Break method and system of circuit breaker based on single-crystal thyristor decoupling control

Abstract

This invention discloses a circuit breaker breaking method and system based on single-thyristor decoupling control, relating to the field of circuit breaker control technology. It includes: upon receiving a breaking command, sending a turn-off signal to the unidirectional thyristor, causing the thyristor to turn off at the natural current zero-crossing point; after sending the turn-off signal, continuously acquiring real-time waveform data of the main circuit current, performing decoupling state analysis on the real-time waveform data, and determining whether the main circuit has entered the decoupling completion state; only when the main circuit is determined to have entered the decoupling completion state, sending a breaking trigger signal to the circuit breaker's breaking mechanism, which then drives the mechanical contacts to perform the breaking action under zero-current conditions. This invention effectively improves the accuracy of decoupling state determination and the safety of breaking.

Description

Technical Field

[0001] This invention relates to the field of circuit breaker control technology, specifically to a circuit breaker tripping method and system based on single-crystal thyristor decoupling control. Background Technology

[0002] In existing technologies, three-phase capacitor switching circuits composed of diode rectifier bridges and unidirectional thyristors are widely used in reactive power compensation and other applications. As semi-controlled devices, thyristors cannot be actively turned off by gate signals; instead, they rely on the characteristic of the alternating current naturally crossing zero twice per cycle, automatically turning off when the current drops below the holding current. Therefore, upon receiving a tripping command, the thyristor gate trigger pulse is usually canceled first, waiting for the current to naturally cross zero to turn off the thyristor, and then the circuit breaker's mechanical contacts are controlled to disconnect the main circuit.

[0003] However, during the operation of the three-phase capacitor switching circuit composed of diode rectifier bridge and single thyristor, after the thyristor is turned off at the natural zero-crossing point of the current, the main circuit is prone to forming a decoupling artifact attenuation waveform due to the coupling effect of internal junction capacitance and stray inductance of the line. This waveform is highly similar to the morphological characteristics of the real decoupling attenuation, and traditional discrimination methods are difficult to effectively distinguish. It is easy to mistakenly judge the main circuit as the decoupling completed state during the artifact attenuation stage, and issue a disconnection command to the circuit breaker in advance. This causes the mechanical contacts to complete the disconnection under non-zero current conditions, which in turn generates an arc that burns the contact structure. At the same time, it generates an operating overvoltage with an excessive amplitude, which affects the safe and stable operation of power equipment. Summary of the Invention

[0004] This invention provides a circuit breaker tripping method and system based on single-crystal thyristor decoupling control, aiming to solve the technical problem in the prior art where the decoupling completion state is easily misjudged due to decoupling artifact interference, resulting in non-zero current tripping.

[0005] In view of the above problems, the present invention provides a circuit breaker disconnection method and system based on single-crystal thyristor decoupling control.

[0006] In a first aspect, the present invention provides a circuit breaker tripping method based on single-crystal thyristor decoupling control, comprising: Upon receiving the disconnection command, a turn-off signal is sent to the unidirectional thyristor, causing the unidirectional thyristor to turn off at the natural zero-crossing point of the current. After the shutdown signal is issued, real-time waveform data of the main circuit current is continuously collected, and decoupling status analysis is performed on the real-time waveform data to determine whether the main circuit has entered the decoupling completion state. Only when the main circuit is determined to have entered the decoupling completion state, a breaking trigger signal is sent to the breaking mechanism of the circuit breaker, and the breaking mechanism drives the mechanical contacts to perform the breaking action under zero current conditions.

[0007] Secondly, the present invention provides a circuit breaker breaking system based on single-crystal thyristor decoupling control, comprising: The thyristor turn-off module is used to send a turn-off signal to the unidirectional thyristor after receiving a turn-off command, so that the unidirectional thyristor turns off at the natural zero-crossing point of the current. The decoupling status determination module is used to continuously collect real-time waveform data of the main circuit current after the shutdown signal is issued, perform decoupling status analysis on the real-time waveform data, and determine whether the main circuit has entered the decoupling completion state. The zero-current disconnection module is used to send a disconnection trigger signal to the circuit breaker's disconnection mechanism only when the main circuit is determined to have entered the decoupling completion state. The disconnection mechanism then drives the mechanical contacts to perform a disconnection action under zero-current conditions.

[0008] One or more technical solutions provided in this invention have at least the following technical effects or advantages: This invention provides a circuit breaker breaking method and system based on single-thyristor decoupling control. Upon receiving a breaking command, the system precisely controls the unidirectional thyristor to reliably turn off at the natural zero-crossing point of the current. It continuously collects real-time waveforms of the main circuit current and performs refined decoupling state analysis, accurately distinguishing the waveform differences between true decoupling attenuation and decoupling artifact attenuation. The circuit breaker breaking mechanism is triggered only after accurately determining that the main circuit has reached the decoupling completion state, ensuring that the mechanical contacts always perform the breaking action under zero-current conditions. This eliminates the erosion damage to the contacts caused by the breaking arc at the source, while avoiding the operational overvoltage risk caused by misjudged breaking, effectively improving the safety and reliability of the three-phase capacitor switching circuit breaking process. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating the circuit breaker tripping method based on single-crystal thyristor decoupling control provided in an embodiment of the present invention. Figure 2 A circuit diagram of a three-phase capacitor switching circuit composed of a diode rectifier bridge and a single crystal thyristor provided for an embodiment of the present invention; Figure 3 This is a schematic diagram of the circuit breaker disconnection mechanism provided in an embodiment of the present invention; Figure 4 A schematic diagram of the circuit breaker breaking system based on single-crystal thyristor decoupling control provided in an embodiment of the present invention; The components represented by each number in the attached diagram are explained below: 1. Disconnecting mechanism; 2. Isolating switch; 3. Operating handle linkage rod; 4. Fuse; Thyristor turn-off module 11, decoupling state determination module 12, zero-current disconnection module 13. Detailed Implementation

[0010] This invention provides a circuit breaker tripping method and system based on single-crystal thyristor decoupling control, which addresses the technical problem in the prior art where the decoupling completion status is easily misjudged due to decoupling artifact interference, leading to non-zero current tripping.

[0011] Example 1, as Figure 1 As shown, this invention provides a circuit breaker disconnection method based on single-crystal thyristor decoupling control, applicable to, for example... Figure 2 The circuit shown is a three-phase capacitor switching circuit composed of a diode rectifier bridge and a unidirectional thyristor. In this circuit, D1, D2, D3, and D4 form a three-phase diode rectifier bridge, L is a current-limiting reactor, and V is a unidirectional thyristor. 1 and 2 are two connection nodes on the DC side of the rectifier bridge, and the reactor and unidirectional thyristor are connected in series between the two connection nodes, forming the main switching control path. The method includes: S100: After receiving the interruption command, it sends a turn-off signal to the unidirectional thyristor, causing the unidirectional thyristor to turn off at the natural zero-crossing point of the current.

[0012] In a bridge-type three-phase capacitor switching circuit composed of a diode rectifier bridge and unidirectional thyristors, the current flowing through the unidirectional thyristor is not a standard sine wave, but a composite waveform resulting from the distribution of the three-phase currents by the rectifier bridge. The zero-crossing point of this composite waveform differs from the zero-crossing point of any single-phase standard sine wave and cannot be obtained by simply detecting the zero-crossing point of a single-phase voltage or current. If the gate trigger pulse is withdrawn based solely on experience or a fixed delay after the disconnection command is issued, the unidirectional thyristor is likely to fail to turn off accurately at the natural current zero-crossing point, thus affecting the reliability of subsequent decoupling state analysis and zero-current disconnection. Therefore, it is necessary to accurately predict the time of the next composite current zero-crossing point based on the three-phase current waveform data and the diode conduction state, so as to withdraw the gate trigger pulse at the accurate time, allowing the unidirectional thyristor to turn off automatically at the natural current zero-crossing point.

[0013] Step S100 in the method provided in this embodiment of the invention includes: Upon receiving the disconnection command, the three-phase current waveform data of the three-phase capacitor is collected in real time. The three-phase current waveform data are preprocessed to obtain a preprocessed multidimensional input feature matrix; The multidimensional input feature matrix is ​​input into the pre-trained zero-crossing prediction model, and the zero-crossing prediction model outputs the time interval from the current time to the time of the next synthetic current zero-crossing. Add the time interval to the current time to obtain the time of the next zero-crossing of the synthesized current; At the moment when the next synthesized current crosses zero, the trigger pulse applied to the gate of the unidirectional thyristor is removed, causing the unidirectional thyristor to turn off automatically at the natural current zero-crossing point.

[0014] First, upon receiving the disconnection command, the three-phase current waveform data of the three-phase capacitor is acquired in real time. The three-phase current waveform data refers to the continuous sampling data of the instantaneous current values ​​of phases a, b, and c of the three-phase capacitor, collected by current sensors, as a function of time. Hall effect current sensors are used to collect the currents of phases a, b, and c of the three-phase capacitor respectively, and a fixed sampling frequency is set to collect the instantaneous current values, forming continuous three-phase current waveform data.

[0015] For example, the sampling frequency is set to 10kHz. After receiving the disconnection command, the instantaneous values ​​of phase a current (+5A), phase b current (-2A), and phase c current (-3A) are collected in real time to form continuous three-phase current waveform data.

[0016] Secondly, the three-phase current waveform data is preprocessed to obtain a preprocessed multidimensional input feature matrix, including: Extract the phase a current sampling sequence, phase b current sampling sequence, and phase c current sampling sequence from the three-phase current waveform data within a preset time window before the current moment; The phase a current sampling sequence, the phase b current sampling sequence, and the phase c current sampling sequence are normalized respectively to obtain the normalized phase a current sampling sequence, the normalized phase b current sampling sequence, and the normalized phase c current sampling sequence. Based on the a-phase current sampling sequence, the b-phase current sampling sequence, and the c-phase current sampling sequence, at each sampling moment, the instantaneous values ​​of the a-phase current, the b-phase current, and the c-phase current are compared to determine the first conducting phase and the second conducting phase at the sampling moment. The first conducting phase is the phase identifier corresponding to the maximum value among the three, and the second conducting phase is the phase identifier corresponding to the minimum value among the three. The phase identifier of the first conducting phase and the phase identifier of the second conducting phase at each sampling time are combined into a conduction state code. The conduction state codes at all sampling times are arranged in chronological order to generate a current conduction state feature sequence. The normalized a-phase current sampling sequence, the normalized b-phase current sampling sequence, the normalized c-phase current sampling sequence, and the current conduction state feature sequence are concatenated along the channel dimension to obtain the multidimensional input feature matrix.

[0017] First, the phase a current sampling sequence, phase b current sampling sequence, and phase c current sampling sequence within a preset time window prior to the current moment are extracted from the three-phase current waveform data. The preset time window is a pre-defined time length used to extract historical current data. The current sampling sequence refers to the set of all current sampling data of a single phase within the preset time window, arranged in chronological order.

[0018] Specifically, from the three-phase current waveform data, the current sampling data of phases a, b, and c within a preset time window before the current moment are extracted to form phase a current sampling sequence, phase b current sampling sequence, and phase c current sampling sequence, respectively. For example, if the preset time window is set to 10ms, the sampling frequency is 10kHz, and the current moment is t1, 100 sampling points within 10ms before t1 are extracted to form phase a current sampling sequence, phase b current sampling sequence, and phase c current sampling sequence.

[0019] Next, the phase a current sampling sequence, the phase b current sampling sequence, and the phase c current sampling sequence are normalized to obtain normalized phase a current sampling sequence, normalized phase b current sampling sequence, and normalized phase c current sampling sequence. The normalization process uses the Min-Max normalization formula to calculate the values ​​for each phase a current sampling sequence, mapping the current values ​​to the [0,1] interval to eliminate the influence of numerical dimensions. The normalized current value = (current instantaneous current value - minimum current sequence value) / (maximum current sequence value - minimum current sequence value).

[0020] For example, the maximum value of the phase a current sampling sequence is +5A and the minimum value is -5A. Each instantaneous value in this phase a current sampling sequence is calculated using the same normalization standard. Taking a phase a current instantaneous value of +5A from the sequence, its normalized value is (+5 - (-5)) / (+5 - (-5)) = 1. Similarly, normalization is performed on all other current sampling data to obtain the normalized phase a current sampling sequence, the normalized phase b current sampling sequence, and the normalized phase c current sampling sequence.

[0021] Furthermore, based on the a-phase current sampling sequence, the b-phase current sampling sequence, and the c-phase current sampling sequence, at each sampling moment, the instantaneous values ​​of the a-phase current, the b-phase current, and the c-phase current are compared to determine the first conducting phase and the second conducting phase at the sampling moment. The first conducting phase is the phase identifier corresponding to the maximum value among the three, and the second conducting phase is the phase identifier corresponding to the minimum value among the three.

[0022] Here, the first conducting phase refers to the phase identifier corresponding to the maximum instantaneous value of the three-phase current (a, b, c) at a single sampling moment. The second conducting phase refers to the phase identifier corresponding to the minimum instantaneous value of the three-phase current (a, b, c) at a single sampling moment. For each sampling moment within the preset time window, the instantaneous values ​​of the currents in phases a, b, and c are compared, and the phase corresponding to the maximum value is selected as the first conducting phase, and the phase corresponding to the minimum value is selected as the second conducting phase.

[0023] For example, at sampling time t1, the instantaneous value of the current in phase a is +5A, the instantaneous value of the current in phase b is -2A, and the instantaneous value of the current in phase c is -3A. At this time, the maximum instantaneous value of the three-phase power current is +5A, corresponding to phase a, therefore the first conducting phase is identified as phase a; the minimum value is -3A, corresponding to phase c, therefore the second conducting phase is identified as phase c. Similarly, the first and second conducting phases are determined for all other sampling times.

[0024] Furthermore, the phase identifier of the first conducting phase and the phase identifier of the second conducting phase at each sampling moment are combined into a conduction state code. The conduction state codes of all sampling moments are arranged in chronological order to generate a current conduction state feature sequence. The conduction state code refers to the code formed by combining the first and second conducting phase identifiers at a single sampling moment. The current conduction state feature sequence refers to the set of conduction state codes of all sampling moments within a preset time window arranged in chronological order.

[0025] Specifically, the first and second conducting phase identifiers at each sampling moment are combined to form a conduction state code. The conduction state codes of all sampling moments within a preset time window are arranged in chronological order to generate a current conduction state feature sequence. For example, the conduction state code at sampling moment t1 is ac, indicating that a is in and c is out. The conduction state codes of 100 sampling moments within the preset time window are arranged in chronological order to form a current conduction state feature sequence of length 100.

[0026] Finally, the normalized phase a current sampling sequence, the normalized phase b current sampling sequence, the normalized phase c current sampling sequence, and the current conduction state feature sequence are concatenated along the channel dimension to obtain the multidimensional input feature matrix. The channel dimension is the category dimension of the feature data. The multidimensional input feature matrix refers to the matrix formed by concatenating the normalized three-phase current sampling sequence and the current conduction state feature sequence along the channel dimension.

[0027] Specifically, the normalized a-phase current sampling sequence, the normalized b-phase current sampling sequence, the normalized c-phase current sampling sequence, and the current conduction state feature sequence are concatenated along the channel dimension to obtain a multi-dimensional input feature matrix. For example, the normalized a-phase, b-phase, and c-phase current sequences with 100 sampling points are concatenated with the current conduction state feature sequence with 100 sampling points to form a 100-row × 4-column multi-dimensional input feature matrix.

[0028] Next, the multidimensional input feature matrix is ​​input into the pre-trained zero-crossing prediction model, and the zero-crossing prediction model outputs the time interval from the current time to the next time the synthesized current crosses zero.

[0029] The pre-training process of the zero-crossing prediction model includes: A bridge-type single-crystal thyristor power capacitor switching simulation circuit was constructed, and simulation operation was carried out under various operating conditions using the bridge-type single-crystal thyristor power capacitor switching simulation circuit. During the simulation, the three-phase current waveform data is recorded, and the zero-crossing time of the composite current flowing through the unidirectional thyristor in each current cycle is marked. For each sampling moment, the time difference between the sampling moment and the next zero-crossing point of the composite current is used as the zero-crossing time label of the sampling moment; Using the three-phase current waveform data within a preset time window as the raw data, and extracting multi-dimensional input feature matrix samples from the raw data, the multi-dimensional input feature matrix samples are combined with the corresponding zero-crossing time labels to form a simulation training sample set. An initial zero-crossing prediction model is constructed, which includes a temporal feature extraction branch, a conduction state encoding branch, and a feature fusion regression branch. The initial zero-crossing prediction model is trained in a supervised manner using the simulation training sample set. During the training process, the mean square error between the prediction time interval and the zero-crossing time label is minimized as the optimization objective until the mean square error converges to below a preset convergence threshold, thus obtaining the zero-crossing prediction model.

[0030] First, a bridge-type single-crystal thyristor power capacitor switching simulation circuit was constructed, and its operation was simulated under various operating conditions. This simulation circuit, built using power electronics simulation software, is a model whose topology and electrical parameters are completely consistent with those of an actual three-phase capacitor switching circuit. Operating conditions refer to the combination of operating parameters such as voltage, load, and current when the three-phase capacitor switching circuit is in operation.

[0031] Specifically, a simulation circuit for switching a three-phase capacitor, consisting of a diode rectifier bridge and unidirectional thyristors, was constructed. Four typical operating conditions were set: rated voltage, light load, heavy load, and grid voltage fluctuation. The simulation circuit was then started and ran continuously. For example, the main circuit parameters of the simulation circuit were: three-phase rated voltage 380V, capacitor capacity 10kVar, and unidirectional thyristor rated current 10A. Simulations were conducted under four conditions: rated operating condition (380V, rated load), light load operating condition (380V, 50% rated load), heavy load operating condition (380V, 120% rated load), and voltage fluctuation operating condition (360V~400V fluctuation, rated load).

[0032] Secondly, during the simulation, the three-phase current waveform data is recorded, and the zero-crossing moment of the composite current flowing through the unidirectional thyristor in each current cycle is marked. The composite current refers to the combined current flowing through the unidirectional thyristor after being distributed by the rectifier bridge from the three-phase current. The zero-crossing moment refers to the moment when the instantaneous value of the composite current changes from positive to negative or from negative to positive.

[0033] During the simulation, three-phase current waveform data are acquired and recorded in real time. The waveform analysis tool in the simulation software is used to mark the zero-crossing moments of the unidirectional thyristor composite current in each current cycle. For example, with the simulation sampling frequency set to 10kHz, continuous sampling data of the three-phase currents (a, b, and c) are recorded. The zero-crossing moments are marked in the simulation waveform, such as the time when the composite current drops from 5A to 0A as 12.0ms and the time when it rises from -3A to 0A as 22.0ms.

[0034] Furthermore, for each sampling time, the time difference between that sampling time and the next zero-crossing point of the composite current is used as the zero-crossing time label for that sampling time. The zero-crossing time label refers to the time difference between a single sampling time and the next zero-crossing point of the composite current; it serves as the supervision label for training the zero-crossing prediction model. By iterating through all sampling times, the time difference between a single sampling time and the next zero-crossing point of the composite current is calculated, and this time difference is used as the zero-crossing time label for the corresponding sampling time. For example, if the sampling time t1 = 10.0 ms and the next zero-crossing point of the composite current is 12.0 ms, the time difference is 2.0 ms, then the zero-crossing time label for that sampling time is 2.0 ms.

[0035] Furthermore, using the three-phase current waveform data within a preset time window as the raw data, multi-dimensional input feature matrix samples are extracted from the raw data. These multi-dimensional input feature matrix samples are then combined with the corresponding zero-crossing time labels to form a simulation training sample set. The simulation training sample set, composed of multi-dimensional input feature matrix samples and corresponding zero-crossing time labels, is a dataset used to train the zero-crossing prediction model. The multi-dimensional input feature matrix samples refer to the feature matrices extracted from the simulated three-phase current waveform data that meet the input requirements of the zero-crossing prediction model.

[0036] Specifically, using the simulated three-phase current waveform data within a preset time window as the raw data, multi-dimensional input feature matrix samples are extracted according to the above preprocessing method. The multi-dimensional input feature matrix samples are then combined with the corresponding zero-crossing time labels to construct a simulation training sample set.

[0037] For example, the preset time window is 10ms, the sampling frequency is 10kHz, and a single sample contains three-phase current data of 100 sampling points. After extracting the corresponding multi-dimensional input feature matrix sample, it is combined with the zero-crossing time label of 2.0ms to form a single training sample. A total of 100,000 training samples are generated to form a sample set.

[0038] Subsequently, an initial zero-crossing prediction model is constructed, which includes a time-series feature extraction branch, a conduction state encoding branch, and a feature fusion regression branch. The initial zero-crossing prediction model is an untrained deep learning regression model with three branches. The time-series feature extraction branch extracts the time-series variation features of the three-phase current; the conduction state encoding branch extracts the classification features of the current conduction state; and the feature fusion regression branch fuses the two types of features and outputs the predicted time interval value.

[0039] Specifically, a deep learning regression model is constructed with the following structure: Temporal feature extraction branch: a single-layer Long Short-Term Memory (LSTM) network with 128 hidden neurons and the tanh activation function; On-state encoding branch: a single fully connected layer with 64 neurons and the ReLU activation function; Feature fusion regression branch: two fully connected layers, with the first layer having 32 neurons and the ReLU activation function, and the second layer having one neuron and the linear activation function; The initial zero-crossing prediction model is an end-to-end regression structure, used to input a multi-dimensional feature matrix and output time interval values.

[0040] Then, the initial zero-crossing prediction model is subjected to supervised training using the simulated training sample set. During training, the optimization objective is to minimize the mean squared error between the predicted time interval and the zero-crossing time label, until the mean squared error converges to below a preset convergence threshold, thus obtaining the zero-crossing prediction model. Supervised training refers to a training method that adjusts model parameters based on labeled training samples. The mean squared error is the average of the squared differences between the predicted value and the true label, used to measure the model's prediction accuracy. The convergence threshold is the maximum mean squared error at which the model training is considered complete.

[0041] Specifically, the initial zero-crossing prediction model is trained in a supervised manner using batch gradient descent. The optimization objective is to minimize the mean squared error between the prediction time interval and the zero-crossing time label. The training batch size is set to 64, and the learning rate is set to 0.001. The training is continued until the mean squared error converges to... At this point, training stops, and the zero-crossing prediction model is obtained. For example, during training, the initial mean squared error is 0.1, and after 500 iterations, the mean squared error decreases to... less than the convergence threshold Training is complete, and a usable zero-crossing prediction model is obtained.

[0042] Based on this, the multidimensional input feature matrix is ​​input into a pre-trained zero-crossing prediction model, which outputs the time interval from the current moment to the next zero-crossing point of the composite current. Alternatively, the multidimensional input feature matrix can be input into the pre-trained zero-crossing prediction model, which, through feature extraction and regression calculation, directly outputs the time interval from the current moment to the next zero-crossing point of the composite current. For example, inputting a 100×4 multidimensional input feature matrix into the trained zero-crossing prediction model will result in an output time interval of 2.0 ms.

[0043] Furthermore, by adding the aforementioned time interval to the current time, the time of the next zero-crossing of the synthesized current is obtained. The time of the next zero-crossing of the synthesized current is the precise moment when the synthesized current of the unidirectional thyristor is about to reach its natural zero-crossing point. The time of the next zero-crossing of the synthesized current = current time + time interval. For example, if the current time is 10.0 ms and the time interval is 2.0 ms, the time of the next zero-crossing of the synthesized current = 10.0 ms + 2.0 ms = 12.0 ms.

[0044] Finally, at the next zero-crossing point of the synthesized current, the trigger pulse applied to the gate of the unidirectional thyristor is removed, causing the unidirectional thyristor to turn off automatically at the current natural zero-crossing point. The unidirectional thyristor gate trigger pulse is the electrical signal applied to the control electrode of the unidirectional thyristor to maintain its conduction. The current natural zero-crossing point is the moment when the synthesized current decreases below the holding current. At the next zero-crossing point of the synthesized current, the trigger pulse applied to the gate of the unidirectional thyristor is immediately removed, and the unidirectional thyristor turns off automatically at the current natural zero-crossing point. For example, at the 12.0 ms natural zero-crossing point of the synthesized current, the trigger pulse to the gate of the unidirectional thyristor is removed, the current flowing through the thyristor drops to 0 A, and the unidirectional thyristor completes its turn-off.

[0045] In this embodiment of the invention, accurate prediction of the zero-crossing point of the synthesized current in a bridge-type single-thyristor circuit is achieved. This zero-crossing prediction method does not rely on the zero-crossing point detection of any single-phase sine wave, but directly extracts features from the three-phase current waveform data and diode conduction states, using a trained zero-crossing prediction model to output an accurate time interval. Compared with fixed-delay or simple threshold methods, this method can adapt to different operating conditions, ensuring that the unidirectional thyristor is always turned off at the natural current zero-crossing point. The accuracy of the turn-off timing provides a reliable zero-current condition for subsequent decoupling state analysis and zero-current mechanical disconnection, avoiding residual current in the main circuit caused by inaccurate thyristor turn-off timing.

[0046] S200: After issuing the shutdown signal, continuously collect real-time waveform data of the main circuit current, perform decoupling state analysis on the real-time waveform data, and determine whether the main circuit has entered the decoupling completion state.

[0047] After a unidirectional thyristor turns off at its natural zero-crossing point, the main circuit current does not immediately decay to zero but undergoes a decoupling decay process. During this process, due to the presence of device junction capacitance, stray inductance, and distributed line parameters, high-frequency oscillation components are superimposed on the current waveform. The decay caused by the actual release of electromagnetic energy is called true decoupling decay, while the waveform distortion caused by measurement noise, quantization error, or non-physical oscillations is called decoupling artifact decay. These two types of decay are extremely similar in waveform morphology and are difficult to distinguish using simple amplitude or period thresholds. If the main circuit is determined to have entered the decoupling completion state during the decoupling artifact decay stage, and a breaking trigger signal is sent to the circuit breaker breaking mechanism, the mechanical contacts will break under non-zero current conditions, generating an arc and operational overvoltage. Therefore, it is necessary to extract effective characteristic parameters from the decoupling decay stage and use a classification model to accurately distinguish between true decoupling decay and decoupling artifact decay, only determining the decoupling completion state after confirming the completion of true decoupling decay.

[0048] Step S200 in the method provided in this embodiment of the invention includes: After the shutdown signal is issued, real-time waveform data of the main circuit current is continuously collected. The first zero-crossing point of the main circuit current after the unidirectional thyristor is turned off is located from the real-time waveform data, and a waveform segment within a preset time period after the first zero-crossing point is extracted as the decoupling attenuation segment. Detect all local maxima with positive values ​​within the decoupling attenuation segment, take the detected local maxima with positive values ​​as positive peaks, and arrange all positive peaks in chronological order to form a positive peak sequence. The amplitude ratio of each pair of adjacent positive peaks in the positive peak sequence is calculated sequentially, and the median of the absolute values ​​of all amplitude ratios is taken as the peak decay rate. The amplitude ratio is the amplitude of the next positive peak divided by the amplitude of the previous positive peak in each pair of adjacent positive peaks. Calculate the time interval between each pair of adjacent positive peaks in the positive peak sequence in turn, take the median of all time intervals as the average oscillation period, and take the reciprocal of the average oscillation period as the waveform oscillation frequency. The peak decay rate and the waveform oscillation frequency are combined into decay feature parameters, and the decay feature parameters are input into a pre-trained decoupling state classification model to output the decay mode category of the decoupling decay segment, wherein the decay mode category is true decoupling decay or decoupling artifact decay. When the attenuation mode category output by the decoupling state classification model is true decoupling attenuation, the main loop is determined to have entered the decoupling completion state.

[0049] First, after issuing the shutdown signal, real-time waveform data of the main circuit current is continuously acquired. The main circuit current refers to the total current flowing in the switching main circuit composed of the diode rectifier bridge, unidirectional thyristors, and three-phase capacitors. Real-time waveform data refers to a sampling sequence of the instantaneous value of the main circuit current changing over time, continuously acquired at a fixed sampling frequency.

[0050] Specifically, starting from the moment the turn-off signal is sent to the unidirectional thyristor, a high-precision current sensor is used to continuously collect the instantaneous value of the main circuit current at a fixed sampling frequency, and the current values ​​at all sampling moments are continuously stored to form real-time waveform data of the main circuit current. For example, if the sampling frequency is set to 10kHz, starting from the moment the turn-off signal is issued, the instantaneous value of the main circuit current at each sampling moment is continuously collected and recorded, and continuous sampling data such as 8A, 5A, 2A, 0A, -1.5A, and 1.2A are obtained in sequence to form real-time waveform data.

[0051] Secondly, the first zero-crossing point of the main circuit current after the unidirectional thyristor is turned off is located from the real-time waveform data, and a waveform segment within a preset time period after the first zero-crossing point is extracted as the decoupling attenuation segment.

[0052] The process of locating the first zero-crossing point of the main circuit current after the unidirectional thyristor is turned off from the real-time waveform data includes: Starting from the moment the shutdown signal is issued, traverse each sampling point in the real-time waveform data along the time-increasing direction; For each sampling point, obtain the instantaneous value of the main circuit current at the sampling point and the instantaneous value of the main circuit current at the previous sampling point; Determine whether the sign of the instantaneous value of the main circuit current at the sampling point is opposite to that of the instantaneous value of the main circuit current at the previous sampling point; When the instantaneous value of the main circuit current at the sampling point has the opposite sign to the instantaneous value of the main circuit current at the previous sampling point, the time corresponding to the sampling point with the smaller absolute value of the instantaneous value of the main circuit current between the sampling point and the previous sampling point is taken as the preliminary zero-crossing candidate time. The preliminary zero-crossing candidate time is verified for its validity. When the verification is successful, the preliminary zero-crossing candidate time is determined as the first zero-crossing time of the main circuit current after the unidirectional thyristor is turned off.

[0053] First, starting from the moment the shutdown signal is issued, each sampling point in the real-time waveform data is traversed along the time-increasing direction. A sampling point refers to each time acquisition node divided according to a fixed sampling frequency; the time-increasing direction refers to the direction of time flow backward from the moment the shutdown signal is issued. Starting from the moment the shutdown signal is issued, each sampling point in the real-time waveform data of the main circuit current is traversed sequentially along the time-increasing direction. For example, starting from 50ms when the shutdown signal is issued, the sampling points corresponding to all subsequent sampling moments are traversed sequentially: 50.1ms, 50.2ms, 50.3ms…

[0054] Secondly, for each sampling point, the instantaneous value of the main circuit current at that sampling point and the instantaneous value of the main circuit current at the previous sampling point are obtained. The previous sampling point refers to the adjacent acquisition node of the currently traversed sampling point on the time axis at the previous moment. The instantaneous current value refers to the real-time value of the main circuit current at the corresponding moment for a single sampling point. For each traversed sampling point, the instantaneous value of the main circuit current corresponding to that sampling point and the instantaneous value of the main circuit current corresponding to the previous adjacent sampling point on the time axis are read synchronously. For example, when traversing sampling point 55.0ms, the instantaneous current value of 1.8A at that moment is read, and the instantaneous current value of -1.2A corresponding to the previous sampling point 54.9ms is read simultaneously.

[0055] Next, determine whether the sign of the instantaneous main circuit current value at the sampling point is opposite to that at the previous sampling point. The sign of the current refers to the positive or negative attribute of the instantaneous current value; a positive value represents forward current flow, and a negative value represents reverse current flow. By comparing the positive and negative attributes of the instantaneous main circuit current values ​​at the current sampling point and the previous sampling point, determine whether there is a switch between positive and negative signs. For example, an instantaneous current value of -1.2A at 54.9ms is negative, while an instantaneous current value of 1.8A at 55.0ms is positive; the signs of the two instantaneous current values ​​are opposite.

[0056] Furthermore, when the instantaneous value of the main circuit current at the sampling point has the opposite sign to the instantaneous value of the main circuit current at the previous sampling point, the time corresponding to the sampling point with the smaller absolute value of the instantaneous value of the main circuit current between the sampling point and the previous sampling point is selected as the preliminary zero-crossing candidate time. The preliminary zero-crossing candidate time refers to the suspected current zero-crossing time temporarily selected after the current sign of adjacent sampling points changes. When the instantaneous current values ​​of adjacent sampling points have opposite signs, the absolute values ​​of the two sampling points are compared, and the time corresponding to the sampling point with the smaller absolute value is selected as the preliminary zero-crossing candidate time. For example, if the absolute value of the current at 54.9ms is 1.2A and the absolute value of the current at 55.0ms is 1.8A, the time with the smaller absolute value, 54.9ms, is selected as the preliminary zero-crossing candidate time.

[0057] Then, the validity of the preliminary zero-crossing candidate time is verified. When the verification is successful, the preliminary zero-crossing candidate time is determined as the first zero-crossing time of the main circuit current after the unidirectional thyristor is turned off.

[0058] The verification of the validity of the preliminary zero-crossing candidate times includes: Using the preliminary zero-crossing candidate time as the dividing point, a waveform segment of a first preset length is extracted from the real-time waveform data as the pre-zero-crossing segment, and a waveform segment of a second preset length is extracted as the post-zero-crossing segment. The pre-zero-crossing segment and the post-zero-crossing segment are then spliced ​​together in chronological order to form the zero-crossing waveform segment to be detected. Acquire historical waveform data of multiple known true zero-crossing times. For each set of historical waveform data of known true zero-crossing times, extract waveform segments of fixed length before and after the zero-crossing time as the original zero-crossing template. Amplitude normalization is performed on all original zero-crossing templates. All original zero-crossing templates after amplitude normalization are aligned on the time axis. The arithmetic mean of all the aligned original zero-crossing templates is taken for each sampling point to obtain the true zero-crossing template waveform. Calculate the waveform similarity between the zero-crossing waveform segment to be detected and the actual zero-crossing template waveform; When the waveform similarity exceeds a preset similarity threshold, the zero-crossing validity verification is confirmed to be successful. When the waveform similarity does not exceed the preset similarity threshold, the preliminary zero-crossing candidate moment is determined to be a false zero-crossing moment caused by noise. The preliminary zero-crossing candidate moment is abandoned, and the remaining sampling points in the real-time waveform data are traversed along the time-increasing direction.

[0059] First, using the preliminary zero-crossing candidate time as the dividing point, a waveform segment of a first preset length is extracted from the real-time waveform data as the pre-zero-crossing segment, and a waveform segment of a second preset length is extracted as the post-zero-crossing segment. The pre-zero-crossing segment and the post-zero-crossing segment are then concatenated in chronological order to form the zero-crossing waveform segment to be detected. The first preset length and the second preset length refer to the pre-set waveform segmenting duration before and after the zero-crossing point. The zero-crossing waveform segment to be detected refers to the complete waveform segment formed by concatenating the pre-zero-crossing segment and the post-zero-crossing segment in chronological order.

[0060] The first and second preset lengths are set based on the power grid frequency cycle and the distortion extension range of the main circuit current zero-crossing waveform. At the same time, the sampling resolution and the characteristics of electromagnetic noise interference on site are taken into account to ensure that the captured waveform segment can completely contain the complete shape of the falling edge before the current crosses zero and the rising edge after the current crosses zero. This can not only meet the feature integrity requirements of waveform template matching, but also avoid introducing irrelevant redundant waveform data due to excessive capture time.

[0061] Specifically, using the preliminary zero-crossing candidate time as a boundary, a waveform of a first preset length is extracted forward as the pre-zero-crossing segment, and a waveform of a second preset length is extracted backward as the post-zero-crossing segment. The two waveform segments are then spliced ​​together in chronological order to obtain the zero-crossing waveform segment to be detected. For example, setting the first preset length to 2ms and the second preset length to 2ms, with 54.9ms as the boundary, the waveform period from 52.9ms to 54.9ms is extracted forward as the pre-zero-crossing segment, and the waveform from 54.9ms to 56.9ms is extracted backward as the post-zero-crossing segment. These segments are then spliced ​​together to form a zero-crossing waveform segment to be detected with a total duration of 4ms.

[0062] Secondly, multiple sets of historical waveform data with known true zero-crossing times are acquired. For each set of historical waveform data with known true zero-crossing times, waveform segments of fixed length before and after the zero-crossing time are extracted as original zero-crossing templates. The original zero-crossing template refers to the standard waveform segments of fixed length before and after the zero-crossing point extracted from the historical waveforms with known true zero-crossing times. This fixed length is set based on the power grid frequency cycle and the complete oscillation duration of the main circuit current zero-crossing transition process, while also matching the waveform truncation scale for zero-crossing validity verification. Multiple sets of historical waveform data with manually confirmed true zero-crossing times are retrieved, and waveform segments of fixed length before and after the zero-crossing point are extracted from each set of data, resulting in multiple sets of original zero-crossing templates. For example, 50 sets of historical waveforms with true zero-crossing times are retrieved, and waveform segments of 2ms before and after the zero-crossing point are extracted from each set, resulting in 50 sets of original zero-crossing templates.

[0063] Next, amplitude normalization is performed on all original zero-crossing templates. These normalized templates are then aligned on the time axis. The arithmetic mean of each sample point of the aligned templates is then calculated to obtain the true zero-crossing template waveform. Time axis alignment means aligning the zero-crossing points of all original zero-crossing templates to the same time reference point. The sample-by-sample arithmetic mean is the sum of all template sample values ​​at the same time position divided by the number of templates.

[0064] Specifically, amplitude normalization is performed on all original zero-crossing templates. All normalized original zero-crossing templates are then aligned to a reference on the time axis. The arithmetic mean of all template values ​​at the same sampling time after alignment is calculated to generate the true zero-crossing template waveform. For example, after normalizing 50 sets of original zero-crossing templates and aligning them to a time reference, the average of the 50 values ​​at each sampling time is taken to fit a standard true zero-crossing template waveform.

[0065] Then, the waveform similarity between the zero-crossing waveform segment to be detected and the real zero-crossing template waveform is calculated. Waveform similarity refers to the degree of matching between the zero-crossing waveform segment to be detected and the real zero-crossing template waveform in terms of time-domain shape and amplitude variation. Cosine similarity or Pearson correlation coefficient can be used for calculation. When using cosine similarity, both the zero-crossing waveform segment to be detected and the real zero-crossing template waveform are considered as L-dimensional vectors. The calculation is completed by dividing the dot product of the two vectors by the product of their magnitudes. The range of cosine similarity is [-1, 1], with a higher similarity value closer to 1. For example, the similarity value between the zero-crossing waveform segment to be detected and the real zero-crossing template waveform in this case is 0.92.

[0066] Then, when the waveform similarity exceeds the preset similarity threshold, the validity verification of the zero-crossing point is confirmed to be successful; when the waveform similarity does not exceed the preset similarity threshold, the preliminary zero-crossing point candidate time is determined to be a false zero-crossing point caused by noise, the preliminary zero-crossing point candidate time is abandoned, and the remaining sampling points in the real-time waveform data are traversed along the time-increasing direction.

[0067] The preset similarity threshold refers to the minimum similarity value set in advance for determining whether a waveform match is qualified. The preset similarity threshold is set comprehensively based on the range of cosine similarity [-1, 1], the normal morphological deviation range of the true zero-crossing waveform of the main circuit under different operating conditions, and the degree of slight waveform distortion caused by electromagnetic sampling noise on site. A false zero-crossing point refers to a non-true current zero-crossing moment caused by electromagnetic noise and sampling interference.

[0068] Specifically, the waveform similarity is compared with a preset similarity threshold. If the waveform similarity exceeds the preset similarity threshold, the validity verification is passed, and the candidate time is determined as the first zero-crossing time. If it does not exceed the threshold, it is determined to be a false zero-crossing caused by noise, the current candidate time is abandoned, and subsequent sampling points are traversed. For example, if the preset similarity threshold is set to 0.85, and the current similarity of 0.92 is greater than 0.85, the verification is passed, and 54.9ms is determined to be the first zero-crossing time of the main circuit current after the thyristor is turned off.

[0069] Finally, a waveform segment within a preset time period after the first zero-crossing is extracted as the decoupling attenuation segment. The preset time period refers to the fixed waveform duration used to extract attenuation characteristics after the first zero-crossing. This preset time period is determined based on the inherent oscillation attenuation time constant corresponding to the stray inductance of the three-phase capacitor switching main circuit and the junction capacitance of the devices, as well as the power grid frequency period. The decoupling attenuation segment refers to the main circuit current waveform segment within the preset time period after the first zero-crossing. Starting from the moment of the first zero-crossing, real-time waveform data of the main circuit within the preset time period is extracted, and this waveform segment is defined as the decoupling attenuation segment. For example, setting the preset time period to 20ms, starting from 54.9ms after the first zero-crossing, the current waveform from 54.9ms to 74.9ms is extracted as the decoupling attenuation segment.

[0070] Next, all positive local maxima within the decoupling attenuation segment are detected. These positive local maxima are taken as positive peak values, and all positive peak values ​​are arranged in chronological order to form a positive peak value sequence. A local maximum is a point in a waveform where the current value at a sampling point is greater than the values ​​of the adjacent sampling points. A positive peak value is the current amplitude corresponding to a local maximum with a positive exponent. The positive peak value sequence is a sequence of values ​​formed by arranging all positive peak values ​​in chronological order.

[0071] Specifically, all sampling points within the decoupling attenuation section are traversed, and all local maxima with positive values ​​are selected. The corresponding current amplitudes are extracted as positive peak values, and all positive peak values ​​are arranged in ascending order of time to generate a positive peak value sequence. For example, the positive local maxima detected within the decoupling attenuation section are 1.2A, 0.9A, 0.65A, and 0.48A, respectively. Arranging them in chronological order, a positive peak value sequence is constructed: 1.2A, 0.9A, 0.65A, and 0.48A.

[0072] Then, the amplitude ratio of each pair of adjacent positive peaks in the positive peak sequence is calculated sequentially, and the median of the absolute values ​​of all amplitude ratios is taken as the peak decay rate. Here, the amplitude ratio is the amplitude of the second positive peak divided by the amplitude of the first positive peak in each pair of adjacent positive peaks. The amplitude ratio refers to the value obtained by dividing the amplitude of the second positive peak by the amplitude of the first positive peak in a positive peak sequence. The peak decay rate is the median of the absolute values ​​of all amplitude ratios.

[0073] Specifically, the amplitude ratio of each pair of adjacent positive peaks in the positive peak sequence is calculated sequentially. The absolute values ​​of all amplitude ratios are then calculated, and the median of the sorted absolute values ​​is taken. This median is the peak decay rate. For example, if the adjacent amplitude ratios of the positive peak sequence are 0.9 / 1.2, 0.65 / 0.9, and 0.48 / 0.65, the calculated ratios are 0.75, 0.722, and 0.738, respectively. Since all are positive, there is no need to take the absolute value; the median of 0.738 is taken as the peak decay rate.

[0074] Then, the time interval between each pair of adjacent positive peaks in the positive peak sequence is calculated sequentially. The median of all time intervals is taken as the average oscillation period, and the reciprocal of the average oscillation period is taken as the waveform oscillation frequency. The time interval refers to the time difference between the corresponding moments of two adjacent positive peaks. The average oscillation period is the median of the time intervals between all adjacent positive peaks. The waveform oscillation frequency is the value obtained by taking the reciprocal of the average oscillation period.

[0075] Specifically, the time interval between adjacent positive peaks in the positive peak sequence is calculated one by one, and the median of all time intervals is taken as the average oscillation period; the waveform oscillation frequency is calculated as follows: waveform oscillation frequency = 1 / average oscillation period. For example, if the time intervals between three adjacent positive peaks are 5ms, 4.8ms, and 5.1ms respectively, the median of 5ms is taken as the average oscillation period; the waveform oscillation frequency = 1 / 5ms = 0.2kHz.

[0076] Subsequently, the peak decay rate and the waveform oscillation frequency are combined into decay feature parameters, and the decay feature parameters are input into a pre-trained decoupling state classification model to output the decay mode category of the decoupling decay segment, wherein the decay mode category is true decoupling decay or decoupling artifact decay.

[0077] The pre-training process of the decoupled state classification model includes: Multiple sets of attenuation characteristic parameter samples were collected from historical decoupling attenuation segment data to form a decoupling training sample feature set. For each set of attenuation feature parameter samples in the decoupling training sample feature set, obtain the corresponding attenuation mode category label to form a decoupling training label set, wherein the attenuation mode category label is true decoupling attenuation or decoupling artifact attenuation. The support vector machine model was used as the initial classification model. The initial classification model is trained in a supervised manner using the decoupled training sample feature set and the decoupled training label set. During the training process, the training objective is to minimize the classification error of the initial classification model for the decay mode category label until the classification error converges to below a preset convergence threshold, thus obtaining the trained initial classification model. The trained initial classification model is then used as the decoupled state classification model.

[0078] First, multiple sets of attenuation characteristic parameter samples are collected from historical decoupling attenuation segment data to form a decoupling training sample feature set. The decoupling attenuation segment data refers to the attenuated oscillation waveform data within a preset time period after the main circuit current first crosses zero after the unidirectional thyristor is turned off. The attenuation characteristic parameter samples are feature data formed by a combination of the peak attenuation rate and the waveform oscillation frequency.

[0079] Specifically, multiple sets of decoupling attenuation segment data under different operating conditions are collected in batches from historical operation records and simulation data. The corresponding peak attenuation rate and waveform oscillation frequency are extracted for each set to form multiple sets of attenuation characteristic parameter samples. All samples are then summarized to form a decoupling training sample feature set. For example, 2000 sets of decoupling attenuation segment waveform data under light load, heavy load, and grid voltage fluctuation are collected. A set of peak attenuation rate and waveform oscillation frequency are calculated for each waveform, and these are summarized to form a decoupling training sample feature set containing 2000 sets of data.

[0080] Secondly, for each set of attenuation feature parameter samples in the decoupling training sample feature set, the corresponding attenuation mode category label is obtained to form a decoupling training label set. The attenuation mode category label is either true decoupling attenuation or decoupling artifact attenuation. The attenuation mode category label refers to the attribute identifier labeled for each set of attenuation feature parameter samples, and is only divided into two categories: true decoupling attenuation and decoupling artifact attenuation. Based on manual verification and waveform benchmark comparison, each set of attenuation feature parameter samples in the decoupling training sample feature set is labeled with the corresponding attenuation mode category label one by one, and all labels are summarized to form the decoupling training label set. For example, out of 2000 feature samples, 1200 are labeled as true decoupling attenuation and 800 are labeled as decoupling artifact attenuation, forming a decoupling training label set in the order of the samples.

[0081] Next, a Support Vector Machine (SVM) model is used as the initial classification model. The SVM model is a machine learning model suitable for small sample sizes and binary classification scenarios, capable of achieving high-precision class division using two-dimensional feature parameters. Using an SVM model as the initial classification model, a Gaussian kernel function is set, and the model input dimension is set to two dimensions, corresponding to the peak decay rate and waveform oscillation frequency, respectively.

[0082] Subsequently, the initial classification model is trained in a supervised manner using the decoupled training sample feature set and the decoupled training label set. During training, the goal is to minimize the classification error of the initial classification model for the decay mode category labels, until the classification error converges to below a preset convergence threshold. This trained initial classification model is then used as the decoupled state classification model. Classification error refers to the deviation between the predicted category output by the initial classification model and the manually labeled category, including the proportion of training samples misclassified. The preset convergence threshold is the maximum allowable classification error value for determining whether the model training has reached the target and for stopping iterative training.

[0083] Specifically, the initial classification model is trained in a supervised manner using a decoupled training sample feature set and a decoupled training label set. The training process aims to minimize the classification error of the decaying mode category labels, continuously iterating and updating the model's internal parameters until the classification error converges to below a preset convergence threshold of 0.01. The iteration stops, and the trained model is designated as the decoupled state classification model. For example, if the initial classification error is 0.25, after 300 iterations, the classification error drops to 0.008, which is less than the preset convergence threshold of 0.01, and the training ends, resulting in a usable decoupled state classification model.

[0084] Furthermore, the peak attenuation rate and the waveform oscillation frequency are combined into attenuation feature parameters, and these attenuation feature parameters are input into a pre-trained decoupling state classification model to output the attenuation mode category of the decoupling attenuation segment. For example, combining a peak attenuation rate of 0.738 and a waveform oscillation frequency of 0.2kHz into attenuation feature parameters and inputting them into the decoupling state classification model will output the attenuation mode category as true decoupling attenuation.

[0085] Finally, when the attenuation mode category output by the decoupling state classification model is true decoupling attenuation, the main loop is determined to have entered the decoupling completion state. If the output category of the decoupling state classification model is true decoupling attenuation, the main loop is determined to have entered the decoupling completion state; if the output category is decoupling artifact attenuation, the main loop is determined to have not completed decoupling, and waveform acquisition and cyclical analysis continue. For example, if the decoupling state classification model outputs true decoupling attenuation, the current main loop is determined to have entered the decoupling completion state.

[0086] In this embodiment of the invention, after the thyristor is turned off, the main circuit current waveform is continuously monitored. A zero-crossing waveform template matching verification method is used to accurately filter false zero-crossings caused by sampling noise, ensuring the accuracy of the initial zero-crossing location. By extracting the peak value change and oscillation period characteristics of the attenuation segment, a support vector machine classification model is used to achieve high-precision differentiation between true decoupling attenuation and decoupling artifact attenuation, eliminating misjudgments during the artifact attenuation stage. Subsequent circuit breaker tripping is only permitted after confirming true decoupling, strictly ensuring that the mechanical contacts are always under zero-current conditions during tripping. This eliminates the erosion damage to the contacts caused by the tripping arc at the source, while suppressing unnecessary operational overvoltages, improving the overall operational stability of the three-phase capacitor switching circuit and extending the equipment's service life.

[0087] S300: Only after determining that the main circuit has entered the decoupling completion state, as shown below... Figure 3 The circuit breaker shown has a tripping mechanism 1 that sends a tripping trigger signal, which drives the mechanical contacts to perform a tripping action under zero current conditions.

[0088] After the main circuit has entered the decoupling completion state, the main circuit current has decayed to zero or near zero. This is the optimal time to execute mechanical contact disconnection. However, the circuit breaker's disconnection mechanism 1 typically includes both electromagnetic tripping and thermal tripping mechanisms. The former is used for short-circuit protection, and the latter for overload protection. In conventional circuit breaker designs, both operate independently of external control signals. To actively control the disconnection, the externally generated disconnection trigger signal needs to be connected to the circuit breaker's electromagnetic-thermal tripping composite mechanism. Utilizing the mechanism's existing driving capability, the mechanical contacts are driven to separate under the condition that the main circuit current is confirmed to be zero. Furthermore, electrical isolation must be completed after the circuit breaker disconnects to prevent residual capacitor charge from posing a danger to maintenance personnel. Therefore, a complete triggering, disconnection, and isolation sequence needs to be designed.

[0089] Step S300 in the method provided in this embodiment of the invention includes: Once the main circuit is determined to have entered the decoupling completion state, a disconnection trigger signal is generated. The disconnection trigger signal is sent to the electromagnetic thermal tripping composite mechanism of the circuit breaker; The electromagnetic thermal tripping composite mechanism responds to the disconnection trigger signal and drives the mechanical contacts to separate under the condition that the main circuit current is zero, thereby completing the arc-free disconnection. After the mechanical contact completes arc-free disconnection, the isolating switch is simultaneously disconnected via the operating handle linkage rod parallel to the circuit breaker disconnection mechanism, thus completing electrical isolation.

[0090] First, upon determining that the main circuit has entered the decoupling completion state, a disconnection trigger signal is generated. This disconnection trigger signal is a fixed-level control command used to issue a disconnection start command to the circuit breaker actuator. At the moment the decoupling state classification model outputs attenuation mode category as true decoupling attenuation and determines that the main circuit has entered the decoupling completion state, a disconnection trigger signal with a preset level format is immediately generated. For example, when the main circuit is determined to reach the decoupling completion state at 74.9ms, a continuously high-level active disconnection trigger signal is immediately generated.

[0091] Next, the tripping trigger signal is sent to the electromagnetic thermal tripping composite mechanism of the circuit breaker. The electromagnetic thermal tripping composite mechanism is a composite execution structure built into the circuit breaker, integrating electromagnetic triggering, thermal overload protection, and mechanical transmission functions. It can receive external electrical control signals and respond to execute mechanical actions. The generated tripping trigger signal is transmitted in real-time to the signal input port of the electromagnetic thermal tripping composite mechanism of the circuit breaker via a dedicated control line. For example, the high-level tripping trigger signal generated at 74.9ms is stably transmitted to the control input terminal of the electromagnetic thermal tripping composite mechanism of the circuit breaker via a control cable.

[0092] Next, in response to the disconnection trigger signal, the electromagnetic thermal tripping composite mechanism drives the mechanical contacts to separate under the condition that the main circuit current is zero, completing the arc-free disconnection. The mechanical contacts are the metal contact components inside the circuit breaker that perform the functions of connecting and disconnecting the main circuit. The zero-current condition means that the instantaneous value of the main circuit current is continuously maintained within the allowable range close to zero, with no effective load current flowing. Arc-free disconnection means that during the separation of the mechanical contacts, no ionizing arc is generated because there is no current breaking down the air insulation.

[0093] Specifically, after receiving the disconnection trigger signal, the electromagnetic thermal tripping composite mechanism continuously collects the real-time value of the main circuit current and confirms for the second time that the main circuit current is stable and meets the zero-current condition. After the verification is passed, it drives the internal transmission component to slowly separate the mechanical contacts, relying on the zero-current condition to eliminate the conditions for arc generation and complete the arc-free disconnection. For example, after receiving the trigger signal, the electromagnetic thermal tripping composite mechanism detects that the main circuit current value is stable and close to 0A for three consecutive sampling cycles, confirming that the zero-current condition is met. Then, it drives the transmission structure to smoothly separate the mechanical contacts, and no arc is generated during the entire disconnection process.

[0094] Finally, after the mechanical contacts complete the arc-free disconnection, the isolating switch 2 is simultaneously disconnected via the operating handle linkage 3, which runs parallel to the circuit breaker disconnection mechanism 1, thus completing electrical isolation. The operating handle linkage 3 is a transmission rod mechanically coupled in parallel with the circuit breaker disconnection mechanism 1, and can synchronously transmit mechanical travel as the disconnection mechanism 1 moves. The isolating switch 2 is an isolation component that provides a clearly visible physical disconnection point in the power distribution circuit. Electrical isolation refers to the safe state of completely severing the electrical coupling connection of the main circuit by forming a visible disconnection gap through mechanical disconnection.

[0095] Specifically, while the mechanical contacts complete the arc-free disconnection, the mechanical stroke of the circuit breaker disconnecting mechanism 1 synchronously drives the matching operating handle linkage 3 to move. The operating handle linkage 3 further drives the isolating switch 2 to move synchronously to the disconnecting position, forming a stable physical disconnection point and completing the electrical isolation of the main circuit. For example, when the circuit breaker mechanical contacts are completely separated, the mechanical stroke of the disconnecting mechanism 1 synchronously drives the operating handle linkage 3 to move. The linkage drives the isolating switch 2 to the disconnecting position, forming a clearly visible disconnection gap in the main circuit and achieving reliable electrical isolation.

[0096] In this embodiment of the invention, the disconnection command is triggered only after the main circuit confirms the actual decoupling, eliminating the risk of premature disconnection caused by artifact attenuation misjudgment. A secondary verification of the zero-current state is performed through an electromagnetic thermal tripping composite mechanism, forming a dual-criteria guarantee to ensure that the mechanical contacts always separate under pure zero-current conditions, completely avoiding contact arcing and operational overvoltage issues. Relying on a mechanical linkage structure, the circuit breaker contacts and isolating switch operate synchronously, achieving arc-free disconnection while rapidly establishing a physical electrical isolation point, improving the safety and operational reliability of the three-phase capacitor switching circuit disconnection process.

[0097] Through the specific implementation methods described above, the embodiments of the present invention achieve the following technical effects: This invention provides a circuit breaker breaking method and system based on single-thyristor decoupling control. First, after the breaking command is issued, a pre-trained zero-crossing prediction model is used to accurately calculate the next zero-crossing point of the composite current from the three-phase current waveforms and conduction states, ensuring that the unidirectional thyristor accurately turns off at the natural current zero-crossing point, thus cutting off the main circuit current first. Second, after the thyristor turns off, the peak decay rate and waveform oscillation frequency are extracted through a positive peak sequence, and a support vector machine classification model is used to effectively distinguish between true decoupling attenuation and decoupling artifact attenuation, avoiding misjudgments caused by noise or parasitic oscillations. A breaking trigger signal is generated only after confirming that true decoupling has been completed, driving the electromagnetic thermal tripping composite mechanism to separate the mechanical contacts under zero-current conditions, effectively eliminating the breaking arc and preventing contact erosion and operational overvoltage. Finally, the isolating switch is simultaneously broken by operating the handle linkage rod, achieving significant electrical isolation and effectively improving the breaking safety and reliability of the circuit breaker in the capacitor switching circuit.

[0098] Example 2, as Figure 4 As shown, the present invention provides a circuit breaker breaking system based on single-crystal thyristor decoupling control, the system comprising: The thyristor turn-off module 11 is used to send a turn-off signal to the unidirectional thyristor after receiving the turn-off command, so that the unidirectional thyristor is turned off at the natural zero-crossing point of the current. The decoupling state determination module 12 is used to continuously collect real-time waveform data of the main circuit current after the shutdown signal is issued, perform decoupling state analysis on the real-time waveform data, and determine whether the main circuit has entered the decoupling completion state. The zero-current disconnection module 13 is used to send a disconnection trigger signal to the circuit breaker's disconnection mechanism only when the main circuit is determined to have entered the decoupling completion state, so that the disconnection mechanism drives the mechanical contacts to perform a disconnection action under zero-current conditions.

[0099] In one embodiment, the thyristor turn-off module 11 is further configured to: Upon receiving the disconnection command, the three-phase current waveform data of the three-phase capacitor is collected in real time. The three-phase current waveform data are preprocessed to obtain a preprocessed multidimensional input feature matrix; The multidimensional input feature matrix is ​​input into the pre-trained zero-crossing prediction model, and the zero-crossing prediction model outputs the time interval from the current time to the time of the next synthetic current zero-crossing. Add the time interval to the current time to obtain the time of the next zero-crossing of the synthesized current; At the moment when the next synthesized current crosses zero, the trigger pulse applied to the gate of the unidirectional thyristor is removed, causing the unidirectional thyristor to turn off automatically at the natural current zero-crossing point.

[0100] The three-phase current waveform data is preprocessed to obtain a preprocessed multidimensional input feature matrix, including: Extract the phase a current sampling sequence, phase b current sampling sequence, and phase c current sampling sequence from the three-phase current waveform data within a preset time window before the current moment; The phase a current sampling sequence, the phase b current sampling sequence, and the phase c current sampling sequence are normalized respectively to obtain the normalized phase a current sampling sequence, the normalized phase b current sampling sequence, and the normalized phase c current sampling sequence. Based on the a-phase current sampling sequence, the b-phase current sampling sequence, and the c-phase current sampling sequence, at each sampling moment, the instantaneous values ​​of the a-phase current, the b-phase current, and the c-phase current are compared to determine the first conducting phase and the second conducting phase at the sampling moment. The first conducting phase is the phase identifier corresponding to the maximum value among the three, and the second conducting phase is the phase identifier corresponding to the minimum value among the three. The phase identifier of the first conducting phase and the phase identifier of the second conducting phase at each sampling time are combined into a conduction state code. The conduction state codes at all sampling times are arranged in chronological order to generate a current conduction state feature sequence. The normalized a-phase current sampling sequence, the normalized b-phase current sampling sequence, the normalized c-phase current sampling sequence, and the current conduction state feature sequence are concatenated along the channel dimension to obtain the multidimensional input feature matrix.

[0101] The pre-training process of the zero-crossing prediction model includes: A bridge-type single-crystal thyristor power capacitor switching simulation circuit was constructed, and simulation operation was carried out under various operating conditions using the bridge-type single-crystal thyristor power capacitor switching simulation circuit. During the simulation, the three-phase current waveform data is recorded, and the zero-crossing time of the composite current flowing through the unidirectional thyristor in each current cycle is marked. For each sampling moment, the time difference between the sampling moment and the next zero-crossing point of the composite current is used as the zero-crossing time label of the sampling moment; Using the three-phase current waveform data within a preset time window as the raw data, and extracting multi-dimensional input feature matrix samples from the raw data, the multi-dimensional input feature matrix samples are combined with the corresponding zero-crossing time labels to form a simulation training sample set. An initial zero-crossing prediction model is constructed, which includes a temporal feature extraction branch, a conduction state encoding branch, and a feature fusion regression branch. The initial zero-crossing prediction model is trained in a supervised manner using the simulation training sample set. During the training process, the mean square error between the prediction time interval and the zero-crossing time label is minimized as the optimization objective until the mean square error converges to below a preset convergence threshold, thus obtaining the zero-crossing prediction model.

[0102] In one embodiment, the decoupling state determination module 12 is further configured to: After the shutdown signal is issued, real-time waveform data of the main circuit current is continuously collected. The first zero-crossing point of the main circuit current after the unidirectional thyristor is turned off is located from the real-time waveform data, and a waveform segment within a preset time period after the first zero-crossing point is extracted as the decoupling attenuation segment. Detect all local maxima with positive values ​​within the decoupling attenuation segment, take the detected local maxima with positive values ​​as positive peaks, and arrange all positive peaks in chronological order to form a positive peak sequence. The amplitude ratio of each pair of adjacent positive peaks in the positive peak sequence is calculated sequentially, and the median of the absolute values ​​of all amplitude ratios is taken as the peak decay rate. The amplitude ratio is the amplitude of the next positive peak divided by the amplitude of the previous positive peak in each pair of adjacent positive peaks. Calculate the time interval between each pair of adjacent positive peaks in the positive peak sequence in turn, take the median of all time intervals as the average oscillation period, and take the reciprocal of the average oscillation period as the waveform oscillation frequency. The peak decay rate and the waveform oscillation frequency are combined into decay feature parameters, and the decay feature parameters are input into a pre-trained decoupling state classification model to output the decay mode category of the decoupling decay segment, wherein the decay mode category is true decoupling decay or decoupling artifact decay. When the attenuation mode category output by the decoupling state classification model is true decoupling attenuation, the main loop is determined to have entered the decoupling completion state.

[0103] The process of locating the first zero-crossing point of the main circuit current after the unidirectional thyristor is turned off from the real-time waveform data includes: Starting from the moment the shutdown signal is issued, traverse each sampling point in the real-time waveform data along the time-increasing direction; For each sampling point, obtain the instantaneous value of the main circuit current at the sampling point and the instantaneous value of the main circuit current at the previous sampling point; Determine whether the sign of the instantaneous value of the main circuit current at the sampling point is opposite to that of the instantaneous value of the main circuit current at the previous sampling point; When the instantaneous value of the main circuit current at the sampling point has the opposite sign to the instantaneous value of the main circuit current at the previous sampling point, the time corresponding to the sampling point with the smaller absolute value of the instantaneous value of the main circuit current between the sampling point and the previous sampling point is taken as the preliminary zero-crossing candidate time. The preliminary zero-crossing candidate time is verified for its validity. When the verification is successful, the preliminary zero-crossing candidate time is determined as the first zero-crossing time of the main circuit current after the unidirectional thyristor is turned off.

[0104] The verification of the validity of the preliminary zero-crossing candidate times includes: Using the preliminary zero-crossing candidate time as the dividing point, a waveform segment of a first preset length is extracted from the real-time waveform data as the pre-zero-crossing segment, and a waveform segment of a second preset length is extracted as the post-zero-crossing segment. The pre-zero-crossing segment and the post-zero-crossing segment are then spliced ​​together in chronological order to form the zero-crossing waveform segment to be detected. Acquire historical waveform data of multiple known true zero-crossing times. For each set of historical waveform data of known true zero-crossing times, extract waveform segments of fixed length before and after the zero-crossing time as the original zero-crossing template. Amplitude normalization is performed on all original zero-crossing templates. All original zero-crossing templates after amplitude normalization are aligned on the time axis. The arithmetic mean of all the aligned original zero-crossing templates is taken for each sampling point to obtain the true zero-crossing template waveform. Calculate the waveform similarity between the zero-crossing waveform segment to be detected and the actual zero-crossing template waveform; When the waveform similarity exceeds a preset similarity threshold, the zero-crossing validity verification is confirmed to be successful. When the waveform similarity does not exceed the preset similarity threshold, the preliminary zero-crossing candidate moment is determined to be a false zero-crossing moment caused by noise. The preliminary zero-crossing candidate moment is abandoned, and the remaining sampling points in the real-time waveform data are traversed along the time-increasing direction.

[0105] The pre-training process of the decoupled state classification model includes: Multiple sets of attenuation characteristic parameter samples were collected from historical decoupling attenuation segment data to form a decoupling training sample feature set. For each set of attenuation feature parameter samples in the decoupling training sample feature set, obtain the corresponding attenuation mode category label to form a decoupling training label set, wherein the attenuation mode category label is true decoupling attenuation or decoupling artifact attenuation. The support vector machine model was used as the initial classification model. The initial classification model is trained in a supervised manner using the decoupled training sample feature set and the decoupled training label set. During the training process, the training objective is to minimize the classification error of the initial classification model for the decay mode category label until the classification error converges to below a preset convergence threshold, thus obtaining the trained initial classification model. The trained initial classification model is then used as the decoupled state classification model.

[0106] In one embodiment, the zero-current interruption module 13 is further configured to: Once the main circuit is determined to have entered the decoupling completion state, a disconnection trigger signal is generated. The disconnection trigger signal is sent to the electromagnetic thermal tripping composite mechanism of the circuit breaker; The electromagnetic thermal tripping composite mechanism responds to the disconnection trigger signal and drives the mechanical contacts to separate under the condition that the main circuit current is zero, thereby completing the arc-free disconnection. After the mechanical contact completes arc-free disconnection, the isolating switch is simultaneously disconnected via the operating handle linkage rod parallel to the circuit breaker disconnection mechanism, thus completing electrical isolation.

[0107] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A circuit breaker tripping method based on single-crystal thyristor decoupling control, characterized in that, The method, applied to a three-phase capacitor switching circuit composed of a diode rectifier bridge and a single-crystal thyristor, includes: Upon receiving the disconnection command, a turn-off signal is sent to the unidirectional thyristor, causing the unidirectional thyristor to turn off at the natural zero-crossing point of the current. After the shutdown signal is issued, real-time waveform data of the main circuit current is continuously collected, and decoupling status analysis is performed on the real-time waveform data to determine whether the main circuit has entered the decoupling completion state. Only when the main circuit is determined to have entered the decoupling completion state, a breaking trigger signal is sent to the breaking mechanism of the circuit breaker, and the breaking mechanism drives the mechanical contacts to perform the breaking action under zero current conditions.

2. The circuit breaker tripping method based on single-crystal thyristor decoupling control according to claim 1, characterized in that, Upon receiving the disconnection command, a turn-off signal is sent to the unidirectional thyristor, causing it to turn off at the natural zero-crossing point of the current, including: Upon receiving the disconnection command, the three-phase current waveform data of the three-phase capacitor is collected in real time. The three-phase current waveform data are preprocessed to obtain a preprocessed multidimensional input feature matrix; The multidimensional input feature matrix is ​​input into the pre-trained zero-crossing prediction model, and the zero-crossing prediction model outputs the time interval from the current time to the time of the next synthetic current zero-crossing. Add the time interval to the current time to obtain the time of the next zero-crossing of the synthesized current; At the moment when the next synthesized current crosses zero, the trigger pulse applied to the gate of the unidirectional thyristor is removed, causing the unidirectional thyristor to turn off automatically at the natural current zero-crossing point.

3. The circuit breaker tripping method based on single-crystal thyristor decoupling control according to claim 2, characterized in that, The three-phase current waveform data are preprocessed to obtain a preprocessed multidimensional input feature matrix, including: Extract the phase a current sampling sequence, phase b current sampling sequence, and phase c current sampling sequence from the three-phase current waveform data within a preset time window before the current moment; The phase a current sampling sequence, the phase b current sampling sequence, and the phase c current sampling sequence are normalized respectively to obtain the normalized phase a current sampling sequence, the normalized phase b current sampling sequence, and the normalized phase c current sampling sequence. Based on the a-phase current sampling sequence, the b-phase current sampling sequence, and the c-phase current sampling sequence, at each sampling moment, the instantaneous values ​​of the a-phase current, the b-phase current, and the c-phase current are compared to determine the first conducting phase and the second conducting phase at the sampling moment. The first conducting phase is the phase identifier corresponding to the maximum value among the three, and the second conducting phase is the phase identifier corresponding to the minimum value among the three. The phase identifier of the first conducting phase and the phase identifier of the second conducting phase at each sampling time are combined into a conduction state code. The conduction state codes at all sampling times are arranged in chronological order to generate a current conduction state feature sequence. The normalized a-phase current sampling sequence, the normalized b-phase current sampling sequence, the normalized c-phase current sampling sequence, and the current conduction state feature sequence are concatenated along the channel dimension to obtain the multidimensional input feature matrix.

4. The circuit breaker tripping method based on single-crystal thyristor decoupling control according to claim 2, characterized in that, The pre-training process of the zero-crossing prediction model includes: A bridge-type single-crystal thyristor power capacitor switching simulation circuit was constructed, and simulation operation was carried out under various operating conditions using the bridge-type single-crystal thyristor power capacitor switching simulation circuit. During the simulation, the three-phase current waveform data is recorded, and the zero-crossing time of the composite current flowing through the unidirectional thyristor in each current cycle is marked. For each sampling moment, the time difference between the sampling moment and the next zero-crossing point of the composite current is used as the zero-crossing time label of the sampling moment; Using the three-phase current waveform data within a preset time window as the raw data, and extracting multi-dimensional input feature matrix samples from the raw data, the multi-dimensional input feature matrix samples are combined with the corresponding zero-crossing time labels to form a simulation training sample set. An initial zero-crossing prediction model is constructed, which includes a temporal feature extraction branch, a conduction state encoding branch, and a feature fusion regression branch. The initial zero-crossing prediction model is trained in a supervised manner using the simulation training sample set. During the training process, the mean square error between the prediction time interval and the zero-crossing time label is minimized as the optimization objective until the mean square error converges to below a preset convergence threshold, thus obtaining the zero-crossing prediction model.

5. The circuit breaker tripping method based on single-crystal thyristor decoupling control according to claim 1, characterized in that, After issuing the shutdown signal, real-time waveform data of the main circuit current is continuously acquired. Decoupling status analysis is performed on the real-time waveform data to determine whether the main circuit has entered the decoupling completion state, including: After the shutdown signal is issued, real-time waveform data of the main circuit current is continuously collected. The first zero-crossing point of the main circuit current after the unidirectional thyristor is turned off is located from the real-time waveform data, and a waveform segment within a preset time period after the first zero-crossing point is extracted as the decoupling attenuation segment. Detect all local maxima with positive values ​​within the decoupling attenuation segment, take the detected local maxima with positive values ​​as positive peaks, and arrange all positive peaks in chronological order to form a positive peak sequence. The amplitude ratio of each pair of adjacent positive peaks in the positive peak sequence is calculated sequentially, and the median of the absolute values ​​of all amplitude ratios is taken as the peak decay rate. The amplitude ratio is the amplitude of the next positive peak divided by the amplitude of the previous positive peak in each pair of adjacent positive peaks. Calculate the time interval between each pair of adjacent positive peaks in the positive peak sequence in turn, take the median of all time intervals as the average oscillation period, and take the reciprocal of the average oscillation period as the waveform oscillation frequency. The peak decay rate and the waveform oscillation frequency are combined into decay feature parameters, and the decay feature parameters are input into a pre-trained decoupling state classification model to output the decay mode category of the decoupling decay segment, wherein the decay mode category is true decoupling decay or decoupling artifact decay. When the attenuation mode category output by the decoupling state classification model is true decoupling attenuation, the main loop is determined to have entered the decoupling completion state.

6. The circuit breaker breaking method based on single-crystal thyristor decoupling control according to claim 5, characterized in that, Locating the first zero-crossing point of the main circuit current after the unidirectional thyristor is turned off from the real-time waveform data includes: Starting from the moment the shutdown signal is issued, traverse each sampling point in the real-time waveform data along the time-increasing direction; For each sampling point, obtain the instantaneous value of the main circuit current at the sampling point and the instantaneous value of the main circuit current at the previous sampling point; Determine whether the sign of the instantaneous value of the main circuit current at the sampling point is opposite to that of the instantaneous value of the main circuit current at the previous sampling point; When the instantaneous value of the main circuit current at the sampling point has the opposite sign to the instantaneous value of the main circuit current at the previous sampling point, the time corresponding to the sampling point with the smaller absolute value of the instantaneous value of the main circuit current between the sampling point and the previous sampling point is taken as the preliminary zero-crossing candidate time. The preliminary zero-crossing candidate time is verified for its validity. When the verification is successful, the preliminary zero-crossing candidate time is determined as the first zero-crossing time of the main circuit current after the unidirectional thyristor is turned off.

7. The circuit breaker breaking method based on single-crystal thyristor decoupling control according to claim 6, characterized in that, The validity verification of the preliminary zero-crossing candidate times includes: Using the preliminary zero-crossing candidate time as the dividing point, a waveform segment of a first preset length is extracted from the real-time waveform data as the pre-zero-crossing segment, and a waveform segment of a second preset length is extracted as the post-zero-crossing segment. The pre-zero-crossing segment and the post-zero-crossing segment are then spliced ​​together in chronological order to form the zero-crossing waveform segment to be detected. Acquire historical waveform data of multiple known true zero-crossing times. For each set of historical waveform data of known true zero-crossing times, extract waveform segments of fixed length before and after the zero-crossing time as the original zero-crossing template. Amplitude normalization is performed on all original zero-crossing templates. All original zero-crossing templates after amplitude normalization are aligned on the time axis. The arithmetic mean of all the aligned original zero-crossing templates is taken for each sampling point to obtain the true zero-crossing template waveform. Calculate the waveform similarity between the zero-crossing waveform segment to be detected and the actual zero-crossing template waveform; When the waveform similarity exceeds a preset similarity threshold, the zero-crossing validity verification is confirmed to be successful. When the waveform similarity does not exceed the preset similarity threshold, the preliminary zero-crossing candidate moment is determined to be a false zero-crossing moment caused by noise. The preliminary zero-crossing candidate moment is abandoned, and the remaining sampling points in the real-time waveform data are traversed along the time-increasing direction.

8. The circuit breaker tripping method based on single-crystal thyristor decoupling control according to claim 5, characterized in that, The pre-training process of the decoupled state classification model includes: Multiple sets of attenuation characteristic parameter samples were collected from historical decoupling attenuation segment data to form a decoupling training sample feature set. For each set of attenuation feature parameter samples in the decoupling training sample feature set, obtain the corresponding attenuation mode category label to form a decoupling training label set, wherein the attenuation mode category label is true decoupling attenuation or decoupling artifact attenuation. The support vector machine model was used as the initial classification model. The initial classification model is trained in a supervised manner using the decoupled training sample feature set and the decoupled training label set. During the training process, the training objective is to minimize the classification error of the initial classification model for the decay mode category label until the classification error converges to below a preset convergence threshold, thus obtaining the trained initial classification model. The trained initial classification model is then used as the decoupled state classification model.

9. The circuit breaker tripping method based on single-crystal thyristor decoupling control according to claim 1, characterized in that, Only after the main circuit is determined to have entered the decoupling completion state, a breaking trigger signal is sent to the circuit breaker's breaking mechanism, which then drives the mechanical contacts to perform a breaking action under zero-current conditions, including: Once the main circuit is determined to have entered the decoupling completion state, a disconnection trigger signal is generated. The disconnection trigger signal is sent to the electromagnetic thermal tripping composite mechanism of the circuit breaker; The electromagnetic thermal tripping composite mechanism responds to the disconnection trigger signal and drives the mechanical contacts to separate under the condition that the main circuit current is zero, thereby completing the arc-free disconnection. After the mechanical contact completes arc-free disconnection, the isolating switch is simultaneously disconnected via the operating handle linkage rod parallel to the circuit breaker disconnection mechanism, thus completing electrical isolation.

10. A circuit breaker breaking system based on single-crystal thyristor decoupling control, characterized in that, For implementing the circuit breaker tripping method based on single-crystal thyristor decoupling control as described in any one of claims 1-9, the system comprises: The thyristor turn-off module is used to send a turn-off signal to the unidirectional thyristor after receiving a turn-off command, so that the unidirectional thyristor turns off at the natural zero-crossing point of the current. The decoupling status determination module is used to continuously collect real-time waveform data of the main circuit current after the shutdown signal is issued, perform decoupling status analysis on the real-time waveform data, and determine whether the main circuit has entered the decoupling completion state. The zero-current disconnection module is used to send a disconnection trigger signal to the circuit breaker's disconnection mechanism only when the main circuit is determined to have entered the decoupling completion state. The disconnection mechanism then drives the mechanical contacts to perform a disconnection action under zero-current conditions.

Images