適應不同氣候情境之流域多水庫多目標發電決策最佳化系統

Abstract

本文旨在開發一套流域多水庫多目標發電決策最佳化系統，以應對極端氣候下的挑戰。針對台灣大甲溪流域，該系統整合模糊類神經網路預測模型，來進行最上游德基水庫的入流量預測，並提出預測區間方法來量化預測的不確定性。藉由建立流域水庫間的逕流模型，來模擬水庫操作的耦合關係。此外，利用模糊邏輯與多準則決策方法，來動態調整決策權重，以及提出評估發電效率與排洪風險的指標，來設計多目標最佳化成本函數。最後，採用二階段逐小時多水庫最佳化架構，整合預測資訊與專家意見，來提升整體的水力發電調度效率、降低排洪風險，並在不同天氣事件中維持著不同的發電決策。

This thesis develops an optimisation system for multi-reservoir, multi-objective power generation decision-making in river basins to address challenges posed by extreme weather. For Taiwan's Dajia river basin, the system integrates upstream inflow forecasting using a fuzzy neural network with meteorological and historical data, quantifies forecast uncertainty via prediction intervals, and models inter-reservoir runoff relationships to simulate operational coupling. In addition, the system uses fuzzy logic and Multi-Criteria Decision Making (MCDM) to dynamically adjust the decision weights, and proposes metrics for evaluating generation efficiency and flood risk to design the multi-objective optimized cost function. Finally, through a two-stage hourly multi-reservoir optimisation framework, the system integrates forecasts and expert opinions to enhance hydropower dispatch, reduce flood risk, and maintain generation during extreme weather.