月前日负荷曲线的概率预测和随机场景模拟

针对现有中长期日负荷曲线预测方法大多为点预测,难以满足电力系统不确定性分析的不足,提出了一种基于因子分析和神经网络分位数回归的月前日负荷曲线概率预测和随机场景模拟方法。采用因子分析技术,在保留日内负荷时序相关性的前提下,对日负荷序列向量降维;提取出少数相互独立的负荷公共因子作为预测变量,以日气象因素、星期类型和前一日公共因子值为输入特征,建立计及相邻日负荷相关性的神经网络分位数回归概率预测模型;以此为基础,利用中期气象预报信息,逐日预测和模拟未来30日的负荷曲线,并生成未来月负荷曲线的随机模拟场景。实际算例结果验证了所提概率预测方法的准确性和高效性,其生成的日负荷曲线模拟场景更好地体现了负荷的时序相关性,能为调度人员提供更准确、全面的月前负荷预测信息。

Probabilistic Forecasting and Stochastic Scenario Simulation of Month-ahead Daily Load Curve

The existing forecasting methods of medium and long term load curves are mostly point prediction methods, which can hardly make up for the inadequacy of the uncertainty analysis of power systems. For this reason, a probabilistic forecasting and stochastic scenario simulation approach based on factor analysis and quantile regression neural network(QRNN)is proposed. By means of the factor analysis method, the dimensionality of daily load time series vector is reduced without sacrificing correlation of intraday load time series. A few independent load common factors are extracted and used as predictor variables. With the daily weather factors, weekday type and the corresponding common factor of the previous day as inputs, the QRNN models of load common factors considering the load correlation between adjacent days are developed. Based on these probabilistic forecasting models, the load curves of next 30 days are predicted and simulated day by day, and stochastic scenarios of future month load curve are generated finally. The accuracy and high efficiency of the proposed approach have been verified by the results of an actual example. The correlation of load time series generated by the proposed approach is more accurate, being able to provide more accurate and all-round month-ahead load forecast information to system operators.