基于支持向量回归的电力系统短期负荷预测

Power System Short- Term Load Forecasting Based on Support Vector Regression

短期负荷预测对电力系统的可靠和经济运行意义重大。电力负荷是一个随机非平稳过程,受自然、社会等诸多因素的影响,用传统支持向量回归(SVR)对其进行准确预测的难度较大。为提升预测精度,在预测中引入区间二型模糊C均值聚类算法对样本进行聚类处理,有效提高了样本类内规律性,达到提升预测精度的效果。同时,考虑传统SVR运算速度慢,采用最小二乘支持向量回归,通过将二次规划问题转换为最小二乘问题,极大地减小了算法复杂度。最后,利用电网真实负荷数据进行了算法仿真比较。仿真结果表明,引入模糊聚类的最小二乘SVR算法预测精度达到了要求,验证了改进算法在短期负荷预测中的有效性和实用性。

The short-term load forecasting plays an important role in power system for its reliable and economic operation. The power load is a random non-stationary process and affected by many factors such as natural and social factors. As a result, it is difficult to forecast the power load using traditional support vector regression （SVR）. To en- hance the prediction accuracy, a Least-Square support vector regression based on interval type-2 FCM algorithm was proposed to improve the forecasting precision. Firstly, the input samples were clustered to improve the regularity with- in clusters. Besides, the Least-Square support vector regression was employed to reduce the computational complexity in traditional support vector regression. Real power load data were utilized for the comparison of different algorithms. The simulation results verify the effectiveness and practicality of our proposed algorithm.