混合平行基因演算法與支持向量機作短期負載預測

Hybrid Parallel Genetic Algorithm and Support Vector Machines for Short Term Load Forecasting

姜大駿(Da-Jun Jiang)

指導教授 ： 曹大鵬

國立臺北科技大學/電資學院/電機工程系研究所/碩士(2014年)

https://doi.org/10.6841/NTUT.2014.00148

摘要

精確的負載預測不僅可以適當的做為發電機排程規劃及調度之參考依據，也能降低機組運轉時的成本及改進供電品質可靠度，最重要的是能避免限電危機或是電力資源浪費。
本論文以平行基因演算法(Genetic Algorithm, GA)結合支持向量機(Support Vector Machines, SVM)中的最小平方支持向量機(Least Squares Support Vector Machines, LSSVM)來求解電力負載預測之問題，本論文乃以最小平方支持向量機為主體，接著使用平行基因演算法來求得最小平方支持向量建立負載預測之模型的各項最佳參數值，且加以改進傳統用於求解建立負載預測模型之最佳參數值方法。
在求解過程中，基因演算法是架構簡單且結果準確率較高的最佳化演算法，因為其編碼方式是採用格雷碼編碼使建立負載預測之模型參數能更加精確，接著經由複製、交配、突變則可以有效提高整體最佳解的搜索能力，而其演算法的缺點為資料龐大時運算時間較久，因此，本研究以基因演算法為基礎，結合平行處裡技術來輔助基因演算法處裡資料龐大而運算時間過長之問題。
最後，本論文依據各項數據資料(電力負載、氣象)為基準，作為實際之負載預測，且將本論文結果與量子基因演算法結合倒傳遞類神經網路(Quantum Genetic Algorithm and Artificial Neural Network,QGA-BPNN)做比較，本方法不僅求得小的負載預測誤差值，且在大部分預測中，因為使用平行處裡技術，使得運算時間也較少，因此，本論文提出較優質的可行方法。

關鍵字

基因演算法 ； 支持向量機 ； 平行化技術 ； 負載預測

並列摘要

The accurate load forecasting can not only be properly considered as the reference of unit commitment and arranging for generation but also lower the operational cost and improve the reliability of supply quality. The most important is that it can avoid the crisis of limit-electricity and the waste of power resource.
In this thesis, the Genetic algorithm (GA）is combined with the Least squares support vector mechine（LSSVM）of Support vector mechine（SVM）to solve the problem of power load forecasting. With the main foundation of LSSVM, parallel GA is adopted to obtain the necessary parameters in the model built by LSSVM for load forecasting,which modifies the original optimized parameters contributed by the conventional model (i.e.LSSVM,BPNN)for predicting load.
During the period of solving, the GA is an optimized algorithm with simple structures and precise results. Thanks to the Grey code introducing, the accuracy of parameters can be promised in the model for load forecasting. Then, through the duplicatation, matching and mutation, the searching ability can be significantly improved in the whole area for optimized solution. However, the disadvantage of this algorithm is that the long calculation time due to the large amount of data. Therefore, the technique of parallel processing is joined to assist the GA for dealing with a great number of data and long calculation time while the GA is taken as the basis in this research.
In the end, several data, including power load and weather parameters utilized in a proposed model, has been viewed as criteria. Compared with QGA-BPNN, the method proposed in this thesis illustrates less prediction error and calculation time than the traditionl existing techniques in most estimation because of parallel process. To sum up, this research indicates a practical and competitive method for predicting load.

並列關鍵字

Genetic Algorithm ； Support Vector Machine ； Parallel Technology ； Load Forecasting

參考文獻

[35] 羅一峰，應用累積元與基因演算法求解配電系統無效電力控制，碩士論文，中原大學，中壢市，2007。

[33] 劉律伸，應用量子基因演算法求解最佳化短期火力機組排程，碩士論文，國立台北科技大學，臺北，2012。

[29] 林大仁，運用支撐向量迴歸與倒傳遞類神經網路建構產品造型設計之預測模型，碩士論文，國立成功大學，台南市，2007

[37] 張淯詠，應用二進制粒子群演算法求解最佳化短期火力機組排程，碩士論文，國立臺北科技大學，臺北，2013。

[6] 邱士齊，混和量子基因演算法與類神經網路作短期負載預測，碩士論文，台北科技大學，臺北，2012