【摘要】 电力负荷预测特别是实时电力负荷预测是电力系统规划的重要组成部分,也是电力系统可靠、经济运行的基础。针对回声状态神经网络在实时负荷预测中存在易受噪声影响、鲁棒性不强、不稳定的问题,提出了将基于模块化回声状态网络的方法应用于实时电力负荷预测中。根据输入时序数据所引起的储蓄池内部状态的相似性对储蓄池空间进行模块划分,将此高维空间划分为多个子模块,针对每一个模块训练一个读出器,最后把各个模块的输出结果集成输出。利用模块化回声状态网络模型,对大客户的实时负荷数据进行预测,并与几种短期负荷预测模型进行精度和稳定性的对比实验,结果表明,模块化回声状态网络在实时负荷预测中既提高了预测精度,又增强了预测的稳定性和泛化性能。更多还原

【Abstract】 Load forecasting, especially real-time load forecasting, is one of important constituent parts in power grid planning, and is also the foundation of reliable and economic operation of power grid. In allusion to such defects of echo state neural network as easy to be interfered by noise, weak robustness and instability while it is applied in real-time load forecasting, it is proposed to apply the modular echo state network(MESN) based method in real-time load forecasting. According to the similarity of internal state within the reservoir caused by the input time series data the modular division of the reservoir space is performed, that is, the high-dimensional space is divided into multi sub-modules and a readout is trained for each module, finally the output of all readouts are integrated. Using MESN model, the real-time load data of big consumers are forecasted and the accuracy and stability of the obtained forecasting results are compared with those of forecasting results by short-term load forecasting models, and comparison result shows that using MESN not only the load forecasting accuracy can be improved, but also both the stability of forecasting and its generalization performance are enhanced.更多还原