Decision Tree Technologies to Power System Monitoring and Security Assessment

IPASJ International Journal of Electrical Engineering (IIJEE) Web Site: http://www.ipasj.org/IIJEE/IIJEE.htm

A Publisher for Research Motivation........ Email: [email protected] Volume 4, Issue 8, August 2016 ISSN 2321-600X

Volume 4, Issue 8, August 2016 Page 15

Abstract Majority of recent large-scale blackouts have been caused by voltage instability. As all states leading to large-scale blackouts are unique, there is no “algorithm” to identify pre-emergency states. Moreover, numerical conventional methods are computationally expensive, which makes it difficult to use for the on-line security assessment. Machine learning techniques with their pattern recognition, learning capabilities and high speed of identifying the potential security boundaries can offer an alternative approach. This paper proposes a novel semi-automated method based on ensemble decision trees (DTs) learning for on-line voltage security assessment. Operating conditions are randomly generated. Multiple DTs are first trained off-line using the resampling cross-validation method. The DT learning algorithm is implemented using C4.5 decision tree, Classification and Regression Tree (CART), bagged CART, Random Forest, Extra Trees and Stochastic Gradient Descent tree. The best model is selected based on its performance. The obtained security model is used on-line to classify the system operating states based on the patterns created in the off-line simulations. If required, the final DT model can produce an alarm for triggering emergency and protection systems. A case study using the IEEE 118-bus system demonstrates the effectiveness of the proposed approach. The results showed that ensemble DT learning approach can identify potentially dangerous states with higher accuracy than other learning techniques such as neural networks and support vector machine.. Keywords: electric power system, emergency, voltage instability, machine learning, security assessment

1. INTRODUCTION Power system security assessment is one of the pressing problems in the modern power engineering. The trends towards liberalization and the need to increase electricity transmission due to growing loads and generation expansion make existing power companies operate their electrical networks in critical conditions, close to their admissible security limits [1]. In such conditions the unforeseen excess disturbances, weak connections, hidden defects of the relay protection system and automated devices, human factors as well as a great amount of other factors can cause a drop in the system security or even the development of catastrophic accidents. In recent years, voltage security problems are one of current key issues in large power systems. The main reason for this are the improvements of protection devices as well as generators speed and voltage regulators and static VAR compensators, which have increased the transient stability limits of power flows, allowing more power to be transferred over longer distances. The reactive compensation problems resulting from higher active power flows and consequently higher reactive losses have led to making the appropriate control of high voltage problematic in extreme situations, leading to voltage instabilities which have caused recent large blackouts in North America in 2003, Russia in 2005, Europe in 2003 and 2006, and India in 2012. For the time being there is a wide spectrum of approaches and tools for the security assessment. All the variety of the methods can be divided into: (1) traditional approaches based on a detailed modeling of potential disturbances in electric power systems and numerical calculations of nonlinear capacity equations [2, 3]; and (2) intelligent approaches, which involve the artificial intelligence algorithms learning on a limited set of power system states, such as artificial neural networks (ANNs), support vector machine (SVM), decision trees (DTs), etc. [1, 4, 5]. This research employs the ensemble methods on the basis of DTs for on-line voltage security assessment. Among attractive aspects of the trees, we mention their ability to uncover the intrinsic mechanism governing physical processes, and to provide a clear description in terms of tractable system parameters. As compared with other machine learning methods, in particular to ANNs and SVM, the decision tree approach produces at least as reliable classifiers. The calculations involved the different DT's such as C4.5 decision tree, Classification and Regression Tree (CART), bagged CART, Random Forest (RF), Extra Trees (ET) and Stochastic Gradient Descent tree (SGD). The effectiveness of their application is confirmed by a great number of calculations on the basis of the IEEE 118 power system. The suggested approach is implemented in the free software environment R intended for calculations with an open-source code.

Decision Tree Technologies to Power System Monitoring and Security Assessment

N. Tomin, V. Kurbatsky, A. Zhukov, D. Sidorov

Melentiev Energy Systems Institute of Siberian Branch of the Russian Academy of Sciences, Irkutsk, Russia




2. PROBLEM STATEMENT

2.1 Power system security and blackouts Security is an ability of electric power system to withstand sudden disturbances without unforeseen effects on the electricity consumers. It is provided by control capabilities of power systems. In the operational practices the required level of security can be achieved by both the preventive control actions (before a disturbance) and the emergency control actions (after disturbance). Control in the pre-emergency conditions is mainly a responsibility of the operational dispatching (security) control. At the same time there can be situations where the speed of power system control by the dispatching personnel appears to be insufficient to avoid dangerous situations. The challenge here is to identify pre-emergency conditions using enormous amounts of data with incomplete and distorted alarm patterns. As all alarm states leading to large-scale blackouts are unique, there is no “algorithm” to identify such states. The problem gets complicated by the fact that the security limit of electric power system constantly changes, therefore fast methods for real time security monitoring are required to analyze the current level of security and accurately trace the limit and detect the most vulnerable regions along it. Several studies identified voltage instability and cascade overload are the major incidents in the progression of blackouts [6]-[8]. The leading idea of the pre-emergency control concept is that the voltage instability following the emergency disturbance which accompanies many system emergencies does not develop as fast as the transient one (typically voltage collapse takes several minutes whereas electromechanical loss of synchronism takes only a few seconds). Thus, when the phase of slow emergency development comes (fig. 1), the balance between generation and consumption is maintained for a long time and this makes it possible to detect the potentially critical states after the contingency.

Figure 1 Specific stages of the system blackout development

However, the available time for emergency control is still below the limit of operator response time and most of the task should rely on automatic devices. The analysis showed that in the phase of initiating events the above-standard disturbances occur [2]. The post-emergency conditions that occur at the end of this phase are off-design for the existing emergency control devices and for the dispatching personnel. Therefore, the existing emergency control systems furnished with the up-to-date automation means and the actions Transmission System Operator may prove ineffective to prevent the subsequent catastrophic development of the emergency. Thus, the following drawbacks of the existing emergency control systems were noted:

1. Lack of emergency control systems for reliable protection against voltage collapse; 2. Low resilience of the emergency control systems; 3. Lack of adaptability and coordination of local devices; 4. Critical redundancy of primary unprocessed data for the operator.

The results of the studies testify to the necessity of the development of next-generation intelligent systems to complement modern emergency control systems, taking into account its “weak points” [9].




2.2Security monitoring and assessment techniques A rather large set of numerical methods are available for security assessment, which are based on more or less accurate analytical models of the power system. Some tools, being based on general purpose power system dynamic simulation packages [10], have a very broad scope; others are based on simplified models and approaches aiming at the representation of only those features relevant for the study of a particular subproblem. In conventional practice, security assessment is obtained by analytically modeling the network and solving load flow equations repeatedly for all the prescribed outages, one contingency at a time. This normal practice is not entirely satisfactory because the computations are lengthy and are particularly so at load values for which the system is in fact insecure against the occurrences of certain contingencies. To reduce the above computational effort of the security assessment most energy management systems use one or more security assessment predictors such as sensitivity matrix, distribution factors, fast decoupled load flows, or performance indicators to reduce the number of critical contingencies to be calculated. These analytical techniques are usually time consuming and therefore are not always suitable for real-time applications. Also, these methods suffer from the problem of misclassification or/and false alarm. Misclassification arises when an active contingency is classified as critical. A great many studies show that the effective solution to this problem can be found on the basis of machine learning methods which normally include ANNs, DTs, deep learning models, etc. [3-5, 9]. Many researchers deem that machine learning methods are indeed able to provide interesting security information for various physical problems and practical contexts. This is related to their capabilities of fast detection of the images, patterns (i.e. typical samples), learning/generalization and, which is important, high speed of identifying the instability boundaries. One of the most successful classes of machine learning methods is the ensemble learning paradigm. They make it possible to form reliable decision rules of classification for a set of potential system states. In this approach, the key idea is to build a universal classifier of power system states, which is capable of tracing dangerous pre-emergency conditions and predicting emergency situations on the basis of certain system security indices. In this case, the detection of dangerous operation patterns is not effective without considering probable disturbance/faults, whose calculation leads to a considerable increase in the computational complexity and a potential decrease in the accuracy for basic algorithms. This leads to the need of finding a way to improve the accuracy of the classifier of power system states. One of such methods is the creation of ensembles (compositions) of the classification models and their training. One of the most successful classes of machine learning methods is the ensemble learning paradigm. They make it possible to form reliable decision rules of classification for a set of potential system states. In this approach, the key idea is to build a universal classifier of power system states, which is capable of tracing dangerous pre-emergency conditions and predicting emergency situations on the basis of certain system security indices. In this case, the detection of dangerous operation patterns is not effective without considering probable disturbance/faults, whose calculation leads to a considerable increase in the computational complexity and a potential decrease in the accuracy for basic algorithms. This leads to the need of finding a way to improve the accuracy of the classifier of power system states. One of such methods is the creation of ensembles (compositions) of the classification models and their training.

3. PROPOSED METHOD The concept of an intelligent system for early detection of pre-emergency states in the electric power system as an option of the preventive operation and pre-emergency control is considered in [11]. The suggested system represents a link between the operational dispatching control and emergency control, and aims to early warn and prevent dangerous conditions and emergency situations before they lead to a large system blackout (Fig. 2).

Operational control

Emergency control

Preventive emergency

control

Prediction + manual control

Disturbance + automatic control

Prediction + preventive control

Figure 2 The proposed approach to system monitoring and control




3.1Decision tree-based technique The decision tree technique is an effective supervised data mining tool to solve the classification problems in high data dimensions (i.e. in the case of PMU data using). For a created database consisting of different cases that are represented by a vector of predictors (or variables) along with an objective, a DT is designed for successful classifications of this objective by using only a small number of these predictors. The decision tree structure is usually binary and there are two types of nodes in such a DT. For each internal node, a question or critical splitting rule is asked to decide which successor the classification process should drop into. The splitting rule could be numerical by comparing the variable value with a threshold, or categorical by checking whether the current value belongs to a specific data set. For each terminal node, a classification result is assigned in terms of the majority class of the objective, e.g., “secure” or “insecure.” Depending on the DT method applied each decision rule will be trained by its subsampling according to the bagging and boosting principles. The final decision on the classification of any power system state is made within the generalized classifier according to different principles – simple majority voting, weighted voting or by choosing the most competent decision sample rule.

3.2The proposed ensemble DTs learning approach Managing a modern grid in real time requires much more automatic monitoring and far greater interaction among human operators, emergency control systems, communications networks and data-gathering sensors that need to be deployed everywhere in power plants and substations. Therefore, the proposed decision tree-based approach is concerned with the real time identification of alarm states that are dangerous for the system security. In this paper, novel semi-automated technique based on ensemble DTs learning is proposed for on-line voltage security assessment (Fig. 3). The purpose of this work is to solve a voltage security assessment problem of a power system with an ensemble DT learning approach by means of classification. The primary principle of the approach lies in the mathematical model learning on the basis of the ensemble method of classification to automatically make a sufficiently accurate assessment of the power system conditions according to the criterion secure/unsecure on the basis of significant classification attributes of a power system state, for example active and reactive power flows, bus voltage, etc. A great amount of such attributes are obtained on the basis of randomly generated data sample consisting of a set of really possible states of electric power system [4]. In the paper, events are all generated by offline simulations using the MATLAB/ Powertrain System Analysis Toolkit (PSAT) environment. We investigated the different DT models with using a general scheme, which is presented in Fig. 3. Multiple DTs models are first trained offline using the resampling cross-validation method (Fig. 4). The DT learning algorithm is implemented using C4.5 decision tree, CART, bagged CART, Random Forest, Extra Trees and SGD tree. For each candidate tuning parameter combination, a DT model is fit to each resampled data set and is used to predict the corresponding held out samples. The resampling performance is estimated by aggregating the results of each hold-out sample set. Resampling methods try to “inject variation” in the system to approximate the model’s performance on future samples. These performance estimates are used to evaluate which combination(s) of the tuning parameters are appropriate. Once the final tuning values are assigned, the final model is refit using the entire training set. The best model from each DT technique is selected to be the candidate model with the largest accuracy or the lowest misclassification cost.

Data generation

Data collection New dataset

Performance estimator(the largest accuracy, the lowest misclassification cost, ROC)

Training and tuning (replacing methods)

DT1 model (i)◘ .

DT1 model (k)

DT1 model (i)◘ .

DT1 model (k)

DT1 model (i)◘ .

DT1 model (k)

DT1 model(bestl)

DT2 model(best)

DTn model(best)

Final DT model

Security index

Voltage, loads, power flow etc.

Feature attributes

Feature attributes

Protection/emergency control actions(reconfigure the automatic control set tings, speed-up of

protections, etc) Figure 3 A general scheme of a semi-automated ensemble DT-based technique for online power system security

assessment




Figure 4 The basic method of the proposed idea.

For the online applications, the final "best of the best" DT model is used to be the candidate technique with the best performance. As result, the system information is periodically checked and updated in order to account for changing system states as accurately as possible so that the off-line trained alarm model may continue to perform well on the new system states. The final DT also can be updated by including newly system conditions and, if required, the tuning parameters and even type of a final DT model can changed after re-checking procedure. The final DT-based model is used on-line to classify the system operating state and, if required, to produce an alarm. If the contingency occurs then alarm security model can trigger some corrective and preventive control actions. The signals coming from the DT-based security model are used to implement the following functions:

1. The “Normal state” signal (high security) selection of a criterion for security control in normal conditions. 2. The “Alarm state” signal (low security) the voltage settings are increased (banks of synchronous capacitors are switched on automatically already at rated

voltage but not at a 15 % voltage decrease); the time settings are decreased (tuned not from the settings of remote backup protection but from the settings of

basic protection). 3. The “Emergency state” signal (system instability); on-load tap changer blocking; generation shedding; system reactive power redistributing.

This paper only addresses the topic of security assessment. Therefore, the detail description of such control actions is beyond the scope of the work done in this paper.

4. RESULTS The feasibility of the approach in a proof-of-concept has been demonstrated on the IEEE 118 power system consisting of more than 118 buses, 54 generators, and 186 transmission lines (Fig. 5a). An open-source environment R with caret package is used as a computing environment for proposed ensemble learning models design and testing. Operating conditions are all generated using MATLAB/PSAT.

4.1Data base generation To obtain the data base composed of 6877 states, each of the prefault normal or heavy load states was combined with the possible disturbances (Fig. 5b). The 490 initial candidate attributes such as active and reactive power flow, voltages used to characterize the power system states. These have been simulated with a variable step the MATLAB/PSAT quasi-dynamic simulation program, which computed the attribute values and allowed us to classify based on the security index the scenarios as normal, alarm, emergency (corr.) and emergency (non-corr.).

Define sets of model parameter values to evaluate; for each parameter set do for each resampling iteration do Hold–out specific samples; Fit the model on the remainder; Predict the hold–out samples; end Calculate the average performance across hold–out predictions end Determine the optimal parameter set;




a) IEEE 118 test system b) Voltage pattern of the 240 stressed and emergency operating conditions

Figure 5 Voltage pattern of the 90 stressed and emergency operating conditions in the IEEE 118 system.

The load model was represented by static characteristics depending on voltage. When critical values of voltage are achieved the load is automatically transferred to shunts. The method of a proportional increase in load at all nodes of the test scheme was optimized for the security analysis in such a way that the initial condition for each emergency disturbance is a stable condition closest to it, from those calculated. Thus, at each stage of an increase in the IEEE test scheme load the random and dependent discrete events (primary disturbances) are modelled by the N-1 reliability principle.

4.3Estimating Performance for Classification In current paper we need to use proper performance measurement metrics for classification problems. We used the following metrics: (1) the overall accuracy of a model indicates how well the model predicts the actual data; and (2) the Kappa statistic is a measure of concordance for categorical data that measures agreement relative to what would be expected by chance. In other words, the Kappa statistic takes into account the expected error rate

4.4DT Training and Performance Ensemble and single trees methods have been built for classifying the power system states, for various candidate attributes and four different security classifications. The models were trained on 6877 samples dataset and tested on 1715 samples. Namely, the following state-of-art classification techniques were tested: C4.5 decision tree, CART, BCART, RF, ET and SGB method. For comparison purpose with other learning techniques, as such multilayer perceptron (MLP), support vector machine (SVM) and self-organized Kohonen network (SOM) were also trained and tested using the proposed method. As previously mentioned, the “optimal” model from each technique is selected to be the candidate model with the largest accuracy. If more than one tuning parameter is “best” then the function will try to choose the combination that corresponds to the least complex model. For example, for the Extra Tree technique, mtry was estimated to be 353 and numRandomCuts = 5 appears to be optimal (Fig. 6).

# Randomly Selected Predictors

Acc

urac

y (R

epea

ted

Cro

ss-V

alid

atio

n)

0.983

0.984

0.985

0.986

0.987

0.988

0.989

0 100 200 300 400

# Random Cuts12

34

5

Figure 6 The relationship between the number of Extra Tree technique components and the resampled estimate of the

area under the cross-validation.

0 50 100 150 200 2500.75

0.8

0.85

0.9

0.95

1

1.05

1.1

States

Volta

ge, p

.u.




Table 1 shows comparison of accuracy achieved by the classification learning techniques. From Table 1, the comparison indicates that Random Forest and Extra Tree models are "best" performance techniques in detecting dangerous states in the IEEE 118 test system. As footnote its to be noted that all obtained accuracy values are close. However, some of the DT models enjoy additional useful properties. For example Extra Trees needs less memory comparing with classical Random Forest, but comparable with Stohastical gradient boosting.

Table 1: Results of an automatic procedure for finding an optimal model of voltage security assessment

Security monitoring models Algorithms Metrics, %

Accuracy Kappa index

Decision trees

С4.5 98,48 97,59 CART 94,98 97,25

Random Forest 98,89 98,22 Extra Trees 98,81 98,21

Stochastic Gradient Descent 98,81 98,26

Neural network models

Kohonen network 93,38 90,07

Multi-layer perceptron 89,91 88,81 Support vector

machine Radial function 97,33 97,01

Fig. 7 shows variable importance for all classes obtained by computing of mean Gini index decrease. The classification trees select voltages under normal states as the most important attributes for security monitoring and assessment. It may be explained by the fact that the voltage sag observed in the power system state reflects proportional increase in load, when the static characteristics of the load model depend on voltage. Under alarm and emergency states, the active and reactive power flow attributes were selected in preference to voltages. A possible explanation lies in the fact that this security criterion is more preventive like. We also demonstrated the feasibility of dealing with incomplete and distorted data. Taking into consideration SCADA malfunctions, the corrupted patterns were used to train ensemble classification trees. The results showed that the test error rate did not changed even if 50% of gaps (Tab. 2).

Figure 7 Variable importance for all classes obtained by computing of mean Gini index decrease.




Table 2: Filling the gaps in data

Percent of gaps Time, s. Test error, %

10 0.0123 0.93 30 0.0411 0.93 50 0.0514 0.93

The results in this Section indicate that the best performance on the test set using the final DT model (Random Forest), the accuracy of which is 98.89%. In other words, there are 1.11% cases misclassified using this tree model. By periodically including new and unknown system states into the database, DTs are updated to learn more useful information for improving robustness and the classification accuracy can be effectively increased. The proposed semi-automated method for online security assessment should work well for all kinds of unforeseen operating conditions no matter how the critical system parameters are distributed.

5. CONCLUSIONS This paper presents a novel semi-automated method for on-line security assessment using DTs. Multiple DTs, such as Random Forest, CART, Extra Trees etc., are first trained offline using the resampling cross-validation method. Resampling the training samples allows us to know when we are making poor choices for the values of DT tuning parameters. The best model from the DT techniques is selected based on its performance. For the on-line applications, the final “the best of the best” DT is used as the candidate technique with the best performance. If required, the final DT checked and updated in order to account for new changing system states as accurately as possible. Operating conditions for the IEEE 118 system are generated to represent the stressed system states from different load system levels using MATLAB/PSAT. Power system security is analyzed using the developed security analysis tool to obtain a security label, such as "normal", "alarm", "emergency (correctable)" and "emergency (non-correctable)" for each case following a severe contingency. The results showed that the set of classification features such as active/reactive power values generated and consumed in each bus system in the pre-emergency condition, carries sufficient information concerning the reliability of the system. However, in the case of voltage instability analysis, the feature space should be extended with readings of voltages at the nodes, as well as with cross-flows of active and reactive power relations. The results showed that ensemble DT learning approach can identify potential dangerous states with higher accuracy than others single learning’s techniques such as MLP, SOM and SVM, and, if required, the final DT model can produce an alarm for triggering emergency and protection systems.

References [1] D. Panasetsky, D. Tomin, N. Voropai, V. Kurbatsky, A. Zhukov, D. Sidorov, “Development of software for

modelling decentralized intelligent systems for security monitoring and control in power systems”, In Proc. of PowerTech Conf., IEEE PES, Eindhoven, pp. 1-6, 2015.

[2] A. B. Osak and A. I. Shalaginov, “Methods for rapid analysis in the problem of security assessment based on short-term forecasting system behavior”, in Proc. of the Int. Scientific Workshop “Methodological problems in reliability of large energy systems”, Saint-Petersburg, pp. 634-643, 2014. (in Russian).

[3] Methods and models for power system reliability studies, Syktyvkar: Komi Scientific Center of Ural Branch of RAS, 2010, 292 p. (in Russian).

[4] L. Wehenkel, Machine Learning Approaches to Power System Security Assessment. PhD dissertation, University of Liege, 1995.

[5] R. Diao et al. “Decision tree-based online voltage security assessment using PMU measurements”, IEEE Trans. Power Syst., Vol. 24, No.2, pp. 832-839, 2009.

[6] IEEE PES CAMS Task Force on Understanding, Prediction, Mitigation and Restoration of Cascading Failures “Initial Review Of Methods For Cascading Failure Analysis In Electric Power Transmission Systems,” In Proc. IEEE PES General Meeting, Pittsburgh, PA USA July 2008 1.

[7] W. Lu, Y. Bésanger, E. Zama, and D. Radu, "Blackouts: Description, analysis and classification," in Proc. of the 6th WSEAS Inter. Conf. on Power Systems, Lisbon, Portugal, 2006.

[8] W.R. Lachs, "Controlling grid integrity after power system emergencies," IEEE Trans. on Power Syst., vol.17, no.2, pp.445-450, 2002




[9] N. Tomin, M. Negnevitsky, Ch. Rehtanz “Preventing Large-Scale Emergencies in Modern Power Systems: AI Approach”, Journal of Advanced Computational Intelligence and Intelligent Informatics, Vol. 18, No.5, pp. 604-614, 2014

[10] F.P. de Mello, J. W. Feltes, T. F. Laskwski, and L. J. Oppel, “Simulating fast and slow dynamic effects in power systems”, IEEE Computer applications in power 5, no. 3. (9), 1999

[11] M. Negnevitsky, N. Voropai, V. Kurbatsky, N. Tomin, and D. Panasetsky, “Development of an Intelligent System for Preventing Large-Scale Emergencies in Power Systems”, IEEE/PES General Meeting, Vancouver, BC, Canada, 2013

AUTHOR

Nikita V. Tomin is a Senior Research Fellow at the Energy Systems Institute of the Russian Academy of Science (ESI SB RAS), Irkutsk, Russia. In 2007 he defended his PhD thesis at the ESI SB RAS. His interests are intelligent state variables forecasting, power system analysis, preventive emergency control and intelligent systems applications in power systems. N.V. Tomin was visiting research fellow in TU Dortmund University (Germany), University of Tasmania (Australia). He is Vice-Chairperson of the IEEE PES Russian (Siberia) Chapter. N.V. Tomin is the author and co-author of more than 100 scientific papers.

Victor G. Kurbatsky (M’08) is Professor, Leading Research Fellow at the Energy Systems Institute of the Russian Academy of Sciences, Irkutsk, Russia. Victor Kurbatsky received his Ph.D degree at SibNIIE (Novosibirsk) in 1984 and Doctor of Technical Sciences at the Energy Systems Institute (Irkutsk) in 1997. His research interests include: electromagnetic compatibility and power quality in electric networks, application of artificial intelligence techniques in power systems. He is the author of several monographs and manuals and more than 300 scientific papers.

Denis N. Sidorov (Ph.D.'99, Dr. habil.'14) was born in Irkutsk, Russia, in 1974. He gained the Ph.D. and Dr. habil. degrees in 1999 and 2014, respectively, from Irkutsk State University, Russia. Since 2000, he worked as a Postdoctoral Research Fellow at Trinity College Dublin and Université de Technologie de Compiègne. He gained his industrial experience at ASTI Holdings Pte Ltd, Singapore. He was Visiting Professor at Tampere University of Technology, Siegen University. Since 2014, he is a Leading Researcher at Energy Systems Institute of Russian Academy of Sciences. His research interests include: DSP, power quality, inter-area oscillations, integral and differential equations theory,

machine learning methods and forecasting. Dr. Sidorov is the author of more than 120 scientific papers and two monographs.

Aleksei V. Zhukov was born in Irkutsk, Russia, in 1991. He gained the MSc degrees in 2013 from Irkutsk State University, Russia. Since 2014, he is an engineer at Energy Systems Institute of Russian Academy of Sciences. His research interests include: machine learning algorithms and methods, intelligent systems applications in power systems and time series forecasting. A.V. Zhukov is the author and co-author of more than 10 scientific papers.




Left Margin 17.8 mm (0.67") Right Margin 14.3 mm (0.56) Top Margin – 17.8 mm (0.7") Bottom Margin – 17.8 mm (0.7")

You should use Times Roman of size 10 for all fonts in the paper. Format the page as one-column: Column Width 86.8 mm (3.42") Column Height – 271.4 mm (10.69") Space/Gap between Columns - 5.0 mm (0.2").

6. TITLE, AUTHORS, BODY PARAGRAPHS, SECTIONS HEADINGS AND REFERENCES

3.1 Title and authors The title of the paper is centered 17.8 mm (0.67") below the top of the page in 24 point font. Right below the title (separated by single line spacing) are the names of the authors. The font size for the authors is 11pt. Author affiliations shall be in 9 pt.

3.2 Body paragraphs The main text for your paragraphs should be 10pt font. All body paragraphs (except the beginning of a section/sub-section) should have the first line indented about 3.6 mm (0.14").

3.2.1 Figures and Tables Place illustrations (figures, tables, drawings, and photographs) throughout the paper at the places where they are first discussed in the text, rather than at the end of the paper. Number illustrations sequentially (but number tables separately). Place the illustration numbers and caption under the illustration in 10 pt font. Do not allow illustrations to extend into the margins or the gap between columns (except 2-column illustrations may cross the gap). If your figure has two parts, include the labels “(a)” and “(b)”.

Figure 1 Testing data- load current (amperes)

3.2.2 Tables Place table titles above the tables.

Table 1: Margin specifications

Margin A4 Paper US Letter Paper

Left 18.5 mm 14.5 mm (0.58 in)

Right 18mm 13 mm (0.51 in)




3.2.3 Sections headings Section headings come in several varieties:

1. first level headings: 1. Heading 1 2. second level: 1.2. Heading 2 3. third level: 1.2.3 Heading 3 4. forth level: (a) Heading 4 5. fifth level: (1) Heading 5

References Number citations consecutively in square brackets [1]. The sentence punctuation follows the brackets [2]. Multiple references [2], [3] are each numbered with separate brackets [1]–[3]. Please note that the references at the end of this document are in the preferred referencing style. Please ensure that the provided references are complete with all the details and also cited inside the manuscript (example: page numbers, year of publication, publisher’s name etc.).

Equations If you are using Word, use either the Microsoft Equation Editor or the MathType add-on (http://www.mathtype.com) for equations in your paper (Insert | Object | Create New | Microsoft Equation or MathType Equation). “Float over text” should not be selected. Number equations consecutively with equation numbers in parentheses flush with the right margin, as in (1). First use the equation editor to create the equation. Then select the “Equation” markup style. Press the tab key and write the equation number in parentheses.

K

kopk

P

pE

12)(

1 (1)

Other recommendations

Equalize the length of your columns on the last page. If you are using Word, proceed as follows: Insert/Break/Continuous.

References [12] A. Bonnaccorsi, “On the Relationship between Firm Size and Export Intensity,” Journal of International Business

Studies, XXIII (4), pp. 605-635, 1992. (journal style) [13] R. Caves, Multinational Enterprise and Economic Analysis, Cambridge University Press, Cambridge, 1982. (book

style) [14] M. Clerc, “The Swarm and the Queen: Towards a Deterministic and Adaptive Particle Swarm Optimization,” In

Proceedings of the IEEE Congress on Evolutionary Computation (CEC), pp. 1951-1957, 1999. (conference style) [15] H.H. Crokell, “Specialization and International Competitiveness,” in Managing the Multinational Subsidiary, H.

Etemad and L. S, Sulude (eds.), Croom-Helm, London, 1986. (book chapter style) [16] K. Deb, S. Agrawal, A. Pratab, T. Meyarivan, “A Fast Elitist Non-dominated Sorting Genetic Algorithms for

Multiobjective Optimization: NSGA II,” KanGAL report 200001, Indian Institute of Technology, Kanpur, India, 2000. (technical report style)

[17] J. Geralds, "Sega Ends Production of Dreamcast," vnunet.com, para. 2, Jan. 31, 2001. [Online]. Available: http://nl1.vnunet.com/news/1116995. [Accessed: Sept. 12, 2004]. (General Internet site)

AUTHOR Taro Denshi received the B.S. and M.S. degrees in Electrical Engineering from Shibaura Institute of Technology in 1997 and 1999, respectively. During 1997-1999, he stayed in Communications Research Laboratory (CRL), Ministry of Posts and Telecommunications of Japan to study digital beam forming antennas, mobile satellite communication systems, and wireless access network using stratospheric platforms. He now with DDI Tokyo Pocket Telephone, Inc.

Documents

Decision Tree Technologies to Power System Monitoring and Security Assessment