基于Blending集成学习模型的电力市场日前出清电价预测

Day ahead clearing price forecasting in power market based on blending ensemble-learning model

中文关键词: 电力市场  日前电价预测  多时间尺度  Blending集成学习  Asinh变换

英文关键词: electricity market  day ahead electricity price forecast  multi time scale  Blending integrated learning  Asinh transform

中文摘要:

      精准的掌握未来电价信息对把握市场的运行状态、支撑市场参与各方进行有效决策、推动市场主体合理优化资源配置具有重要意义。因此基于Blending集成学习机制构建了一种面向于日前电价预测的综合模型。该模型充分考虑电价的高波动性特点，采用Ashin变换，减小了输入数据波动性对预测模型的影响；选取SVM、LightGBM、EWNN、SARIMAX 4种较为成熟的单一电价预测模型作为初级学习器，进而保证了基于Blending集成学习模型的预测精度。选用美国PJM电力市场实际运行数据对上述构建的电价预测模型进行验证，通过对预测结果的对比分析，表明构建的基于Blending 集成学习机制的电价综合预测模型集成了多种传统预测模型的优点，具有较好的准确性与稳定性。

英文摘要:

      Accurate grasp of future electricity price information is of great significance to obtain the operation state of the market, support all parties involved in the market to make effective decisions, and promote market players to reasonably optimize resource allocation. Hence, a comprehensive model is constructed for day ahead price forecasting based on blending integrated learning mechanism. The model fully considers the high volatility of electricity price and adopts Ashin transform to reduce the impact of input data volatility on the prediction model. Four mature single electricity price forecasting models, SVM, Lightgbm, EWNN and SARIMAX are selected as primary learners to ensure the forecasting accuracy based on blending integrated learning model. The actual operation data of American PJM power market is selected to verify the above electricity price prediction model. The comparative analysis of the prediction results shows that the electricity price comprehensive prediction model based on blending integrated learning mechanism integrates the advantages of a variety of traditional prediction models and has good accuracy and stability.