- 學位論文
應用類神經網路進行考慮無效電力預測之電驛晶片設計
Application of Artificial Neural Networks for Relaying Chip Design Considering Reactive Power Forecasting
摘要
摘 要 電能是人類生活中重要的能源之一，更是國家重要發展的指標，而隨著國民所得的提升，生活品質的要求亦越來越受到重視，各類形式的電氣設備紛紛出爐，對電力能源的需求日益遽增。若對電力系統之負載可以有效預測，則可以避免設備因過載而發生設備毀損。如此，便可節省人力物力，並且有效降低營運成本。本論文將提出一使用類神經網路來預測未來虛功之方法，以提供未來無效電力控制之協調。 本文將使用多層前餽式網路的方法來預測無效電力，而輸入的部分考慮的為無效電力值(分為預測前之3小時、4小時以及5小時的輸入量)、日期以及時間，利用這些輸入資料配合類神經網路，來預測我們所需要的無效電力值。 本論文將使用台灣電力公司所提供之歷史資料，時間範圍為2006年2月20日至2006年2月24日為止，總共5 24筆資料，作為類神經網路訓練以及預測之資料。並使用Matlab以及Gene Hunter等模擬軟體，建構一個前饋式類神經網路，分別進行對無效電力訓練以及預測，並將其較好的測試結果之權重值取出，將其饋入DSP Builder，再透過DSP Builder的計算，對無效電力做一預測，並計算實際值與預測值的誤差量，最後可在場可程式閘陣列( Field Programmable Gate Array, FPGA )中實現。
並列摘要
Abstract The electric energy is one of the important energy in human life. It is also an important index for the country development. With the increase of the people income, the life quality is being also paid attention to. Due to new electric equipments, the demand for the electric energy increases gradually. If the load of the power system can be predicted effectively, it can avoid the equipment damage because of overload. Therefore, it can save the manpower and materials, and reduce the operating cost further effectively. This thesis used the feed-forward neural network to perform the short-term reactive power demand forecasting for future reactive power coordination. This thesis used the multi-layer feed-forward neural network to predict the reactive power. The inputs of the neural network consider for the reactive power demands (3 hours, 4 hours and 5 hours ahead to the present time), date and time, and the output of the neural network is the concerned reactive power demand at the next time. This thesis used the historical information that Taiwan Power Company provided. It starts February 20, 2006 till February 24, 2006, It includes 524 data sets. The data sets were trained and tested by neural networks. The commercial packages Matlab and Gene Hunter were used. The feed-forward neural network was adopted. The trained weighting factors in the neural network served as parameters for DSP Builder for constructing neural network for forecasting reactive power. The accuracy is verified by errors. Finally, the neural network is realized by Field Programmable Gate Array (FPGA).