A method of estimating the output fluctuation of many photovoltaic power generation systems dispersed in a wide area

The largest output fluctuation is an index used to quantify the disturbance of a power grid caused by wind power plants and photovoltaic power generation systems connected to it. In order to develop its estimation method, we investigate the relationship between the largest output fluctuation and the standard deviation of a newly proposed random variable generated by differencing the output variation of photovoltaic power generation systems. Output fluctuation coefficients are defined and estimated using measured data of photovoltaic power generation systems located at 52 places. An approximation equation is presented to predict geographical correlations between the proposed random variables for photovoltaic power systems by using the distances between their locations. Finally, it is shown that the largest output fluctuation of many photovoltaic power generation systems dispersed in a wide area is predictable by using the output fluctuation coefficients and the equation approximating the geographical correlations of output fluctuations. ©2008 Wiley Periodicals, Inc. Electr Eng Jpn, 166(4): 9–19, 2009; Published online in Wiley InterScience ( www.interscience.wiley.com ). DOI 10.1002/eej.20723     

An operating method using prediction of photovoltaic power for a photovoltaic–diesel hybrid power generation system

This paper describes a novel operating method using prediction of photovoltaic (PV) power for a photovoltaic–diesel hybrid power generation system. The system is composed of a PV array, a storage battery, a bidirectional inverter, and a diesel engine generator (DG). The proposed method enables the system to save fuel consumption by using PV energy effectively, reducing charge and discharge energy of the storage battery, and avoiding low‐load operation of the DG. The PV power is simply predicted from a theoretical equation of solar radiation and the observed PV energy for a constant time before the prediction. The fuel consumption of the proposed method is compared with that of other methods by a simulation based on measurement data of the PV power at an actual PV generation system for 1 year. The simulation results indicate that the amount of fuel consumption of the proposed method is smaller than that of any other methods, and is close to that of the ideal operation of the DG. © 2005 Wiley Periodicals, Inc. Electr Eng Jpn, 151(3): 8–18, 2005; Published online in Wiley InterScience ( www.interscience.wiley.com ). DOI 10.1002/eej.20116     

基于布谷鸟-Elman算法的光伏发电预测

提高光伏发电功率预测的精度对于保证电网的安全稳定运行、提高光伏资源的开发和利用率具有重要的意义。文中提出了一种基于天气相似度以及改进布谷鸟算法优化Elman神经网络的光伏发电短期功率预测模型。首先在选取相似日上,提出一种基于距离和角度趋势的相似度计算方法,选出与待预测日相似度更高的相似日。其次,利用改进后的布谷鸟算法对Elman神经网络的权值和阈值进行优化并构建光伏发电短期功率预测模型。最后将文中提出的光伏发电预测模型与传统Elman神经网络模型的预测结果及实际输出值进行比较,结果表明改进布谷鸟算法优化Elman神经网络的光伏发电短期功率预测模型预测精度更高。     

计及出力波动的光伏电站无功支撑能力评估

在满足配电网稳定运行的前提下,充分利用光伏电站并网逆变器的额定容量,挖掘光伏的无功支撑能力,能够促使配电网经济运行,因此开展光伏电站无功支撑能力评估具有重要的现实意义。首先,提出了基于动态时间弯曲算法的无功支撑能力评估方法和评估指标体系,实现实时态的无功支撑能力评估;然后,提出了基于深度置信网络模型的光伏出力预测方法和预测指标体系,结合预测数据实现未来态的无功支撑能力预测。算例分析表明,预测的光伏电站未来态无功支撑能力与实际未来态评估结果相吻合,验证了所提评估、预测方法的有效性。     

基于MBI-PBI-ResNet的超短期光伏功率预测

为了增强光伏并网的稳定性,提高光伏发电功率预测精度,提出一种基于相似日聚类、群分解(swarm decomposition, SWD)和MBI-PBI-ResNet深度学习网络模型的光伏发电功率超短期预测方法。首先,使用快速傅里叶变换(fast fourier transform, FFT)提取太阳辐照度的期望频率,将其作为聚类特征向量,并根据此聚类特征向量采用自适应仿射传播聚类(adaptive affinity propagation clustering, AdAP)实现相似日聚类。其次,对每一类相似日分别使用群分解算法进行分解,以提取原始数据的多尺度波动规律特征。最后,利用MBI-PBI-ResNet来实现对天气环境多变量关联影响下的时序特征挖掘以及对多尺度分量的局部波形空间特征和长时间依赖时序特征的同时挖掘,并对不同类型特征进行综合集成来实现光伏发电功率超短期预测。研究结果表明：所提方法在光伏发电功率超短期预测领域相较于其他深度学习方法预测精度提高了3%以上,说明此方法在光伏发电功率超短期预测领域具有较高的预测精度和较强的泛化能力。     

高渗透率分布式光伏接入的新型电力系统净功率预测

在双碳背景下,分布式光伏发电的大规模增加以及并网接入,对新型电力系统带来了巨大的挑战。高渗透率分布式光伏出力与电力负荷因受天气因素的影响,具有较强的不确定性和波动性,这在一定程度上增加了配电网净功率的预测难度。为了提高配电网净功率的预测精度,文章提出了Attention-双向GRU神经网络配电网净功率预测方法。文章首先对光伏出力特性、用户侧负荷特性、以及配电网净功率影响因素进行分析,充分掌握净功率受分布式光伏出力和用户侧负荷变化规律的影响。然后将Attention机制融入到双向GRU神经网络中建立了配电网净功率预测模型。其中,Attention机制赋予输入特征不同的关注度,双向GRU神经网络能够学习到净功率的时序特征,二者的完美结合,大大提升了净功率预测模型的表示能力和泛化能力。实验结果表明,文章提出的方法大大提高了配电网净功率预测精度,且性能优于对比模型。     

基于混合神经网络的光伏电量预测模型的研究

精确的光伏发电量预测对光伏发电系统的安全运行有重要的作用。然而,由于太阳能的不稳定性、间歇性和随机性,现有光伏发电量的短期预测模型存在预测误差大、泛化能力低等问题。因此,提出一种混合神经网络和注意力机制的分布式光伏电站电量短期预测模型(A-HNN)。利用残差长短期记忆网络与扩展因果卷积相结合提取数据的时间和空间特征,加入注意力机制增强特征选择,给出一种改进的混合神经网络模型。根据发电量数据时间序列本身的特性,选取以日为周期的时间序列数据。最后,通过实验与近期其他模型对比,结果表明在同等条件下此混合模型可以大幅提高光伏发电量预测的精度。     

基于深度信念网络的光伏电站短期发电量预测

为了解决现有光伏电站短期发电量预测方法存在的预测模型复杂、预测误差较大、泛化能力较低的问题,提出一种基于深度信念网络的短期发电量预测方法。首先综合考虑影响光伏出力的环境因素和光伏板的运行参数以及光伏电站历史发电量数据,对深度信念网络进行训练和学习。在此基础上,采用重构误差的方法确定深度信念网络隐含层层数。最后针对某光伏电站短期发电量进行预测算例分析,验证了该预测模型能主动选择样本抽象特征、自动确定隐含层层数,对短期发电量预测精度较高。对比前馈反向传播(Back Propagation, BP)神经网络预测模型与长短期记忆网络(Long/Short Term Memory, LSTM)预测模型,结果表明所提方法运算量低、预测精度高,且增加神经网络的深度比改进神经网络神经元对预测效果更有效。     

基于Attention-LSTM的光伏超短期功率预测模型

超短期光伏发电功率预测有利于电网的调度管理,提高电力系统运行效率及经济性.针对传统长短时记忆(LSTM)神经网络在处理长序列输入时易忽略重要时序信息的缺陷,文章提出了一种结合注意力机制(Attention)与LSTM网络的功率预测模型.采用皮尔森相关系数法(Pearson)分析了实验的历史数据集,剔除无关变量,对数据集进行了降维处理,简化了预测模型结构.在此基础上将Attention机制与LSTM网络相结合作为预测模型.At-tention机制通过对LSTM的输入特征赋予了不同的权重,使得预测模型对长时间序列输入的处理更为有效.以某地光伏电站实测数据对文中所提模型进行训练和对比验证,所提出的预测模型能够更充分地利用历史数据,对长时间输入序列中的关键信息部分更为敏感,预测精度更高.     

Analysis of possible introduction of PV systems considering output power fluctuations and battery technology,employing an optimal power generation mix model

This paper presents an evaluation of the impact of extensive introduction of photovoltaic (PV) systems and stationary battery technology into the optimal power generation mix in the Kanto and Kinki regions. The introduction of solar PV systems is expected to be extensively deployed in the Japanese household sector and utility companies in order to address the concerns of energy security and climate change. Considering this expected large‐scale deployment of PV systems in electric power systems, it is necessary to investigate the optimal power generation mix which is technologically capable of controlling and accommodating the intermittent output‐power fluctuations inherent in PV systems. Against this background, we develop both a solar photovoltaic power generation model and an optimal power generation mix model, including stationary battery technology, which can be used to explicitly analyze the impact of PV output fluctuations at a detailed time interval resolution such as 10 minutes for 365 consecutive days. Simulation results reveal that PV introduction does not necessarily increase battery technology due to the cost competitiveness of thermal power plants in the load‐following requirement caused by PV systems. Additionally, on the basis of sensitivity analysis on PV system cost, dramatic cost reduction proves to be indispensable for PV to supply bulk electricity similarly to thermal and nuclear power plants. © 2012 Wiley Periodicals, Inc. Electr Eng Jpn, 182(2): 9–19, 2013; Published online in Wiley Online Library ( wileyonlinelibrary.com ). DOI 10.1002/eej.22329     

基于EMD-LSTM的光伏发电预测模型

随着能源消费结构的改变,可再生能源发电的消纳比例逐渐上升。文中以光伏发电功率为研究对象,分析了不同天气状态下的发电功率曲线特性及不同气象因素与光伏发电出力的相关性,进而提出了一种经验模态分解-长短期记忆神经网络(EMD-LSTM)方法融合的光伏发电功率预测模型。首先对预处理后的光伏发电功率历史序列进行重构,并对重构后的出力序列进行EMD分解,针对分解得到的各子序列分别建立长短期记忆神经网络模型,最后将各子序列预测模型得到的结果叠加得到光伏发电功率预测值。采用国内某地区光伏发电的实际出力数据对模型进行了检验,与滑动平均自回归模型(ARIMA)、支持向量机模型(SVM)、LSTM等预测模型相比,文中所提出的模型预测误差小,能有效提高光伏发电功率的预测精度。     

Compensation of power fluctuations of wind power generation by means of biomass gas turbine generator and flywheel energy storage equipment

Power generation using natural energy contains electric power fluctuations. Therefore, in order to put such power generation systems to practical use, compensation for system power fluctuations is needed. In this paper, we propose a power compensation method using a biomass gas turbine generator and flywheel energy storage equipment. The gas turbine generator is used for compensation of low‐frequency power fluctuations in order to decrease the required flywheel capacity. The usefulness of the proposed system is confirmed by experiments using a test plant. © 2009 Wiley Periodicals, Inc. Electr Eng Jpn, 170(3): 1–8, 2010; Published online in Wiley InterScience ( www.interscience.wiley.com ). DOI 10.1002/eej.20896     

A method of single-phase PWM for an independent power supply in photovoltaic power generation systems

We propose a method of single-phase PWM for an independent power supply in photovoltaic power generation systems. This new PWM is derived by comparing levels of signal waves with one of carrier waves which have bipolar swing different from unipolar swing in the conventional PWM. In this PWM, we can use a battery with lower voltage in combination with the photovoltaic power generation; fundamental level of output voltage is raised by about 11%, though poor in quality of waveforms. © 1998 Scripta Technica. Electr Eng Jpn, 122(4): 55–62, 1998     

A method for load frequency control using battery in power system with highly penetrated photovoltaic generation

It is generally believed that large battery systems will be needed to store surplus electric energy due to the high penetration of renewable energy (RE) such as photovoltaic generation (PV). Since the main objective of high penetration of RE is to reduce CO₂ emissions, reducing kWh output of thermal generation that emits large amounts of CO₂ in power systems should be sufficiently considered. However, thermal generation plays an important role in load frequency control (LFC) of power systems. Therefore, if LFC could be performed with batteries and hydropower generation, the kWh output of thermal generation could be reduced significantly. This paper presents a method of LFC using batteries in a power system with highly penetrated PVs. An assessment of the effect of the proposed method considering mutual smoothing effect of highly penetrated PVs is made. © 2013 Wiley Periodicals, Inc. Electr Eng Jpn, 184(4): 22–31, 2013; Published online in Wiley Online Library ( wileyonlinelibrary.com ). DOI 10.1002/eej.22425     

Application of unified voltage analysis for high‐ and low‐voltage distribution systems by power flow calculation using indication method between nodes

It has been noted that the voltage of connection points rises according to the reverse power flow when grid‐connected photovoltaic systems are concentrated in distribution systems in residential areas. When this happens, the photovoltaic system may control the power generation output to maintain a suitable voltage for the connection point. Designing a demand area power system aiming at free access to a distributed power supply for energy‐effective practical use requires a precise understanding of this problem. When analyzing photovoltaic systems mainly connected to low‐voltage systems, we looked for a method of analysis in which the high‐voltage systems and the low‐voltage single‐phase three‐wire systems are unified. This report concerns use of the indication method between nodes using power flow calculation, for the purpose of developing a technique of analyzing unified high‐voltage systems and low‐voltage single‐phase three‐wire systems. © 2004 Wiley Periodicals, Inc. Electr Eng Jpn, 147(3): 49–62, 2004; Published online in Wiley InterScience ( www.interscience.wiley.com ). DOI 10.1002/eej.10255     

基于风光混合模型的短期功率预测方法研究

准确地预测风力发电及光伏发电的输出功率对提高风光互补供电系统的调度质量具有重要意义。建立了基于BP神经网络的风光混合预测模型,将现有技术中分两次预测的风电功率和光伏功率采用同一个预测模型,同时实现整个区域风电场及光伏电站的输出功率预测,在简化预测方法的同时提高预测准确度。通过某海岛的风电及光伏电站的实际数据验证,计算分析了预测误差。结果表明该方法具有较高的预测精度,对风光混合的功率预测具有一定的学术价值和工程实用价值。     

Positive Effect of Constant Leading Power Factor Operation of PV in Power System

A voltage rise problem in distribution networks has been discussed as the foremost concern with respect to the spread of large numbers of photovoltaic systems. We focus on the latent ability of the present distribution network and photovoltaic systems to find a low‐cost solution to the problem and consider a solution to mitigate the voltage rise using the photovoltaic system's constant leading power factor operation. Previously, based on simulations using an aggregated model of a real distribution network, we proposed that a combination of a photovoltaic system's constant leading power factor operation and LDC makes it possible to maintain the line voltage and LTC tap position adequately. In this paper, additionally, we confirm some effects of the proposed method in an aggregated model of a distribution network and a trunk power system. One of them is that the proposed method reduces the frequency of restricting output power from photovoltaic systems and changing the LTC and SVR tap position, although photovoltaic systems rapidly fluctuate. Another is that the proposed method cannot make a significant impact on a trunk power system in a voltage class exceeding 6.6 kV.     

基于PFA-MBAS-BP神经网络模型的光伏发电短期预测

针对光伏发电短期预测模型的输入变量多且关系复杂、BP神经网络稳定性差且易陷入局部最优解等问题,建立了一种基于主因子分析法(PFA)和优化天牛须搜索算法(MBAS)的改进BP神经网络光伏发电短期预测模型。该模型首先对光伏历史发电数据和气象数据进行降维简化分析,利用主因子分析法对影响光伏发电的主要因素进行相关性分析,选取主因子作为预测模型输入量。然后利用MBAS算法的空间寻优搜索,选取BP神经网络训练的最优权值阈值。最后,利用实测历史数据对不同预测模型进行仿真对比。仿真结果表明,所建立的改进模型的预测精度可达92.5%,图像数据拟合程度高且适用多种天气类型的光伏发电预测。     

基于连续潮流的配电网供电能力评估

随着非全相运行的分布式电源大量接入配电网,配电网固有的三相不平衡特征更加突出,传统配电网供电能力评估因忽略配电网三相不平衡特征导致结果不准确。为了准确分析三相不平衡特征对配电网最大供电能力评估的影响,建立了以配电网供电负荷参数最大为目标函数,考虑了支路热约束和节点电压等状态变量和分布式电源的有功和无功功率等控制变量的含分布式电源三相不平衡配电网供电能力评估模型。选择电压跌落情况最严重的相作为连续参数,确保预测-校正过程的的连续潮流法求解的结果更加精确。最后,采用拓展的IEEE33节点配电系统进行仿真验证,表明文中所提的模型和求解方法是有效的。     

基于Wide & Deep-XGB2LSTM模型的超短期光伏功率预测

为了充分利用电网自身的海量历史数据进行光伏功率预测,提出一种宽度&深度(Wide&Deep)框架下融合极限梯度提升(XGBoost)算法和长短时记忆网络(LSTM)的Wide&Deep-XGB2LSTM超短期光伏功率预测模型.对历史数据进行特征提取,获得时间、辐照度、温度等原始特征,在此基础上进行特征重构,通过交叉组合和挖掘统计特征构造辐照度×辐照度、均值、标准差等组合特征,并通过Filter法和Embedded法进行特征选择.在TensorFlow框架下通过算例对比验证了所提模型及特征工程工作对光伏功率预测性能的提升效果.