分布式SVR在短期负荷预测中的研究

Research on Distributed SVR in Short-term Load Forecasting

摘要:

准确的负荷预测，可以合理安排机组启停，降低发电成本，特别是短期负荷预测对电力系统控制、运行和规划都有重要意义。传统的预测方法不能及时准确地反映需求响应，在Hadoop环境下利用分布式支持向量回归机(Support Vector Regression，SVR)实现负荷预测，同时使用基于均匀设计的自调用SVR(UD-SVR)方法进行参数寻优，进一步提高本文实现的分布式SVR算法精度。通过真实的电力负荷数据集验证该算法，实验数据来自我国西部某地级市连续424天的真实用电量数据。结果表明，本文改进后的算法用于短期电力负荷预测是可行的，不仅预测准确度又在原有基础上明显提高，并且随着数据量的增加，计算速度也大幅提高，减小了负荷预测时间。

Abstract:

Accurate load forecasting can reasonably arrange the start and stop of the unit and reduce the cost of power generation, especially short-term load forecasting is of great significance for power system control, operation and planning. However, the traditional forecasting methods can not reflect the demand of users timely and accurately. Load forecasting using Distributed Support Vector Regression (SVR) in Hadoop environment, at the same time, the self-call SVR (UD-SVR) method based on uniform design is used to optimize the parameters, and the accuracy of the distributed SVR algorithm implemented in this paper is further improved. The algorithm is validated by a real power load data set. The experimental data comes from real electricity consumption data for 424 consecutive days in a prefecture-level city in western China. The results show that the improved algorithm is feasible for short-term power load forecasting. Not only the prediction accuracy is improved on the original basis, but also the calculation speed is greatly improved with the increase of data volume, which reduces the load forecasting time.