Converter transformer saturation protection method and system based on fusion network

Abstract

The method comprises the following steps: collecting three-phase currents and neutral point zero sequence currents of a converter transformer network side; inputting the three-phase currents and the neutral point zero sequence currents of the converter transformer network side into a trained fusion network LSTM-Transformer model to obtain a DC bias prediction value; and when a single condition criterion or a composite condition criterion is met according to the DC bias prediction value, immediately locking the saturation protection; wherein the single condition criterion comprises that the DC bias prediction value is less than a set upper limit and the discrete slope of the DC bias prediction value is less than a set slope threshold within a set time; and the composite condition criterion comprises that the proportion of the number of DC bias prediction values less than a set threshold in the total number of all DC bias prediction values is greater than a set proportion threshold. The application effectively utilizes the locking criterion when only the magnetizing inrush current or the magnetizing inrush current exists in the converter transformer, and prevents the saturation protection from malfunctioning.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of direct current transmission converter transformer protection, and more particularly relates to a converter transformer saturation protection method and system based on a fusion network. BACKGROUND

[0002] The extra-high voltage direct current transmission technology has the characteristics of long distance, large capacity, low loss and good controllability, and is an effective way to realize the intensive development of clean energy and the optimal allocation of energy across regions. The extra-high voltage direct current transmission system adopts a three-phase split converter transformer. During the operation of the converter transformer, the direct current bias magnetization phenomenon may occur, which leads to the persistent saturation of the core of the converter transformer and the generation of local overheating. Therefore, the converter transformer is configured with saturation protection to prevent the influence of direct current bias magnetization on the core of the converter transformer.

[0003] In actual operation, the saturation protection indirectly calculates the size of the neutral point direct current by calculating the peak value of the neutral point magnetizing current on the grid side. The magnetizing inrush current during the charging and switching of the converter transformer is easily affected by the closing angle and residual magnetism, which may lead to the misoperation of the saturation protection and seriously affect the operation reliability of the saturation protection. In the past three years, the magnetizing inrush current has been too large in many converter stations, leading to the operation of the saturation protection of the converter transformer and further causing the tripping of the equipment.

[0004] Therefore, there is a need for a method for identifying the magnetizing inrush current and direct current bias magnetization of a converter transformer, and on this basis, the saturation protection in the prior art is improved. SUMMARY

[0005] To solve the problems in the prior art, the application provides a converter transformer saturation protection method and system based on a fusion network. When only the magnetizing inrush current or the magnetizing inrush current is the main component exists in the converter transformer, the saturation protection is effectively blocked to prevent the misoperation of the saturation protection caused by the excessive magnetizing inrush current.

[0006] The application adopts the following technical solutions.

[0007] The first aspect of the application provides a converter transformer saturation protection method based on a fusion network, comprising:

[0008] The three-phase current and the neutral point zero sequence current on the grid side of the converter transformer are collected;

[0009] The three-phase current and the neutral point zero sequence current on the grid side of the converter transformer are input into the trained fusion network LSTM-Transformer model to obtain a direct current bias magnetization prediction value;

[0010] According to the direct current magnetic bias prediction value, when the single working condition criterion or the composite working condition criterion is met, the saturation protection is immediately locked; wherein, the single working condition criterion includes that the direct current magnetic bias prediction value is less than the set upper limit within the set time, and the discrete slope of the direct current magnetic bias prediction value is less than the set slope threshold; the composite working condition criterion includes that the proportion of the number of the direct current magnetic bias prediction value less than the set threshold in the total number of all direct current magnetic bias prediction values is greater than the set proportion threshold.

[0011] Preferably, the training of the fusion network LSTM-Transformer model comprises:

[0012] Step A.1, determine a plurality of excitation inrush current simulation scenarios and parameters with a set closing angle as a step;

[0013] Step A.2, determine a plurality of direct current magnetic bias simulation scenarios and parameters within a set neutral point current range of the converter transformer network side with a set neutral point current step; the plurality of excitation inrush current simulation scenarios and parameters, the plurality of direct current magnetic bias simulation scenarios and parameters constitute a training set;

[0014] Step A.3, construct a feature vector x t with each state vector in the training set before time t;

[0015] x t ={f t ,v t ,k t ,s t ,p t} T

[0016] In the formula:

[0017] f t , v t , k t , s t , p t all represent each state vector before time t;

[0018] Step A.4, normalize and segment the feature vector, and input the segmented time sequence x n ={X1, X2, …, X n} into the fusion network LSTM-Transformer model for feature learning, wherein the normalization and segmentation processing are respectively represented by the following formulas,

[0019]

[0020] In the formula:

[0021] x t ' represents the normalized feature vector;

[0022] min(x t ), max(x t ) are respectively a vector composed of minimum values of each state vector in the feature vector and a vector composed of maximum values of each state vector;

[0023] x1′, x′2, …, x′ n+k-1 are respectively normalized feature vectors at the first time, the second time, …, the n+k-1 time;

[0024] n is the number of time periods;

[0025] k is the number of time points contained in each time period;

[0026] X1…X n represent a sequence of normalized feature values of the neutral point current of the converter transformer in the n time periods;

[0027] Step A.5, according to the feature learning result of the previous step and in combination with the feature vector of the current step, outputting the DC bias magnetic field prediction value.

[0028] Preferably, the fusion network LSTM-Transformer model comprises an LSTM module, a fusion layer and a multi-head self-attention mechanism, which are respectively represented by the following formulas,

[0029]

[0030] In the formula,

[0031] represents the output vector of the LSTM network;

[0032] represents the current hidden layer vector;

[0033] G t represents the output vector of the multi-head attention operation;

[0034] g(h t-1 ) represents the vector after the sigmoid function operation of the hidden layer vector;

[0035] L represents the long short-term memory recursive operation, F represents the feature fusion operation, and T represents the multi-head attention operation.

[0036] Preferably, the single working condition criterion is used to lock the saturation protection when it is judged that the neutral point zero sequence current is the excitation inrush current, and is represented by the following formula,

[0037]

[0038] In the formula,

[0039] F(t) is a DC bias prediction value;

[0040] a is a first threshold value;

[0041] k1 represents a discrete slope;

[0042] N is the total number of DC bias prediction values;

[0043] F(t=t i ) represents the DC bias prediction value at time t i ;

[0044] b is a second threshold value;

[0045] t s represents a set time.

[0046] Preferably, the set time is 6s, and the three-phase current of the converter transformer bushing and the neutral point zero sequence current are collected every 0.02s to input into a fusion network for prediction; the first threshold value is set to 0.1, and the second threshold value is set to 0.05.

[0047] Preferably, when the composite working condition criterion is applied, it further includes that the DC bias content is lower than the maximum DC current value allowed to flow through the neutral point.

[0048] Preferably, the composite working condition criterion is used to lock out the saturation protection when it is judged that the magnetizing inrush current and the DC bias exist simultaneously in the neutral point zero sequence current, but the proportion of the magnetizing inrush current is greater than a critical value, which is expressed by the following formula,

[0049]

[0050] In the formula,

[0051] F(t=t i ) represents the DC bias prediction value at time t i ;

[0052] c is a third threshold value;

[0053] IF[*] is a conditional function, which is 1 when the condition * is met, and 0 when the condition * is not met;

[0054] N is the total number of DC bias prediction values;

[0055] n* is the number of DC bias prediction values less than the third threshold value;

[0056] m is a fourth threshold value;

[0057] k2 represents the proportion of the prediction value point of the DC bias prediction value less than the third threshold value to the total prediction value point.

[0058] Preferably, the third threshold is set to 0.05, and the fourth threshold is set to 0.93.

[0059] The second aspect of the present application provides a converter transformer saturation protection system based on a fusion network, which runs the converter transformer saturation protection method based on the fusion network, comprising:

[0060] A signal acquisition module is configured to acquire three-phase currents and a neutral point zero sequence current of a converter transformer bushing on a network side;

[0061] A model training module is configured to train a fusion network LSTM-Transformer model, and calculate a DC bias prediction value according to the actually acquired three-phase currents and the neutral point zero sequence current of the converter transformer bushing on the network side.

[0062] A logic judgment module is configured to compare the DC bias prediction value according to a set criterion condition, and immediately lock the saturation protection when any condition is met.

[0063] The third aspect of the present application provides a computer readable storage medium, which stores a computer program, and the program is executed by a processor to realize the steps of the method.

[0064] The fourth aspect of the present application provides a computer device / system, comprising:

[0065] A memory is configured to store a computer program / instruction;

[0066] A processor is configured to execute the computer program / instruction to realize the steps of the method.

[0067] The fifth aspect of the present application provides a computer program product, comprising a computer program / instruction, and the computer program / instruction is executed by a processor to realize the steps of the method.

[0068] Compared with the prior art, the present application has at least the following beneficial effects: the present application provides a converter transformer saturation protection method and system based on a fusion network LSTM-Transformer, which utilizes a neural network algorithm to effectively identify a pure field magnetizing inrush current, a pure DC bias, and a field magnetizing inrush current combined with a DC bias, and can effectively avoid saturation protection action caused by a pure field magnetizing inrush current or a field magnetizing inrush current as the main component, thereby improving the reliability of saturation protection action. BRIEF DESCRIPTION OF DRAWINGS

[0069] Figure 1 is a flowchart of a converter transformer saturation protection method based on a fusion network LSTM-Transformer provided by the present application;

[0070] Figure 2is a fusion network LSTM-Transformer closed-loop criterion schematic diagram provided by the embodiment of the application;

[0071] Figure 3 is a fusion network LSTM-Transformer data training schematic diagram provided by the embodiment of the application;

[0072] Figure 4 is a simple excitation inrush current prediction schematic diagram provided by the embodiment of the application;

[0073] Figure 5 is a simple DC bias magnetic field prediction schematic diagram provided by the embodiment of the application;

[0074] Figure 6 is a DC bias magnetic field and excitation inrush current composite prediction schematic diagram provided by the embodiment of the application;

[0075] Figure 7 is a field example waveform prediction schematic diagram provided by the embodiment of the application. DETAILED DESCRIPTION

[0076] In order to make the purpose, technical scheme and advantages of the present application clearer, the technical scheme of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. The embodiments described in the present application are only a part of the embodiments of the present application, not all the embodiments. Based on the spirit of the present application, other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0077] The present application provides a method for preventing the maloperation of the converter transformer saturation protection during the excitation inrush current, which increases the closed-loop criterion of the excitation inrush current component on the basis of the original converter transformer saturation protection criterion, and locks the saturation protection when the neutral point zero sequence current of the converter transformer meets the fusion data network prediction relationship. The innovation of the present application is as follows: 1. The excitation inrush current and the distorted current caused by the DC bias magnetic field are identified by the change of the predicted value of the neutral point zero sequence current in the fusion network LSTM-Transformer model; 2. Whether the predicted value of the neutral point zero sequence current meets the simple excitation inrush current criterion is calculated; 3. Whether the predicted value of the neutral point zero sequence current meets the criterion of the composite excitation inrush current of the DC bias magnetic field and the excitation inrush current which is the main component is calculated.

[0078] As shown in Figure 1 , the embodiment 1 of the present application provides a converter transformer saturation protection method based on the fusion network LSTM-Transformer, which includes the following steps:

[0079] Step 1, the three-phase current and the neutral point zero sequence current of the converter transformer are collected, and the timing is started when the converter transformer saturation protection starting condition is met.

[0080] Preferably, according to the neutral point zero sequence current of the converter transformer, the peak current of each unit power frequency period is obtained, and the saturation protection starts timing when the average value of the 6 cycle peak currents is greater than the inverse time curve setting value.

[0081] Step 2, input the actually collected converter transformer grid side bushing three-phase current I acA , acB , acC and the grid side neutral point zero sequence current I fz0 into the trained fusion network LSTM-Transformer model to obtain a direct current bias prediction value F(t), which represents the probability that the model predicts the neutral point zero sequence current to be direct current bias or the content of direct current bias in the neutral point zero sequence current.

[0082] The distorted current when the converter transformer has a magnetizing inrush current and direct current bias is different. The transformer has a magnetizing current which is symmetrical on the positive and negative half axes without direct current bias, and only contains odd harmonics. When the transformer winding has direct current flowing through it, the direct current and alternating current magnetic flux are superimposed, and the half cycle magnetic flux consistent with the direction of the direct current bias is greatly increased, and the other half cycle magnetic flux is reduced, resulting in the magnetizing current working in the saturation of the core magnetization curve, resulting in a magnetizing current waveform that is extremely asymmetrical on the positive and negative half waves.

[0083] According to the different characteristics of the distorted current, a prediction model based on the fusion network LSTM-Transformer is obtained by inputting typical direct current bias and magnetizing inrush current waveforms. As shown in Figure 3 , the training of the fusion network LSTM-Transformer model includes the following steps:

[0084] Step A.1, set a typical magnetizing inrush current scenario, consider the comprehensiveness of the closing angle, take 1 degree as a step, and design 360 simulation scenarios and parameters as the training set of the network;

[0085] Step A.2, set a typical direct current bias scenario, set the injected converter transformer grid side neutral point current to be 0-300A, take 1A as a step, and obtain 300 simulation scenarios and parameters as the training set of the network;

[0086] It is worth noting that other steps can also be selected according to actual needs, such as 2 degrees, 3 degrees, 2A, etc., to obtain different numbers of magnetizing inrush current or direct current bias simulation scenarios, which are based on the spirit of the present application and fall within the protection scope of the present application.

[0087] Step A.3, construct a feature vector x t from each state vector in the training set before time t, which is represented by the following formula,

[0088] x t ={f t ,v t ,k t ,s t ,p t} T

[0089] In the formula:

[0090] f t v t k t s t p t Each represents a state vector before time t.

[0091] Step A.4: Normalize the feature vector and segment it, then process the segmented time series x. n ={X1, X2, ..., X n The input is fed into the LSTM-Transformer fusion network model for feature learning, where normalization and segmentation are represented by the following formulas:

[0092]

[0093] In the formula:

[0094] x t ' represents the normalized eigenvalue;

[0095] min(x t ), max(x) t ) are the vectors formed by the minimum values ​​of each state vector in the eigenvector and the vectors formed by the maximum values ​​of each state vector, respectively.

[0096] x1′、x′2、……、x′ n+k-1 These are the normalized feature vectors for the first time step, the second time step, ..., the (n+k-1)th time step, respectively;

[0097] n is the number of time periods;

[0098] k is the number of moments contained in each time period;

[0099] X1…X n This represents the sequence of normalized eigenvalues ​​of the neutral point current on the converter substation side over n time periods.

[0100] Step A.5: In the feature layer, based on the feature learning results of the previous step and combined with the feature vector of the current step, output the DC bias prediction value.

[0101] Preferably, in the feature layer, the expressions for the LSTM module, the fusion layer, and the multi-head self-attention mechanism are as follows:

[0102]

[0103] In the formula:

[0104] This represents the output vector of the LSTM network;

[0105] This represents the current hidden layer vector;

[0106] G t This represents the output vector of the multi-head attention operation;

[0107] g(h t-1 This represents the vector obtained after the hidden layer vector is processed by the sigmoid function.

[0108] L represents recursive operation based on long short-term memory, F represents feature fusion operation, and T represents multi-head attention operation.

[0109] Step 3: Based on the predicted DC bias value, when the criterion for blocking saturation protection is met, immediately block the saturation protection.

[0110] In a preferred but non-limiting embodiment of the present invention, a criterion for blocking saturation protection is added to the original converter transformer saturation protection criterion, as follows: Figure 2 As shown, it includes a single operating condition criterion (lock-in criterion 1) and a composite operating condition criterion (lock-in criterion 2).

[0111] The single-condition criterion is for the condition of simple inrush current, and is used to block saturation protection when the neutral point zero-sequence current component is determined to be simple inrush current. Specifically, it includes: blocking saturation protection when the DC bias prediction value F(t) is continuously less than the first threshold a within a set time and the discrete slope k1 is less than the second threshold b. Under the fused network LSTM-Transformer, the DC bias prediction value of the simple inrush current condition is equal to 0 or close to 0, and the slope hardly changes.

[0112] More preferably, the single operating condition criterion is expressed by the following formula:

[0113]

[0114] In the formula:

[0115] a is the first threshold, and more preferably, it is set to 0.1;

[0116] b is the second threshold, and more preferably, it is set to 0.05;

[0117] N represents the total number of predicted DC bias values, i.e., the total number of predicted value points.

[0118] F(t=t i ) represents t i DC bias prediction value at time;

[0119] k1 represents the discrete slope;

[0120] t s This indicates the set time.

[0121] In an exemplary but non-limiting implementation, the time interval is set to 6 seconds. The three-phase current of the converter transformer side bushing and the zero-sequence current of the neutral point are collected every 0.02 seconds and input into the fusion network for prediction. A total of 300 prediction points are obtained within 6 seconds. At this time, t... s =6, N=300.

[0122] The composite operating condition criterion is for operating conditions involving both inrush current and DC bias. It is used to lock out saturation protection when both inrush current and DC bias are present in the neutral point zero-sequence current, but the inrush current is the dominant component. Specifically, saturation protection is locked out when the proportion k2 of the number of predicted DC bias prediction values ​​F(t) less than the third threshold c within a set time period is greater than the fourth threshold m. Under the fused network LSTM-Transformer, when the initial inrush current of the composite operating condition is large, the predicted DC bias value is close to 0. As the inrush current decays, the predicted DC bias value gradually increases, and the higher the DC bias component, the faster the rate of change of the predicted DC bias value.

[0123] More preferably, under the combined excitation inrush current and DC bias conditions, when applying the combined condition criterion, the DC bias content should be lower than the maximum DC current value allowed to flow through the neutral point, for example, 30A.

[0124] More preferably, the composite operating condition criterion is expressed by the following formula:

[0125]

[0126] In the formula:

[0127] k2 represents the proportion of prediction points where the DC bias prediction value is less than the third threshold out of the total prediction points;

[0128] c is the third threshold, and more preferably, it is set to 0.05;

[0129] IF[*] is a conditional function. It takes the value 1 when the condition * is met and 0 when the condition * is not met.

[0130] n* represents the number of predicted DC bias values ​​that are less than the third threshold;

[0131] m is the fourth threshold.

[0132] More preferably, the value of m in the composite operating condition criterion is set to 0.93. According to simulation experimental data, m = 0.93 corresponds to the proportion of DC bias prediction values ​​less than 0.05 when the residual magnetism of a certain phase is 1.0, the closing angle is 0°, the residual magnetism of the other two phases is 0, the single-phase excitation inrush current is the largest, and the DC current injected into the grid side neutral point is 30A.

[0133] Embodiment 2 of the present invention provides a converter variable saturation protection system based on a fused network LSTM-Transformer, characterized in that it includes:

[0134] The signal acquisition module is used to acquire the three-phase current and neutral point zero-sequence current of the bushing on the converter transformer side.

[0135] The model training module is used to train the fusion network LSTM-Transformer model and calculate the DC bias prediction value based on the actual collected three-phase current of the bushing on the converter transformer side and the neutral point zero-sequence current.

[0136] The logic judgment module is used to compare the predicted DC bias value with the set criteria. When any condition is met, the saturation protection is immediately locked.

[0137] Embodiment 3 of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods described in Embodiment 1.

[0138] Embodiment 4 of the present invention provides a computer device / equipment / system, comprising:

[0139] Memory, used to store computer programs / instructions;

[0140] A processor for executing the computer program / instructions to implement the steps of any of the methods described in Embodiment 1.

[0141] Embodiment 5 of the present invention provides a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of any of the methods described in Embodiment 1.

[0142] Application examples:

[0143] To further illustrate the technical solution of the present invention and the beneficial technical effects it brings, application examples are introduced below.

[0144] like Figure 4As shown, the transformer is closed under no-load conditions when the initial phase angles of phase A voltage are 0°, 90°, 180°, and 270°. The neutral point inrush current waveform is shown in the figure, where the vertical axis represents the instantaneous current value. The generated data is then identified and predicted by the network through the fusion network LSTM-Transformer, and the corresponding results are marked as scatter plots. Figure 4 The simulation results show that, considering the influence of changes in the closing phase angle, the predicted values ​​of the fusion network LSTM-Transformer model are all 0 or close to 0, consistent with the data labels of the excitation inrush current sample set.

[0145] like Figure 5 As shown, the DC currents are 0, 100, 200, and 300 A. The simulation models of DC bias and the predicted neutral point current under these four different current values ​​are shown in the figure, where the vertical axis represents the instantaneous current value. Due to the significant difference between the inrush current and the DC bias waveform, the predicted values ​​of the fused network LSTM-Transformer model are all 1 or close to 1, consistent with the data labels of the DC bias sample set. This preliminarily proves that the fused network LSTM-Transformer can correctly identify DC bias.

[0146] like Figure 6 The figure shows the DC bias superimposed on inrush current condition. The closing angle of phase A is set to 0°, and the DC currents are 100A and 300A respectively. The simulation models of DC bias and inrush current composite under these two different scenarios, along with their neutral point current waveforms and judgment results, are shown in the figure. The vertical axis represents the instantaneous current value. As can be seen from the figure, since DC bias is almost non-existent in the first few seconds, the model judges it as 0 for the first part of the time, consistent with the data label of the inrush current sample set. As the DC bias content increases and the inrush current content decreases, the waveform gradually approaches the DC bias waveform, so the algorithm gradually approaches 1 in probability. As time continues, the DC bias increases sharply, and the predicted value of the fused network LSTM-Transformer model is gradually judged as 1 or close to 1, consistent with the data label of the DC bias sample set.

[0147] like Figure 7 The image shows the waveform prediction results for a field engineering example of DC bias and inrush current at a certain station. Figure 7 It can be seen that the DC bias prediction value of a certain station is stable at 0.08, and the slope of the discrete point is 0.02, which meets the single working condition criterion described in Embodiment 1 of the present invention. It can normally block the saturation protection and avoid protection tripping. The DC bias prediction value of a certain station is 0.92. At this time, the saturation protection is normally open and operates according to the time calculated by the inverse time curve.

[0148] Compared with the prior art, the beneficial effects of the present invention include at least the following: The present invention provides a converter transformer saturation protection method and system based on the fusion network LSTM-Transformer. By using a neural network algorithm, it can effectively identify simple inrush current, simple DC bias, and combined inrush current and DC bias scenarios, which can effectively avoid saturation protection action caused by simple inrush current or inrush current as the main component, and improve the reliability of saturation protection action.

[0149] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0150] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A converter transformer saturation protection method based on a fused network, characterized in that, include: Collect the three-phase current and neutral point zero-sequence current on the converter transformer side; The three-phase current and neutral point zero-sequence current on the converter transformer side are input into the trained fusion network LSTM-Transformer model to obtain the DC bias prediction value. The fusion network LSTM-Transformer model includes an LSTM module, a fusion layer, and a multi-head self-attention mechanism. Training the fusion network LSTM-Transformer model includes: Using the set closing angle as the step size, multiple inrush current simulation scenarios and parameters are determined; Within the set neutral point current range on the converter grid side, multiple DC bias simulation scenarios and parameters are determined using a set neutral point current step size. The training set consists of multiple inrush current simulation scenarios and parameters, and multiple DC bias simulation scenarios and parameters. The feature vector is constructed from the state vectors of the training focused before a set time. The feature vectors are normalized and segmented, and the segmented time series are input into the LSTM-Transformer fusion network model for feature learning. Based on the feature learning results of the previous step and combined with the feature vector of the current step, the DC bias prediction value is output. Based on the predicted DC bias value, saturation protection is immediately locked when a single operating condition criterion or a composite operating condition criterion is met. The single operating condition criterion includes that the predicted DC bias value is less than the set upper limit within a set time period and the discrete slope of the predicted DC bias value is less than the set slope threshold. The composite operating condition criterion includes that the proportion of the number of predicted DC bias values ​​less than the set threshold in the total number of all predicted DC bias values ​​is greater than the set proportion threshold. When applying the composite operating condition criterion, it also includes that the DC bias content is lower than the maximum DC current value allowed to flow through the neutral point.

2. The converter transformer saturation protection method based on fused network according to claim 1, characterized in that: In the training of the fusion network LSTM-Transformer model, normalization and piecewise processing are expressed by the following formulas, respectively. In the formula: For feature vectors; Represents the normalized eigenvectors; , These are the vectors formed by the minimum values ​​of each state vector in the eigenvector and the vectors formed by the maximum values ​​of each state vector, respectively. , ... They are respectively the first moment, the second moment, ... the second moment. n + k The normalized eigenvector at time -1; n Number of time periods; k The number of moments contained in each time period; … express n The normalized characteristic value sequence of neutral point current on the converter transformer side within a time period, the segmented time sequence x n = { X 1, X 2, ..., X n } 3. The converter transformer saturation protection method based on fused network according to claim 1, characterized in that: The LSTM module, fusion layer, and multi-head self-attention mechanism are represented by the following formulas. In the formula: This represents the output vector of the LSTM network; This represents the current hidden layer vector; This represents the output vector of the multi-head attention operation; This represents the vector obtained after processing the hidden layer vector using the sigmoid function. This indicates recursive operations based on Long Short-Term Memory. Indicates feature fusion operation, This represents multi-head attention operation.

4. The converter transformer saturation protection method based on fused network according to claim 1, characterized in that: The single operating condition criterion is used to block saturation protection when the neutral point zero-sequence current is determined to be inrush current, and is expressed by the following formula. In the formula: This is the predicted value for DC bias. a The first threshold; k 1 represents the discrete slope; N This represents the total number of predicted DC bias values. express t i The predicted DC bias value at time; b The second threshold; This indicates the set time.

5. The converter transformer saturation protection method based on fused network according to claim 4, characterized in that: The set time is 6 seconds. The three-phase current of the converter transformer side bushing and the zero-sequence current of the neutral point are collected every 0.02 seconds and input into the fusion network for prediction. The first threshold is set to 0.1 and the second threshold is set to 0.

05.

6. The converter transformer saturation protection method based on fused network according to claim 1, characterized in that: The composite operating condition criterion is used to lock out saturation protection when it is determined that both inrush current and DC bias exist in the neutral point zero-sequence current, but the proportion of inrush current is greater than the critical value. It is expressed by the following formula. In the formula: express t i The predicted DC bias value at time; c The third threshold; IF[*] is a conditional function. It takes the value 1 when the condition * is met and 0 when the condition * is not met. N This represents the total number of predicted DC bias values. The number of predicted DC bias values ​​that are less than the third threshold; m The fourth threshold; k 2 indicates the proportion of predicted value points where the DC bias prediction value is less than the third threshold out of the total predicted value points.

7. The converter transformer saturation protection method based on a fused network according to claim 6, characterized in that: The third threshold is set to 0.05, and the fourth threshold is set to 0.

93.

8. A converter transformer saturation protection system based on a fusion network, operating the converter transformer saturation protection method based on a fusion network as described in any one of claims 1-7, characterized in that, include: The signal acquisition module is used to acquire the three-phase current and neutral point zero-sequence current of the bushing on the converter transformer side. The model training module is used to train the fusion network LSTM-Transformer model and calculate the DC bias prediction value based on the actual collected three-phase current of the bushing on the converter transformer side and the neutral point zero-sequence current. The logic judgment module is used to compare the predicted DC bias value with the set criteria. When any condition is met, the saturation protection is immediately locked.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When executed by a processor, the program implements the steps of the method described in any one of claims 1-7.

10. A computer device / equipment / system, characterized in that, include: Memory, used to store computer programs / instructions; A processor for executing the computer program / instructions to implement the steps of the method according to any one of claims 1-7.

11. A computer program product, comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method described in any one of claims 1-7.

Images