Mathematical models to estimate the quality of monitoring software systems for electrical substations

Transcription

1 Mathematical models to estimate the quality of monitoring software systems for electrical substations MIHAIELA ILIESCU 1, VICTOR URSIANU 2, FLORICA MOLDOVEANU 2, RADU URSIANU 2, EMILIANA URSIANU 3 1 Faculty of Engineering and Management of the Technological Systems 2 Faculty of Automatic Control and Computers University POLITEHNICA of Bucharest Splaiul Independentei nr. 313, Bucharest ROMANIA 3 Institute of Mathematical Statistics and Applied Mathematics Romanian Academy Calea 13 Septembrie nr. 13, Bucharest ROMANIA [email protected]; [email protected] Abstract: General aspects regarding software quality are discussed. The analysis of quality aspects is made for a software monitoring system of an electrical substation. Mathematical models are proposed to ensure software quality including Bayesian networks, Markov chains. Using a monitoring system developed in Romania as case study, we proposed modeling this system using SCL language and Bayesian networks in order to estimate its reliability. Some software modules were developed in order to implement the mathematical models proposed. Key-Words: - software quality, software quality assurance, IEC61850, Smart Grid, energy field, Bayesian networks, Markov chains, fault prediction, reliability, substation monitoring, IED. 1. Introduction The quality of a software product is given by the ability to be used efficiently, effectively and comfortably by certain users for a set of goals in the specified conditions. Ensuring the software quality is very important in order to have a reliable product with as few failures and errors as possible. In the last years there is an increasing number of electrical substations that are entirely monitored and controlled using software systems. The quality problem of monitoring software systems from the energy field is of great issue. The national and international economic development depends on the power field. There is a common assent of the specialists that the power field and the future Smart Grid cannot work without monitoring the electrical equipment within substations. In the worst case, the situation may generate a power unbalance in a certain geographical area and cause important material damage for consumers. 2. Analysis of Quality Aspects for a Monitoring System of an Electrical Substation 2.1. General architecture A monitoring system of an electrical substation requires the existence of electronic devices that monitor the electrical equipment parameters - IED (Intelligent Electronic Device), one or more computers (servers) that acquire monitored data and save it into a database and one or more computers (clients) that run a client software which displays the monitored parameters. Generally, such a system has a client-server infrastructure (Fig. 1). The server software applications receive in realtime the values of monitored parameters for electrical equipment from the IEDs, process these values and then save it into a database. The client software application acquires this data from the database and presents the information to the user (the operating personnel from the electrical substation). ISBN:

2 Client Computer (control room) Server Computer LAN IED1.. IEDn Fig. 1 The general architecture 2.2. Problems that may cause system failures The problems that may cause a malfunction of such a monitoring system of an electrical substation may vary, from the delay of data transmission, wrong data display, to the failure of one component or of the whole system. The information obtained with the help of the monitoring system is useful to local and national energy dispatcher. Each delay in pointing out a fault of the monitored electrical equipment may affect the energy balance in a certain geographical area. 3. Case Study: EMCSIT system In the next paragraphs there will be presented a case study regarding EMCSIT system 3.1. The particularities of the system EMCSIT is a complex online system that monitors an electrical substation developed entirely in Romania [1]. The EMCSIT system has a client - server architecture and consists of IEDs (Intelligent Electronic Device) placed in every relay cabin inside the electrical substation. They are connected to sensors and transducers, mounted on the electrical equipment of the substation and transmit the received data to local servers. The monitoring system includes computers (referred as local servers), which are connected to the monitoring equipment (IEDs), receive the values of the monitored parameters, process these values and save the results in a central Oracle database server. The user can view all these data by using the EMCSIT client-software (Fig. 2) from the local network. Fig. 2. The EMCSIT Station software The EMCSIT SuperServer application runs one instance of server-type software for every IED (not visibile for the user), by using more execution threads. At specific periods of time (for example every one minute), the EMCSIT SuperServer interrogates all the IEDs that are connected to the local server and receives the data regarding the monitored parameters of electrical equipment. After the data package from the IED has been received, it is processed and saved in the database installed onto the central server of every substation inside the component of the electrical substation. During the system development, the number of faults detected in every software module inside the system was gathered and the failure rate of every system component was determined Testing the EMCSIT system In order to test the software before putting into operation, we can generate input data (like those acquired from an IED) and send it to the server software. The server will process the values and save into the database. After that, the client software will acquire the values and present the equipment state according to these values. For testing the server software, the simulator generates test data using technical specifications of the monitored parameters for the electrical equipment: minimum value, maximum value and the pre-alarm value. Using these limits, we know what equipment state corresponds to a set of input values and is easy to verify if the server software performs correct. Generating input data leads to many test cases: the test vectors represent combinations of selected values for input data. ISBN:

3 The Pairwise testing [2] ensures that every possible combination among the selected values for every pair of input variables is tested at least once. Empirical studies have shown that using this method may help detecting approximately 70% of all the faults in the software. The simulator generates test cases via the Pairwise testing method, and sends these data to the server software. The time interval when the test data packages are sent and the number of analogue inputs can both be modified; the simulator can generate data packages for 16 analogue inputs maximum. By using the log files generated by both the simulator and the server software, correlations can be made in order to observe whether all the test data has been received and to observe the state of equipment for these values. This way, the accuracy of calculations made in the server software can be verified. 4. IEC Markov chain and Bayesian network 4.1. SCL Language The most recent and important standard that defines communication in electrical substations is the IEC61850 standard. Substation Configuration Language is the language and representation format specified by IEC61850 for describing the equipment configuration of an electrical substation. The complete representation of data in the SCL language enables the possibility of interoperability between equipment from an electrical substation via file exchange. The great advantage of using IEC61850 is the interoperability between IEDs, which belong to different manufacturers. The language has a standard naming convention for data, equipment is automatically described via the SCL language, it allows virtual modeling for logical equipment and offers a common language for configuration of equipment The mathematical model of Markov chain The idea of modelling the monitoring system using a Markov chain is based on the fact that future states depend on the present states and are independent of the ones from the past. Based on our experience in developing and using monitoring software systems, such a system should be associated at a certain point with one of the following states: 1. Good State (S1): overall, the system works according to specifications; 2. Acceptable State (S2): there are some missing data packages (they are not received by the server software) in certain time intervals; the errors that appear sporadically do not call for intervention by the substation personnel; 3. Bad State (S3): the system is unavailable most of the time; data packages are often missed, delay in displaying the acquired data; the user has to interfere in order to restart one or more software systems; it may become necessary to restart the client computer, one or more IEDs; 4. Unacceptable State (S4): the majority or even all system components are not functioning or are not functioning according to the specifications; the system is no longer available; In order to estimate the system s current state, we can use: - Information from log files recorded by the software monitoring system; - Information from the log files recorded by the database server; - Information regarding system resources used by servers, etc. The proposed method for estimating the current state and for estimating the transition time between states is based on the analysis of log files. In the case of the EMCSIT monitoring system used as case study, the log files created by the server software contain data about the receiving date of every data package, the ID of the package, whether the package was valid or not and the state of the monitored electrical equipment. By analyzing the log file, it can be observed whether there are missing packages between those that were supposed to be received at pre-established moments of time. By observing the system behavior for a certain period of time, we discovered that transition from one state to another is connected with the number of lost data packages (not received). Therefore, we propose that the system can be classified in four discrete states: If less than one data package per day is missed (not received by the server software), then the system is in the S1 state; If no more than one data package is missed every six hours or an average of four data packages are missed per day, then the system is in the S2 state. The system is also in this state if maximum two data packages are missed consecutively; ISBN:

4 If there are maximum 24 missed data packages per day (an average of one data package every hour) or more than three consecutively missed packages, then the system is in the S3 state; In worse situations than the aforementioned ones, the system is in the S4 state. The Markov model can be used to estimate transition times between the states of the monitoring system, respectively when the system will reach the unacceptable state [3]. If two intermediate states (between the good one and the unacceptable one) are introduced in this Markov model, it can help the development team to promptly intervene for fixing the errors, before the system reaches an unacceptable state. If a software application is in state (S i ), the passage time between state S i and S j state can be calculated - FPT ij. There are several ways to calculate first passage times - FPT. One of the methods studied is the state combination method (isj) presented in fig. 3. This method combines all intermediate states (from initial state i to the final state j) in one state (S) and calculate the transition rate from the initial state and the final state j noted FPTij. To do so, the following transition rates are used: λ is and λ Sj. Transition rates are calculated as number of cases when the system went from state i to state j divided by the total number of cases the system has changed its state. The (λ ij ) and (λ ji ) rates are obtained from calculations using initial modeling of the states, before combining the intermediary states into S state. Transition rates in (S i ) din (S) are calculated using the state combination method using the following formulas: λ = (1) λ is λ iz Si = p zλzi p z (2) λ = (3) λ js λ jz Sj = pzλzj p z (4) where z S is a state, and p z is the probability that the system should be in state z. The probability of being in a certain state is calculated for this system as the number of cases in which the software has been in a particular state divided to the total number of cases registered. For FPT ij calculation, where FPT ij = (average time) the transition time from state (i) to state (j) the following formula is used: λsi + λsj + λis ( FPT) ij = (5) λ λ + λ λ + λ λ Si ij Sj ij After (FPT) ij is computed, the result is obtained in the same time unit as the input data that was collected. For example, (FPT) ij = 0,3 with i=1 and j=3 using input data from 4 months of tests, then assuming the system is initially in state S1 (good state) and according to the Markov modeling, it will reach state S3 (bad state) in 0,3 months. The advantage of calculating transition times between system states was proven in different domains including software engineering. Considering that a specific software application runs without problems, using the information regarding defect rates collected during test period, the moment of time when the system will fail can be estimated with a certain degree of confidence. Sj is Fig. 3. State representation for FPT calculation 4.3. Bayesian networks A Bayesian network (BN) is a directed graph without cycles representing probabilistic relationships between the variables of an array. These variables are represented as nodes: {X1, X2,,Xn} and the relations between them are represented by arcs: parent node is the variable that determines the change of the child variable; child node is the variable that is influenced by the parent variable; ISBN:

5 The intensity and the way in which the influences between variables show, are given by the conditional probability table of each node separately. This allows us to calculate probability values for the random variable associated to the node, depending on the values for parents. Each node has a conditional probability table associated with it. The conditional probabilities are based on the information from the past. A conditional probability is mathematically written as P (x p1, p2,, pn) and it is likely that X variable is in x state if the parent P1 is in p1 state, P2 is in p2 state,, respectively Pn is in pn state. The conditional probabilities are based on the information from the past. A conditional probability is mathematically written as P (x p1, p2,, pn) and it is likely that X variable is in x state if the parent P1 is in p1 state, P2 is in p2 state,, respectively Pn is in pn state. In the case of a BN (Fig. 4), in order to calculate the fault probability of the parent nodes marked with Ai, Bayes formula is applied for the calculation of the conditioned probability: where n is the node that directly influence the node i, i=1,...,n. The probability that the child node B defects is: (6) (7) The BNs can estimate the probability of errors for a software system based on the history of the fault data for the constituent modules. Our purpose is to model a monitoring system for an electrical substation through a BN [4]. This will allow us to calculate the fault probability of each node and for the entire system. For the calculations regarding the fault probability of the nodes (IEDs) and of the monitoring system, information regarding the number of faults must be acquired on a certain period of time for each IED. Each child node is characterized by a certain fault rate (number of faults in a given period of time). The BN (as an orientated graph) supports any kind of relationship between the nodes. Depending on the relationships between the nodes, the fault probability of the node is calculated [6], [7].. 5. Developed software 5.1. Diagrams The Diagrams application was developed in order to create virtual entities in a substation that can be later used for Bayesian network modeling. The user can connect different types of entities (electrical equipment), save the configuration in XML files or load other substation configurations. Each entity has specific attributes such as: name, voltage level, type (transformer, circuit breaker, disconnector, etc.), neighbors, physical coordinates, etc. (Fig. 5).Equipment can be lately integrated in electrical cells. This feature along with the possibility to define neighbors is of great help in modeling the electrical substation using Bayesian networks Parser SCL and BN Calculus In order to describe an electrical substation according to IEC61850 standard, the SCL (Substation Configuration Language) language, Visual SCL application and the self-developed Diagrams application were used [5], [8]. Fig. 4. Example of a Bayesian network (BN) Fig. 5. Diagrams application ISBN:

6 Fig. 6. Define the node types for the BN Visual SCL application allows the user to save the substation schema in the XML standard format type specific to IEC For the purpose of using the information from SCD files, Parser SCL software module was developed. It reads the information from the SCD files written in the configuration language of an electrical substation - SCL and delivers as output, a data structure that can be later used to defined the nodes of a Bayesian network (BN). The output data is later used by the BN Calculus module for estimating system's reliability (Fig. 6). The nodes and fault information regarding IEDs will be used for estimating the fault probability of the entire monitoring system. The information obtained (electrical equipment and the associated IEDs) is used for the mathematical model of the BN associated to the monitoring system. Connections between the IEDs and the electrical equipment must be made in the BN Calculus module. After that, the user must complete the fault data for each IED. When this data is not available for a certain IED, the fault probability can be set manually. Finally, the software module calculates the fault probability of the monitoring system as well as the a-posteriori fault probability of each IED and also of the subsystems (if being defined). In the case study, these subsystems are represented by local servers from relay cabins. 6. Conclusion Mathematical models like Markov chains, Bayesian networks are used for estimating the reliability of monitoring systems for electrical substations. Using as case study, a system that was developed in Romania and installed in electrical substations (EMCSIT), a Bayesian network is modelled using information from SCL file (IEC61850). The failure probability of the system is calculated using software applications developed by the authors. The obtained results were used by the developer of EMCSIT system to improve its hardware components and software modules. In the future, we intend to develop a software application that implements the Markov theory presented in this article in order to predict the failure time based on the log files from data recorded in substations. References: [1] C. Moldoveanu, V. Ursianu, M. Avramescu and others, Intelligent electronic system for continuous monitoring and diagnostic of high voltage substations, International Conference on Condition Monitoring, Diagnosis and Maintenance, Bucharest, [2] K. Naik, P. Tripathy, Software testing and Quality assurance, Wiley Ed., [3] C. Moldoveanu, R. Ursianu, E. Ursianu, V. Ursianu and others, Determine the optimal moments for investigating technical state of primary equipments for the purpose of assuring the safety levels imputed by the National Company Transelectrica România, International Conference on Condition Monitoring and Diagnosis, Beijing, [4] V. Ursianu, R. Ursianu, E. Ursianu, Bayesian Networks to predict Software Quality, SPSR Conference of Statistics and Probabilistics Romanian Society, Bucharest, [5] Visual SCL, Applied Systems Engineering website: [6] V. Ursianu, Quality assurance solutions for electrical substations monitoring and control software systems, Ph.D. Thesis, [7] M. Iliescu, V. Ursianu, Fl. Moldoveanu, M. Avramescu, R. Ursianu, Quality Assurance Solutions for Monitoring and Control Software Systems of Electrical Substations, WSEAS International Conference on Data Networks, Communications, Computers, Malta, [8]. Popescu. D., Amza C.G., Laptoiu D., Amza G., Competitive Hopfield Neural Network Model for Evaluating Pedicle Screw Placement Accuracy, Strojniški vestnik - Journal of Mechanical Engineering, 58(2012)9, ISSN ISBN: