Method and apparatus for disturbance classification in power transmission lines based on machine learning

Abstract

The present disclosure provides a method and apparatus for determining an interference condition in an electric power transmission line. The method includes obtaining (302) a plurality of sample values ​​corresponding to an electrical parameter measured in each phase. The method further includes determining (304) a plurality of magnitudes of the electrical parameter corresponding to each phase based on the corresponding plurality of sample values ​​and determining (306) a plurality of difference values ​​for each phase based on the corresponding plurality of magnitudes. The method includes processing (308) the plurality of difference values ​​using machine learning techniques to determine an interference condition. The interference condition is one of a load change condition, a power swing condition, and an electrical fault condition. The method also includes performing (310) at least one of a protection function and a control function based on the interference condition.

Description

Background Art

[0001] Embodiments of the present disclosure relate generally to interference detection, and more particularly to machine learning techniques for detecting and classifying interference conditions in power transmission lines.

[0002] The power transmission system experiences various electrical disturbances such as electrical faults, power swings, and load changes due to various intrinsic or extrinsic factors. For example, power swings may occur when a group of generators speed up relative to each other, and electrical faults may occur when electrical parameters change from their nominal values, affecting the unit and causing disruption to normal operation. Load changes occur when electrical parameters, usually those at a point in the electrical system, change due to a load change on the entire supply output.

[0003] Infrastructure devices and consumer equipment are protected during disturbance conditions generated due to various operating conditions. For example, during a fault condition, a portion of the grid may need to be disconnected by a protective relay. However, certain power swing conditions in the absence of a fault condition can also operate the protective relay, resulting in a power outage. Power swings are the primary cause of many power outages around the world. Regardless of the cause, once the system loses power, it often takes several hours to restore the power system. This has a huge economic impact on utilities and consumers, and must be prevented in modern power grids. Reliable determination of fault conditions and other disturbance conditions is very important for distance relay applications to prevent false operation and power outages during power swing conditions.

[0004] Typically, a blind zone (or concentricity characteristic) is used to distinguish between normal faults and power swings. Alternatively, other techniques (such as but not limited to the swing center voltage (SCV) technique, the resistance technique, and the superimposed current and voltage signal technique) are used to detect and distinguish fault disturbances. However, these techniques are ineffective for disturbances near the swing center point or in three-phase transmission lines. Additional blind zones are used around the system line corners to detect disturbances near the swing center point. Such methods require complex system-specific settings and better phasor estimation quality.

[0005] Recently, the integration of renewable energy sources in power system architecture has put new demands on the power handling and fault tolerance capabilities of power systems. Power fluctuations change the grid inertia, source-to-line impedance ratio (SIR), and swing frequency. These conditions require elaborate procedures that require intensive computations. A new disturbance detection scheme needs to be developed for future low-inertia power systems. Summary of the invention

[0006] According to one aspect of the present disclosure, a method for determining an interference condition in a power transmission line is disclosed. The method is performed by an intelligent electronic device (IED) communicatively coupled to a measuring device connected to the power transmission line. The method includes obtaining a plurality of sample values ​​corresponding to an electrical parameter measured in each phase among one or more phases. The electrical parameter in the one or more phases is measured by a measuring device connected to one end (single end) of the power transmission line. The method further includes determining a plurality of magnitudes (such as root mean square (RMS) values) of the electrical parameter corresponding to each phase based on the corresponding plurality of sample values. The method also includes determining a plurality of difference values ​​for each phase based on the corresponding plurality of magnitudes (such as RMS values). Each difference value in the plurality of difference values ​​represents a rate of change of the magnitude (such as RMS value) of the electrical parameter of the corresponding phase. The method includes processing a plurality of difference values ​​corresponding to the one or more phases using a machine learning technique to determine an interference condition in the power transmission line. The interference condition is one of a load change condition, a power swing condition, and an electrical fault condition. The method also includes performing at least one of a protection function and a control function based on the interference condition.

[0007] According to another aspect of the present disclosure, an intelligent electronic device (IED) for determining an interference condition in a power transmission line is disclosed. The IED is communicatively coupled to a measuring device connected to the power transmission line. The IED includes a data acquisition unit, which is communicatively coupled to a measuring device connected to one end (single end) of the power transmission line, and the data acquisition unit is configured to obtain a plurality of sample values ​​corresponding to an electrical parameter measured by the measuring device in each of one or more phases. The IED further includes a signal processing unit, which is communicatively coupled to the data acquisition unit, and the signal processing unit is configured to determine a plurality of magnitudes (such as root mean square (RMS) values) of the electrical parameter corresponding to each of the one or more phases based on the plurality of sample values. The signal processing unit is further configured to determine a plurality of difference values ​​for each of the one or more phases based on the corresponding plurality of magnitudes (such as RMS values). Each difference value in the plurality of difference values ​​represents a rate of change of the magnitude (such as RMS value) of the electrical parameter of the corresponding phase. The IED further includes a machine learning unit communicatively coupled to the signal processing unit and configured to process the plurality of difference values ​​corresponding to the one or more phases using a machine learning technique to determine a disturbance condition in the power transmission line. The disturbance condition is one of a load change condition, a power swing condition, and a fault condition. The IED also includes a control unit communicatively coupled to the machine learning unit and configured to perform at least one of a protection function and a control function based on the disturbance condition. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] These and other features and aspects of embodiments of the present invention will be better understood when the following detailed description is read with reference to the accompanying drawings, in which like reference numerals represent like parts throughout the drawings, wherein:

[0009] Figure 1 is a schematic diagram of a power transmission system having an intelligent electronic device (IED) according to an exemplary embodiment;

[0010] Figure 2 is a diagram of an IED configured to perform interference detection according to aspects of the present disclosure;

[0011] Figure 3 is a flow chart illustrating a method for interference detection in a power transmission line according to an exemplary embodiment;

[0012] FIG. 4A to FIG. 4C is a diagram illustrating phase signals corresponding to a load change condition, a power swing condition, and a power fault condition according to an embodiment of the present disclosure;

[0013] Figure 5A is a diagram illustrating a root mean square signal of a phase voltage determined according to an exemplary embodiment;

[0014] Figure 5B is a diagram illustrating a method according to an exemplary embodiment Figure 5A A view of a smoothed version of the curve;

[0015] Figure 6 is a diagram illustrating the accuracy of a proposed machine learning technique for twenty-five-fold cross validation according to an exemplary embodiment;

[0016] Figure 7 is a diagram illustrating producer accuracy of the proposed machine learning technique according to an exemplary embodiment;

[0017] Figure 8 is a diagram illustrating user accuracy of the proposed machine learning technique according to an exemplary embodiment; and

[0018] Fig. 9 is a schematic diagram illustrating a gradient boosting classification technique according to an exemplary embodiment. DETAILED DESCRIPTION

[0019] As will be described in detail below, an intelligent electronic device (IED) having the capability to determine interference conditions in a power transmission line and a corresponding interference determination method are presented. More specifically disclosed embodiments relate to machine learning techniques for determining interference conditions in a power transmission line.

[0020] In the present disclosure, the term "phase" and the phrase "phase signal" are equivalent and used interchangeably. Similarly, the phrases "transmission line" and "power transmission line" are equivalent and used interchangeably in this document. The phrase "sample value" refers to a real value representing an electrical parameter and is obtained by sampling the electrical parameter. In a non-limiting sense, the electrical parameter may refer to a voltage signal and / or a current signal. The sample value corresponds to the electrical parameter at one moment. The phrase "root mean square value" specifically refers to the square root of the average value of the sample value of the reference time window. However, embodiments of the present disclosure may also employ sample energy values ​​or sample power values ​​corresponding to multiple sample values ​​in the time window. The term "smoothing" refers to low-pass filtering of the sample value within the time window. The phrase "difference signal" refers to a discrete signal having multiple difference values. In one embodiment, the phrase "difference value" refers to the difference between two consecutive root mean square values. The phrase "ensemble learning" refers to a class of machine learning techniques that employ multiple learning models in multiple stages. The phrases "boost" and "boost" are equivalent and used interchangeably. The phrase "residual model" refers to one of a plurality of learning models that are trained to progressively reduce the modeling error in successive stages. The residual model is trained based on the residual error of the previous stage. The residual error refers to the difference between the prediction and the ground truth of the current stage. The "protection action" performed by the intelligent electronic device (IED) includes blocking the operation of the IED, thereby providing immunity to power swing conditions. Similarly, the "control action" performed by the IED includes unlocking the protection relay, enabling the protection relay to operate, thereby ensuring protection from the fault.

[0021] Figure 1 1 is a schematic diagram of a power transmission system 100 having a plurality of intelligent electronic devices (IEDs), such as an IED 120, according to an exemplary embodiment. The power transmission system 100 includes a first transmission line 106 and a second transmission line 108 connected between a first bus 102 and a second bus 104. Further, the first bus 102 is coupled to a first power source 110 and the second bus 104 is coupled to a second power source 112. It may be noted that in some embodiments of the system 100, the second bus 104 may be coupled to a load, and in such embodiments, the numeral 112 represents the load. In the illustrated embodiment, a disturbance condition 118 of the second transmission line 108 is shown. However, the disturbance condition in the first transmission line may also be managed by the same techniques disclosed herein, wherein an IED is associated with the corresponding line.

[0022] It may be noted herein that the interference condition 118 may be one of, but is not limited to, a load change, a power swing, and a fault in at least one phase of the power transmission system. The power transmission system 100 includes a measuring device 122 located at at least one end of a transmission line. In the illustrated embodiment, the measuring device 122 is disposed near the first bus 102. The measuring device 122 may include a voltage transformer 114 or other suitable voltage sensor, and the measuring device is configured to measure a voltage signal. In an alternative embodiment, the measuring device 122 may be disposed near the second bus 104. In one embodiment, when the measuring device 122 is equipped with computing resources, the measuring device 122 may be configured to perform sampling of the voltage signal and generate a plurality of sample values. In a three-phase electrical system, a plurality of sample values ​​corresponding to each of the three phases are generated by the measuring device 122. The IED 120 is coupled to the measuring device 122 and is configured to receive a plurality of sample values ​​corresponding to one or more phases and determine the type of interference in the power transmission system 100. In one embodiment, the plurality of samples corresponds to a plurality of voltage samples obtained from one or more phase voltage signals. Although in Figure 1 In the embodiment of the present invention, the measuring device 114 shown is a voltage transformer, but the method disclosed in this article is applicable to other electrical parameters such as current, power, etc.

[0023] Figure 2 is a diagram of an IED 200 configured to perform interference determination according to aspects of the present disclosure. It can be noted that the IED 200 corresponds to Figure 1 The IED 120 of FIG. 1 may refer to an electric power system device configured to perform one or more of a measurement function, a protection function, and a control function. As an example, the IED may be a protection relay configured to protect a line when an electrical disturbance or fault occurs. In one embodiment, the IED 200 includes a data acquisition unit 202, a signal processing unit 204, a machine learning unit 206, a control unit 208, and a memory unit 210 that are communicatively coupled to each other via a communication bus 212.

[0024] The data acquisition unit 202 is communicatively coupled to the transmission line 214 via the measurement device and is configured to receive a plurality of sample values ​​corresponding to measurements in one or more phases among the plurality of phases of the transmission line 214. In one embodiment, the data acquisition unit 202 is configured to acquire an analog signal from the measurement device 122, and is configured to perform sampling of the voltage signal and generate a plurality of sample values. In a three-phase transmission line, a sample value from at least one phase among the plurality of phases may be obtained by the data acquisition unit 202. The data acquisition unit 202 includes a module for obtaining a voltage signal from the measurement device ( Figure 1The data acquisition unit 202 may further include a circuit for receiving sample values ​​(see reference numeral 122 in the figure). In an alternative embodiment, the data acquisition unit 202 may also include a sampling and digitizing circuit to convert an analog voltage signal sensed by a voltage sensor in the measuring device into a sample value. The data acquisition unit 202 may also be configured to perform signal conditioning and preprocessing on the received sample values. In a non-limiting example, the signal conditioning and preprocessing operations may include normalization operations, noise reduction operations, and bandwidth limiting operations. In yet another embodiment, the data acquisition unit 202 receives sample values ​​of the measured signal from a measuring device (e.g., measuring device 122) or other measuring units communicatively coupled to such a measuring device.

[0025] The signal processing unit 204 is communicatively coupled to the data acquisition unit 202 and is configured to determine a plurality of difference values ​​required for the machine learning unit 206 to determine the interference condition. Specifically, in one embodiment, the signal processing unit 204 is configured to generate a plurality of values ​​indicating the magnitude of the measured electrical quantity based on a plurality of sample values ​​of each phase among the one or more phases, which may be a root mean square value (RMS). Each value in the plurality of RMS values ​​is given, for example, by the following formula,

[0026]

[0027] Among them, V rms (k) is the RMS value for sample index k, which is the value obtained using the N sample values ​​V corresponding to the phase indexed by p. p (l) to calculate.

[0028] Although the magnitude of the measured electrical quantity can be determined as an RMS value, other indicators that quantify the magnitude (or amplitude) of the electrical parameter measured in the corresponding one phase can be used. It can be noted herein that in alternative embodiments, a sample energy value or a sample power value can be determined instead of an RMS value. In one embodiment, the sample energy value is determined as the square sum of the sample values, and the sample power value is determined as the average value of the square sum of the sample values. Specifically, in one example, when the supply voltage is fifty hertz, each phase is sampled at a sampling rate of one thousand hertz, and twenty samples are considered to calculate the RMS value. In another example, when each phase of a supply voltage with a frequency of sixty hertz is sampled at a sampling rate of 1200 hertz, the same number of samples can be considered. Alternatively, when a supply voltage of sixty hertz is sampled at a sampling rate of one thousand hertz, seventeen samples are considered to calculate the RMS value. Further, when each new sample value is available for the signal processing unit 204, a new RMS value is determined. When twenty samples are considered to calculate the RMS value, a new RMS value in each phase is determined by considering the previous nineteen sample values ​​of the corresponding phase and the new sample value. In one embodiment, a smoothed version of the plurality of RMS values ​​corresponding to each of the plurality of phases is determined by filtering the corresponding RMS values. The signal processing unit 204 performs smoothing on the plurality of RMS values ​​in each of the plurality of phases using a low pass filter.

[0029] The plurality of difference values ​​calculated by the signal processing unit 204 represent a rate of change corresponding to each of the plurality of phases based on the plurality of RMS values. In one embodiment, the signal processing unit 204 is configured to generate a difference signal having a plurality of difference values ​​corresponding to the plurality of RMS values ​​in each phase. Each difference value is determined as the difference between two consecutive RMS values. The difference value is given by:

[0030] ΔV rms (k) = V rms (k)-V rms (k-1)

[0031] Where, ΔV rms is the difference, V rms (k) and V rms (k-1) is a continuous value of the RMS value. In alternative embodiments, when a higher sampling frequency is selected, the difference values ​​can be generated as second-order differences (i.e., the difference between the first RMS value and the third RMS value), or as higher-order differences (i.e., the difference between the first RMS value and subsequent RMS values, without considering multiple RMS values ​​in between).

[0032] The signal processing unit 204 is further configured to detect an interference condition based on the difference value. In one embodiment, each difference value is compared with a predetermined threshold value, and interference is detected based on the result of the comparison. Specifically, when the difference value is greater than the predetermined threshold value, an interference condition is detected. In another embodiment, the difference value in each phase is compared with a predetermined threshold value, and when the difference value corresponding to any phase of the plurality of phases is greater than the predetermined threshold value, an interference condition is detected.

[0033] After the interference condition is detected, the type of interference is determined. To this end, the signal processing unit 204 selects a plurality of difference values ​​corresponding to the interference condition. Specifically, in one embodiment, the signal processing unit 204 is configured to select the first three difference values ​​in each phase immediately after the interference condition is detected. In an alternative embodiment, a different number of difference values ​​may be selected in one or more phases after the interference condition is detected. Further, the signal processing unit 204 is configured to transmit the selected difference values ​​to the machine learning unit 206.

[0034] The machine learning unit 206 is communicatively coupled to the signal processing unit 204 and is configured to process the difference received from the signal processing unit 204. The difference received from the signal processing unit 204 may correspond to one or more phases. According to some embodiments, the machine learning unit 206 is configured to use an integrated machine learning technique to process a plurality of difference values ​​corresponding to a plurality of phases to determine the type of interference condition. In other embodiments, the machine learning unit 206 may also use other machine learning techniques to process the plurality of difference values. The type of interference condition includes one of a load change condition, a power swing condition, and a transmission line fault condition. The integrated machine learning technique may include one of a plurality of techniques, such as but not limited to one of a bagging technique or one of a boosting technique. Specifically, the boosting technique may include one of an adaptive boosting technique or a gradient boosting technique. In an embodiment, a fast version of a gradient boosting technique (referred to herein as an "extreme gradient boosting technique") is used to determine the interference condition. In one embodiment, the machine learning model required to determine the interference condition is a priori calculated via an offline activity (e.g., by a Monte Carlo simulation).

[0035] In one embodiment, a Monte Carlo simulation is performed to train a gradient boosting learning model using a plurality of labeled data sets. Each data set in the plurality of labeled data sets includes a phase signal and a corresponding interference condition as an output category. Each data set may also include a corresponding set of three difference values. In a specific example, hundreds of labeled data sets for each output category are used to train the gradient boosting learning model. Appropriate parameters are selected for the gradient boosting learning model to obtain satisfactory performance. The parameters of the gradient boosting learning model include, but are not limited to, the number of residual models, target parameters, and evaluation indicators.

[0036] The control unit 208 is communicatively coupled to the machine learning unit 206 and is configured to perform at least one of a protection function and a control function based on the interference condition. In one embodiment, when the machine learning unit determines a fault condition, the protection function is initiated. In this case, the control unit 208 is configured to unlock the protection function and allow the IED to isolate the transmission line from the fault. In another embodiment, when the machine learning unit 206 determines a power swing condition, the control unit 208 is configured to lock the IED as one of the control actions. In this embodiment, the IED cannot operate and the power swing condition is not isolated from the grid. Appropriate actions can be taken based on the interference condition. For illustration, information can be output based on the determined interference condition and / or a control action can be automatically triggered based on the determined interference condition. The control action can be selected based on the determined interference condition.

[0037] Figure 3 is a flow chart of a method 300 for interference determination according to an exemplary embodiment. The method 300 is for determining interference conditions in a transmission line and performing a protection action or a control action based on the determined interference conditions. Example protection actions include, but are not limited to, operating a switchgear (e.g., a circuit breaker) using an IED during an extreme power swing due to a fault condition. The control action may include blocking the operation of the switchgear using an IED during a self-restoring power swing condition or unblocking the operation of the switchgear during a fault condition.

[0038] Method 300 includes obtaining a plurality of sample values ​​corresponding to one or more of a plurality of phases of a power transmission line, as shown in step 302. Obtaining step 302 includes measuring a voltage signal in one or more phases of a transmission system at a location of interest by a measuring device. In the present embodiment, the measuring device includes one or more voltage sensors such as voltage transformers. The voltage signal is sampled by the measuring device, a dedicated measurement unit communicatively coupled to the measuring device, or a data acquisition unit of an IED and the sampled voltage signal is converted into a plurality of sample values. In one example, the power transmission system is a three-phase system, and three voltage signals corresponding to a three-phase transmission line are measured by the measuring device. In addition, the plurality of voltage signals are sampled at a sampling rate of 1 kHz to generate a plurality of sample values ​​corresponding to each of the plurality of phases. Although step 302 is explained here with reference to a voltage signal, the plurality of sample values ​​may also correspond to any other electrical parameter other than a voltage parameter. Specifically, the plurality of sample values ​​may correspond to a current signal in one or more phases of the transmission system, or to a combination of a current signal and a voltage signal.

[0039] At step 304, the method also includes determining multiple values ​​indicating the magnitude of the measured electrical quantity corresponding to each phase in the multiple phases based on multiple samples, which can be a root mean square (RMS) value. In one embodiment, each of the multiple RMS values ​​is determined based on multiple samples (sample values) corresponding to each phase. In some embodiments disclosed herein, approximately twenty sample values ​​are considered in each phase to calculate an RMS value in the corresponding phase. In one embodiment, a smoothed version of multiple root mean square values ​​corresponding to one phase is considered for processing by subsequent steps of method 300. A smoothing operation can be performed on multiple RMS values ​​using a suitable low-pass filter. Although the magnitude of the measured electrical quantity can be determined as an RMS value, other indicators that quantify the magnitude (or amplitude) of the electrical parameter measured in the corresponding one phase can be used.

[0040] The method 300 further includes determining at step 306 a plurality of differences representing the rate of change of the magnitude (e.g., RMS value) of the electrical parameter measured from each phase signal in the plurality of phase signals. The plurality of differences are determined based on a plurality of RMS values. In one embodiment, the difference is derived as the difference between two consecutive magnitudes (e.g., RMS values). In an alternative embodiment, the difference may be determined as the difference between non-adjacent magnitudes (e.g., RMS values). The method 300 also includes processing the differences corresponding to the plurality of phases using machine learning techniques to determine the interference condition of the power transmission line, as shown in step 308. The difference utilized in this step may be a difference associated with interference (or interference condition). As described above, interference conditions may be detected by comparing each difference with a predetermined threshold. Further, the difference used to determine the type of interference condition may be some specific values ​​(two differences, three differences, etc.) together or excluding the difference associated with the interference condition.

[0041] The machine learning technique used for processing may be an integrated machine learning technique. The integrated machine learning technique includes, but is not limited to, an adaptive boosting technique and a gradient boosting technique. As previously described, the interference condition includes one of a load change condition, a power swing condition, and a transmission line fault condition. The method 300 also includes performing at least one of a control action or a protection action based on the interference condition. Specifically, in one embodiment, the action performed by the IED includes operating a switchgear connected to the transmission line to lock or unlock the IED based on the interference condition at step 310. At least one control action or protection action may be automatically performed.

[0042] FIG. 4A to FIG. 4C4 is a diagram illustrating three voltage signals used in some embodiments of the present disclosure. The three voltage signals correspond to a load change condition, a power swing condition, and a power fault condition, respectively. The diagrams 400, 408, 416 correspond to signals in the same phase of a power transmission line. Figure 4A 4 is a graph illustrating voltage signal changes during a load change condition in a three-phase transmission line. Graph 400 includes an x-axis 402 representing time in seconds and a y-axis 404 representing voltage changes due to a disturbance condition. Graph 400 includes a curve 406 representing voltage changes during a load change disturbance condition. Figure 4B 4 is a graph illustrating voltage signal changes during a power swing condition in a three-phase transmission line. Graph 408 includes an x-axis 410 representing time in seconds and a y-axis 412 representing voltage changes due to a disturbance condition. Graph 408 includes a curve 414 representing voltage changes during a power swing disturbance condition. Figure 4C 4 is a view 416 illustrating a change in a voltage signal under a fault condition in a three-phase transmission line. The view 416 includes an x-axis 418 representing time in seconds and a y-axis 420 representing a change in voltage due to a disturbance condition. The view 416 includes a curve 422 representing a change in voltage during a three-phase fault disturbance condition. It can be noted herein that the curves 406, 414, and 422 have similar patterns and are not suitable for classification using threshold-based techniques or by a setting-based approach. However, the phase voltage change rate of the curve 422 corresponding to the three-phase fault is higher than the voltage change rate corresponding to the power swing and load change conditions. As explained in detail with reference to subsequent figures, this property is used to determine a disturbance condition in different embodiments of the present disclosure.

[0043] Figure 5A 5 is a diagram illustrating a root mean square (RMS) signal of a phase voltage determined according to an exemplary embodiment. The diagram 500 includes an x-axis 502 representing time in seconds and a y-axis 504 representing a root mean square (RMS) value in kilovolts (kV). The diagram 500 further includes a curve 506 representing an RMS voltage signal corresponding to one of the three phase voltage signals. It can be observed that the curve 506 exhibits variations in short time intervals and may not be suitable for determining the difference required to detect a disturbance condition.

[0044] Figure 5B is a diagram illustrating a method according to an exemplary embodiment Figure 4A 508 is a view of a smoothed version of the curve of . View 408 includes an x-axis 510 representing time in seconds and a y-axis 512 representing root mean square (RMS) values ​​in kilovolts (kV). View 408 further includes Figure 5A506. Curve 514 represents a filtered version of the RMS voltage signal corresponding to one of the three phase voltage signals. It can be observed that curve 514 exhibits smooth variations and is more suitable for determining the difference required to detect interference conditions.

[0045] Figure 6 6 is a view 600 illustrating the accuracy of the proposed machine learning model according to an exemplary embodiment. The view 600 includes an x-axis 602 representing an index of cross validation and a y-axis 604 representing the model accuracy of the extreme gradient boosting model in units of percentage values. The view 600 also includes a plurality of columns 606, one column for each cross validation test. The view 600 includes results from twenty-five cross validation tests. The columns 606 represent the model accuracy values ​​for the corresponding validation tests. It can be observed that the model accuracy values ​​vary between 97.5% and 100%, confirming the satisfactory performance during disturbance detection in the power transmission line.

[0046] Figure 7 700 is a diagram illustrating producer accuracy corresponding to the proposed machine learning technique according to an embodiment of the present disclosure. The term "producer accuracy" as used herein refers to an efficiency term defined based on omission errors. In one embodiment, the efficiency of the producer is calculated by missing misclassifications from the correct category and indicates the situation in which each category is classified. The diagram 700 includes an x-axis 702 representing the category of interference conditions and a y-axis 704 representing the average producer accuracy in percentage values. The diagram 700 also includes three bars 706, 708, 710 representing the producer accuracy for fault conditions, power swing conditions, and load change conditions, respectively. The bar height represents the average of the producer accuracy values ​​obtained from multiple cross-validation tests for the corresponding interference conditions. It can be observed that the efficiency of the interference condition is higher than 97%, which can be considered satisfactory for interference detection in power transmission lines.

[0047] Figure 8800 is a diagram illustrating the user accuracy of the proposed machine learning technology according to an exemplary embodiment. The term "user accuracy" used herein refers to an efficiency term defined based on misclassification error. Specifically, user accuracy is calculated as the correct classification of the total number of predicted classifications relative to the category. The user's efficiency indicates the probability that a sample from the classified category belongs to the actual category. View 800 includes an x-axis 802 representing the interference category. View 800 also includes a y-axis 804 representing the average user accuracy. The average value of user accuracy is obtained by averaging the user accuracy obtained from multiple cross-validation tests. View 800 includes three columns 806, 808, 810 representing the average user accuracy corresponding to three interference categories (i.e., fault conditions, power swing conditions, and load change conditions), respectively. It can be observed that for all interference conditions, the value of the average user accuracy exceeds 97.5%. It can also be observed that the proposed method can accurately classify power system interference and can be used to block or unlock distance relays during power swings.

[0048] As disclosed herein above, models based on one or more machine learning techniques may be utilized. One of the models according to one of the embodiments is described below. Fig. 9 9 is an illustration of a gradient boosting classification model 900 according to an exemplary embodiment. Model 900 illustrates a data set 902 provided as an input to a gradient boosting classification technique. During the training phase, data set 902 is a labeled training set, while during the deployment phase, data set 902 is a real-time data set obtained from a power transmission line.

[0049] Model 900 includes a plurality of models 904, 910, 916 that are sequentially determined using a data set 902 as input data. During the training phase, the data set 902 is used to determine a first model 904 and a corresponding first predicted output 908. A first residual 906 is determined based on the first predicted and ground truth outputs. In the next step, the data set 902 is used together with the first residual 906 as ground truth to determine a second model 910 and a corresponding second residual 914. A second residual 912 is determined based on the second prediction and the first residual 906. Similarly, subsequent models among the plurality of models and their corresponding predictions and residuals are determined. In the illustrated embodiment, a third model 916 is determined as the final model and a corresponding third prediction 920 is determined as the final prediction. Further, an estimate 922 of the data set 902 is determined based on a plurality of predictions 908, 914, ... 920. In one embodiment, a final prediction 922 is determined by combining a plurality of predictions obtained from a plurality of models. During the deployment phase, the plurality of models 904, 910, ... 916 determined during the training phase are used to process the data set 902 to determine a plurality of predictions 908, 914, ... 920. In this embodiment, the data set 902 represents real-time measurement data from the transmission line. The final prediction 922 represents the interference condition determined based on the plurality of predictions 908, 914, 920.

[0050] In one example, a total of five thousand eight hundred and ninety-eight samples covering one thousand nine hundred and sixty-six disturbance conditions are used to train the model 900. The disturbance conditions are simulated using PSCAD / EMTDC software. The disturbance conditions include eight hundred and sixteen power swing conditions, nine hundred and fifty faults, and two hundred load change conditions. The disturbance conditions used for training cover the maximum possible combination of power system disturbances in terms of three-phase faults (fault location, fault resistance, initial angle, and swing frequency), power swing changes, and load changes. A cross-validation technique is used to determine the optimal tuning parameters for the extreme gradient boosting classification model.

[0051] The embodiments disclosed in the present disclosure enable faster setup of free interference detection, interference classification and taking operational actions such as power swing blocking and unlocking methods for modern power systems. Such features can be used for both three-phase and two-phase systems and are desirable for any advanced IED (such as digital protection relays). The disclosed interference detection technology only requires measurement of electrical parameters at one end with a low sampling frequency (single-ended method) and can also be implemented in available digital / digital relays. The proposed method can be applied to voltage signals only. In this case, no current signal is required and the problem of current transformer saturation is completely avoided.

Claims

1. A method (300) for determining a disturbance condition in a power transmission line, wherein: The method (300) is performed by an intelligent electronic device (IED) communicatively coupled to a measuring device, wherein the measuring device is connected to the power transmission line, and the method (300) comprises: Obtaining (302) a plurality of sample values ​​corresponding to an electrical parameter measured in each of one or more phases, wherein the electrical parameter in the one or more phases is measured by the measurement device connected to one end of the power transmission line; determining (304) a plurality of magnitudes of the electrical parameter corresponding to each based on the corresponding plurality of sample values; determining (306) a plurality of difference values ​​for each phase based on the corresponding plurality of magnitudes, wherein each difference value in the plurality of difference values ​​represents a rate of change of magnitude of the corresponding phase; processing (308) the plurality of differences corresponding to the one or more using a machine learning technique to determine a disturbance condition in the power transmission line, wherein the disturbance condition is one of a load change condition, a power swing condition, and an electrical fault condition; and performing (310) at least one of a protection function and a control function based on the interference condition, Wherein, the machine learning technique is based on an integrated machine learning technique, wherein the integrated machine learning technique includes an extreme gradient boosting classification model trained to determine the interference condition.

2. The method (300) of claim 1, wherein: Each difference value of the plurality of difference values ​​is a difference between two consecutive magnitudes.

3. The method (300) of claim 1, wherein: The disturbance condition is not a load change condition or an electrical fault condition.

4. The method (300) of claim 1, wherein: The magnitudes are root mean square, RMS values.

5. The method (300) of claim 1, wherein: Determining (306) the plurality of differences includes: detecting the interference condition based on comparing each of the plurality of differences to a predetermined threshold; and One or more difference values ​​immediately following detection of the interference condition are selected for processing by the machine learning technique.

6. The method (300) of claim 5, wherein: The three difference values ​​immediately following detection of the interference condition are selected for processing by the machine learning technique.

7. The method (300) of any one of claims 1 to 6, wherein: The plurality of sample values ​​correspond to a plurality of voltage samples obtained from sampling the voltage signal.

8. The method (300) of claim 1, wherein: Performing at least one of a protection function and a control function based on the disturbance condition includes locking operation of the switchgear using the IED during a self-restoring power swing condition or unlocking operation of the switchgear during a fault condition.

9. The method (300) of claim 1, wherein: The extreme gradient boosting classification model includes multiple residual models.

10. The method (300) of any one of claims 1-6 and 8-9, wherein: Determining (306) the plurality of difference values ​​includes performing a smoothing operation on the plurality of magnitude values ​​corresponding to each of the one or more phases by filtering the corresponding magnitude values.

11. The method (300) of claim 10, wherein: The plurality of difference values ​​are determined by calculating differences between consecutive samples of smoothed versions of the plurality of magnitudes.

12. The method (300) of any one of claims 1-6, 8-9 and 11, wherein: The one or more phases correspond to a three-phase electrical system.

13. The method (300) of any one of claims 1-6, 8-9 and 11, wherein: The plurality of sample values ​​are obtained by sampling at a sampling rate of one kilohertz in corresponding phases.

14. An intelligent electronic device (IED) (200) for determining a disturbance condition in a power transmission line (106, 108), the IED (200) comprising: a data acquisition unit (202) communicatively coupled to a measurement device (122) connected to one end of a power transmission line, and configured to obtain a plurality of sample values ​​corresponding to an electrical parameter measured in each of one or more phases, wherein the measurements are performed by the measurement device; A signal processing unit (204) is communicatively coupled to the data acquisition unit (202) and is configured to: determining a plurality of magnitudes of the electrical parameter corresponding to each of the one or more phases based on a corresponding plurality of sample values; determining a plurality of differential values ​​for each of the one or more phases based on the corresponding plurality of magnitudes, wherein each differential value of the plurality of differential values ​​represents a rate of change of magnitude of the corresponding phase; a machine learning unit (206) communicatively coupled to the signal processing unit (204) and configured to process the plurality of difference values ​​corresponding to one or more using a machine learning technique to determine a disturbance condition in the power transmission line, wherein the disturbance condition comprises one of a load change condition, a power swing condition, and a transmission line fault condition; and a control unit (208) communicatively coupled to the machine learning unit (206) and configured to perform at least one of a protection function and a control function based on the interference condition, in, The machine learning unit (206) is further configured to train an integrated machine learning model for determining interference; and The control unit (208) is further configured to train an extreme gradient boosting classification model for determining the interference condition.

15. The IED (200) of claim 14, wherein: Each difference value of the plurality of difference values ​​is a difference between two consecutive magnitudes.

16. The IED (200) of claim 14, wherein: The disturbance condition is not a load change condition or an electrical fault condition.

17. The IED (200) of claim 14, wherein: The magnitudes are root mean square, RMS values.

18. The IED (200) of claim 14, wherein: The signal processing unit (204) is further configured to: detecting the interference condition based on comparing each of the plurality of differences to a predetermined threshold; as well as One or more difference values ​​immediately following detection of the interference condition are selected for processing by the machine learning technique.

19. The IED (200) of claim 18, wherein: The three difference values ​​immediately following detection of the interference condition are selected for processing by the machine learning technique.

20. The IED (200) according to any one of claims 14 to 19, wherein: The plurality of sample values ​​corresponds to a plurality of voltage samples obtained from the voltage signal.

21. The IED (200) of claim 14, wherein: The control unit (208) is configured to lock operation of the switchgear using the IED during a self-restoring power swing condition, or to unlock operation of the switchgear during a fault condition.

22. The IED (200) of claim 14, wherein: The machine learning unit (206) is configured to determine a plurality of residual models.

23. The IED (200) according to any one of claims 14-19 and 21-22, wherein: The signal processing unit (204) is further configured to generate a smoothed version of the plurality of magnitudes corresponding to each of the plurality of phase voltage signals by filtering the corresponding magnitudes.

24. The IED (200) of claim 23, wherein: The signal processing unit (204) is configured to determine the plurality of difference values ​​based on smoothed versions of the plurality of magnitudes.

25. The IED (200) of any one of claims 14-19, 21-22 and 24, wherein: The data acquisition unit (202) is configured to obtain the plurality of samples from one or more phases of a three-phase electrical system.

26. The IED (200) of any one of claims 14-19, 21-22 and 24, wherein: The data acquisition unit (202) is configured to sample the electrical parameter at a sampling rate of one kilohertz for each of the one or more phases.

Images