基于遗传优化的最小二乘支持向量机风电场风速短期预测

摘要

风电场短期风速的准确预测能为风电并网运行的规划、调度、运行和控制提供及时有效的信息.支持向量机基于结构风险最小化原理,从整体上考虑曲线的平滑度对数据进行拟合,对风速预测时能及时跟踪其变化趋势.针对支持向量参数难以确定问题,采用遗传算法对最小二乘支持向量机惩罚系数C和核参数σ<'2>寻优,在对参数遗传编码时,通过对数变换编码提高了搜索灵敏度,加快了模型收敛速度.最终利用现场连续150 h实测风速样本,对其中最后12 h进行预测,结果与广义回归神经网络(GRNN)相比,表明LS-SVM有更好的泛化能力,且取得了相对误差绝对值的平均值为8.32%的良好效果.%Timely and effective information can be obtained and then applied to the planning, scheduling, operation and control of wind power system, provided that the short-term wind speed can be accurately forecasted in wind farms.Support vector machine algorithm is established based on structural risk minimization principles.It considers smoothness of the regression curve entirety on the whole in regression model and predicts wind speed and tracks trend in time.To sovle the problem that the parameters of SVM are difficult to determine, genetic algorithm is employed to optimize the penalty factor C and kernel parameter σ2 of support vector machines.In the genetic coding of the parameters, the search sensitivity is improved and the model convergence speed is accelerated through the logarithmic transformation.Finally, prediction of the last 12-hour samples of 150-hour wind speed samples is done, and compared with the general regression neural network (GRNN) method, LS-SVM achieves better generalization ability and its average absolute value of relative error is only 8.32％.