直流微电网中存在多个控制器,而不同控制器参数的组合对系统的整体稳定性影响也存在差异。为提升系统稳定性,提出了一种基于小信号稳定性的直流微电网多控制器参数全局优化方法。首先,推导了系统的小信号模型,对系统进行特征值分析和参与因子分析,确定了PI参数、下垂系数、虚拟惯性系数等关键控制参数的稳定域;其次,建立了包含系统主导极点实部、阻尼比和储能最大输出功率的目标函数,并采用正交实验法获得样本数据,利用综合赋权法对多目标进行赋权;然后,采用结合灰狼优化的改进粒子群算法对系统的关键参数进行优化,结果表明参数优化后的系统特征值更加远离虚轴,阻尼比增大,储能最大输出功率增加,系统稳定性得到了提升;最后,通过仿真验证了所提方法的有效性和优越性。

DC microgrid systems have multiple controllers, and the combination of different controller parameters has different effects on the stability of the entire system. To improve system stability, a multi-controller parameter global optimization method based on small-signal stability is proposed. First, the small-signal model of the system is deduced, and eigenvalue and participation factor analyses of the system are performed to determine the stability domain of key control parameters such as PI parameters, droop coefficients, and virtual inertia coefficients. The objective function, including the maximum real part of the eigenvalue, damping ratio, and maximum output power of the energy storage, is established. The sample data are obtained by using an orthogonal experiment, and the multi-objective is weighted by using a comprehensive weighting method. The key parameters of the system are then optimized using improved particle swarm optimization based on the grey wolf algorithm, thereby yielding the optimization results. The results show that the eigenvalues of the system after parameter optimization are farther away from the imaginary axis, the damping ratio increases, and the maximum output power of the energy storage increases, which improves system stability. Finally, the effectiveness and superiority of the proposed method are verified by building a model on the MATLAB Simulink simulation platform.

多标签特征选择是针对多标签数据的特征选择技术，提高多标签分类器性能的重要手段。提出一种基于流形学习的约束Laplacian分值多标签特征选择方法（Manifold-based Constraint Laplacian Score，M-CLS）。方法分别在数据特征空间和类别标签空间定义两种Laplacian分值：在特征空间利用逻辑型类别标签的相似性对邻接矩阵进行改进，定义特征空间的约束Laplacian分值；在标签空间基于流形学习将逻辑型类别标签映射为数值型，定义实值标签空间的Laplacian分值。将两种分值的乘积作为最终的特征评价指标。实验结果表明，所提方法性能优于多种多标签特征选择方法。

Multi-label feature selection is a feature selection technique based on multi-label data, which is an important means to improve the performance of multi-label classifiers. This paper proposes a Manifold-based Constraint Laplacian Score（M-CLS） feature selection method. The proposed method defines two kinds of Laplacian scores in two different spaces, namely the data feature space and label space. In the feature space, the similarity of the logical labels is used to modify the adjacency matrix, and a constraint laplacian score is defined. In the label space, the logical labels are extended to the numeric labels through manifold learning, and a Laplacian score is defined based on the real values of labels. The product of the two scores is the final feature evaluation index. Experiments show that the proposed method outperforms several multi-label feature selection methods.

短期负荷预测为综合能源系统安全稳定运行提供保障,但负荷波动的不确定性及多种能量相互耦合增大了预测难度。基于此,提出一种基于综合相关性指标和SA-BiGRU的综合能源系统多元负荷预测模型。考虑到不同气象因素对多元负荷的影响,采用综合相关性指标计算气象因素与负荷间的相关性,提出多元负荷三项耦合乘积挖掘能源间交叉耦合关系,并构建特征矩阵作为预测模型输入。同时,利用自适应k-means将原始输入数据划分为不同负荷场景,降低预测复杂度;在双向门控循环单元网络中引入自注意力机制,为输入特征赋予不同权重,从而增强模型对重要特征的区分能力。最后,采用算例与现有模型进行对比分析,结果表明所提出的多元负荷预测方法具有更高的预测精度和更短的预测时间。

Short-term load forecasting provides a guarantee for the safe and stable operation of IES, but the uncertainty of load fluctuations and the coupling of multiple energy sources increase the difficulty of prediction. To address this, this paper proposes a multi-dimensional load forecasting model for the integrated energy system based on the comprehensive correlation index and SA-BiGRU. Firstly, the comprehensive correlation index is used to calculate the correlation between meteorological factors and loads, and a multi-dimensional load coupling feature matrix is constructed to explore the cross-coupling relationship between energy sources,then the coupled load features are constructed as model inputs. At the same time, It divides the input into different load scenarios using adaptive k-means clustering, in order to reduce modeling complexity. And a self-attention mechanism is incorporated into the bidirectional GRU network to differentiate the importance of input features for enhanced distinction. Finally, compared with other forecasting models, the results show that the proposed method has higher accuracy and shorter forecasting time.

Several variants of the long short-term memory (LSTM) architecture for recurrent neural networks have been proposed since its inception in 1995. In recent years, these networks have become the state-of-the-art models for a variety of machine learning problems. This has led to a renewed interest in understanding the role and utility of various computational components of typical LSTM variants. In this paper, we present the first large-scale analysis of eight LSTM variants on three representative tasks: speech recognition, handwriting recognition, and polyphonic music modeling. The hyperparameters of all LSTM variants for each task were optimized separately using random search, and their importance was assessed using the powerful functional ANalysis Of VAriance framework. In total, we summarize the results of 5400 experimental runs (≈ 15 years of CPU time), which makes our study the largest of its kind on LSTM networks. Our results show that none of the variants can improve upon the standard LSTM architecture significantly, and demonstrate the forget gate and the output activation function to be its most critical components. We further observe that the studied hyperparameters are virtually independent and derive guidelines for their efficient adjustment.