- 學位論文
應用粒子群算法優化於量子和殘差長短期記憶混合模型的日前時空風速預測
Day-Ahead Spatiotemporal Wind Speed Forecasting Based on a Hybrid Model of Quantum and Residual Long Short-Term Memory Optimized by Particle Swarm Algorithm
摘要
風速的間歇性行為給風能融入電網帶來了重大困難，因為它可能會導致風力渦輪機輸出功率的突然變化，進而使系統不穩定，甚至可能導致停電。對於大型風能計畫來說，這一點尤其重要，因為風能發電的不穩定和不可預測性可能會對電網的可靠性產生顯著影響。本研究提出了一種新的混合模型，用於日前時空風速預報，該模型利用殘差長短期記憶（LSTM）和量子神經網絡（QNN），並通過粒子群優化（PSO）進行改進。利用PSO調節殘差LSTM的結構參數以及其超參數（時間序列、時間延遲、隨機放棄率和學習率）。將基於QAOA的量子嵌入層添加到優化後的殘差LSTM神經網絡中，以提高所提出模型的準確性。測試結果表明，所提出的方法優於許多機器學習技術和深度學習算法，並證明其準確性。
並列摘要
The intermittent behavior of wind speed raises substantial difficulties for the integration of wind energy into the electrical grid since it could lead to abrupt variations in the power output of wind turbines, which may cause the system to become unstable and perhaps result in power outages. This is especially important for large-scale wind energy projects since the erratic and unpredictable nature of wind energy generation can significantly affect the reliability of the grid. This study presents a novel hybrid model for day-ahead spatiotemporal wind speed forecasting that utilizes quantum neural network (QNN) with residual long short-term memory (LSTM) and is improved by particle swarm optimization (PSO). The residual LSTM's structural parameter as well as its hyperparameters (time series, time lag, dropout rate, and learning rate) are tuned employing PSO. A QAOA-based quantum embedding layer is added to the optimized residual-LSTM neural network to increase the proposed model's accuracy. The test results show that the proposed approach outperforms many machine learning techniques and deep learning algorithms and proves to be accurate.