二次分解策略组合Informer的短期电力负荷预测方法

针对电力负荷数据存在的波动性、非平稳性而导致预测精度低的问题,提出一种具有二次分解重构策略的深度学习电力负荷预测模型。首先,对负荷数据进行基于局部加权回归的周期趋势分解(STL)-改进的自适应噪声完备集合经验模态分解(ICEEMDAN)二次分解,通过计算样本熵和最大信息数对分量进行重构;然后在Informer模型中引入非平稳性机制,并融合卷积神经网络对重构分量进行预测;最后,将各分量的预测结果线性相加,得到最终预测结果。实验结果表明,所提方法在3个评价指标上的预测误差均低于所对比模型,证明该预测方法可以有效降低数据的非平稳性并提高预测精度。

基于时序相关性的城镇交通流预测方法研究

智能交通系统中的关键一环就是交通流预测,城镇路网复杂于高速路网的关键是路网空间特征更灵活,且空间特征对交通流输出干扰大。于是为得到更精确且灵活的城镇路网空间特征,提出了基于时序相关性的预测方法。这种方法以路网输出的交通流时间序列为依据,采用基于密度的聚类算法思想优化预测模型的训练数据集。同时提出改进剪枝算法优化预测模型,以满足短时交通流预测所要求的时效性。实验结果表明,使用新模型使预测误差下降25%,且时效性提高45%。

考虑短期负荷影响的DeepESN电力市场实时电价预测研究

精准的电价预测有助于宏观调控的实施。但能源结构转型导致大规模可再生能源并网,因此会导致电价降低和产生波动,降低时序预测序列的相关性,加大实时电价的预测难度。针对这一问题,采用自相关函数和最大信息数计算电价自身和电价与电量关联性,为模型输入提供依据,并在此基础上应用具有深度储备池特性的深度回响网络进行实时电价预测。研究结果表明:电价与电量、电价自身具有较强相关性,应考虑自身与电量因素;深度回响网络能够显著提升预测模型的精度,并且具有较强的鲁棒性。

AN EME BLIND SOURCE SEPARATION ALGORITHM BASED ON GENERALIZED EXPONENTIAL FUNCTION

This letter investigates an improved blind source separation algorithm based on Maximum Entropy （ME） criteria. The original ME algorithm chooses the fixed exponential or sigmoid ftmction as the nonlinear mapping function which can not match the original signal very well. A parameter estimation method is employed in this letter to approach the probability of density function of any signal with parameter-steered generalized exponential function. An improved learning rule and a natural gradient update formula of unmixing matrix are also presented. The algorithm of this letter can separate the mixture of super-Gaussian signals and also the mixture of sub-Gaussian signals. The simulation experiment demonstrates the efficiency of the algorithm.

MAXIMUM INFORMATION AND OPTIMUM ESTIMATING FUNCTION

In order to construct estimating functions in some parametric models, this paper introducestwo classes of information matrices. Some necessary and sufficient conditions for the informationmatrices achieving their upper bounds are given. For the problem of estimating the median,some optimum estimating functions based on the information matrices are acquired. Undersome regularity conditions, an approach to carrying out the best basis function is introduced. Innonlinear regression models, an optimum estimating function based on the information matricesis obtained. Some examples are given to illustrate the results. Finally, the concept of optimumestimating function and the methods of constructing optimum estimating function are developedin more general statistical models.