风光水互补系统时间序列变量概率预报框架

风光水互补系统实时调度受风速、太阳辐射强度、径流和电力负荷等时间序列变量的不确定性影响，如何准确预报这些变量并量化预报的不确定性是风光水互补系统面临的关键难题。为此，本研究提出一种基于深度学习模型的时间序列变量概率预报框架。首先，从时间序列数据中挖掘特征输入并采用相关系数对生成的特征进行初选；其次，基于深度学习模型和高斯过程回归构建时间序列变量概率预报模型，同时分别通过0-1规划思想和贝叶斯优化算法实现特征组合优化和超参数优选；进而，从确定性预报、概率预报和可靠性3个方面对预报模型进行全面评价；最后，以雅砻江流域风光水互补先期试点示范基地作为研究对象，分别在径流、风速、光伏和负荷4个数据集上与目前7个不同的时间序列变量预报模型进行全面对比，验证本研究提出预报框架的精度和概率综合性能。

A probabilistic forecasting framework of time series variables for wind-solar-hydropower hybrid systems

The real-time dispatch of wind-solar-hydropower (WSH) hybrid system is affected by the uncertainty of time-series variables such as wind speed,solar radiation intensity,runoff,and power load.How to accurately forecast these variables and quantify the uncertainty is the key problem faced by the WSH hybrid system.In order to solve the problem,a probabilistic forecasting framework for time series variables based on a deep learning model is proposed by this study.First,the feature input is mined from the time series data and the correlation coefficient is used to select the generated features.Then,based on deep learning model and Gaussian process regression,the time series variable probabilistic forecasting model is constructed,and the feature combination optimization and hyperparameter optimization are realized through the 0-1 planning idea and the Bayesian optimization algorithm respectively.The forecasting model is comprehensively evaluated from three aspects:deterministic forecasting,probabilistic forecasting and reliability.Finally,taking the WSH complementation pilot demonstration base in the Yalong River Basin as the research object,the framework proposed in this study is compared with the current seven state-of-the-art time-series variable forecasting models on the four datasets of runoff,wind speed,photovoltaics and power load,respectively,to verify the accuracy and probabilistic comprehensive performance.