Professional Documents
Culture Documents
Power System Operation and Control A Data-Driven Approach
Power System Operation and Control A Data-Driven Approach
Power System Operation and Control A Data-Driven Approach
Data-Driven Approach
Parvaiz Ahmad Ahangar Shameem Ahmad lone Neeraj Gupta
2022 International Conference on Intelligent Controller and Computing for Smart Power (ICICCSP) | 978-1-6654-7258-6/22/$31.00 ©2022 IEEE | DOI: 10.1109/ICICCSP53532.2022.9862029
Abstract—Renewable energy is becoming more popular around These environmentally friendly power sources serve as a
the world, especially wind power and solar photovoltaic (SPV) replacement for existing electric power systems by acting as a
systems with data interfaces and IoT sensors that generate number of small-scale power generation sources with storage
significant volumes of data. In addition to serving as a monitoring
device, the data provided by such devices can be used to improve technologies. These DERs are dispatchable sources that can
system reliability and efficiency by providing real-time data. be connected to a traditional distribution system. Nonetheless,
When compared to traditional model-based operation, data- their integration has a significant impact on current utility
driven based optimal renewable power operation is an emerging power systems and attracts attention for its seamless operation
method for ensuring trouble-free power system operation. The [3]. The power system operation is described as the state
data-driven method is effective for studying the impact of
rapid distributed generation systems integration on utility power of the power system network in a given period, such as
system functioning. In data-driven approach, Machine-learning a day, an hour, or a snapshot, as determined by generator
(ML) is an emerging technology for addressing the optimal func- output, load demand, transmission topology, and power flow.
tioning power system networks. Data-driven operated distributed These patterns have a significant impact on the operation and
energy resources (DER) provide real-time management of our planning of electricity generation systems. Because renewable
dependable power supply through suitable forecasting methods
and hence give rise to the smart grid idea. In this proposed work, energy is stochastic in nature and is dependent on a various
our objective is to apply a data-driven based strategy to the smart factors such as weather and household load, it poses a number
grid in order to ensure the smooth operation and control of both of issues in terms of power system management and planning
utility and renewable-rich power system. and hence their accurate forecasting is required.
Index Terms—Distributed generation system, solar power, data A lot of research has been done in the past, but it is mostly
driven technology, machine learning, support vector machine, based on traditional and empirical experience, which did not
Generative adversarial network, battery energy storage. Load produce precise results. [4]. According to the literature review,
forecasting. the stochastic optimal planning model of MG, such as [1]
has not taken into account the randomness of solar and load.
I. I NTRODUCTION Similarly, in [2] has tackled power system stability evaluation
Modern power grids are equipped with many digitalization using traditional methods, but it falls short of addressing data-
technologies and encompass different Distributed energy re- driven ML techniques and the emerging Io T technology. Un-
sources (DERs), such as solar panels, wind farms, and energy like hydro power facilities, which have seasonal fluctuations,
storage. owing to utilization of sensors and data measuring stochastic renewable energy makes balancing generation and
units can generate significant volumes of data [1], [2]. Tra- demand difficult. As a result, by performing accurate load and
ditional power systems are currently under severe strain as generation forecasting analysis and constructing appropriate
a result of high consumer demand for energy, and they are dynamic models based on such analysis using ML and Io
emitting significant volumes of carbon dioxide, such as in T, this difficulty can be mitigated. As smart devices function
thermal power plants that use coal as an input, rendering our across the whole network, a vast amount of data is generated
climate unfit for human habitation. Since we do not have super and shared across grid components. AI and ML-based algo-
thermal power plants in our country, we rely on traditional rithms are now effective in overcoming the challenges posed
power plants for roughly 65 percent of our electricity demand. by traditional approaches such as how to detect faults and
There is a growing demand for renewable energy to reduce how to optimize the operation of an interconnected power
our reliance on traditional power plants and to protect the systems to achieve higher reliability and efficiency [5], [6].
environment, and this demand is supposed to increase at As illustrated in Fig. 1, there are various types of machine
a quicker rate due to strong demand for clean electricity. learning techniques accessible today, such as supervised and
unsupervised learning, which may be further classified into
classification, regression, and clustering.
This paper emphasizes on novel technology and improvement
978-1-6654-7258-6/22/$31.00 ©2022 IEEE over traditional methods and hence a data-driven based optimal
Authorized licensed use limited to: VISVESVARAYA NATIONAL INSTITUTE OF TECHNOLOGY. Downloaded on March 30,2023 at 10:50:45 UTC from IEEE Xplore. Restrictions apply.
CLASSIFICATION Data
Go next
Capturing
SUPERVISED
LEARNING
REGRESSION
MACHINE LEARNING Study the Train ML Evaluate
UNSUPERVISED problem Algorithm Solution
LEARNING
CLUSTERING
Analyze
Errors
Fig. 1. Classification of Machine Learning techniques. Fig. 2. Typical illustration of Machine learning based approach.
Authorized licensed use limited to: VISVESVARAYA NATIONAL INSTITUTE OF TECHNOLOGY. Downloaded on March 30,2023 at 10:50:45 UTC from IEEE Xplore. Restrictions apply.
Seperate signal from Use training ML model the collected data can be trained and analyzed using learning
Weather data noise, correlate data for real-time analysis
Use ML model to train IoT data
Prediction
algorithms and a response model can be developed which can
Energy Market Preprocessing Training/Modelling
Visualization
help us in knowing the necessary parameters of power system.
Financial Data clean & Contexualize
data
Through forecasting analysis based on data-driven approach,
Optimization
Activity data
we can mitigate system contingencies and thereby stability can
Real Time
ML algorithm be improved by taking prior steps required for keeping power
bank
IoT Data
system frequency and voltage intact and hence avoiding large
frequency and voltage excursions.
Fig. 3. Overview of data-driven based power system operation. III. DATA - DRIVEN BASED POWER SYSTEM DYNAMIC
MODELS
tained through measuring instruments and AMI and a suitable Power grid topology
Solar and wind generation Load on Buses
Simulation
simulation model can be developed from where we can get Conventional Load
Solar and Wind data
information about various operation modes. Day to Day Operation
Curtailment
Authorized licensed use limited to: VISVESVARAYA NATIONAL INSTITUTE OF TECHNOLOGY. Downloaded on March 30,2023 at 10:50:45 UTC from IEEE Xplore. Restrictions apply.
Researchers are working tirelessly in removing barriers erected huge computation, which can bring several challenges as
by conventional power systems. They are simpler to put into they make training extremely slow, also they make it difficult
practise when it comes to their application and are working as to achieve a better solution [22], [24]. Therefore, dimension
” assistants” in power system areas as a decision tool. Machine reduction is vital for keeping only required information and
learning can be segmented into supervised, unsupervised, and eliminating the redundancy.
reinforcement learning same is indicated in figure 01. Among
the machine learning algorithms, supervised learning is applied Reducing dimensionality may filter out the noise present and
in areas where we have labelled data. Whereas unsupervised unnecessary details thereby making training faster. Moreover,
learning did not have labelled data and works on its own dimensionality reduction assists in visualizing the data on
intelligence. Similarly reinforcement learning works on obser- 2D/3D plots and hence important insights can be achieved by
vations based approach. In order to fit data points better during analyzing these patterns. Different algorithms work for dimen-
training process, Neural networks have been found to be sion reduction like: PCA, t-SNE and LDA. t-SNE (Distributed
effective at fitting non-linear relationships, which are prevalent stochastic neighbour embedding) takes a high dimension data
in renewable energy because the data used for training in set and reduces it to low dimensional graph and retains a lot
unsupervised learning is not labelled, we must first use an of original information [25]. It does so by giving every data
unsupervised method to find an appropriate clustering solution. point a location in 2D or 3D map.
The trained data for real-time power systems is typically high- 3) Dimension reduction: Dimension reduction is second
dimensional, highly correlated, and difficult to comprehend. type of unsupervised learning problem. Reducing the dimen-
Furthermore, when renewable penetration is large, the number sions of data not only allows us to compress the data for
of operation patterns increases. As a result, the algorithm saving computer memory or disk space, but it also allows us to
adopted must be capable of handling sophisticated data anal- speed up our learning algorithms. We can do dimensionality
ysis. In general, the following steps are followed to complete reduction using 2D to 1D transformation without losing the
this activity: required features.
• Preprocessing 4) Visualization: Second application of dimensionality re-
• Clustering duction is to visualize the data. For a lot of machine learning
• Dimension reduction applications, it really helps us to develop effective learning
• Visualization algorithms, for understanding our data better, there is some
1) Preprocessing: Each operation mode is a complicated way of visualizing the data better, and so, dimensionality
vector associated with irrelevant data, starting with the pre- reduction offers us, often, another useful tool to do so. Let’s
processing stage. Because there may be some garbage data say we have collected a large volume of data set of may
owing to smart metre failures and measurement errors, the generating units and connected consumers in a power system
preprocessing stage is called to squeeze the information and network. The different features/attributes can be represented
remove the worthless data in order to improve computing as:
efficiency. Principal component analysis (PCA) algorithm is X1 = Total generation of desired location
generally suggested for data compression analysis wherein a X2 = Total load demand........
matrix is selected. For instance, Let K= (k1 k2....Km)N × N X3 = Solar generation.........
represents the incoming matrix of operation modes, where X4 = Wind generation..........
column matrix represents operation mode vector, M = no. of • and so on.. and suppose we may have huge data say 50
operation modes, where each operation mode has N Dimen- attributes like X ∈ R50 where each generating unit can
sions. In preprocessing only valuable data is processed and be represented as X (i) ∈ R50 it turns out that when we
garbage data is removed. look at output of dimensionality reduction algorithm, it
2) Clustering: Clustering is considered as the most im- does not astride physical meaning to new features.
portant tool in unsupervised machine learning for making
grouping of data based on similar attributes. It dissociates V. O PTIMIZING RENEWABLE POWER AND ENERGY
complexity of data sets based on their similarity into smaller STORAGE
groups with the result, the data required for making appro- Data-driven enabled ML and optimization techniques can
priate prediction is reduced. In clustering phase wherein we be applied in entire power system operation, such as electric
have an unlabelled data set and we need an algorithm which power data analytics, energy forecasting, Renewable scenario,
can intuitively group the data into synchronous subsets or optimal ESS, MGs and distribution networks [26]. Present-day
into identical groups. K- Means algorithm is regarded as power systems encounter numerous exigencies owing to fast
more suitable clustering algorithm. Moreover, Partitioning and accommodation of RESs. Meanwhile environmental impact
hierarchical methods are the two main clustering tools. assessment(EIA) forecasts that by 2050 renewables will
K Means is an iterative algorithm and it does two things. contribute half of world electricity. Since their output widely
First is a cluster assignment step, and second is a move depends on weather condition and hence their prediction and
centroid step. For finding the informative and useful patterns control are essential for grid reliability. Thanks to massively
under massive operation scenario of power system, we require deployed sensors and big energy data, ML along with Deep
Authorized licensed use limited to: VISVESVARAYA NATIONAL INSTITUTE OF TECHNOLOGY. Downloaded on March 30,2023 at 10:50:45 UTC from IEEE Xplore. Restrictions apply.
Learning is being purely implemented to predict Renewable
Generation and electric load demands. The perfect forecasting
Load
of PV and wind power is also of vital importance for making Solar PV
Converter Wind turbine
strategic development in renewable power areas [27], [28]. Converter
Authorized licensed use limited to: VISVESVARAYA NATIONAL INSTITUTE OF TECHNOLOGY. Downloaded on March 30,2023 at 10:50:45 UTC from IEEE Xplore. Restrictions apply.
ML technique is to exploit the available data using suitable [11] Miettinen, J. Holttinen, H. Ammala, J. Piironen, M. Wind power
algorithms. Due to availability of high data through advanced forecasting at Transmission System Operator’s control room. In Pro-
ceedings of the 2015 IEEE Power Energy Society General Meeting,
metering infrastructure (AMI), ML techniques give excellent Denver, CO, USA, 26–30 July 2015; pp. 1–5.
results and here we do not need explicit system models and [12] J. De La Ree, V. Centeno, J. S. Thorp, and A. G. Phadke,” Syn-
control stability can be ensured. To avoid repeating manual and chronized phasor measurement applications in power systems,” IEEE
Trans.smart Grid, vol. 1, no.1, pp.20-27, jun.2010.
traditional work in power grids, Auto ML is a fascinating so- [13] R. C. Green, L. Wang, and M. Alam, ” Applications and trends of
lution, which aims at automatically implementing the process high performance computing for electric systems: focussing on smart
of applying ML through an ML model. ML being one of the grid,” IEEE Trans. smart grid, vol. 4, no.2,pp. 922-931, jun.2013.
[14] P. Kunder et al.,”definition and classification of power system stabil-
most exciting recent technologies enable Learning algorithms ity,” IEEE Trans. power syst., vol. 19, no.3, pp.1387-1401, Aug.2004.
to mimic how the human brain learns. Currently, ML has so [15] N. G. Baltas, P. Mazidi, J. Ma, F. de Asis Fernandez, and P.
disseminated due to growth of web and growth of automation Rodriguez,” A comparative analysis of decision trees, support vector
machines and artificial neural networks for on-line transient stability
due to which we have large volumes of data than ever before assessment,” in proc. Int. Conf. Smart Energy Syst. Technol., Sevilla,
in all fields of engineering including power system. Normally, Spain, Sep. 2018, PP.1-6
we use different types of learning algorithms based on their [16] Li, F. TWang, Q. ang, Y. Xu, Y. Integrating Model-Driven and Data-
Driven Methods for Power System Frequency Stability Assessment
application. In our proposed case study, following algorithms and Control. IEEE Trans. Power Syst. 2019, 34, 4557–4568. [Cross-
are proposed and likely to be implemented: Ref]
[17] Nazih, A. S., Fawwaz, E., Osama M, A. (2011). Medium-term electric
• Linear regression load forecasting using multivariable linear and nonlinear regression.
• Support vector machine (SVM) Smart Grid and Renewable Energy, 2011.
• Decision trees [18] L. Zhu, C. Lu, Z. Y. Dong, and C. Hong,” Imbalance learn-
ing machine- based power system short-term voltage stability as-
• K-means clustering sessment,” IEEE Trans.Ind.Inform., vol.13, no.5, pp. 2533-2543,
• Random forest Oct.2017.
[19] Y. Zhang, Y. Xu, and Z. Y. Dong,” Robust Ensemble data analytics
In future work, we will extend this technique to distribution for incomplete PMU measurements-based power system stability
side of power system for better operation and control of system assessment,” IEEE Trans. Power Syst., vol. 33, no.1, pp. 1124-1126,
Jan.2018.
network as it is the distribution side in a power system which [20] C. Ren, Y. Xu, and Y. zhang, ”Post-disturbance transient stability as-
encounters frequent frequency and voltage excursions. The sessment of power system towards optimal accuracy-speed tradeoff,”
proposed work will be carried out in MATLAB and PYTHON Protection control modern power system., vol.3, no.1, pp. 194-203,
Jun.2018.
environment. [21] L. Duchense, E. Karangelos, and L. Wehenkel,” Recent Developments
in machine learning for energy systems reliability management,” Proc.
IEEE, in Press.
R EFERENCES [22] D. R. Gurusinghe and A. D. Rajapakse, ” Post disturbance transient
stability status prediction using synchrophasor measurements,” IEEE
[1] An optimal multivariable constrained nonlinear (MVCNL) stochastic Trans.power syst. vol. 31, no. 5, pp. 3654-3664, sep-2018.
microgrid planning and operation problem with renewable penetration [23] C. A. Jensen, M. A. El-Sharkawi, and R. J. Marks,” Power system
author Kumar, Maneesh and Tyagi, Barjeev, journal IEEE Systems security assessment using neural networks: feature selection using
Journal, volume 14, number 3, pages 4143–4154, year 2020, publisher fisher discrimination,” IEEE Trans. power system., vol. 16, no. 4, pp.
IEEE. 757-763, Nov. 2001.
[2] Haq, M. R., Ni, Z. (2019). A new hybrid model for short-term [24] Z. Chen, X. Xiao, C. Li, Y. Zhang, and Q. Hu,” Real time transient
electricity load forecasting. IEEE Access, 7, 125413-125423. stability status prediction using cost-sensitive extreme learning ma-
[3] J. L. Sawin, F. Sverrisson, J. Rutovitz, S. Dwyer, S. Teske, H. E. chine,” neural comput. Appl., vol. 27, no. 2, pp. 321-331 Feb.2016.
Murdock, R. Adib, F. Guerra, L. H. Blanning, and V. Hamirwasia, [25] L. Zheng, W. Hu, Y. Zhou, Y. Min, X. Xu, C. Wang, and R. Yu,” Deep
“Renewables 2018-global status report,” REN21-Renewables Now, belief network based non-linear representation learning for transient
Paris, France, Tech. Rep. REN21, 2019. stability assessment,” in Proc. IEEE Power energy soc. Gen. Meeting,
[4] J. Tan and Y. Zhang, ”Coordinated Control Strategy of a Battery Chicago, IL, USA, Jul. 2018, pp.1-5.
Energy Storage System to Support a Wind Power Plant Providing [26] F. Shi, X. Hu, Z. Yu, Y. Huang, and G. Lu,” Association rules analysis
Multi-Timescale Frequency Ancillary Services,” IEEE Transactions between power system operating states and transient stability margin,”
on Sustainable Energy, vol. 8, no. 3, pp. 1140-1153, Jul. 2017. in Proc. IEEE Inovative smart grid Technol., Asia, Auckland, New
[5] V. Gevorgian and B. O’Neill, ”Advanced grid-friendly controls Zealand, Dec. 2017, pp. 1-5.
demonstration project for utilitys cale PV power plants,” National [27] M.Alajmi, S.Aljahdali, S.Alsaheel, M. Fattah and M.Alshehri,” Ma-
Renewable Energy Lab.(NREL), Golden, CO (United States), Jan. chine learning as an efficient diagnostic tool for abnormality detection
2016. and localisation in SPV” EPiC Series in computing, Vol.63, pp.21-33,
[6] B. Kroposki et al.,” achieving a 100% Renewable grid: operating elec- 2019.
tric power systems with extremely high levels of variable renewable [28] M. Sun, J. Cremer, and G. Strbac, ” A Novel Data-Driven Scenario
energy,” IEEE Power energy Mag., vol.15,no.2,pp.61-73, Mar. 2017. generation framework for transmission expansion planning with high
[7] E.Ela, M.Milligan, and B.Kirby, ”Operating reserves and variable renewable penetration, ” Appl. energy, vol. 228, pp. 546-555, Oct.
generation,” Nat.Renew.Energy Lab., Golden, CO, USA, Aug 2011. 2018.
[8] P. Prakash and D. K. Khatod,”optimal siting and sizing techniques for [29] A. Bagheri, J. Wang, and C. Zhao, ”Data-driven stochastic transmis-
distributed generation in distribution systems: A Review,” Renewable sion expansion planning,” IEEE Trans. Power Syst., vol. 32, no. 5,pp.
sustain energy Rev., vol. 57,pp. 111-130, 2016. 3461-3470, sep. 2017.
[9] A.Ehsan and Q.Yang,”optimal integration and planning of renewable [30] S. Marsland, Machine Learning: An Algorithm Perspective, 2nd ed,
distributed generation in the power distribution networks: A Review CRC Press, Boca Raton, FL, 2015.
of analytical techniques,” Appl. energ, vol. 210,pp. 44-49, 2018.
[10] A. G. Exposito, A. J. Conejo, and C. Canizares, Electric energy
systems: Analysis and operation. Boca Raton, FL,USA: CRC Press,
2016.
Authorized licensed use limited to: VISVESVARAYA NATIONAL INSTITUTE OF TECHNOLOGY. Downloaded on March 30,2023 at 10:50:45 UTC from IEEE Xplore. Restrictions apply.