結合徑向基底函數網路與粒子群演算法於小型風機之最大功率追蹤

Maximum Power Point Tracking (MPPT) of Small Wind Power Generators Based on Radial Basis Function Neural Network (RBFNN) and Particle Swarm Optimization (PSO)

姚家翔(Chia-Shiang Yao)

指導教授 ： 李俊耀

中原大學/工學院/電機工程研究所/碩士(2012年)

https://doi.org/10.6840/cycu201200069

摘要

由於風力發電機具非線性特性，且最大功率點易隨風速負載變化改變，故本文旨於在不同風速及負載變化狀況下，利用本研究所提出之最大功率追蹤進行風機控制。
首先，本文提出兩個徑向基底函數網路模型，分別為估測風速類神經網路及估測功率類神經網路，並以粒子群最佳化方法獲得最佳神經元分佈。其中，估測風速網路透過發電機轉速及輸出功率得到估測風速；估測功率網路根據估測風速及負載，建立控制電壓與輸出功率之對應關係。而本研究將使用此兩網路模型結合粒子群演算法，輸出最佳控制電壓，進行最大功率追蹤控制。
其次，為評估本文所提方法在人工風場實驗環境下之各項效能，本研究實際量測一小型風力發電機在不同風速負載下之發電機轉速及輸出功率等各項數據。並經由本研究之控制架構，使風力發電機運轉於最佳轉速，以達到最大功率輸出目的。
最後，分別在固定風速、弦波風速、急遽變化風速、隨機變化風速、負載變化及風速負載皆變動等六種不同情況下，進行最大功率追蹤控制，並與傳統擾動觀察法進行比較。實驗結果顯示，在風速及負載變動下，本文提出之控制架構皆能在短時間內使風機運轉於最大功率輸出點，且降低控制系統造成的損耗。

關鍵字

最大功率點 ； 徑向基底函數網路 ； 粒子群演算法 ； 最大功率追蹤

並列摘要

This thesis proposes a maximum power point tracking (MPPT) to control wind generators with unsteady wind speed and load, because wind generators are nonlinear and the maximum power point would change under such circumstances.
First, two radial basis function neural network models are proposed, one for estimating wind power and the other for estimating power, and then particle swarm optimization is used to obtain their best neuron distribution. In that, the network for estimating wind speed operates with inputting the rotational rate and the output power of the wind generator; the network for estimating power forms the correspondence between the controlled votage and output power according to the estimated wind speed and load. This thesis will employ these two models combined with particle swarm optimization to output the best controlled votage to proceed the MPP tracking.
Second, to evaluate the performance of MPPT when being experimented under a artificial wind farm, this thesis measures signal data (the rotational rate, the output power, etc) of a small-scale wind generator with different wind speed and load. By the MPPT proposed in this thesis, the wind generator can operate at the optimal rotational speed to output the maximum power.
Finally, the MPPT is experimented under six conditions, including with fixed, sine-waved, dramatically-changed, or randomly-changed wind speed, or variable load, or both being unstable, and then compared with traditional perturb & observe (P&O) method. The experiment results indicate that with variable wind speed and load, the MPPT in this thesis can shortly make the wind generator operate at the MPP and reduce the loss of the tracking system.

並列關鍵字

maximum power point (MPP) ； maximum power point tracking (MPPT) ； radial basis function neural network (RBFNN) ； particle swarm optimization (PSO)

參考文獻

[15] 黃意明，風力發電之類神經網路最大功率追蹤，中原大學電機工程所，碩士論文，民國95年。

[21] 于俊傑，風力發電最大功率追蹤技術之研究，中原大學電機工程所，碩士論文，民國93年。

[4] A. Miller, E. Muljad, and D. S. Zinger “A variable speed wind turbine power control,” IEEE Transactions on Energy Conversion, vol. 12, no. 2, pp. 181-186, 1997.

[5] J. H. R. Enslin, ”Maximum power point tracking: a cost saving necessity in solar energy systems,” Conference of IEEE IECON, vol. 2, no.1, pp.1073-1077, 1990.

[6] K. H. Hussein, I. Muta, T. Hoshino, and M. Osakada, “Maximum photovoltaic power tracking: an algorithm for rapidly changing atmospheric condition,” IEE Proceedings-Generation Transmission Distribution, vol. 142, no. 1, pp. 59-64, 1995.