基于Attention和残差网络的非侵入式负荷监测

非侵入式负荷监测（Non-Intrusive Load Monitoring, NILM）可以从家庭电表的总功率读数,估算出各用电器的功率。由于对于同一类用电器,其状态种类、各状态持续时长、各状态的功率波形都不同,这使得基于特征工程和聚类的模型的泛化能力不强;回归模型的分解功率难以迅速跟踪真实功率。针对这些问题,文中将回归问题转化为在序列每个时刻的多分类问题,并提出基于Attention和残差网络的非侵入式负荷监测模型。该模型基于具有编码器和解码器的seq2seq框架,首先通过嵌入矩阵将高维稀疏one-hot向量映射为低维稠密向量;在编码部分,通过双向GRU从前后两个方向提取序列信息,引入Attention机制计算序列中当前时刻最重要的信息,引入残差连接学习残差部分输入输出之间的差异;在解码部分,用回归层组合BiGRU解码结果,取经过softmax函数处理的最大概率功率类别作为结果。该模型在选取REFIT数据集中表现良好,其中测试集与训练集完全独立,表明训练好的模型可以直接应用在新的住宅用户中。     

RPROP神经网络在非侵入式负荷分解中的应用

为了解决常用家电设备投切状态辨识问题,提出一种以神经网络为辨识模型的方法,增强其快速辨识能力。首先,从负荷印记出发,针对各用电设备的稳态电流谐波特性,建立用电设备特征标签。然后,采用弹性BP(Resilient back propagation,RPROP)神经网络,将输入数据特征向输出层非线性映射,实现快速收敛至全局最优点。训练中采用多种设备组合方式,进行用电设备特征辨识。最终,以五类常用用电设备进行实验,实验结果表明该算法能够有效地识别家用设备的工作状态组合,且对功率相近、谐波具有较小差异的用电设备工作状态也具有很好的辨识能力。     

Challenges for real-world applications of nonintrusive load monitoring and opportunities for machine learning approaches

The rapid increase of power consumption calls for efficient and effective energy usage and conservation strategies in buildings. One of the requirements of achieving such a goal is load monitoring of residential appliances. Among the available load monitoring frameworks, nonintrusive load monitoring (NILM) which is used to estimate the appliance-level power usage from the aggregated signals from smart meters, has the potential to be widely deployed. This paper presents an up-to-date review of NILM methods and the challenges existing in each step of NILM. Then this paper reviews two state-of-the-art machine learning based NILM methods including Hidden Markov Model and Deep Learning techniques. Finally, this paper discusses areas for future research and development of NILM in real-world applications, where machine learning approaches can play a more significant and even decisive role.     

基于生成对抗网络和混合时空神经网络的入侵检测

针对网络入侵检测领域存在检测准确率低的问题,研究异常流量样本少和分类器性能不佳时的入侵检测模型,提出一种基于改进生成对抗网络和混合时空神经网络的入侵检测模型。改进生成对抗网络通过学习异常流量样本的分布特性,生成具有特定标签的人工异常流量样本;融合卷积神经网络和双向长短时记忆神经网络提取攻击流量的时空融合特征,利用注意力机制对时空融合特征进行加权,构建混合时空神经网络对网络流量进行分类预测。在UNSW-NB15数据集上对所提模型进行仿真实验,准确率和F1分数分别为92.93%和94.81%,表明所提模型能够有效改善原始数据集中的类别不平衡性问题,提高对异常流量样本的检测能力和网络入侵的检测准确率。     

Recurrent neural networks for short-term load forecasting

Forecasting the short-term load entails the construction of a model, and, using the information available, estimating the parameters of the model to optimize the prediction performance. It follows that the more closely the chosen model approximates the actual physical generating process, the higher the expected performance of the forecasting system. In this paper it is postulated that the load can be modeled as the output of some dynamic system, influenced by a number of weather, time and other environmental variables. Recurrent neural networks, being members of a class of connectionist models exhibiting inherent dynamic behavior, can thus be used to construct empirical models for this dynamic system. Because of the nonlinear dynamic nature of these models, the behavior of the load prediction system can be captured in a compact and robust representation. This is illustrated by the performance of recurrent models on the short-term forecasting of the nation-wide load for the South African utility, ESKOM. A comparison with feedforward neural networks is also given     

Cascaded artificial neural networks for short-term load forecasting

An application of artificial neural networks (ANNs) to short-term load forecasting is presented in this paper. An algorithm using cascaded learning together with historical load and weather data is proposed to forecast half-hourly power system load for the next 24 hours. This cascaded neural network algorithm (CANNs) includes peak, minimum and daily energy prediction as additional input data for the final forecast stage. These additional input data are predicted using the first (ANNs) model. The networks are trained and tested on the electric power system of Kuwait. The absolute average forecasting error is reduced from 3.367% to 2.707% by applying CANNs as compared to the conventional ANNs. Simulation results indicate that the developed forecasting approach is effective and point to the potential of the methodology for economic applications     

Short-term load forecasting using neural networks

An artificial neural network (ANN) model for short-term load forecasting (STLF) is presented. The proposed model is capable of forecasting the next 24-hour load profile at one time, as opposed to the usual ‘next one hour’ ANN models. The inputs to the ANN are load profiles of the two previous days and daily maximum and minimum temperature forecasts. The network is trained to learn the next day's load profile. Testing of the model with one year of data from the Greek interconnected power system resulted in a 2.66% average absolute forecast error.     

Artificial neural networks for load flow and external equivalents studies

In this paper an artificial neural network (ANN) based methodology is proposed for (a) solving the basic load flow, (b) solving the load flow considering the reactive power limits of generation (PV) buses, (c) determining a good quality load flow starting point for ill-conditioned systems, and (d) computing static external equivalent circuits. An analysis of the input data required as well as the ANN architecture is presented. A multilayer perceptron trained with the Levenberg–Marquardt second order method is used. The proposed methodology was tested with the IEEE 30- and 57-bus, and an ill-conditioned 11-bus system. Normal operating conditions (base case) and several contingency situations including different load and generation scenarios have been considered. Simulation results show the excellent performance of the ANN for solving problems (a)–(d).     

结合图半监督与广义回归神经网络的非侵入式海洋平台负荷监测

海洋平台微电网所处环境复杂,对其自动化和智能化要求较高,目前缺少对其负荷实时智能监测和管理的方法。从非侵入式负荷监测的角度,考虑海洋平台的经济性要求和特殊的工业环境,提出结合图半监督与广义回归神经网络的非侵入式海洋平台负荷监测方法。采用图半监督学习算法自动标记训练数据集,减少了人工标记数据的工作量,使系统能自动完成数据标记。并与半监督聚类算法对比分析,表明图半监督学习算法对数据标记具有更高的正确率。再利用广义回归神经网络较强的非线性分类能力,提升负荷识别的识别精度和减少计算复杂度。Matlab/Simulink仿真结果表明,所提出的负荷识别算法不仅减少了人工干预而且具有高精度的识别率。     

Application of fuzzy neural networks and artificial intelligence for load forecasting

An integrated evolving fuzzy neural network and simulated annealing (AIFNN) for load forecasting method is presented in this paper. First we used fuzzy hyper-rectangular composite neural networks (FHRCNNs) for the initial load forecasting. Then we used evolutionary programming (EP) and simulated annealing (SA) to find the optimal solution of the parameters of FHRCNNs (including parameters such as synaptic weights, biases, membership functions, sensitivity factor in membership functions and adjustable synaptic weights). We knew that the EP has a good capability for searching for globe optimal value, but a poor capability for searching for the local optimal value. And, the SA only had a good capability for searching for a local optimal value. Therefore, we combined both methods to obtain both advantages, and so improve the shortcoming of the traditional ANN training where the weights and biases are always trapped into a local optimum. Finally, we use the AIFNN to see if we could improve the solution quality, and if we actually could reduce the error of load forecasting. The proposed AIFNN load forecasting scheme was tested using data obtained from a sample study including 1 year, 1 month and 24 h time periods. The result demonstrated the accuracy of the proposed load forecasting scheme.     

负荷持续曲线的神经网络模型

鉴于前馈人工神经网络的诸多优点，本文提出了一种新的基于ＡＮＮ的负荷持续曲线模型，从而不用给定某个特定的显式数学表达式。该模型以实际的小时负荷数据及统计性的特征参数对ＡＮＮ的输入数据；对实际作了抽样处理，以免输入数据量太多；针对常规的误差反向传播算法的缺点，并考虑到遗传算法是一种鲁棒性好的优化方法，本文采用遗传算法来进行ＡＮＮ的学习可以方便地直接采用相对误差作目标函数，同时可以搜索到全局最优解。     

Harmonic source monitoring and identification using neural networks

Neural networks are applied to make initial estimates of harmonic sources in a power system with nonlinear loads. The initial estimates are then used as pseudomeasurements for harmonic state estimation, which further improves the measurements. This approach permits measurement of harmonics with relatively few permanent harmonic measuring instruments. Simulation tests show that the trained neural networks are able to produce acceptable estimates for varying harmonic sources and that the state estimator will generally pull these estimates closer to the correct values. The process successfully identified and monitored a suspected harmonic source that had not previously been measured     

神经网络预测中长期电力负荷对比研究

为避免传统方法预测中长期电力负荷建模的复杂性,根据电力负荷历史数据,研究了基于LM算法的BP网络、RBF网络在中长期电力负荷预测中的应用,通过神经网络对训练样本的学习,自动提取影响中长期电力负荷的诸多因素。从训练速度、预测误差等方面分析对比了两种神经网络预测能力,仿真和实例数据表明了两种神经网络在中长期电力负荷预测方面的可行性和良好效果。     

神经网络预测中长期电力负荷对比研究

为避免传统方法预测中长期电力负荷建模的复杂性,根据电力负荷历史数据,研究了基于LM算法的BP网络、RBF网络在中长期电力负荷预测中的应用,通过神经网络对训练样本的学习,自动提取影响中长期电力负荷的诸多因素.从训练速度、预测误差等方面分析对比了两种神经网络预测能力,仿真和实例数据表明了两种神经网络在中长期电力负荷预测方面的可行性和良好效果.     

基于深度卷积和门控循环神经网络的传感器运动识别

用于可穿戴传感器的人体运动识别任务的传统机器学习方法通常需要手工提取特征,可以自动提取人体运动数据特征的深度神经网络正成为新的研究热点。目前将卷积神经网络(CNN)和长短期记忆(LSTM)循环神经网络组合而成的DeepConvLSTM在识别精度方面有着优于其他识别方法的表现。针对带有长短期记忆循环单元的神经网络的训练较为困难的问题,提出了一种基于卷积神经网络和门控循环单元(GRU)的融和模型,并在3个公开数据集(ACT数据集、UCI数据集和OPPORTUNITY数据集)上与卷积神经网络和DeepConvLSTM进行了性能对比。实验结果显示,该模型在3个公开数据集上的识别精度都高于卷积神经网络,与DeepConvLSTM相当,但是收敛速度比DeepConvLSTM更快。     

一种基于DBN的高光谱遥感图像分类方法

高光谱遥感数据越来越普及并为人们广泛使用,基于高光谱数据的地面物体精确分类是高光谱遥感技术的核心应用之一。对高光谱数据进行提特征提取是进行地物分类的有效方法。深度学习是机器学习研究中的新领域,它多隐层的多层感知器结构使其能够学习到对数据有更本质的刻画的特征,在图像分类和可视化领域取得了更好的成绩。深度置信网（deep belief network ,DBN）是深度学习网络中常见的模型。利用高光谱数据的高维特性,搭建基于DBN的高光谱图像分类模型,结合高光谱数据的空间结构对地物进行分类。实验表明,基于DBN的高光谱图像分类方法可以得到更好的分类效果。     

Power system dynamic load modeling using artificial neural networks

The dynamic characteristics of power system loads are critical to obtaining quality operating point-prediction and stability calculations. The composition of components at a load bus makes the aggregated behavior too complicated to be expressed by a simple form. Armed with the theorems recently developed on the approximation capability of artificial neural networks, the authors devise a load model to describe the complex dynamic behavior of loads. Real field data are used to train and test this model. The results verify that this model can emulate load dynamics well and should therefore be suitable as a representation of load for stability analysis     

Short-term load forecasting based on artificial neural networks parallel implementation

This paper presents the development and application of advanced neural networks to face successfully the problem of the short-term electric load forecasting. Several approaches including Gaussian encoding backpropagation (BP), window random activation, radial basis function networks, real-time recurrent neural networks and their innovative variations are proposed, compared and discussed in this paper. The performance of each presented structure is evaluated by means of an extensive simulation study, using actual hourly load data from the power system of the island of Crete, in Greece. The forecasting error statistical results, corresponding to the minimum and maximum load time-series, indicate that the load forecasting models proposed here provide significantly more accurate forecasts, compared to conventional autoregressive and BP forecasting models. Finally, a parallel processing approach for 24 h ahead forecasting is proposed and applied. According to this procedure, the requested load for each specific hour is forecasted, not only using the load time-series for this specific hour from the previous days, but also using the forecasted load data of the closer previous time steps for the same day. Thus, acceptable accuracy load predictions are obtained without the need of weather data that increase the system complexity, storage requirement and cost.     

Fuel constrained economic emission load dispatch using hopfield neural networks

This paper presents a new approach using Hopfield neural networks for solving fuel constrained economic emission load dispatch problems of thermal generating units. This is a multi-objective minimization technique that includes the standard load constraints as well as the fuel constraints. The generation schedule is compared to that which would result if fuel constraints were ignored. The comparison shows that fuel consumed can be adequately controlled by adjusting the power output of various generating units so that the power system operates within its fuel limitations and within contractual constraints. It has been found that one of the two objective functions (fuel cost and emission level) may be increased while other may be decreased to serve the same power demand but this may well compensate for the penalty that might be otherwise imposed for not maintaining the fuel contract. Numerical results for an example system have been presented to illustrate the performance and applicability of the proposed method.     

Unit commitment using load forecasting based on artificial neural networks

This paper consists of two parts. While the first part shows the application of artificial neural networks to load forecasting using new input-output models, the second part utilizes the results from the first part in unit commitment. Based on the forecasts provided, unit commitment schedules are obtained for both hourly and daily load variations. Issues related to both problems are discussed along with an illustration of the two-step method using data obtained from a local utility. While a generation schedule such as this is not only invaluable to power system planners and operators, it is shown that this two-step process paves the way for an artificial intelligence (AI) type of method for the unit commitment problem based on the same inputs as the load forecasting method. For the chosen inputs, the simulations here show an average error of 4.3% and 3.1% in the case of the daily (twenty-four hours ahead) and hourly (one hour ahead) load forecast, respectively.