作者：董利江1, 李伟2, 王瑜3, 钱白云2, 沈中信1, 马勤勇2, 孔筱叶4
作者单位：1. 新疆电力建设调试所, 新疆 乌鲁木齐 830011;
2. 国网新疆电力公司电力科学研究院, 新疆 乌鲁木齐 830000;
3. 华北电力大学电子工程系, 河北 保定 071003;
4. 四川中测辐射科技有限公司, 四川 成都 610000
关键词：参数优化;风速预测;相关向量机;遗传算法;空间重构
摘要：
该文利用混沌理论中的相空间重构方法，对基于相关向量机的风速预测模型的训练样本进行构建，然而通过混沌理论求出的相空间参数(嵌入维数E和时间延迟)往往不是预测模型的最优解。针对预测模型超参数优化问题，提出一种基于遗传算法的多参数优化方法，即对E、以及相关向量机核参数进行同步优化。该方法首先基于遗传算法搜索相关向量机预测模型参数(E、、)的全局最优解，进而建立预测模型；然后对待预测风速时间序列进行预测；最后以2组实际风速数据为例进行实验研究，并与对比模型方法（只优化参数）进行对比。结果表明：该文模型不仅具有较低的预测误差，而且可提高预测效率，缩短预测时间。
Multi-parameter synchronous optimization of genetic algorithm for RVM short-term wind speed prediction model
DONG Lijiang1, LI Wei2, WANG Yu3, QIAN Baiyun2, SHEN Zhongxin1, MA Qinyong2, KONG Xiaoye4
1. Xinjiang Electric Power Construction Commission, Urumqi 830011, China;
2. Electric Power Research Institute of Xinjiang Electric Power Company, State Grid Corporation of China, Urumqi 830000, China;
3. Department of Electronic Engineering, North China Electric Power University, Baoding 071003, China;
4. Sichuan Radiation Technology Co., Ltd., Chengdu 610000, China
Abstract: Phase space reconstruction method in chaos theory is applied to construct the training sample of wind speed prediction model of relevance vector machine. However, phase space parameters (embedding dimension E and time delay τ) calculated by chaos theory are usually not the optimal solution of prediction model. As for hyperparameter optimization problems of prediction model, a multi-parameter optimization method based on genetic algorithm is proposed to optimize E, τ and σ (kernel function parameter) synchronously. Firstly, this method searches the global optimal solution of relevance vector machine prediction model parameters (E, τ,σ) based on genetic algorithm, then establishes the prediction model. Secondly, it treats the prediction of the time series of the predicted wind speed. Finally, two groups of actual wind speed data are adopted for wind speed time series and compared with contrast model method (optimize parameter σ only). The results show that the model does not only have higher prediction accuracy, but also can improve prediction efficiency and shorten prediction time.
Keywords: parameters optimization; wind speed prediction; RVM; GA; space reconstruction
2018, 44(7): 13-18 收稿日期: 2017-10-19;收到修改稿日期: 2017-11-13
基金项目: 国家电网公司科技项目资助（TSHT-FW-2017-20）
作者简介: 董利江(1986-),男,新疆乌鲁木齐市人,工程师,硕士,研究方向为汽轮机性能测试。
参考文献
[1] 丁华杰, 宋永华, 胡泽春, 等. 基于风电场功率特性的日前风电预测误差概率分布研究[J]. 中国电机工程学报, 2013, 33(34):136-144.
[2] 陈玲, 赖旭, 刘霄, 等. WRF模式在风电场风速预测中的应用[J]. 武汉大学学报(工学版), 2012, 45(1):103-106.
[3] TIPPING M E. Sparse bayesian learning and the relevance vector machine[J]. Journal of Machine Learning Research, 2001(1):211-244.
[4] SUN G Q, CHEN Y, WEI Z N, et al. Day-ahead wind speed forecasting using relevance vector machine[J]. Journal of Applied Mathematics, 2014(6):1-6.
[5] BAO Y, WANG H, WANG B N. Short-term wind power prediction using differential EMD and relevance vector machine[J]. Neural Computing and Applications, 2014, 25(2):283-289.
[6] TAKENS F. Detecting strange attractors in turbulence[J]. Lecture notes in Math, 1981, 898:366-381
[7] 韩中合, 徐伟轩. 基于相空间重构的支持向量机的风电场风速预测[J]. 可再生能源, 2012, 30(6):31-34.
[8] 王富强, 王东风, 韩璞. 基于混沌相空间重构与支持向量机的风速预测[J]. 太阳能学报, 2012, 33(8):1321-1326.
[9] 孙国强, 卫志农, 翟玮星. 基于RVM与ARMA误差校正的短期风速预测[J]. 电工技术学报, 2012, 27(8):187-193.
[10] 白裔峰, 肖建, 于龙, 等. 基于结构风险最小化的加权偏最小二乘法[J]. 计算机应用, 2007, 27(4):939-941.
[11] MACKAY J C. Bayesian interpolation[J]. Neural Computation, 1992, 4(3):415-447.
[12] 叶林, 陈政, 赵永宁, 等. 基于遗传算法-模糊径向基神经网络的光伏发电功率预测模型[J]. 电力系统自动化, 2015, 39(16):16-22.
[13] SELAKO A, CVIJETINOVIC D, MILOVIC L, et al. Hybrid PSO-SVM method for short-term load forecasting during periods with significant temperature variations in city of Burbank[J]. Applied Soft Computing, 2014, 16(3):80-88.
[14] 朱霄珣, 韩中合. 基于PSO参数优化的LS-SVM风速预测方法研究[J]. 中国电机工程学报, 2016, 36(23):6337-6342.
[15] 严欢, 卢继平, 覃俏云, 等. 基于多属性决策和支持向量机的风电功率非线性组合预测[J]. 电力系统自动化, 2013, 37(10):29-34.
[16] 牛东晓, 魏亚楠. 基于FHNN相似日聚类自适应权重的短期电力负荷组合预测[J]. 电力系统自动化, 2013, 37(3):54-57.
