风电功率概率区间预测方法研究

Wind power forecasting is essential to the management of power system. However, traditional short wind power point forecasting can hardly be accurate due to the randomness of wind energy. Providing prediction intervals by giving a range of potential outcomes for a given probability is important to assess the potential uncertainties. In this project, a short probabilistic prediction interval method are studied. Firstly uncertainty and randomness of short wind power point forecasting are analyzed. Based on statistical distributing law of the error for wind power point forecasting, probabilistic prediction interval model with non-parameters bootstrap is proposed, which is classified as the indirect method. It generates a range of power under a given probability and provides benefit reference with uncertain information for the system of wind power point forecasting. Finally probabilistic prediction interval model classified as direct method is presented to directly obtain the uncertain interval. The model employs the kernel extreme learning machine (KELM) to establish probabilistic prediction interval model. It adopts the PSO algorithm with assessment constraint to optimize the output weights of KELM. The comparative analysis between the direct model and indirect model is also carried out for generalization characteristic. This project will provide more accurate theoretical basic of the power balance and power market transactions, which is of great theoretical significance for maximizing the utilization of wind energy.

风电功率预测能为电网的规划和运行提供重要依据，然而传统预测方法多为精确点预测，结果会有不同程度的误差，概率区间预测方法能有效描述风电输出功率的不确定性，因而逐步受到重视。项目探索短期风电功率概率区间预测方法。首先分析短期精确点功率预测的不确定误差原因，然后统计现有短期精确预测结果误差分布规律，提出非参数bootstrap的间接法风电概率区间预测模型，生成风电功率在不同置信水平下可能的波动区间，是对现有风电功率确定性预测精度的有益补充。最后提出直接法概率预测模型，通过采用在评价指标约束下的粒子群优化核极限学习机智能模型直接预测风电功率不确定性概率区间，并对比间接法和直接法预测模型的普适性。本项目为风电系统接入电网的功率平衡和调度、电力市场交易等提供更准确的信息，对于最大化利用风电具有重要和深远意义。

风电输出功率的间歇性和波动性，给电力系统运行和稳定性带来挑战，风电功率预测能为电网的规划和运行提供重要依据。传统预测方法多为精确点预测，会有不同程度的误差，项目研究短期风电功率概率区间预测方法，更有效描述风电输出功率的不确定性，给出不同置信概率下预测功率的波动范围，为合理、科学的调度风电提供更多信息。项目分析了点功率预测的不确定误差原因，揭示了风电功率预报的误差来源和规律。统计预测误差分布规律，提出了基于点预测误差统计规律分析的间接法风电功率概率区间预测模型。从数据序列的相关性和概率信息角度选择贝叶斯、马尔科夫等不同方法建立风电功率确定性点预测模型，从而提高点预测模型精度；然后采用参数法、非参数方法统计提取预测误差规律进而生成概率区间预测模型。项目提出了基于马尔科夫链、贝叶斯等方法的分位数回归、核密度估计、以及基于bootstrap方法的概率区间预测模型，获得风电功率在不同置信水平下可能的波动区间，是对现有风电功率确定性预测精度的有益补充。项目同时提出了直接法概率区间预测模型，对比分析了间接法和直接法预测模型的特点。直接法概率区间预测模型采用评价指标约束下的智能学习预测模型，直接预测风电功率不确定性概率区间，通过粒子群、人群算法等群智能算法优化模型参数。项目提出了结合分位数理论的核极限学习机粒子群优化方法概率区间预测模型、结合经验模态分解的核极限学习机人群优化方法概率区间预测模型、基于粗糙集约减的贝叶斯网络概率区间预测模型。将深度学习理论应用于功率概率区间预测，提出了基于VMD分解、相空间重构和PSO优化的卷积神经网络概率区间预测模型。本项目的研究可为风电系统接入电网的功率平衡和调度、电力市场交易等提供更准确的风电功率不确定信息，对于最大化利用风电具有重要价值。