基于可信度加权组合的电价波动模式日前预报方法

Day-ahead Prediction Approach of Electricity Price Fluctuation Pattern based on Credibility Weighted Combination

准确的日前电价预测对电力市场参与者的优化决策具有重要意义。目前,大多数日前电价预测方法并不区分每天电价的波动模式而采用统一模型进行预测,当被预测日的波动模式与历史数据出现较大差异时无法保证预测的准确性。根据不同的日波动模式采用相似历史数据进行分类建模是解决此问题的有效途径,这就需要建立针对历史数据不同波动模式的分类识别模型和针对未来波动模式的日前预报模型。为此,文章提出一种针对分类预测的电价日波动模式日前加权组合预报方法。第一,采用K-means算法对日电价序列进行聚类分析,在分析聚类结果特性的基础上提取反映每日波动模式差异的特征向量,利用支持向量机分类(support vector machine for classification,SVC)方法建立电价数据日波动模式的识别模型;第二,利用多种常规方法建立日前电价预测模型对日前电价进行预测,并将预测结果输入日波动模式识别模型得到对应的模式识别结果;第三,根据多个方法波动模式预测结果对历史数据表现出来的不同精度,设计了基于可信度的组合机制,实现考虑预测准确性的加权组合预测,从而得到最终的日波动模式预测结果。利用美国PJM电力市场电价数据进行的仿真分析表明,提出的日前电价波动模式预测方法能得到较为准确的模式预测结果;利用电价波动模式日前预报进行分类预测的精度相对统一预测有显著提高。

Accurate forecasting of day-ahead electricity price is of great significance to the decision-making optimization for electricity market participants. At present,most day-ahead electricity price forecasting methods do not distinguish daily fluctuations of electricity prices and adopt unified models. When the forecasted day＇s fluctuation pattern is different from historical data,the forecasting accuracy cannot be guaranteed. Based on different fluctuation patterns,classification modeling that uses similar historical data is an effective way to solve this problem. As a result,it is necessary to establish a classification recognition model for different patterns of historical data and a prediction model for future fluctuations. For this reason,this paper proposes a day-ahead weighted combination forecasting method for daily fluctuations of electricity price that serves for classification forecasting. Firstly,the K-means algorithm is used to cluster the daily electricity price series. On the basis of clustering analysis results,the feature vector reflecting the difference of daily fluctuation patterns is extracted.The support vector machine for classification（ SVC） method is used to establish a recognition model for daily fluctuation patterns of electricity price data. Secondly,multiple conventional methods are applied to establish day-ahead electricity price forecasting models,and the forecasting results are input into the daily fluctuation pattern recognition model to obtain corresponding pattern recognition results. On the basis of the different pattern forecasting precisions of historical results from multiple models,a credibility based combination mechanism is designed to realize weighted combination forecasting considering the prediction accuracy,and the final forecasting results of the daily fluctuation model is obtained. Simulation analysis using the electricity price data of the PJM electricity market in the United States shows that the proposed forecasting method for future electricity price fluctuations can obtain relatively accurate forecasting results; and compared with unified forecasting the accuracy using this method for classification forecasting of electricity price is improved.