态势感知+CNN-LSTM网络基于CNN-LSTM网络的电力系统态势感知预测水平研究

需积分: 5 45 浏览量 2022-10-10 18:12:21 上传评论收藏 3.92MB PDF 举报

资源推荐

资源详情

资源评论

IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 17, NO. 10, OCTOBER 2021 6951

Toward the Prediction Level of Situation

Awareness for Electric Power Systems Using

CNN-LSTM Network

Qi Wang , Member, IEEE,SiqiBu , Senior Member, IEEE, Zhengyou He, Senior Member, IEEE,

and Zhao Yang Dong

, Fellow, IEEE

Abstract—Situation awareness (SA) has been recog-

nized as a critical guarantee for the stable and secure oper-

ation of electric power systems, especially under complex

uncertainties after renewable energy integration. In this

article, an artiﬁcial-intelligence-powered solution is pre-

sented to reach a full realization of SA covering percep-

tion, comprehension, and prediction, the last of which is

more advanced but challenging and hence has not been

discussed in any literature before. A novel SA model is

proposed by aggregating two powerful deep learning struc-

tures: convolutional neural network (CNN) and long short-

term memory (LSTM) recurrent neural network. The pro-

posed CNN-LSTM model has superiority to achieve collab-

orative data mining on spatiotemporal measurement data,

i.e., to learn both spatial and temporal features simultane-

ously from phasor measurement units data. Two functional

branches are designed within the SA model: a contingency

locator to detect the exact fault location at present and a

stability predictor to predict stability status of the system in

the future. Test results have shown high performance (ac-

curacy) of the model even on a low level of data adequacy.

Manuscript received June 26, 2020; revised August 23, 2020 and

November 14, 2020; accepted December 20, 2020. Date of publica-

tion December 28, 2020; date of current version June 30, 2021. This

work was supported in part by the National Natural Science Founda-

tion of China for the Research Project under Grant 51807171, in part

by the Guangdong Science and Technology Department for the Re-

search Project 2019A1515011226, in part by the Hong Kong Research

Grant Council for the Research Project under Grant 25203917, Grant

15200418, and Grant 15219619, in part by the Department of Electrical

Engineering, The Hong Kong Polytechnic University for the Start-up

Fund Research Project under Grant 1-ZE68, in part by the funding

of Chengdu Guojia Electrical Engineer ing Company, Ltd. under Grant

NEEC-2019-B01, and in part by the UNSW Digital Grid Futures Institute

seed fund. Paper no. TII-20-3082. (Corresponding author: Siqi Bu.)

Qi Wang is with the School of Electrical Engineering, Southwest

Jiaotong University, Chengdu, Sichuan 611 756, China, and also with

the Department of Electrical Engineering, The Hong Kong Polytechnic

University, Kowloon, Hong Kong (e-mail: clarkstarcraft@gmail.com).

Siqi Bu is with the Hong Kong Polytechnic University Shenzhen Re-

search Institute, Centre for Advances in Reliability and Safety, Research

Institute for Smart Energy, and the Department of Electrical Engineering,

The Hong Kong Polytechnic University, Kowloon, Hong Kong (e-mail:

siqi.bu@gmail.com).

Zhengyou He is with the School of Electrical Engineering, South-

west Jiaotong University, Chengdu, Sichuan 611 756, China (e-mail:

hezy@swjtu.cn).

Zhao Yang Dong is with the School of Electrical Engineering and

Telecommunications, The University of New SouthWales, Sydney, NSW

2033, Australia (e-mail: joe.dong@unsw.edu.au).

Color versions of one or more ﬁgures in this article are available at

https://doi.org/10.1109/TII.2020.3047607.

Digital Object Identiﬁer 10.1109/TII.2020.3047607

The proposed SA model can promisingly facilitate very fast

postfault actions by the system operators to prevent the

power system from any unstable operational status.

Index Terms—Convolutional neural network (CNN), deep

learning, long short-term memory (LSTM) recurrent neural

network, power system stability, situation awareness (SA),

spatiotemporal data mining.

I. INTRODUCTION

TABLE and secure operation of electric power systems is

the foremost premise to guarantee constant and efﬁcient

power supply. Nevertheless, the actual operational states of

power systems are full of uncertainties, such as stochastic load

ﬂuctuations and unexpected fault contingencies, which tremen-

dously increase the observational difﬁculty of system stability

and security. Moreover, the growing integrations of renewable

energy sources (RESs), such as wind and solar power which

are generally hard to forecast, have brought about far more

uncertainties to the operation of modern power systems. As

a result, the system operators may not be fully cognizant of

what they need to know in all probability, which can lead to

catastrophic blackouts eventually.

Situation awareness (SA), which originated in the military and

aviation domains, has been recently recognized as a promising

solution to the above issue in power systems [1]. SA plays a vital

role to bridge between the system operators’ rational perception

and the system’s real operational states [2]. The consensus in the

research community has divided the function of SA into three

levels [3]: 1) perception of system’s measurements, 2) compre-

hension of system’s current states, and 3) projection of system’s

future status. Although certain studies have been devoted t o SA

on the perception and comprehension levels [4]–[9], there is

still no report for a viable solution on the more advanced but

challenging projection level (which is also called the prediction

level in this article for an intuitive understanding).

On the other hand, owing to the successful application of

wide-area measurement devices such as phasor measurement

units (PMUs), substantial measurement data are presented for

the condition description and assessment of power systems. This

opens up fabulous feasibility to utilize the data mining scheme

in the SA of power systems [10]–[13]. The high-speed and

easy-deployment features of PMUs allow these measurement

See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: Wuhan University. Downloaded on July 04,2022 at 03:51:10 UTC from IEEE Xplore. Restrictions apply.

6952 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 17, NO. 10, OCTOBER 2021

data to be produced in a real-time and widespread manner.

Therefore, such measurement data always contains rich but

implicit information in both temporal and spatial representation.

Most of the studies have only focused on mining from either

temporal or spatial perspective [14]–[18]. However, few of them

have utilized both of them simultaneously. And the lack of col-

laboratively spatio-temporal analysis will limit SA’s application

prospect, especially on the prediction level.

Recently with the booming development of artiﬁcial intel-

ligence (AI), deep learning has shown its outstanding ability

towards spatiotemporal data mining as well as various prediction

tasks [19], [20]. This article presents an AI-powered solution

to the SA of power systems toward the prediction level. A

novel SA model is developed by aggregating two powerful

deep learning models—convolutional neural network (CNN)

and long short-term memory (LSTM) recurrent neural network

(RNN), to fully utilize the spatio-temporal measurement data

and hence to facilitate a full realization of SA covering percep-

tion, comprehension, and prediction. The main contributions of

this article are outlined as follows.

1) With the aggregation of the CNN and LSTM, the proposed

SA model can interpret and learn both spatial and temporal

features simultaneously from measurement data. By a well-

designed data preprocessing procedure, the measurement data

from PMUs can be mapped as a spatiotemporal input form to

feed into the CNN-LSTM model. This indicates the spatiotem-

poral perception ability achieved by the SA model.

2) The proposed SA model is also endowed with the ability to

comprehend the system’s current status and the ability to predict

the system’s future status, at the same time. This is achieved by

two branches of functional output, i.e., a contingency locator

to ﬁnd the exact location where an earth fault occurs at present

(comprehension), and a stability predictor to forecast whether

the system is stable or not after the fault in the future (prediction).

3) After being well-tuned, the SA model only relies on a

low level of data adequacy in both the temporal and spatial

scales. In other words, the SA model can still maintain its high

performance (accuracy) when making perception and prediction

with a very short period of postfault sequential data and even

with partial measurements due to missing PMU conditions.

The rest of this article is organized as follows. Section II

is a literature review section on the related works. Section III

presents the methodology of the proposed SA model. Section IV

is the case study which shows the prominent performance of the

SA model. Section V concludes this article.

II. R

ELATED WORKS

Some signiﬁcant and related works are reviewed in this sec-

tion. These works are classiﬁed into two parts: the state-of-the-

art SA in power systems and the state-of-the-art AI for data

mining in power systems.

A. SA in Power Systems

As deﬁned by Ensley [2], SA has been regarded as a critical

role in many areas that involve human decision and operation

under complex and changeable s ituations. In the ﬁeld of power

industry, SA has attracted increasing attention over the past few

years [1]. According to Ensley’s three-level model of SA [2],

the implementation of SA in power systems is also expected to

be unfolded in the following three levels [3].

The ﬁrst level is the perception of information and key ele-

ments, i.e., measurement data with respect to time and space. At

this level, studies mainly focus on the aspects of measurement

deployment and data acquisition, such as the optimal PMU

placement [4], noisy signal identiﬁcation [5], and missing or

bad data handling [6].

The second level is the comprehension of what the perceived

data indicates, to reveal the system’s current states, existing

faults, and potential risks to the system operators. Researchers

have made a series of progress at this level, including security as-

sessment [7], anomaly detection [8], and robust state estimation

under a vulnerable cyber environment [9].

The third level is the projection (or prediction) of the future

status of the system. This is the most advanced and challenging

level because the difﬁculty in predicting the future and the

accurate prediction results can be only produced based on the

impeccable accomplishment of the ﬁrst two levels. However, no

study has been found to provide a feasible solution toward this

level so far.

Therefore, there is a pressing need to explore a possibility

to further develop the SA at the prediction level, in order to

facilitate a complete, sufﬁcient, and accurate SA for power

system operators. Moreover, the realization of SA toward the

prediction level provides a viable solution to t he challenging

stability prediction problems (e.g., transient stability prediction)

in power systems.

B. AI for Data Mining in Power Systems

Over the years, AI is making remarkable advances, which

brings a golden opportunity to apply data mining techniques in

modern electric power systems [19]. In order to take advantage

of the real-time PMU data, earlier studies have made pioneering

explorations on the basis of conventional machine learning

approaches, such as decision tree [10], random forest (RF) [11],

support vector machine (SVM) [12], and ensemble learning [13].

Positive contributions have been made by these works in solv-

ing such dynamic security problems that have resisted many

model-based approaches before. But the performance of these

conventional machine learning approaches can be limited by

increasing uncertainties such as RES integration, until the advent

of deep learning [20].

Different structures of the deep learning model have been

given several attempts for various data mining tasks in power

systems under uncertain environments, including multilayer per-

ceptron (MLP) [14], CNN [15], [16], RNN [17], and generative

adversarial network [18]. Among them, CNN and RNN are the

most common models to mine spatial and temporal features

from data, respectively. CNN is originally designed to handle

array-like data such as images or audio s pectrograms [21]. A

typical CNN consists of two kinds of layers—convolutional

layers and pooling layers. The convolutional layers are equipped

with several convolution kernels that extract feature maps from

Authorized licensed use limited to: Wuhan University. Downloaded on July 04,2022 at 03:51:10 UTC from IEEE Xplore. Restrictions apply.

WANG et al.: TOWARD THE PREDICTION LEVEL OF SITUATION AWARENESS 6953

Fig. 1. Flowchart of the proposed SA model and its application.

the pixel-correlated arrays. The pooling layers (usually max

pooling) then merge those similar features in each feature

map [22]. RNN is another skillful model when processing

sequential data such as text and speech. RNN has a specialized

recurrent structure which can maintain memory over every time

step of the input sequence [23]. A variant of RNN, which is

known as LSTM network, has ﬁxed the gradient vanishing or

exploding problems by a meticulously designed cell structure.

As a result, LSTM network has proved to be more powerful in

harnessing long-term memory [24].

The excellent performance of deep learning has facilitated a

number of data mining studies based on the wide-area PMU data.

Some of them utilized the spatial connectivity of the data [15],

[18], while others concentrated on the temporal correlation [14],

[16], [17]. However, few studies have made simultaneous min-

ing on both the spatial and temporal features. This will limit

the data mining ability of AI when dealing with the transient

stability prediction problems in power systems because power

system transient stability is a system-level issue which requires

exploration on the spatial connectivity, and at the same time

also a dynamic process which requires analysis on the temporal

correlation.

III. M

ETHODOLOGY

A. Framework Overview

The framework and workﬂow of the proposed SA model is

outlined in Fig. 1. First, all the measurement data from PMUs

are obtained and preprocessed to uncover the spatio-temporal

patterns inside the data. As the main body of the workﬂow,

the CNN-LSTM model is aggregated by two deep learning

modules—the CNN module and the LSTM module. The spatial

features within the data are extracted and learned by the CNN

module, while the temporal features are handled by the LSTM

module. The aggregation of these two modules is realized by a

time-distributed operation.

The CNN-LSTM model is designed to concurrently accom-

plish two tasks, i.e., contingency location and stability predic-

tion. By feeding a very short period of postfault data, the model

can not only locate the exact fault l ocation at present, but also

predict the stability status of the system in the future. In this

way, all three levels of SA (i.e., perception, comprehension, and

prediction) can be achieved. As for the ﬁeld application, the

proposed SA model can alert the system operator in a prompt

and accurate manner, in order to implement very fast postfault

actions (PFA) subsequently.

B. Data Preprocessing

1) Matrix Mapping: As the ﬁrst component of the CNN-

LSTM model, the CNN module requires the shape of input data

to be 2-D array-like. Therefore, it is necessary to map the 1-D

measurement data into the expected 2-D form. Some pioneer

studies have presented feasible solutions on the PMU data

preparation for the CNN model. For example, Gupta et al. [16]

proposed a novel operation to map the 1-D PMU data into a 2-D

heatmap representation, to facilitate the instability prediction

function of the CNN model. Inspired by [16], this article creates

an improved mapping operation named matrix mapping, to map

the 1-D measurement data (i.e., 1-D vector which consists of

all PMU measurements) into a 2-D measurement matrix. The

matrix mapping is operated as

X =[x

i,j

]

m×m



i,j

= 0 i = j

i,j

= a

− a

i = j

(1)

where X is the m × m measurement matrix, x

i,j

denotes every

entry of X for i = 1, 2,...,m, and j = 1, 2,...,m, m is the

total number of measurements. Besides, a

and a

are the

respective ith and jth measurement from the original vector

,...,a

The matrix mapping uses a simple but effective way to obtain

the spatial relationships of every PMU measurement. These

spatial relationships are stored pairwise as every entry of the 2-D

measurement matrix. By representing the spatial r elationships,

the 2-D measurement matrix is ﬁlled with implicit spatial fea-

tures which are well prepared to be captured by the subsequent

CNN module. In addition, since the difference between two mea-

surements tends to be small, such matrix mapping is conducive

to form a uniform scale of the 2-D measurement matrix, which

adapts better to the shared weights property of CNN.

2) Label Encoding: Most of the deep learning models (in-

cluding the proposed CNN-LSTM model) require all input

variables and output variables to be numerical. However, the

label data are in fact categorical rather than numerical which

cannot be directly utilized as the output of the CNN-LSTM

model. Therefore, it is necessary to encode the label data into a

numerical form to fulﬁll the output requirements. To be speciﬁc

Authorized licensed use limited to: Wuhan University. Downloaded on July 04,2022 at 03:51:10 UTC from IEEE Xplore. Restrictions apply.

剩余10页未读，继续阅读

评论收藏

内容反馈

RachelMuZy

粉丝: 117
资源: 3

态势感知+CNN-LSTM网络 基于CNN-LSTM网络的电力系统态势感知预测水平研究

大型水电站电力监控系统网络安全态势感知系统应用与研究.pdf

电力监控系统网络安全态势感知研究与建设.pdf

电力企业网络安全态势感知体系建设.pdf

智慧电力网络安全态势感知能力建设与提升.docx

基于人工智能代理的电力负荷态势感知及调控方法研究.pdf

基于LSTM的网络安全态势预测模型研究

基于CNN-LSTM融合网络的溢流早期预测深度学习方法.pdf

基于卷积神经网络-长短期记忆网络(CNN-LSTM)的多输入多输出预测，CNN-LSTM回归预测 评价指标包括:R2、MAE

基于LSTM网络预测模型的实验室开放管理策略研究.docx

基于卷积神经网络-长短期记忆网络(CNN-LSTM)多维时间序列预测，CNN-LSTM回归预测，多输入单输出模型 评价指标包

具备安全态势感知能力的安全管理平台--转

智慧电力网络安全态势感知能力建设与提升--数据驱动从传统运维转型智慧运维.pdf

【LSTM时序预测】基于贝叶斯网络优化CNN-LSTM实现股票价格预测附matlab代码+仿真结果和运行方法zip

基于贝叶斯算法(bayes)优化卷积神经网络-长短期记忆网络(CNN-LSTM)多变量时间序列预测，bayes-CNN-LSTM

基于卷积神经网络-长短期记忆网络结合注意力机制(CNN-LSTM-Attention)多变量时间序列预测，CNN-LSTM-At

基于麻雀算法(SSA)优化卷积神经网络-长短期记忆网络(CNN-LSTM)分类预测，SSA-CNN-LSTM多特征输入模型

基于PSO-LSTM的网络安全态势感知预测方.pdf

基于贝叶斯(bayes)优化卷积神经网络-长短期记忆网络(CNN-LSTM)分类预测，bayes-CNN-LSTM多特征输入模型

基于麻雀算法(SSA)优化卷积神经网络-长短期记忆网络(CNN-LSTM)回归预测，SSA-CNN-LSTM多输入单输出模型

基于卷积神经网络-长短期记忆网络CNN-LSTM多维时间序列预测，CNN-LSTM多变量回归预测，多输入单输出模型 评价指标

基于贝叶斯(bayes)优化卷积神经网络-长短期记忆网络(CNN-LSTM)回归预测，bayes-CNN-LSTM多输入单输出模

基于麻雀算法(SSA)优化卷积神经网络-长短期记忆网络(CNN-LSTM)多维时间序列预测，SSA-CNN-LSTM回归预测多输

论文研究-基于多流CNN-LSTM网络的群体情绪识别.pdf

基于鹈鹕算法(POA)优化卷积神经网络-长短期记忆网络(CNN-LSTM)回归预测，POA-CNN-LSTM多输入单输出模型

基于贝叶斯(bayes)优化卷积神经网络-长短期记忆网络(CNN-LSTM)回归预测（Matlab完整源码和数据）

基于鲸鱼算法(WOA)优化卷积神经网络-长短期记忆网络(CNN-LSTM)分类预测，WOA-CNN-LSTM多特征输入模型

完整车牌号识别程序，可以识别车牌和颜色，可以集成到项目中 支持win7+

ChatGPT教程（终极版）最全整理

博客中Kmeans以及FCM算法数据（免积分）

神经网络回归预测--气温数据集

最新资源

态势感知+CNN-LSTM网络基于CNN-LSTM网络的电力系统态势感知预测水平研究

基于卷积神经网络-长短期记忆网络(CNN-LSTM)的多输入多输出预测，CNN-LSTM回归预测评价指标包括:R2、MAE

基于卷积神经网络-长短期记忆网络(CNN-LSTM)多维时间序列预测，CNN-LSTM回归预测，多输入单输出模型评价指标包

基于卷积神经网络-长短期记忆网络CNN-LSTM多维时间序列预测，CNN-LSTM多变量回归预测，多输入单输出模型评价指标

完整车牌号识别程序，可以识别车牌和颜色，可以集成到项目中支持win7+