TN-C-S system electric leakage fault secondary discrimination method based on abrupt change ratio and random forest algorithm

Abstract

The invention relates to a TN-C-S system electric leakage fault secondary discrimination method based on a sudden change ratio and a random forest algorithm, and belongs to the technical field of electric leakage fault discrimination. According to the method, a TN-C-S system loop is analyzed, and by analyzing the quantitative relation between residual current abrupt change and unbalanced current abrupt change and zero line impedance and grounding impedance before and after a grounding fault occurs and based on the principle that the current abrupt change vector ratio is not changed during load switching, fault first-level judgment based on the current abrupt change vector ratio is carried out; obtaining a first-level discrimination result; and if the primary judgment result is a fault, carrying out secondary judgment by using a random forest algorithm by taking fundamental waves and third harmonic waves of residual current and unbalanced current abrupt change quantities as fault identification characteristic quantities so as to judge whether the fault is a leakage current fault or not. According to the method, reliable fault judgment under the condition of relatively low calculation capability can be realized; moreover, the method can greatly improve the reliability and sensitivity of the TN-C-S system middle protection and total protection, and provides a new thought and direction for the future related research.

Description

Technical Field

[0001] The invention belongs to the technical field of leakage fault identification, and in particular relates to a secondary identification method for leakage faults in a TN-CS system based on a mutation ratio and a random forest algorithm. Background Art

[0002] With the rapid development of the new energy industry, solar photovoltaic power generation has been widely used in the power sector, but the large-scale access of photovoltaic power sources to the grid has also brought new challenges to leakage protection. The actual operating environment of the low-voltage distribution system containing photovoltaic power sources is complex, and the residual current detection in the substation is easily interfered by multiple factors such as normal line leakage current, three-phase unbalanced current, common mode current, and external environment. When a leakage fault occurs, the residual current components are complex and diverse, and the amplitude is randomly variable, making it more difficult to accurately detect leakage faults.

[0003] China's previous low-voltage power distribution system was mainly based on the TN-C system. When a zero-break fault occurs, the voltage of the metal shell of the equipment increases, which is prone to electric shock accidents. However, the TN-C system cannot install RCD to prevent electric shock. Therefore, the TN-C system is gradually replaced by the TN-CS system. The TN-CS system has a repeated grounding point on the PEN line for zero-break protection, and the part after the PE line and the N line are separated can be equipped with a terminal protection to achieve leakage protection. The problem is that part of the unbalanced current flows into the repeated grounding point and flows back to the neutral point of the system through the earth. The total leakage protection and intermediate protection feel the residual current of the downstream line, which can reach several amperes, making the total protection and intermediate protection unable to be put into operation.

[0004] At present, AC residual current devices (RCDs) are widely used in my country's low-voltage distribution networks to achieve leakage protection. The leakage protection device can only operate correctly when a 50Hz sinusoidal AC residual current is generated in the line. The harmonics introduced by the grid-connected photovoltaic power generation system will affect the normal operation of the AC RCD or even make it completely ineffective. To this end, some scholars have conducted research on complex residual current detection and proposed to apply type B RCD, magnetic modulation type, and full current sensitive RCD to complex residual current detection with multiple current components. This method requires setting a fixed action threshold for leakage protection, but in actual low-voltage distribution networks, the inherent residual current amplitude of a low-voltage distribution station area during normal operation can reach 5A or even more than 20A, which can easily cause RCD malfunction. In recent years, in order to overcome the influence of inherent residual current on leakage fault detection, leakage current change method, adaptive threshold method, amplitude and phase detection method, and double mutation characteristic method have been proposed one after another.

[0005] At present, this type of method unilaterally considers the impact of normal leakage current or three-phase unbalanced current on leakage detection, which relatively improves the fault detection accuracy. However, it does not comprehensively consider the interference of normal leakage current, three-phase unbalanced current, common mode current and external environment on leakage fault detection, and is difficult to apply to low-voltage distribution networks containing photovoltaic power sources, and lacks anti-interference ability and universality.

[0006] Therefore, how to overcome the shortcomings of the existing technology is an urgent problem to be solved in the field of leakage fault identification technology. Summary of the invention

[0007] The purpose of the present invention is to solve the deficiencies of the prior art and to provide a secondary discrimination method for leakage faults in a TN-CS system based on mutation ratio and random forest algorithm.

[0008] To achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0009] The two-level identification method of leakage fault in TN-CS system based on mutation ratio and random forest algorithm includes the following steps:

[0010] The TN-CS system loop is analyzed. By analyzing the quantitative relationship between the residual current mutation and unbalanced current mutation before and after the ground fault occurs and the neutral line impedance and ground impedance, the first-level fault discrimination based on the current mutation vector ratio is performed based on the principle that the current mutation vector ratio remains unchanged when the load is switched, and the first-level discrimination result is obtained.

[0011] If the primary judgment result is a fault, the residual current, the fundamental wave of the unbalanced current mutation, and the third harmonic are used as fault identification feature quantities, and the random forest algorithm is used for secondary judgment to determine whether it is a leakage current fault.

[0012] Furthermore, preferably, the specific method of the first-level discrimination is:

[0013] (1.1) Calculate the ratio M of the residual current mutation amount and the unbalanced current mutation amount in real time according to formula (16);

[0014]

[0015] in, is the unbalanced current mutation amount; is the residual current mutation amount;

[0016] (1.2) Calculate the current moment in real time according to formula (17)

[0017]

[0018] Among them, K is a constant complex number, which is approximately a constant real number; is the fault current change value;

[0019] (1.3)Judgement If it is greater than the fault judgment threshold, it is judged as a fault, otherwise it is judged as no fault.

[0020] Furthermore, it is preferred that

[0021] in, is the A phase current; is the B phase current; is the C phase current; is the residual current; is the unbalanced current on the transformer side; is the grounded neutral current;

[0022] Furthermore, it is preferred that

[0023] in, is the unbalanced current on the transformer side; is the unbalanced current on the load side; Three-phase to ground leakage current sum.

[0024] Furthermore, preferably, the specific method of the secondary discrimination is:

[0025] (2.1) Through artificial experiments, the ground fault is simulated, the fault current amplitude is collected, and the original sample set is obtained.

[0026] (2.2) Select the normalized data of residual current mutation, fundamental wave of unbalanced current mutation and third harmonic as fault identification feature;

[0027] (2.3) Randomly classify the normalized data sample set to obtain training set samples and test set samples;

[0028] (2.4) Using the training set, the normalized data of the residual current mutation, the fundamental wave of the unbalanced current mutation, and the third harmonic are used as the input of the RF classification model, and the corresponding working state is used as the output to train and construct the RF classification model; and using the test set to test the constructed RF classification model; the working states are normal state, ground fault, load shedding fault, and leakage current fault;

[0029] (2.5) Collect the data of the residual current mutation, the fundamental wave of the unbalanced current mutation, and the third harmonic to be secondary judged, normalize them, and input them into the RF classification model constructed in step (2.4) for secondary judgment to obtain the corresponding working state, so as to distinguish whether it is a leakage current fault.

[0030] Furthermore, preferably, the normalization adopts a maximum-minimum normalization method.

[0031] Furthermore, preferably, the sample ratio of the training set and the test set is 7:3.

[0032] Furthermore, preferably, in the RF classification model, the number of RF algorithm decision trees n_estimators is 42, and the maximum number of features max_features considered by the optimal model is 4.

[0033] Furthermore, preferably, in the RF classification model, the maximum depth max_depth of the decision tree is 15.

[0034] The present invention aims at the problem that the TN-CS grounding system in the low-voltage distribution substation containing photovoltaic power sources cannot be put into operation due to the large inherent residual current, and proposes a multi-level discrimination method for leakage faults based on residual current-unbalanced current mutation vector ratio and random forest algorithm. First, by analyzing the quantitative relationship between the residual current mutation amount and the unbalanced current mutation amount and the zero line impedance and the grounding impedance before and after the grounding fault occurs, based on the principle that the current mutation vector ratio remains unchanged when the load is switched, a first-level fault discrimination method based on the current mutation vector ratio is proposed. Secondly, the fundamental wave and the third harmonic of the residual current and the unbalanced current mutation amount are used as fault identification feature quantities, and the random forest algorithm is used for secondary discrimination to achieve reliable fault discrimination under low computing power conditions. This method can greatly improve the reliability and sensitivity of the TN-CS system protection and total protection, and provide new ideas and directions for future related research.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] Aiming at the problem that the TN-CS grounding system cannot be put into operation due to the large inherent residual current in the low-voltage distribution substation containing photovoltaic power sources, the present invention proposes a secondary discrimination method for leakage faults in the TN-CS system based on mutation ratio and random forest algorithm. A large number of simulation and experimental results show that in the TN-CS system with a large inherent residual current amplitude, the traditional leakage fault detection method cannot distinguish the residual current changes caused by load switching and faults, and the detection success rate of leakage faults is less than 1%; while in the single-branch system, the grounding fault transition resistance of the method proposed in the present invention is 6.67Ω, corresponding to a fault current of 33mA, and the effective value of the residual current mutation caused by load switching is about 357mA, and the |ΔI calculated by the CMVRM method HK | is only 0.4mA; the effective value of the residual current mutation caused by the fault input is about 8mA, |ΔI HK| is 24.9mA, and the residual current changes caused by faults and load switching can be 100% distinguished through simple threshold judgment. In addition, in view of the problem that the primary judgment method cannot fully prove that the abnormal point is indeed a leakage fault, a secondary judgment method based on RF is proposed, which can achieve more than 85% reliable detection of leakage faults when the inherent residual current amplitude is higher than the fault current amplitude. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a flow chart of the secondary discrimination method of the leakage fault of the TN-CS system based on the mutation ratio and the random forest algorithm of the present invention;

[0038] Figure 2 Simplify the equivalent circuit for the TN-CS system;

[0039] Figure 3 Construct a flow chart for the RF algorithm;

[0040] Figure 4 This is the test circuit diagram in the application example;

[0041] Figure 5 It is the effective value of residual current and fault current. DETAILED DESCRIPTION

[0042] The present invention is further described in detail below in conjunction with embodiments.

[0043] Those skilled in the art will appreciate that the following examples are only used to illustrate the present invention and should not be considered to limit the scope of the present invention. If no specific techniques or conditions are specified in the examples, the techniques or conditions described in the literature in the art or the product specifications are used. If the manufacturer of the materials or equipment used is not specified, they are all conventional products that can be purchased.

[0044] Example 1

[0045] The two-level identification method of leakage fault in TN-CS system based on mutation ratio and random forest algorithm includes the following steps:

[0046] The TN-CS system loop is analyzed. By analyzing the quantitative relationship between the residual current mutation and unbalanced current mutation before and after the ground fault occurs and the neutral line impedance and ground impedance, the first-level fault discrimination based on the current mutation vector ratio is performed based on the principle that the current mutation vector ratio remains unchanged when the load is switched, and the first-level discrimination result is obtained.

[0047] If the primary judgment result is a fault, the residual current, the fundamental wave of the unbalanced current mutation, and the third harmonic are used as fault identification feature quantities, and the random forest algorithm is used for secondary judgment to determine whether it is a leakage current fault.

[0048] Example 2

[0049] The two-level identification method of leakage fault in TN-CS system based on mutation ratio and random forest algorithm includes the following steps:

[0050] The TN-CS system loop is analyzed. By analyzing the quantitative relationship between the residual current mutation and unbalanced current mutation before and after the ground fault occurs and the neutral line impedance and ground impedance, the first-level fault discrimination based on the current mutation vector ratio is performed based on the principle that the current mutation vector ratio remains unchanged when the load is switched, and the first-level discrimination result is obtained.

[0051] If the primary judgment result is a fault, the residual current, the fundamental wave of the unbalanced current mutation, and the third harmonic are used as fault identification feature quantities, and the random forest algorithm is used for secondary judgment to determine whether it is a leakage current fault.

[0052] The specific method of primary discrimination is:

[0053] (1.1) Calculate the ratio M of the residual current mutation amount and the unbalanced current mutation amount in real time according to formula (16);

[0054]

[0055] in, is the unbalanced current mutation amount; is the residual current mutation amount;

[0056] (1.2) Calculate the current moment in real time according to formula (17)

[0057]

[0058] Among them, K is a constant complex number, which is approximately a constant real number; is the fault current change value;

[0059] (1.3)Judgement If it is greater than the fault judgment threshold, it is judged as a fault, otherwise it is judged as no fault.

[0060]

[0061] in, is the A phase current; is the B phase current; is the C phase current; is the residual current; is the unbalanced current on the transformer side; is the grounded neutral current;

[0062]

[0063] in, is the unbalanced current on the transformer side; is the unbalanced current on the load side; Three-phase to ground leakage current sum.

[0064] The specific method of secondary discrimination is:

[0065] (2.1) Through artificial experiments, the ground fault is simulated, the fault current amplitude is collected, and the original sample set is obtained.

[0066] (2.2) Select the normalized data of residual current mutation, fundamental wave of unbalanced current mutation and third harmonic as fault identification feature;

[0067] (2.3) Randomly classify the normalized data sample set to obtain training set samples and test set samples;

[0068] (2.4) Using the training set, the normalized data of the residual current mutation, the fundamental wave of the unbalanced current mutation, and the third harmonic are used as the input of the RF classification model, and the corresponding working state is used as the output to train and construct the RF classification model; and using the test set to test the constructed RF classification model; the working states are normal state, ground fault, load shedding fault, and leakage current fault;

[0069] (2.5) Collect the data of the residual current mutation, the fundamental wave of the unbalanced current mutation, and the third harmonic to be secondary judged, normalize them, and input them into the RF classification model constructed in step (2.4) for secondary judgment to obtain the corresponding working state, so as to distinguish whether it is a leakage current fault.

[0070] The normalization adopts the maximum-minimum normalization method.

[0071] The sample ratio of the training set and the test set is 7:3.

[0072] In the RF classification model, the number of RF algorithm decision trees n_estimators is 42, and the maximum number of features max_features considered by the optimal model is 4.

[0073] In the RF classification model, the maximum depth of the decision tree is 15.

[0074] Example 3

[0075] In the TN-CS system, there is at least one repeated grounding point between the user's incoming line and the upper line. Figure 2 shown.

[0076] The residual current protection of the low-voltage distribution network is divided into three levels: total protection, intermediate protection and final protection. The total protection is installed at the transformer outlet, and the action setting is generally 300-500mA; the intermediate protection is installed on the trunk or branch line between the total protection and the final protection (household protection), and the action setting is 50-100mA; the final protection is for residential distribution protection or protection of a single electrical equipment, and the action setting is generally 30mA. In the TN-CS system, there is at least one repeated grounding point at the user's incoming line and the upper line. A part of the neutral line current flows back to the transformer side through the repeated grounding point, which is manifested as a residual current value of several amperes. For the sake of simplicity, the present invention analyzes the TN-CS system with only one repeated grounding point. Since the line impedance is much smaller than the load impedance, the 3-phase line impedance is ignored; the neutral line voltage to the ground is low, and the amplitude of the leakage current to the ground is small, so its influence is ignored. The simplified system circuit is as follows Figure 2 shown.

[0077] Figure 2 Middle: Z Aload , Z Bload , Z Cload is a three-phase load; Z Nline is the N line impedance; Z AG is the leakage impedance of phase A to ground, Z BG is the leakage impedance of phase B to ground, Z CG is the leakage impedance of C relative to ground; Z OG is the transformer neutral point to ground impedance; Z SG is the impedance of the repeated grounding point to ground; Z H is the impedance to ground of the C phase leakage fault; is the power frequency current phasor, where is the unbalanced current on the load side.

[0078] Low-voltage distribution network lines are generally short, and the N line impedance is below 0.1Ω. According to Chinese national standards, the impedance of repeated grounding points in the TN-CS system is no more than 10Ω, and the neutral point impedance is no more than 4Ω, which is much smaller than the ground leakage impedance Z of the three-phase line and electrical equipment. AG , Z BG , Z CG Therefore, it can be considered that the unbalanced current on the load side flows entirely through the N line and the repeated grounding point, such as Figure 2 The blue dotted part.

[0079] By analyzing the quantitative relationship between the residual current and unbalanced current mutations before and after the ground fault occurs and the neutral line impedance and ground impedance, and based on the principle that the current mutation vector ratio remains unchanged during load switching, a first-level fault discrimination method based on the current mutation vector ratio is proposed:

[0080] When there is no leakage fault in the system, Figure 2 Analysis shows that:

[0081]

[0082] Where: is the unbalanced current on the transformer side; is the unbalanced current on the load side; Three-phase to ground leakage current sum.

[0083] Arranged:

[0084]

[0085] Then M is

[0086] M=Z Nline / (Z OG +Z SG +Z Nline ) (Formula 3)

[0087] The current flowing through the repeated grounding point for the neutral line;

[0088] It can be seen from formula (3) that M is a complex number, which is determined by the neutral line impedance and the total grounding impedance. When the system structure and line parameters remain unchanged, the change of M is mainly caused by the slow change of non-electrical quantities such as temperature and humidity, such as the change of soil resistivity caused by rain and snow, so M will not change suddenly in a short time.

[0089] After the fault occurs, the ground leakage impedance Z is ignored. AG , Z BG , Z CG The influence of fault current And three-phase to ground leakage current By Z OG and Z SG , Z Nline Two paths flow back to the transformer neutral point, such as Figure 2 As shown by the red dotted line in the middle. At this time, the flow through Z SG , Z Nline The fault current component that returns to the transformer for:

[0090]

[0091] Because Z Nline Much smaller than Z OG and Z SG , ignoring Z Nline The influence of , formula (4) can be approximately simplified as

[0092]

[0093] Where: k is a constant complex number; ZOG and Z SG Approximately resistive.

[0094] Then through Z SG , Z Nline The three-phase to ground leakage current component flowing back to the neutral point of the transformer for:

[0095]

[0096] Depend on Figure 2 Analysis shows

[0097]

[0098] Where: is the A phase current; is the B phase current; is the C phase current; is the residual current; is the unbalanced current on the transformer side; is the grounded neutral current;

[0099] Kirchhoff current analysis of node Q yields:

[0100]

[0101] Combining equations (2), (4)-(8), the residual current of the TN-CS system can be obtained as

[0102]

[0103] From formula (9), we can see that when the fault does not occur:

[0104]

[0105] After the failure occurs:

[0106]

[0107] Where: is the sudden change of the three-phase to ground leakage current; is the unbalanced current mutation amount; is the fault current variation.

[0108] Then the residual current mutation before and after the fault is for

[0109]

[0110] Within the protection range of the general protection and medium protection, the number of electrical appliances is limited, the phase voltage fluctuation caused by the switching of a single load can be ignored, and the line-to-ground leakage current remains basically unchanged. The leakage current of household appliances to the ground generally does not exceed 1mA. When the number of electrical appliances switched on and off at the same time is small, The change is small, below 10mA. Based on the above analysis, ignore It is classified as a system error, and the residual current mutation amount for

[0111]

[0112] In the low-voltage distribution line, a set of typical line parameters is taken, the unit length resistance is 0.2163Ω / km, the inductance is 0.143Ω / km, the N line length is 0.25km, and the Z line length is 0.25km. OG Take 12Ω, Z SG Taking 4Ω, the modulus of M is 0.0128 from formula (3), which is much smaller than 1. Therefore, ignoring the effect of M on the denominator, formula (13) is simplified to

[0113]

[0114] In the formula, K is a constant complex number, which is approximately a constant real number; is the fault current change value.

[0115] In practice, Z OG Generally around 1Ω, Z SG Size Z OG In a system with unchanged circuit structure, Z OG The smaller and Z SG The larger it is, the closer K is to 1. The closer Non-electrical quantities such as rain and snow may cause Z SG becomes smaller, considering the worst case, Z SG Value and Z OG equal, at this time, the value range of K is (1,2], The calculated value is at least 50% is enough to reflect the fault current. Analysis, namely:

[0116]

[0117] It can be seen from formula (15) that for a single-branch system, when no fault occurs, the current caused only by load switching is zero, that is, load switching does not affect the change of fault current. At this time:

[0118]

[0119] M does not change suddenly, so the M value at this moment is calculated by formula (16) and combined with the value at the next moment Using formula (15), we can get the next moment True effective value

[0120] In summary, the fault current change value is calculated using the following formula:

[0121]

[0122] A method for leakage protection using two mutation quantities, residual current and unbalanced current, is obtained, which is defined as current mutation vector ratio method (CMVRM).

[0123] The basic structure and fault determination method of the method are as follows:

[0124] 1) Calculate the ratio of the residual current mutation amount to the unbalanced current mutation amount in real time according to formula (16);

[0125] 2) Calculate the current moment in real time according to formula (17)

[0126] 3) Greater than the fault judgment threshold I Hth , it is judged as a fault, otherwise it is judged as no fault.

[0127] The residual current, the fundamental wave of the unbalanced current mutation, and the third harmonic are used as feature quantities, and the random forest algorithm (RF) is used for secondary discrimination, as follows:

[0128] The idea of ​​building the RF classification model is as follows Figure 3 As shown, the specific construction process is as follows:

[0129] 1) Based on the classification method of the Bagging algorithm, the Bootstrap method is used to randomly sample m times with replacement in the original training sample set S to form m sample subsets for training the base classifier, and the number of samples in each subset is guaranteed to be the same, expressed as: {S1, S2, ..., S m},S m is the mth sample, and m is the number of samplings. When performing Bootstrap sampling, a subset S can be obtained. t The probability p that (t=1,2,...,m) does not contain a certain sample is:

[0130]

[0131] When m→∞, we have:

[0132]

[0133] The 37% of sample data that was not selected is called out-of-bag (OOB). When generating each decision tree, an OOB error estimate can be calculated at the same time. The average of the OOB error estimates of all decision trees in the forest can be used to obtain the generalization error estimate of RF.

[0134] 2) For all sample subsets, the CART algorithm is used to construct corresponding decision trees and combine them to form a random forest, which is represented by {C1, C2, ..., C m}; C m is the mth decision tree sample.

[0135] 3) Use the random subspace idea to split the nodes of each decision tree in the random forest generated in step 2). Randomly extract from N feature variables with equal probability The sub-variables form the split feature variable subset of the node, and use the minimum Gini index value principle in the CART algorithm to select an optimal split feature variable and optimal split value to split the node until each feature variable is used as a split node. The Gini coefficient can be defined as

[0136]

[0137] Where: t i is a currently selected characteristic variable; K is t i The corresponding number of categories; p k is the probability that the sample point belongs to the kth class. Further, in determining the optimal split feature variable t i Based on this, suppose a subset S t According to t i Split into two subsets S t1 and S t2 , then the optimal split value a can be calculated by the following formula:

[0138]

[0139] Where |S t |、|S t1 | and |S t2 | are sample sets S t , S t1 and S t2 The number of samples.

[0140] 4) After each decision tree is built from top to bottom, keep the integrity of the tree without pruning it, and test all decision trees using the test set X.

[0141] 5) Vote for the classification results obtained from the test and select the category with the most votes as the final category of the test set. The voting idea can be expressed as

[0142]

[0143] Where: f RF (x) is the classification result of RF for the test set sample X; I(·) is the number of cases that satisfy the expression in the brackets; The output result of the lth decision tree is i; c is the number of random forest categories; i is a category in c categories.

[0144] In summary, the two-level identification method of leakage fault in TN-CS system based on random forest algorithm has the following specific process:

[0145] 1) Simulate the ground fault through artificial experiments, collect the fault current amplitude, and obtain the original sample set.

[0146] 2) Select the normalized data of residual current mutation, fundamental wave of unbalanced current mutation and third harmonic as fault identification feature quantity and input them into RF classification model;

[0147] The fault identification feature variables are normalized to the same interval using the data normalization method. Data normalization is to scale the data so that it falls into a small specific interval. Through data normalization, fault feature components of different units can be converted into dimensionless values. Inputting the converted values ​​into the classification model can simultaneously improve the convergence speed and accuracy of the model. The present invention normalizes the selected fault diagnosis components and maps them to the [0,1] interval:

[0148]

[0149] Where: a i is the selected fault classification feature variable; bi is the normalized feature variable; N is the number of samples.

[0150] 3) Randomly classify the normalized data sample set to obtain training set samples and test set samples.

[0151] 4) Use the training set to build the RF classification model, and use the test set to test the RF classification model and output the test results; combine the actual labels of the original data sample set to obtain the final result of the model test (recognition accuracy);

[0152] 5) Collect the data of the residual current mutation, the fundamental wave of the unbalanced current mutation, and the third harmonic to be judged for the second time, normalize them, and input them into the RF classification model constructed in step 4) for second judgment, so as to identify whether it is a leakage current fault.

[0153] Application Examples

[0154] (1) Primary discrimination verification

[0155] To verify the feasibility of the method, a laboratory Figure 4 In the closed test system shown, the resistance of a 100kW three-phase adjustable resistor box is adjusted to simulate the load switching of a three-phase line. Figure 4 In the figure, CTR and CT0 are current transformers for measuring residual current and unbalanced current respectively; R Aload , R Bload , R Cload is a three-phase load; C A , C B , C C is the three-phase capacitance to ground, with a value of 1μF; R SG is the grounding resistance of the repeated grounding point, which is 4Ω; R Nline is the zero line resistance, with a value of 0.05Ω; R OG is the neutral point grounding resistance, take 1Ω; R H is the ground fault transition resistance, with a value of 6.67Ω, corresponding to a fault current of 33mA. r | and fault current variation like Figure 5 shown.

[0156] Depend on Figure 5 It can be seen that the effective value of the residual current mutation caused by load switching is about 357mA. Only 0.4mA; the effective value of the residual current mutation caused by the fault is about 8mA. is 24.9mA, through a simple threshold I Hth =5mA, the residual current change caused by the fault and load switching can be distinguished. If it is greater than, it is judged as a fault, otherwise it is not a fault. Because 8mA>5mA, it is judged as a fault.

[0157] However, since it is impossible to completely prove that the abnormal point is indeed a leakage fault, in order to ensure the reliability of power supply, a second judgment is required after the first judgment.

[0158] (2) Secondary discrimination verification

[0159] Taking into account the reliability of power supply, the working status of the actual normal operating system is divided into three categories: normal state, fault state, and abnormal state; the processing methods for the three are no operation, tripping, and alarm respectively.

[0160] The number of decision trees n_estimators of the RF algorithm, the maximum number of features max_features considered by the optimal model, and the maximum depth of the decision tree max_depth are important parameters of the RF model, which have a great impact on the leakage fault detection effect of the entire model. The RF model parameter settings are shown in Table 1.

[0161] Table 1 Optimal parameter settings

[0162]

[0163]

[0164] RF is used to classify the first-level suspected fault points into fault and abnormal conditions. When training the model, 200 randomly selected abnormal points and 200 fault points are used as training samples, and 111 of each constitute the test set. Different feature vectors are selected for training, and different combinations of feature quantities are input into the RF model for training and testing. The fault detection success rate is shown in Table 2.

[0165] Table 2 Fault detection success rate under different feature quantities

[0166] Eigenvector Number of features Classification success rate / % Total detection success rate / % <![CDATA[K 012345 ]]> 24 100 86.55 <![CDATA[K 024 ]]> 12 77.1 60.17 <![CDATA[K 135 ]]> 12 100 82.21

[0167] K 012345 represents all eigenvectors composed of odd harmonics and even harmonics, K 024 The eigenvector K represents the even harmonic composition. 135 Eigenvector representing the composition of odd harmonics.

[0168] In Table 2, the classification success rate indicates the classification success rate of fault points and abnormal points; the total detection success rate indicates the total fault detection success rate. In combination 1, the number of feature quantities is 24, that is, all odd harmonics and even harmonic components are used as input quantities. RF can distinguish all fault points and abnormal points, and the classification success rate can reach 100%, and the total detection success rate is 86.55%; in combination 2, when only 12 feature quantities of even harmonics are used as input, the classification success rate is only 60.17%; in combination 3, when only 12 feature quantities of odd harmonics are used as input, the fault detection success rate can also reach 82.21%. The above shows that the main difference between fault points and abnormal points lies in odd harmonics.

[0169] In summary, the method proposed in the present invention can achieve reliable detection of more than 85% of leakage faults when the inherent residual current amplitude is higher than the fault current amplitude.

[0170] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A two-level identification method for leakage fault in TN-CS system based on mutation ratio and random forest algorithm, characterized in that: The steps include: The TN-CS system loop is analyzed. By analyzing the quantitative relationship between the residual current mutation and unbalanced current mutation before and after the ground fault occurs and the neutral line impedance and ground impedance, the first-level fault discrimination based on the current mutation vector ratio is performed based on the principle that the current mutation vector ratio remains unchanged when the load is switched, and the first-level discrimination result is obtained. If the primary judgment result is a fault, the residual current, the fundamental wave of the unbalanced current mutation, and the third harmonic are used as fault identification feature quantities, and the random forest algorithm is used for secondary judgment to determine whether it is a leakage current fault.

2. The secondary identification method for leakage fault in TN-CS system based on mutation ratio and random forest algorithm according to claim 1 is characterized in that: The specific method of primary discrimination is: (1.1) Calculate the ratio M of the residual current mutation amount and the unbalanced current mutation amount in real time according to formula (16); in, is the unbalanced current mutation amount; is the residual current mutation amount; (1.2) Calculate the current moment in real time according to formula (17) Among them, K is a constant complex number, which is approximately a constant real number; is the fault current change value; (1.3)Judgement If it is greater than the fault judgment threshold, it is judged as a fault, otherwise it is judged as no fault.

3. The secondary identification method for leakage fault in TN-CS system based on mutation ratio and random forest algorithm according to claim 1 is characterized in that: in, is the A phase current; is the B phase current; is the C phase current; is the residual current; is the unbalanced current on the transformer side; is the grounded neutral current.

4. The two-level identification method for leakage fault in TN-CS system based on mutation ratio and random forest algorithm according to claim 1 is characterized in that: in, is the unbalanced current on the transformer side; is the unbalanced current on the load side; Three-phase to ground leakage current sum.

5. The secondary identification method for leakage fault in TN-CS system based on mutation ratio and random forest algorithm according to claim 1 is characterized in that: The specific method of secondary discrimination is: (2.1) Through artificial experiments, the ground fault is simulated, the fault current amplitude is collected, and the original sample set is obtained. (2.2) Select the normalized data of residual current mutation, fundamental wave of unbalanced current mutation and third harmonic as fault identification feature; (2.3) Randomly classify the normalized data sample set to obtain training set samples and test set samples; (2.4) Using the training set, the normalized data of the residual current mutation, the fundamental wave of the unbalanced current mutation, and the third harmonic are used as the input of the RF classification model, and the corresponding working state is used as the output to train and construct the RF classification model; and using the test set to test the constructed RF classification model; the working states are normal state, ground fault, load shedding fault, and leakage current fault; (2.5) Collect the data of the residual current mutation, the fundamental wave of the unbalanced current mutation, and the third harmonic to be secondary judged, normalize them, and input them into the RF classification model constructed in step (2.4) for secondary judgment to obtain the corresponding working state, so as to distinguish whether it is a leakage current fault.

6. The secondary identification method for leakage fault in TN-CS system based on mutation ratio and random forest algorithm according to claim 5 is characterized in that: Normalization uses the maximum-minimum normalization method.

7. The secondary identification method for leakage fault in TN-CS system based on mutation ratio and random forest algorithm according to claim 5 is characterized in that: The sample ratio of the training set and the test set is 7:

3.

8. The secondary identification method for leakage fault in TN-CS system based on mutation ratio and random forest algorithm according to claim 5 is characterized in that: In the RF classification model, the number of RF algorithm decision trees n_estimators is 42, and the maximum number of features max_features considered by the optimal model is 4.

9. The secondary identification method for leakage fault in TN-CS system based on mutation ratio and random forest algorithm according to claim 8 is characterized in that: In the RF classification model, the maximum depth of the decision tree is 15.

Images