﻿ 电力系统短期负荷的WGBR预测方法研究 Wavelet Gradient Boosting Regression Method Study in Short-Term Load Forecasting

Smart Grid
Vol.05 No.04(2015), Article ID:15915,8 pages
10.12677/SG.2015.54023

Hong Wang, Gendai Gu

College of Mathematics and Science, North China Electric Power University, Baoding Hebei

Received: Jul. 30th, 2015; accepted: Aug. 17th, 2015; published: Aug. 20th, 2015

ABSTRACT

The authors proposed gradient boosting regression method based on wavelet transform considering the influence of weather factors and the characteristics of the load and meteorological data. The load and meteorological data were decomposed into several subsequences in different band by wavelet transform respectively. Forecasting the load subsequence by building different gradient boosting regression model, lastly, the final forecasting result is attained via adding all child-load-serials forecasting results. It has been showed by load data of a city in north China that the method achieved good prediction accuracy.

1. 引言

GBR算法是Friedman提出的将Boosting算法扩展以解决回归问题的预测算法，在多个领域有着广泛应用，如生物研究、搜索排名和机器学习等。它对数据具有很强的鲁棒性，能够很好地处理不干净的、含噪声的数据，支持不同的损失函数，并对非线性数据有很强的预测能力。国内外研究人员将GBR算法应用于太阳能领域[2] 的预测、汽车保险损失成本预测[3] 、碳钢的土壤腐蚀规律研究[4] 等。但在电力系统负荷预测领域应用还很少。

2. GBR算法

GBR本质上是一种利用个基函数的加法展开式对目标函数进行逼近的方法.表达式[3] 如下所示：

(1)

(2)

(3)

(4)

(5)

2) for to do

3) 计算负梯度

(6)

4) 通过输入向量及负梯度拟合一个回归树模型，即可得到

5) 通过最小化

(7)

6) 更新模型

(8)

7) end for

3. WGBR预测模型

3.1. 预测模型

(9)

3.2. GBR的实现环境及参数说明

Figure 1. Short-term load forecasting model

Table 1. List of input features for different series

4. 实例结果分析

(10)

(11)

Table 2. Parameter values of WGBR and GBR models

Table 3. Statistics of average relative error for a week

Table 4. Load forecasting result on the Nineteenth day by the WGBR and the GBR methods

Figure 3. Comparison of forecasting load errors for a week

1.5%，取得了较好的预测精度。表4是8月19日(周四)各时的预测值与真实负荷值。图4是周一各时的预测值与真实值的曲线图，能够较清晰地看出WGBR的预测效果。

5. 结论

Wavelet Gradient Boosting Regression Method Study in Short-Term Load Forecasting[J]. 智能电网, 2015, 05(04): 189-196. http://dx.doi.org/10.12677/SG.2015.54023

1. 1. 葛少云, 贾鸥莎, 刘洪 (2012) 基于遗传灰色神经网络模型的实时电价条件下短期电力负荷预测. 电网技术, 1, 224-229.

2. 2. Prakash, A. and Singh, S.K. (2014) Towards an efficient regression model for solar energy prediction. 2014 Innovative Applications of Computational Intelligence on Power Energy and Controls with Their Impact on Hu-manity (CIPECH), 18-23. http://dx.doi.org/10.1109/CIPECH.2014.7019040

3. 3. Guelman, L. (2012) Gradient boosting trees for auto insurance loss cost modeling and prediction. Expert Systems with Applications, 39, 3659-3667. http://dx.doi.org/10.1016/j.eswa.2011.09.058

4. 4. 鲁庆, 穆志纯 (2014) 应用提升回归树研究碳钢的土壤腐蚀规律. 中南大学学报(自然科学版), 6, 1879-1886.

5. 5. 高云龙, 潘金艳, 吉国力, 高峰 (2011) 基于Boosting梯度下降理论的时间序列建模方法. 中国科学: 技术科学, 7, 929-943.

6. 6. 乐斌 (2004) Boosting算法研究及其在光谱分析中的应用. 硕士论文, 浙江大学, 杭州.

7. 7. 李航 (2012) 统计学习方法. 清华大学出版社, 北京.

8. 8. 吴喜之, 马景义, 吕晓玲, 闫洁 (2009) 数据挖掘前沿问题. 中国统计出版社, 北京.

9. 9. Friedman, J.H. (2001) Greedy function approximation: A gradient boosting machine. The Annals of Statistics, 5, 1189- 1232. http://dx.doi.org/10.1214/aos/1013203451

10. 10. 徐科, 徐金梧, 班晓娟 (2001) 基于小波分解的某些非平稳时间序列预测方法. 电子学报, 29, 566-569.

11. 11. Hastie, T., Tibshirani, R. and Friedman, J. (2009) Elements of statis-tical learning: Data mining inference, and prediction. Springer-Verlag New York Inc., New York. http://dx.doi.org/10.1007/978-0-387-84858-7

12. 12. 王兴玲, 李占斌 (2005) 基于网格搜索的支持向量机核函数参数的确定. 中国海洋大学学报(自然科学版), 35, 859-862.