REQUIREMENTS OF STATE ESTIMATION IN SMART DISTRIBUTION GRID

Transcription

1 3 rd International Conference on Electricity Distriution Lyon, 5-8 June 05 Paper 09 REQUIREMENTS OF STATE ESTIMATION IN SMART DISTRIBUTION GRID Anggoro PRIMADIANTO Wei Ting LIN David HUANG Chan-Nan LU NSYSU - Taiwan NSYSU - Taiwan NSYSU - Taiwan NSYSU - Taiwan anggoro.primadianto@gmail.com lwtnsysu@gmail.com choits84@gmail.com cnl@ee.nsysu.edu.tw ABSTRACT Advanced analytical applications for gloal controls and optimization of distriution systems have not een widely adopted y utilities ecause of lacing real time complete system models. As more Intelligent Electronic Devices (IED) with two-way communications are eing employed, the amount of quasi real time data gathered at different rates y various automation systems is increasing. With effective integration of data from IED, smart meters, feeder and sustation automation systems, a series of steady state snapshots distriution state estimations (DSE) will e a ey feature to enale real-time optimization, adaptive protection and control, pricing signal, demand response, and many other smart grid features. This paper riefly reviews the state of the art of DSE, focusing on the requirements for smart distriution grid applications, the effects of distriution networ characteristics and ad data detection capaility. INTRODUCTION Utilities have faced many changes in distriution networ. There are more distriuted generators (DG), Plug-in Electric Vehicle (PEV), IED, and phasor surement unit (PMU) in distriution system. The introduction of DG is in accordance with one of the smart grid visions presented y European Commission Community Research, to provide connection access to all networ users, particularly for renewale power sources and high efficiency local generation with zero or low caron emissions []. Distriution system states have ecome more dynamic due to the integrations of many intermittent distriuted generation and loads. Due to these changes, utility companies have to control and protect the idirectional power flow networ, to improve power quality and may have to defer investment ecause of environmental concerns. In order to deal with these issues, a near real time distriution networ model ecomes necessary for more resilient control. SCADA systems have een installed in distriution networ to monitor us voltage, ranch power flow, and other parameters at certain nodes in the distriution feeders. Economic constraint maes it impossile to install the surement devices in every place where it is needed. Additionally, surement data are suject to error due to the nature of surement device and communication prolem. Recent widespread deployment of advanced metering infrastructure (AMI) and sensors in the distriution networ provides the utilities with a etter insight aout networ conditions as well as its customer ehaviour []. With adequate integrations of heterogeneous data, DSE could provide a solution to further increase the system model accuracy. DSE is the first step to efficient system operations. Bad data detection is important for effective state estimation (SE), which detects the existence of gross errors in the surement data, identifies and eliminates them. DSE enales real-time optimization, adaptive protection and control, pricing signal, demand response, and many other smart grid features. DSE results provide more accurate information for spatial load forecast that can e used in networ planning. Many efforts have een made to design DSE that is comparale to well estalished SE techniques used in the transmission system control centre. However, SE methods used in transmission system are not directly applicale to the distriution networ due to different characteristics of the two systems. The distriution system has the following characteristics: ) High r/x ratio ) Three phase unalance system 3) Radial topology 4) Limited amount of real time surements In order to assist effective system operation, DSE needs to deliver accurate distriution model and handles all the aforementioned distriution grid characteristics efficiently. Most of the proposed DSE algorithms are ased on weighted least square (WLS) approaches that differ mainly in the choice of state variales and surements used. Bus voltage magnitude and phase angle (polar form) were used as state variales in [3] while us injection current surement ased formulation with us voltages expressed in rectangular form was adopted in [4]. Baran and Kelley introduced a ranch current ased DSE algorithm in [5] that uses ranch current in rectangular form as state variales. This method further refined in [6] y Wang and Schulz. Some alternative techniques have een presented as well. Ghosh et al. [7] proposed a proailistic approach that treats the real surements as solution constraints. A load estimator algorithm that simplifies the networ into several surement areas was proposed in [8]. In [9], Naa et al. proposed a DSE algorithm ased on particle swarm optimization. This paper presents a comparative investigation of DSE etween ranch current and us voltage ased WLS methods. Comparison of required DSE features such as solution accuracy, computation time, and ad data detection capaility in different networ topologies are investigated as well as the effects of the inclusion of voltage surements in the estimation. DISTRIBUTION STATE ESTIMATOR The relation etween a surement ( z) and the state variales (x) can e expressed as follow: z = h(x) + r () CIRED 05 /5

2 3 rd International Conference on Electricity Distriution Lyon, 5-8 June 05 Paper 09 Where h(x) is the function relating the surement to the state variale vector x. r is the surement residual. The estimated system state is otained y minimizing the following ojective function: min J(x) = m i= w i (z i (h i (x)) x = [z h(x)] T W[z h(x)] () - W is the surement weight matrix with W ii =. ii is ii the standard deviation of surement i. The following normal equation is solved iteratively to compute the state variale updates and x + = x + x. [G(x )] x = H T (x ) W [z h(x )] (3) Where G(x ) = H T (x ) W H(x ). H(x ) is the surement function Jacoian matrix. To include zero injection surements, equality constraints can e included y adding a plenty term to (). Two existing DSE algorithms compared in this study are descried in the followings. Bus Voltage Based DSE In [3], a state estimation method using three phase feeder model with us voltage magnitude and phase angle as state variales was proposed. The method is ased on the WLS approach and can handle power, voltage, and current surements. Another WLS state estimator using us voltage in rectangular form as state variales was proposed in [4]. The rectangular us voltage algorithm uses a us injection current surement formulation in which the power surements are converted to their equivalent current surements. The entries of surement Jacoian matrix ecome constant and equal to the admittance matrix elements. Voltage magnitude surements are converted to their equivalent rectangular form using phase angle information otained from the calculated voltage value, such that the corresponding surement Jacoian terms are either zero or unity. The following equations are used to convert the power surements into their equivalent current surements. Branch power flow: V cal ) I m = ( (P+jQ) m = Re(I m ) + jim(i m ) (4) Where I m is the equivalent ranch current surement from us to us m. (P + jq) m is the cal power flow surement at ranch m. V is the estimated us voltage at us. Bus power injection: I (P+jQ) = ( cal ) V = Re(I ) + jim(i ) (5) is the equivalent us injection current surement of actual power injection surement at us, (P + jq). I Voltage magnitude: V = V ( V cal V cal ) = Re(V ) + jim(v ) (6) V is equivalent us voltage surement. Branch Current Based DSE In [5], a popular ranch current ased DSE method for radial and wealy meshed distriution feeders, using rectangular form ranch current as state variales was introduced. Equations (4) and (5) are used to convert the power surements into equivalent current surements which are calculated at each iteration allowing the corresponding power surement s Jacoian matrix entries to e either unity or zero. The current ased algorithm was improved further in [] that treats the current surements more efficiently and nonlinear Jacoian terms are avoided. After eing neglected in the early development of current ased DSE, voltage surement was incorporated to the algorithm in [] and [3]. Wang and Schulz [6] proposed a revised current ased SE, using the magnitude and phase angle of ranch current as state variales. In addition, meter placement issue was also discussed. It was shown in [6] that among all surement types, voltage magnitude surement is the least effective surement to enhance accuracy of the current ased DSE. Weight Transformation In aove mentioned rectangular us voltage ased and ranch current ased DSE methods, actual surements are transformed into their equivalent surements. According to [4], the equivalent surement variance can e calculated as follows: m y = ( F ) z ii i i= = ( F ) z + ( F ) z ( F ) z mm (7) m where y = F(z, z,, z m ) is the surement equivalent function. Assuming the surements to e independent with each other and applying equation (4)-(6) into equation (7), the equivalent variance are shown in Tale. CIRED 05 /5

3 3 rd International Conference on Electricity Distriution Lyon, 5-8 June 05 Paper 09 Tale. Calculation of surement variances (P + jq) I (P + jq) m I m V V eqv r = Variance (R eqv = eqv ) e f (e + f ) P + (e + f ) Q f eqv x = (e + f ) P + ( e ) (e + f ) Q eqv r = e f (e + f ) P + (e + f ) Q f eqv x = (e + f ) P + ( e ) (e + f ) Q eqv r = e e + f V, eqv x = f e + f V Where e and f are the real and imaginary parts of estimated voltage at us. For every equivalent surement, eqv r and eqv x are the equivalent variances of its real and imaginary parts. CASE STUDY The performance of existing polar node voltage ased (NVP), rectangular node voltage ased (NVR), and rectangular ranch current ased (BCR) WLS DSE formulations are compared. The performance of DSE methods is evaluated y using Taiwan Power Company (TPC) 3-phase unalance 39 us test system. The line parameter is j 0.0 Ω/m Figure. TPC 39 us test system Three surement types are included in the tests. Distriution automation systems provide real-time surements containing us voltage and feeder ranch real and reactive power surements. Zero injections are considered as virtual surements and treated as equality constraints. Two types of pseudo-surements are considered. Pseudo-surements ased on AMI data that include voltage magnitude and us power injection surements. In the simulations, surement s standard deviation ( i ) is calculated as follow: i = e i 3 00 where e i is the percentage of surement error. Performance Indices In order to compare the accuracy and execution time of different methods,,000 samples with random error ased on normal distriution are generated for each test case. Root n square errors (RMSE) are calculated and used for accuracy comparison. RMSE = n i= (v i v ti ) (9) n where v i is estimated voltage magnitude and phase angle at us i, v ti is the true voltage value otained from load flow solution. n is the numer of uses. Test Results Test aims to sure the level of accuracy and execution time of different DSE methods. In addition to the performance comparison, effectiveness of AMI ased power injection and voltage magnitude pseudosurements are also investigated. It is assumed that us power injection data ased on customer meter readings are availale every 5 minutes and therefore have less error compared to those calculated ased on monthly illing data in conjunction with typical load profiles. Voltage magnitude surements are also availale from the AMI readings. Tale shows the surement types, locations and standard deviations of different test cases. Tale. Test Cases in Test Case No. 3 Type Location PQ flow (FTU) -, PQ inj (Pseudo s.) all load uses 0. PQ flow (FTU) -, PQ inj (AMI) all load uses 0.05 PQ flow (FTU) -, PQ inj (AMI) all load uses 0.05 V (AMI) all load uses 0.05 Figures to 4 show the RMSE and execution times of tested methods. Among them NVP and NVR ased methods have higher solution accuracy than BCR ased method, particularly on the estimated voltage magnitude. On the other hand, BCR and NVR ased methods outperform NVP ased method in the execution time. From Figures and 3, it is also evident that the accuracy has increased with the introduction of power injection and voltage magnitude from AMI. (8) CIRED 05 3/5

4 second second p.u. p.u. degree 3 rd International Conference on Electricity Distriution Lyon, 5-8 June 05 Paper Figure. RMSE of us voltage angle estimation (Test ).40E-04.0E-04.00E-04.80E-04.60E-04.40E-04.0E-04.00E-04 Figure 3. RMSE of us voltage magnitude (Test ) Figure 4. Average execution time (Test ) Test investigates the effect of using variance transformation for the surements used in NVR and BCR ased methods. Test case uses the original surement variance as the weight in the calculation. While test case 4 uses variance transformation formula presented in Tale to find the surement weights. Tale 3. Test Cases in Test Case No. 4 Type Location Weight Matrix PQ flow (FTU) -, 3-4 inj all load (Pseudo s.) uses ii PQ flow (FTU) -, 3-4 Variance PQ inj all load Transformation (Pseudo s.) uses (Tale ) Fig. 5 shows the accuracy comparison with and without variance transformation in surement conversion. It can e seen that variance recalculation is required in order to improve the station accuracy. The cost to the increase of estimation accuracy is highlighted in Fig. 6. The execution time is increased due to the recalculation of weight matrix and the gain matrix in each iteration. In this study, weight matrix is recalculated only in the first and second iteration. Average numers of iterations required in the NVR and BCR ased methods to reach the final solution are and 3 respectively..6e-04.4e-04.e-04.0e-04.8e-04.6e-04.4e-04.e-04.0e-04 NVR Case I BCR Figure 5. RMSE of us voltage magnitude (Test ) NVR Case I Case IV Case IV BCR Figure 6. Average execution time (Test ) Test 3 investigates the difference in ad data detection for radial and wealy meshed feeder topologies. For this purpose, a loop has een introduced in the test feeder y adding a new.7 m ranch etween us and us 3 of CIRED 05 4/5

5 3 rd International Conference on Electricity Distriution Lyon, 5-8 June 05 Paper 09 the TPC test system shown in Fig.. Bad data was introduced to surement PQ 6 y replacing it with 300% of its actual value. NVR ased method with weight transformation is used in this test. Wealy meshed networ (Test case 6) has higher surement redundancy as compared to the radial networ (Test case 5). Tale 4 shows that the NVR ased DSE can correctly locate the ad surement data in the wealy meshed networ ut it failed to do so in radial networ, despite the relatively high error used in this test. Bad data detection is ased on normalized residuals. The result suggests that ad data detection is one of the crucial challenges in the DSE that needs to e addressed since most of the distriution feeders are radial. Tale 4. Bad Data Detection Test PQ flow (=0.0) PQ inj (=0.05) Feeder Topology Bad Data Case 5 Case 6 -, -5, -4, 4-4, 4-4 Radial Largest r N Location PQ CONCLUDING REMARKS All load uses Wealy meshed PQ 6 (300% error) PQ 6 This paper presents a comparison etween three existing WLS ased DSE formulations, namely NVP, NVR, and BCR ased methods. Based on the results otained from 000 simulations in each test case, it is found that NVP and NVR ased methods deliver etter solutions. However, the average execution time of NVP is higher than those of NVR and BCR ased methods, maing it less attractive for distriution networ applications since distriution networ has much higher numer of usses. The use of AMI data in DSE provides a more accurate solution which is not only due to the lower us power injection surements error, ut also the inclusion of voltage magnitude surements. Test results also suggest that the application of surement variance transformation in NVR and BCR ased methods would improve the solution accuracy, although it requires additional computation and results in a longer execution time. Bad data detection is one of the important issues in DSE. More roust technique needs to e developed to deal with ad surement data in oth radial and wealy meshed distriution networs. REFERENCES [] Community Research of European Commision, European Technology Platform SmartGrids, Office for Official Pulications, 006. [] GTM Research, "The Soft Grid 03-00: Big Data & Utility Analytics for Smart Grid," 03. [3] M. E. Baran and A. W. Kelley, "State Estimation for Real Time Monitoring of Distriution System," IEEE Transactions on Power Systems, vol. 9, no. 3, pp , 994. [4] C. N. Lu, J. H. Teng and W. H. E. Liu, "Distriution System State Estimation," IEEE Transactions on Power Systems, vol. 0, no., pp. 9-40, 995. [5] M. E. Baran and A. W. Kelley, "A Branch Current Based State Estimation Method for Distriution Systems," IEEE Transaction on Power Systems, vol. 0, no., pp , 995. [6] H. Wang and N. N. Schulz, "A Revised Branch Current Based Distriution System State Estimation Algorithm and Meter Placement Impact," IEEE Transactions on Power Systems, vol. 9, no., pp. 07-3, 004. [7] A. K. Ghosh, D. L. Lueman, M. J. Downey and R. H. Jones, "Distriution Circuit State Estimation Using Proailistic Approach," IEEE Transactions on Power Systems, vol., no., 997. [8] I. Dzafic, M. Gilles, R. A. Jar, B. C. Pal and S. Henselmeyer, "Real Time Estimation of Loads in Radial and Unsymmetrical Three-Phase Distriution Networs," IEEE Transactions on Power Systems, vol. 8, no. 4, 03. [9] S. Naa, T. Genji, T. Yura and Y. Fuuyama, "A Hyrid Particle Swarm Optimization for Distriution State Estimation," IEEE Transactions on Power Systems, vol. 8, no., 003. [0] W.-M. Lin and J.-H. Teng, "State Estimation for Distriution Systems with Zero-Injection Constraints," IEEE Transactions on Power Systems, vol., no., pp , 996. [] W. M. Lin and J. H. Teng, "A Highly Efficient Algorithm in Treating Current Measurements for the Branch Current Based Distriution State Estimation," IEEE Transactions on Power Delivery, vol. 6, no. 3, pp , 00. [] M. E. Baran, J. Jaesung and T. E. McDermott, "Including Voltage Measurements in Branch Current State Estimation for Distriution Systems," Power & Energy Society General Meeting, 009. [3] J. H. Teng, "Using Voltage Measurement to Improve the Result of Branch Current Based State Estimator for Distriution Systems," Generation, Transmission and Distriution, vol. 49, no. 6, 00. [4] Hui Li, Ming-hao Yang, "A ranch-current-ased state estimation for distriution systems nonsurement loads," Power Engineering Society General Meeting, 004. CIRED 05 5/5