由於風力渦輪發電機存在著非線性的特性,其最大功率點運轉位置將隨著風力狀況的改變而有所不同。為了使風力發電系統在任何風速下皆可操作在最大功率點,並避免實際運轉時使用風力計,本論文引用類神經網路最大功率追蹤控制理論應用於風力發電系統上。 本文先建立各不同風速下風力發電系統之負載特性曲線模型,便於設計控制程式與探討所提最大功率追蹤之可行性。除實作完成傳統電流型擾動觀察法、三點權位比較法、含風速計轉速控制法外,並與本文所使用類神經方法相互比較。系統實作上應用最大功率追蹤技術搭配數位信號處理器(TMS320C240)調整降壓式轉換器之責任週期,使風力渦輪發電機運轉在最大功率輸出。 為比較驗證上述四種控制方法,本論文以一部三相、12極、100瓦的風力渦輪發電機作為實際測試對象。風力渦輪發電機包括三片直徑1.17公尺葉片之小型風力渦輪機與一台永磁式同步發電機。從實驗結果得知,本論文所使用之類神經最大功率追蹤控制器,在不同的風速下,不需使用風速計亦可有效地追蹤至系統之最大功率點,且可解決舊有方法擾動的問題。
The wind-turbine generation system (WTGS) exhibits a nonlinear characteristic and thus its maximum power point varies with changing atmospheric conditions. In order to operate the WTGS at maximum power points under different wind speeds and to avoid using anemometer in practical applications, the thesis adopts neural network base maximum power points tracking (MPPT) control theory in the WTGS. In the thesis, load characteristic models of the WTGS under different wind speeds are first built up for design of control rules and feasibility studies of the proposed MPPT methods. We realize and compare the traditional current-type perturbation & observation (P&O) algorithm, three-point-weighting comparison (TPWC) algorithm, variable-speed wind turbine power control method, as well as the proposeed neural networks based MPPT method. In the practical system implementations, the MPPT methods are integrated with the TMS320C240 digital signal processor to adjust the duty ratios of the buck converter to operate the WTGS at maximum power outputs. To compare and verify the effectiveness of the four MPPT control methods mentioned above, a practical WTGS has been used. The WTGS includes a small wind turbine with three 1.17m-diameter blades and a three-phase, 12-pole, 100W, small permanent-magnet synchronous generator. The experimental results show that the neural networks based MPPT method can reach maximum power points in different wind-speed conditions without using anemometer, and it can solve the oscillation problems around the maximal power output point in traditional P&O and TPWC algorithms.