基于Markov链的园区随机功率多场景预测模型

Prediction model of stochastic power in industrial parks based on Markov chain

为了促进园区分布式能源的消纳,针对园区随机功率的预测问题提出了一种基于马尔科夫(Markov)链的随机功率多场景预测模型。首先,针对园区新能源及负荷的随机特征,分别采用差分自回归移动平均(ARIMA)模型及Markov链对其建模;其次,针对园区负荷随生产、季节等因素周期性波动的特点,采用后验信息自适应调整Markov概率矩阵以提高其预测精度;然后,为了提高预测时域内多步预测的精度,考虑多步预测场景及其概率提出了一种基于场景树的多场景预测模型,以便更有效地利用Markov概率矩阵;最后,由园区历史功率数据进行了算例分析。结果表明,相比未调整的Markov模型,当自适应调整时间为7 d时,负荷功率的预测误差最低,为0.0345(标幺值)。相比于常用的极大似然估计法,所提多场景预测模型误差的加权平均值更低。

To promote the consumption of distributed energy in industrial parks,a multi-scenario prediction model based on Markov chain is proposed for stochastic power prediction.Firstly,ARIMA model and Markov chain are respectively used to build the load model of a park according to the random characteristics of renewable energy and the load.Then,posterior information is used to adjust the Markov probability matrix adaptively to improve its prediction accuracy which is affected by the periodical fluctuation of load varying with production,season or other factors.To improve the accuracy of multi-step prediction in a prediction horizon,a multi-scenario prediction model based on scenario tree is proposed,considering the multi-step prediction scenarios and their probabilities.The model can make more effective use of Markov probability matrix.Finally,a case study is implemented based on historical power data of an industrial park.The results show that compared with the unadjusted Markov model,the adaptive model is of a lower load prediction error,and the prediction error of the one with 7-day adjustment time is as low as 0.0345 p.u.Compared with the commonly used Maximum Likelihood Estimation,the proposed multi-scenario prediction model is of a lower weighted average error.