Definition. Purpose. Definition SET-1520 Sähköenergiatekniikan uudet sovellukset. State estimator Lecture 1 - PDF

7.. SET-5 Sähköenergiatekniikan uudet sovellukset Definition State estimator Lecture Distribution Network State Estimation Antti Mutanen Tampere University of Technology 7.. Program for calculating the

Please download to get full document.

View again

of 10
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.

Entertainment & Media

Publish on:

Views: 16 | Pages: 10

Extension: PDF | Download: 0

7.. SET-5 Sähköenergiatekniikan uudet sovellukset Definition State estimator Lecture Distribution Network State Estimation Antti Mutanen Tampere University of Technology 7.. Program for calculating the state estimates State estimator forms a comprehensive view of the state of the network by combining measurements, laws of electric networks and network models Estimator incorporates a computer-implemented mathematical model that simulates the real physical network as accurately as possible. o Contains a detailed network model (network topology, line impedances etc.) o Complies with the laws of electric networks o Does load flow calculation and optimization o Optimization is done to find the most likely state of the network given the available measurements Definition Purpose What is state estimation? State estimation is a methodology that provides the best possible approximation for the state of the system by processing the available information In power systems, the states are Voltage magnitudes and angles Current magnitudes and angles Power flows Circuit breaker statuses outputs Why state estimation is needed? To estimate unmeasured variables To improve overall accuracy To detect bad measurements To detect invalid topological information Available information is Network model (network topology, line lengths and impedances) Measurements (voltage, current and power measurements) Circuit breaker status information inputs 7.. Motivation 5 Existing distribution network state estimators 7 Why state estimation is needed in smart grids? One central purpose of smart grid is to maximize the utilization degree of electricity networks and electricity production capacity by leveraging the latest information technology, two-way communication and system intelligence operational margins become smaller. State estimation is an important part of any network control system o In order to control the network, you need to know what is the state of the network. In smart grids, the amount of distribution automation and active control increases o Active distribution network management functions such as voltage level management, control of distributed generation, reactive power regulation, feeder reconfiguration and restoration, and demand side management require accurate real-time estimates of network voltages and line flows. Especially the increase of distributed generation is an important driver for state estimation development. The current state estimation methods used in distribution network are relatively simple because: The amount of measurements is small Distribution networks are radial in nature o In radial networks, load flows can be calculated using simple backward/forward sweep method instead of the more complex iterative Newton Raphson method used for meshed networks. Backward/forward sweep method a) Sum currents using Kirchhoff s current law b) Calculate voltage losses and voltages using Ohm s law Motivation 6 Existing distribution network state estimators 8 Benefits of Smart Grids Smart grids not only require more accurate state estimation but they also enable it. Smart grids bring more measurements to distribution networks o Remotely readable measurements are being installed in disconnector stations, distributed generators, distribution substations and even on customer connection points. Basic state estimation. Use load profiles to get initial load estimates. Calculate load flow using load estimates and network model st level state estimate. Calculate the difference between estimated and measured feeder load flows. Divide the difference to the load estimates in relation to their variances 5. Recalculate the load flow Figure. Basic flow chart for distribution network state estimation. 7.. Example 9 Limitations in existing distribution network state estimators st level state estimate P feeder =69 kw Difference to measurements P diff = 5 kw-69 kw = 8 kw Divide estimation error to load points in relation to load estimate variances ( ) Restricted to radial networks Handles correctly only power measurements When current measurements have to be converted to power measurements by making assumptions on voltage and power factor Voltage measurements can be used only in forward sweep phase to enhance voltage estimates downstream from the measurement New state estimation methods are needed to fully benefit from the additional measurements being added to the networks With modern computers, more complex and more accurate state estimation methods can used Example can be taken from the transmission network state estimation. Example (continues) Transmission system state estimation Now, if we recalculate load flow with these corrected load values we get P feeder =57 kw 5 kw We could iterate calculation until convergence is achieved or we could calculate the differential coefficient of the network losses for each load node (k i ) and take increased losses into account directly. Power system state estimation has long roots in transmission network side Mathematical theory was formulated during the 97 s Has been in limited use since the 98 s Became popular during the 99 s Nowadays, is a vital part of any transmission system control centre Formulation Voltage magnitudes and angles are used as state variables o All other variables (current and power flows) can be calculated from the state variables Now, if we recalculate load flow we get P feeder =5.7kW 7.. Transmission system state estimation Weighted least squares (WLS) estimation 5 Transmission system state estimators have some problems when applied to distribution networks Decoupling of real and reactive powers can not be used to speed up computation because assumption X R is not valid for distribution networks Transmission system state estimators have problems handling current measurements, which are common in distribution networks Solution: select current as state variable instead of voltage With current based state estimation: State equations are simpler Computation is faster and more robust Current measurements are easier to handle Compared to traditional transmission system state estimation Weighted least squares (WLS) estimation Measurement location simulations: simulation network 6 Weighted least squares estimation is commonly used in transmission system state estimation Minimizes the sum of the squares of the differences between estimated and measured values Each measurement is weighted with the measurements accuracy i.e. accurate measurements are trusted more than inaccurate measurements IEEE,8 kv:n test network, 6 nodes, load,5 MW /, MVAr, length,5 km 8..8 7.. Measurement location simulations (/) 7 Alternative to additional measurements 9 Adding current measurements along the feeder is an effective way to improve state estimation accuracy best locations for current measurements are at the beginning of highly load branches Jännitteen Mean keskivirhe error (%) (%) Voltage estimation error Jännitevirhe No Ei measurements mittauksia x x current virtamittaus meas Solmupiste Node Instead of adding measurements, state estimation accuracy can also be improved by improving the load profile accuracy State estimation accuracy is linearly dependent on the pseudo measurement (load profile) accuracy The adjacent figure shows how state estimation accuracy improves when load profiling errors (RSD) decrease Average voltage estimation error (%) Voltage b) Pseudo estimation measurements error Base case -5 % RSD -5 % RSD Node Measurement location simulations (/) 8 Conclusions In WLS methods, voltage measurements can used to improve not only voltage estimates but also current and power flow estimates Best locations for current measurements are at the branch ends Jännitteen Mean keskivirhe error (%) (%) Voltage estimation error Jännitevirhe No Ei measurements mittauksia % accuracy tarkkuus % accuracy tarkkuus,5 %% accuracy tarkkuus, %% accuracy tarkkuus Solmupiste Node Accurate state estimates are needed in smart grids as the amount of distribution automation and active control increases and becomes more complex Active distribution network management functions such as voltage level management, control of distributed generation, reactive power regulation, feeder reconfiguration and restoration, and demand side management all require accurate real-time estimates of network voltages and line flows. Smart grids increase the amount of measurements in distribution networks and thus smart grid enables more accurate state estimation New state estimation methods are emerging to handle the additional measurements and non-radial networks. The emerging methods use o Weighted least squares estimation o Branch currents as state variables 7// SET-5 Sähköenergiatekniikan uudet sovellukset Introduction (/) Lecture Load Modelling Antti Mutanen Tampere University of Technology 7.. Finland has a long history in load profiling. Finnish electric utilities started to co-operate in load research during the 98 s and in 99 Finnish Electricity Association (nowadays Sener) published customer class load profiles for 6 different customer classes. 8 load profiles for housing 8 load profiles for agriculture 9 load profiles for industry load profiles for public and private services Each load profile contains expectation and standard deviation values for every hour of the year Load profiles are presented either as topographies or as index series Kuorman odotusarvo expectation Keskihajonta std Introduction (/) Defects in current load profiles (/) Since load profiles have been available for a long time, Finnish network companies have come up with several applications utilizing load profiles. Load profiles are used extensively in distribution network calculation, for example in: Load flow calculation Planning calculation State estimation Tariff planning. Each individual customer is linked to one predefined customer class load profile by using the information from customer information system (CIS). CISs contain all the available information on each customer s electrical connection, type and electricity consumption. All the customers are linked to the geographic network model in the network information system (NIS). This enables network calculations using the load profiles. The existing load profiles have several defects.. They are very old. During the past years electricity consumption has experienced significant changes, the amount of heat pumps and airconditioners has multiplied, the use of entertainment electronics has increased and electricity consumption in recreational dwellings has changed. Furthermore, in the future, the changes will be even bigger if plug-in hybrids and customer-specific distributed generation become popular.. Sampling errors. Small sample sizes were used when calculating current load profiles.. Geographical generalization. Load profiles are defined in a national level. Some of the accuracy is lost due to geographical generalization and within-country differences in electricity consumption are left unmodelled. 7// Defects in current load profiles (/) 5 AMR and load profiling 6. Customer classification. DNOs have limited information on the type of the customers. The type of the customer is usually determined through a questionnaire when the electricity connection is contracted. However, the customer type may later change for instance because of a change in the heating solution. 5. Outliers. Some customers may have such an exceptional behaviour that they do not fit in any of the predefined customer class load profiles. 6. Insufficient information about load temperature dependencies. New load profiles are needed to make network calculation more accurate! Previously, load profiling required expensive and time-consuming load research projects, but now automatic meter reading (AMR) is providing huge amounts of information on electricity consumption. All the previously mentioned problems could be solved with the help of AMR measurements. The customer classification and load profiling could be done according to actual consumption data. Since AMR data is collected continuously, the classification and load profiles would remain up-to-date at all times. The classification and accuracy of the load profiles could be checked automatically for instance once a year. The load profiles could also be calculated separately for each DNO or region, thus avoiding the errors caused by geographical generalization. Outliers could be detected and individual load profiles could be formed for the outliers. Individual load profiles could also be calculated for some of the largest customers to improve the load estimation accuracy. Load profiling procedure 7 Temperature dependency calculation (/) 8 When forming load profiles from AMR measurements, the following steps are taken: Data validation, editing and estimation (VEE) Temperature dependency calculation and normalization Dimension reduction Outlier filtering Data based customer classification (clustering) o Customers are grouped into groups that behave similarly o Clustering methods like k-means, fuzzy k-means, ISODATA, hierarchical clustering and Gaussian mixture models may be used. Calculation of load profiles o Expectation values and standard deviation are calculated for each hour of the year. Outdoor temperature has a significant effect on the electricity consumption, especially when electric heating is presents. Measured load for a single customer with electric heating. Figure shows loads and temperatures for week in years 5 and 6. P( t) ( Tave E T ( t) ) E P( t) where P(t) is the outdoor temperature dependent part of the load P at time t, T ave is the outdoor temperature (daily average), E[T(t)] is the expectation value of the outdoor temperature at time t (long-term daily average temperature), is the temperature dependency parameter [%/ C] and E[P(t)] is the expectation value of the load at time t. Load (W) 8 6 Load 6 Load 5 Electric load vs. temperature Temperature 6 Temperature Temperature (C) Time (h) 7// Temperature dependency calculation (/) 9 Significance of correlation Calculating temperature dependency parameters from AMR-data Temperature dependency parameters must be calculated for at least seasons. Linear regression is used to calculate temperature dependency parameters. Daily energy consumption fluctuates according to weekday and month. This fluctuation can be taken into account by selecting the regressand and regressor as follows: The correlation is significant if the value t in equation below is larger than the value of t picked from t-distribution with a chosen probability level. The reference value for t is picked from one tailed t-distribution table with N degrees of freedom. Dependent variable (regressand): the percent error between the daily energy consumption and the average daily energy consumption on a similar day (same day of the week and month). Determining variable (regressor): difference between the daily average temperature and the average temperature on a similar day. The significance of relationship between the daily energy and outdoor temperature can be assessed with the correlation coefficient and the Student s t- test. E (%) T ( C) Example Question: If sample size is, how large must the correlation coefficient r be to make the relationship between two variables significant? Solution: First, choose the significance level. For example p=.5. The reference value for t is looked from one tailed t-distribution table with 8 degrees of freedom and significance level p=.5. We get t =,7. Now, we can insert t and N into above equation and solve the limits for correlation coefficient. We get that the relationship is significant, if r .6 or r .6 Outlier filtering Dimension reduction Some customers have a very unusual behaviour, their consumption pattern does not match with any other customer there is no point trying to cluster these, instead they should be filtered from the data set before clustering Outlier filtering can be made according to statistical rules Fixed or varying threshold for maximum distance can be set Distance metrics: Euclidian distance power (p.u.) Monthly outlier Since the next phase (clustering) is very compute-intensive it is often beneficial to reduce the dimension of the data before clustering Methods for dimension reduction include: Principal component analysis (PCA) coefficient Selitysasteet of determination yksittäin Individual Self organizing maps (SOMs) kumulatiivinen cumulative Feature vectors Pattern vectors % time (h) pääkomponentit Principal components 7// K-means example Example results from AMR data clustering. User set up the number of clusters they d like. (e.g. k=5). Randomly guess K cluster Center locations. Each data point finds out which Center it s closest to. (Thus each Center owns a set of data points). Each centre finds the centroid of the points it owns 5. and jumps there 7 customers from public service customer group, clustered into 5 subgroups (only first week of January is shown in figures). Cluster 5 5 n k=5 Cluster Cluster n k= 5 5 n k= Cluster 5 Cluster n k= 5 5 n k= 6. Repeat until convergence (no more changes) Method comparison 5 Method comparison 6 Case. domestic customers Case. industrial customers a) Clustering vs. Individual load profiles b) AMR data based clustering vs. Grouping according to CIS information (industry class codes) 7// Conclusions 7 Load profiling accuracy can be improved with AMR measurements. The benefits are: With AMR measurements, it is possible to update load profiles and keep them up-to-date at all times No more sampling errors since all the customers are measured Area or DNO specific load profiles can be formed instead of national load profiles Customer classification can be done according to actual customer behaviour Individual load profiles can be formed for outliers and large customers Temperature dependency parameters can be calculated. 5
Related Search
Similar documents
View more...
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks

We need your sign to support Project to invent "SMART AND CONTROLLABLE REFLECTIVE BALLOONS" to cover the Sun and Save Our Earth.

More details...

Sign Now!

We are very appreciated for your Prompt Action!