Detection and identification of bad data in AC/DC hybrid systems with LCC and MMC

Based on the CIM/XML and CIM/E documents exported from the regional dispatching system, this paper focuses on data generation and starts by converting the exported documents into raw input data for state estimation. Considering the interactions between the AC system, LCC, MMC, and LCC-MMC interfaces, a unified iterative method is proposed to model the state estimation of the 500 kV subnetwork. Subsequently, Gaussian noise is added to the original measurement data, and the maximum residual test method is employed for detecting and identifying bad data. Finally, the effectiveness of the proposed models for AC/DC state estimation and the detection and identification of bad data are validated through simulation data.


Introduction
With the extensive integration of clean energy sources such as wind power, nuclear power, and hydrogen energy into the grid, LCC-HVDC and MMC-HVDC have become important means of energy transmission due to their advantages of low loss, asynchronous interconnection, and strong controllability [1].The hybrid AC/DC grid represents a new stage in the evolution of power grids, and accurate measurement and monitoring are essential tasks to ensure the reliable operation of the power system.The detection and identification of bad data can timely discover and diagnose faults, improve the stability of the grid, optimize grid operation, and enhance network security.
The traditional Maximum Normalized Residual (LNR) method was improved by integrating the Gaussian Mixture Model (GMM) algorithm, reducing the time required for bad data detection and identification, and improving the robustness and estimation accuracy of the algorithm [2].However, the data dimension, parameter quality, and model complexity under large-scale network Gaussian mixture models may lead to a series of problems.In the context of ultra-large-scale AC power systems, the LNR test was applied to group suspicious measurements, thereby improving the efficiency of bad data identification [3].While maintaining accuracy, this method significantly reduces processing time.However, its applicability to hybrid AC/DC grids with LCC-HVDC and MMC-HVDC needs further verification.Bad data detection algorithms from the perspective of false data injection in hybrid AC/DC systems are analyzed, providing insights into the state estimation of such systems [4][5].
This paper takes CIM/XML and CIM/E documents exported by a regional dispatch system as a starting point.It successfully converts CIM/XML documents into data in a specified format, models

CIM/XML data conversion
The topology parameters and measurement data of the AC/DC system need to be provided for state estimation, which are included in the CIM/XML and CIM/E documents exported by the scheduling system.Through Python programming to achieve document parsing, topology conversion, and AC/DC data generation conversion processes, CIM/XML documents are converted into text data that can be directly read by the power system [6].The generation process is shown in Figure 1.
The CIM/E document contains redundant measurement information for the entire power system, which can be collected according to certain rules to obtain input state estimation data.At this point, the transformed data meets the prerequisites for state estimation and bad data detection and identification in large-scale power systems.

State estimation
The unified iterative algorithm is used to model the AC DC hybrid system, and the state variables on the AC and DC sides are used together as solving variables in state estimation.The modeling of the AC/DC steady-state system is shown in Figure 2. and Qsi, j, k, n in the figure represent the active and reactive power injected into the nodes of the AC/DC system; Usi, j, k, n are the voltage amplitudes of each node; δsi, j, k, n are the voltage phase angles of each node.Pj, k and Qj, k are the active and reactive power output from the AC system to the corresponding DC system; Pdcj and Qdcj are the active and reactive power absorbed by the LCC converter station; Pck and Qck are the active and reactive powers absorbed by MMC; Pdck is the DC power output by MMC; Uck∠δsk-δk is the voltage phasor on the AC side of MMC; Udcj, k and Idcj, k are DC voltage and current; Xcj is the equivalent reactance of LCC reactive compensation; kTj is the transformation ratio of LCC converter transformer; Xj is the LCC commutation reactance; Rk and Xk are equivalent impedances of MMC.
For AC systems, AC nodes are divided into three types: pure AC nodes, LCC nodes, and MMC nodes.The measurement equations mainly include the node voltage measurement equation, the node injection power measurement equation, and the line transmission power measurement equation.Specifically, the injection power measurement equation for DC nodes has changed.
The state variables of the LCC DC system are Udc, Idc, cosθ, cosφ, and kT.The measurement equation is composed of the following: 1) LCC DC voltage and current measurement equation where Udci m and Idci m are the DC voltage and current measurement values; Udci and Idci are the calculated values of DC voltage and current; vUdci and vIdci are the DC voltage and current measurement errors.
2) LCC DC power measurement equation where Pdci m is the measured DC power value; vPdci is the DC power measurement error.
3) Measurement of transmission power of DC lines where Pdcij m is the measured value of the active power of the DC line; vPdcij is the measurement error of the active power of the line.4) AC side power transmission measurement equation s s dc dc where Psi m is the measured value of active power on the AC side; vPsi is the measurement error of active power on the AC side; Qsi m is the measured value of reactive power on the AC side; vQsi is the measurement error of reactive power on the AC side.5) Pseudo measurement equation where θi is the control angle of the inverter.For the rectifier, the control angle is the triggering angle α.
For inverters, the angle control is the turn off angle μ. kr is a constant with a value of 0.995.
where nc is the number of converter stations in the DC network; Nj is the number of converter groups in the j converter station.where Pdci m is the measured DC power value; vPdci is the DC power measurement error.
3) Measurement of transmission power of DC lines where Pdcij m is the measured value of the active power of the DC line; vPdcij is the measurement error of the active power of the line.
∋ ( where Psi m is the measured value of active power on the AC side; vPsi is the measurement error of active power on the AC side; Qsi m is the measured value of active power on the AC side; vQsi is the measurement error of active power on the AC side.
where nc is the number of converter stations in the DC network; Nj is the number of converter groups in the j converter station.
Among them, 4 control equations are determined based on the actual control method, and 2 of them are pseudo measurement equations; Idci s , Udci s , θi s , and kTi s are the controller reference values.

Bad data detection and identification
Data is often contaminated by non-Gaussian errors during transmission and transformation, and bad data processing aims to detect, identify, and eliminate measurement data with serious errors.Serious errors are usually caused by device failures, network attacks, or communication out of sync [7].In the detection of bad data using the least squares method, χ 2 tests and Maximum Normalized Residual Test (LNR) are used to detect, identify, and correct defective data.
1) Chi square distribution (χ 2 ) Inspection χ 2 -test is a statistical method based on chi square distribution for detecting bad data.Testing the relationship between the state estimation objective function and χ 2 to compare the distribution and χ 2 .The value is evaluated within the (m-s) degree of freedom and C% confidence interval, where m is a numerical measurement, s is the number of states, and C% is selected by the SCADA system, typically using 95% [8].
If J(x) calculated using Equation ( 1) is greater than the estimated χ 2 .This method detects the presence of bad data, where ri is the residual, calculated by zi-hi(x).
2) Maximum Normalized Residual Test LNR is a hypothesis testing method used to identify bad data.The process of integrating LNR into the least squares state estimation algorithm after detecting the presence of bad data in the chi square test is described as follows: (1) χ 2 is used to check for bad data.
(2) The LNR for all measurements is calculated based on the following equation: where S is the residual sensitivity matrix given by the following equation: (3) The threshold thr is defined for identifying suspicious measurements: , , where the thr value is between 1.5 and 4, usually using a value of 3.
(4) max() for all suspicious measurements is selected to eliminate or correct.
(5) The process is restarted from the state estimator block using the updated measurement set.

Simulation
Generating required data based on CIM/XML and CIM/E documents exported from a regional dispatch system.The AC/DC network consists of 188 AC nodes, and 230 AC branches, as well as an LCC threeterminal DC system and an LCC/MMC hybrid three-terminal DC system.The local topology is shown in Figure 3.
1) The correctness of the unified iterative calculation for state estimation of the hybrid AC/DC system containing LCC and MMC, as established in this paper, is verified.The generated raw data for AC/DC state estimation converges after 5 unified iterations.
2) In the face of potential noise interference in real-world scenarios, given the fewer measurements in the DC system, the probability of encountering bad data is lower.The simulation simulates a situation with significant noise interference in the AC system, resulting in an 8.1% rate of bad data in AC measurements (number of bad data points/total number of measurements).Simultaneously, the AC measurement errors do not affect the DC measurement data, and there is no phenomenon of interference data transferring between AC and DC systems.
3) By exporting data from the dispatch system, this paper achieves state estimation and data correction for hybrid AC/DC power systems, utilizing a large amount of data as a driving force.The functionality of obtaining the power system state and identifying bad data measurement points is realized through the analysis, mining, and modeling of data.

where
equations are determined based on the actual control method, and 2 of them are pseudo measurement equations; Idci s , Udci s , θi s , and kTi s are the controller reference values.The state variables of the MMC DC system are Udc, Idc, M, and δ.The measurement equation mainly consists of the following four equations: 1) MMC DC voltage and current measurement equation Udci m and Idci m are the DC voltage and current measurement values; Udci and Idci are the calculated values of DC voltage and current; vUdci and vIdci are the DC voltage and current measurement errors

Figure 3 .
Figure 3. Topology diagram of the hybrid AC-DC interconnected system.Simulation data can be divided into two categories: 1) using SCADA raw measurement data from CIM/E documents, and 2) applying a 10% Gaussian noise error to the raw measurement data of AC nodes in SCADA.