基于聚类和Bayesian推断的市场出清电价离散概率分布预测

电力市场中，市场出清电价(market clearing price, MCP)受到众多因素的共同作用，具有较强的随机性和不确定性，常规的MCP单值预测模型未充分利用历史数据反映的不确定性信息，预测结果无法体现MCP的随机变化特性，预测精度也有限。该文提出一种基于免疫遗传机制的聚类方法，用以实现历史数据输入输出映射关系的划分，并结合贝叶斯概率法则建立MCP离散概率分布的预测模型。对美国PJM市场数据的仿真结果显示，该文的预测模型能较好地反映MCP的不确定性特点，且具有较高的预测精度。

Forecasting of MCP 's Discrete Probability Distribution Based on Clustering and Bayesian Method

As being affected by too many factors, the energy market clearing price (MCP) in power markets is stochastic and volatile. Traditional MCP forecasting models usually make no good use of the uncertain information contained in the history data, the estimation can't reflect the volatility and the precision is limited. This paper proposes a linear clustering method based on the immune-genetic algorithm (IGA) and uses the method to divide the mapping relationship between the inputs and outputs in the history data. Besides, on the base of clustering, a forecasting model of MCP's discrete probability distribution by the Bayesian method is established. Simulation experiment on history data from American power market PJM shows the forecasting method proposed in this paper can perfectly reflect the stochastic volatility of MCP and has a good precision.