregressive forecasting: the temporal lag between predicted and observed values. This lag, in which the predicted curve merely represents a shifted replica of the actual data, can significantly undermine the model’s practical utility. To mitigate this issue, we integrate a PID correction layer that dynamically adjusts the forecast outputs in real time. Experimental results demonstrate notable improvements, with the root mean squared error (RMSE) reduced from 3.1 to 2.05 and the coefficient of determination (R²) increased from 0.91 to 0.96. These findings underscore the potential of PID-based correction in strengthening AI-driven time series models and offer a promising direction for future advancements in error compensation mechanisms.