Vacuum circuit breaker inrush current suppression method and system based on voltage phase angle

Abstract

The application provides a voltage phase angle-based vacuum circuit breaker inrush current suppression method and system, relates to the technical field of power system protection, and comprises the following steps: obtaining a voltage signal and historical closing data of a circuit breaker; calculating a voltage phase angle difference value and analyzing historical data to obtain characteristic parameters; inputting the phase angle difference value, a voltage amplitude, a system frequency and the historical characteristic parameters into a double-hidden layer neural network to output an optimal closing time and a protection triggering sequence; when a closing instruction is received, deciding whether to immediately or delay trigger a protection component according to whether it is the optimal time; controlling the circuit breaker body to close after all multi-stage protection components are turned on; and adding actual inrush current data to the historical data to optimize the neural network model and realize inrush current continuous suppression.

Description

Technical Field

[0001] This invention relates to power system protection technology, and more particularly to a method and system for suppressing inrush current in vacuum circuit breakers based on voltage phase angle. Background Technology

[0002] Vacuum circuit breakers often generate inrush current during the closing process, which can easily lead to system fluctuations and equipment damage. Traditional inrush current suppression methods mainly rely on fixed-value pre-charge resistors or simple synchronous closing techniques. However, when faced with complex loads and variable power grid conditions, the suppression effect is limited and lacks adaptability, resulting in insufficient or excessive inrush current suppression, causing closing delays and making it difficult to adapt to different operating conditions.

[0003] In existing technologies, closing strategies mostly rely on simple preset rules, which cannot be dynamically adjusted based on real-time grid parameters and historical data. Although phase angle-based synchronous closing has been applied, it requires high accuracy in phase angle calculation and suppression timing control, and lacks intelligent decision-making mechanisms. Therefore, it cannot cope with transient disturbances in the power grid and changes in load characteristics, and its suppression effect has significant limitations.

[0004] Existing multi-level protection structures typically employ fixed triggering sequences, which cannot be dynamically adjusted according to actual inrush current conditions, leading to incoordination in protection coordination. Furthermore, the lack of real-time monitoring and closed-loop control during the closing process hinders timely response to abnormal situations. In addition, current technologies rarely consider the matching relationship between the characteristics of the protection components themselves and the closing control strategy, making it difficult to achieve optimal performance from the protection components. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for suppressing inrush current in vacuum circuit breakers based on voltage phase angle, which can solve the problems in existing technologies.

[0006] This invention provides a method for suppressing inrush current in a vacuum circuit breaker based on voltage phase angle, comprising:

[0007] The voltage signals and historical closing data at both ends of the vacuum circuit breaker body are acquired and input to the controller.

[0008] The controller calculates the voltage phase angle difference between the two ends of the vacuum circuit breaker body based on the voltage signal; the controller analyzes the historical closing data to obtain historical characteristic parameters;

[0009] The controller inputs the voltage phase angle difference, the voltage amplitude at both ends of the vacuum circuit breaker body, the system frequency, and the historical characteristic parameters into a double hidden layer neural network, and outputs the optimal closing time and the hierarchical protection trigger sequence.

[0010] When receiving a closing command, if the current time is the optimal closing time, the controller sends trigger signals to the multi-level protection components connected in parallel at both ends of the vacuum circuit breaker body in sequence according to the hierarchical protection trigger sequence; if the current time is not the optimal closing time, the controller delays until the optimal closing time before sending the trigger signal.

[0011] The controller closes the vacuum circuit breaker body after all the multi-level protection components are turned on, thus completing the inrush current suppression control.

[0012] The controller adds the actual inrush flow data of this closing operation to the historical closing data to further optimize the dual hidden layer neural network.

[0013] Optionally,

[0014] The steps by which the controller analyzes the historical closing data to obtain historical characteristic parameters include:

[0015] The historical closing data includes inrush waveform data, load parameter data, and environmental parameter data;

[0016] Wavelet transform is performed on the surge waveform data to obtain multi-scale decomposition coefficients, and the temporal distribution characteristics of the surge peak are extracted based on the multi-scale decomposition coefficients.

[0017] The environmental stress factor is obtained by calculating the temperature stress coefficient, humidity stress coefficient, and air pressure stress coefficient based on the environmental parameter data and multiplying them together.

[0018] The information entropy of the time distribution feature, the environmental stress factor, and the load parameter data are calculated respectively. The weight coefficient of each feature is determined according to the magnitude of the information entropy, and the weighted combination is used to generate a fusion feature matrix.

[0019] A radial basis function (RBF) neural network is constructed, comprising an input layer, a hidden layer, and an output layer. The number of nodes in the input layer is the same as the feature dimension of the fused feature matrix, and the hidden layer uses a Gaussian radial basis function as the activation function. The fused feature matrix and corresponding surge characteristic parameters from historical samples are input into the RBF neural network for training. The center position and width parameters of the radial basis function are optimized using the least squares method. The trained RBF neural network is then used to perform feature mapping on the fused feature matrix and output the historical feature parameters.

[0020] Optionally,

[0021] The steps by which the controller inputs the voltage phase angle difference, the voltage amplitude across the vacuum circuit breaker body, the system frequency, and the historical characteristic parameters into a dual-hidden-layer neural network, and outputs the optimal closing time and the graded protection trigger sequence, include:

[0022] The voltage phase angle difference is normalized to obtain a normalized phase angle difference. The deviation of the voltage amplitude is calculated based on the rated voltage to obtain the voltage deviation feature. The deviation of the system frequency is calculated based on the nominal frequency to obtain the frequency deviation feature. A historical feature vector containing inrush peak value and transient recovery voltage change rate is constructed based on the historical feature parameters.

[0023] A dual-hidden-layer neural network is constructed. The input layer of the dual-hidden-layer neural network receives the normalized phase angle difference, voltage deviation features, frequency deviation features, and historical feature vectors. The number of nodes in the first and second hidden layers of the dual-hidden-layer neural network is adaptively adjusted according to the square root of the product of the number of nodes in the input layer and the number of nodes in the output layer.

[0024] A combined optimization objective function is constructed based on minimizing the peak inrush current and minimizing the rate of change of transient recovery voltage. The dual-hidden-layer neural network is trained based on the combined optimization objective function, and the network weight parameters are updated using the backpropagation algorithm of the motive term.

[0025] The normalized phase angle difference, voltage deviation features, frequency deviation features, and historical feature vectors acquired in real time are input into the trained dual-hidden-layer neural network, and the optimal closing time is calculated based on the combined optimization objective function.

[0026] A graded protection trigger sequence is generated based on the optimal closing time, and the trigger time interval between adjacent protection levels in the graded protection trigger sequence is controlled to be greater than the preset minimum coordination time.

[0027] Optionally,

[0028] The steps of constructing a combined optimization objective function based on minimizing the peak inrush current and minimizing the transient recovery voltage rate of change, training the double-hidden-layer neural network based on the combined optimization objective function, and updating the network weight parameters using the backpropagation algorithm with motive terms include:

[0029] A transfer function for the vacuum circuit breaker system is constructed. Based on the transfer function, the standard deviation between the real-time state and the steady-state value of the system is calculated to obtain the transient process deviation. The rate of change of active power and reactive power of the load connected to the vacuum circuit breaker is calculated, and multiplied by the first sensitivity coefficient and the second sensitivity coefficient respectively, and then summed to obtain the weight adjustment amount.

[0030] A time-varying weight coefficient is constructed based on the transient process deviation. The time-varying weight coefficient decreases exponentially as the transient process deviation increases and increases exponentially over time. The time-varying weight coefficient is combined with the weight adjustment amount to obtain the inrush weight coefficient.

[0031] The product of the inrush current weighting coefficient and the inrush current peak characteristic is used as the short-term inrush current target term; the voltage weighting coefficient is obtained by subtracting the inrush current weighting coefficient from the unit value, and the product of the voltage weighting coefficient and the transient recovery voltage characteristic is used as the medium-term voltage target term; the voltage deviation is calculated to obtain the long-term stability target term;

[0032] The operating force and breaking current of the vacuum circuit breaker are collected. The mechanical life loss is obtained by calculating the exponential function of the operating force and the electrical life loss is obtained by calculating the power function of the breaking current. Life constraint terms are constructed based on the mechanical life loss and the electrical life loss.

[0033] The short-term inrush current target, medium-term voltage target, long-term stability target, and lifetime constraint are weighted and combined to construct an optimization objective function. The dual-hidden-layer neural network is trained using a backpropagation algorithm with momentum term, which is a weighted combination of the previous cycle update and the current gradient. The weight coefficients are dynamically adjusted based on the inner product of adjacent gradient vectors until the rate of change of the optimization objective function is less than a preset rate of change threshold for a consecutive preset number of cycles.

[0034] Optionally,

[0035] The steps of generating a graded protection trigger sequence based on the optimal closing time and controlling the trigger time interval between adjacent protection levels in the graded protection trigger sequence to be greater than the preset minimum coordination time include:

[0036] Construct a protection feature vector that includes voltage phase angle difference, inrush current peak characteristics, and transient recovery voltage change rate;

[0037] The ratio of the expected inrush amplitude to the preset benchmark inrush amplitude is calculated based on the optimal closing time and used as the graded weighting coefficient.

[0038] Using the Euclidean distance between the graded weight coefficient and the protection feature vector as the clustering criterion, a dynamic clustering method is adopted to divide the protection components into multiple protection levels, and the initial triggering time of the first protection level is determined based on the optimal closing time.

[0039] Calculate the trigger time interval for each protection level, minimize the sum of squared deviations between the trigger time interval and the preset minimum coordination time, and generate an initial graded protection trigger sequence, wherein the trigger time interval between adjacent protection levels in the initial graded protection trigger sequence is greater than the minimum coordination time;

[0040] Real-time inrush current data is collected during the closing process of the vacuum circuit breaker. The difference between the real-time inrush current data and the expected inrush current amplitude is calculated to obtain the inrush current deviation. The product of the square of the inrush current deviation and the adaptive coefficient is used as the trigger sequence adjustment amount.

[0041] When the inrush deviation exceeds the preset inrush threshold, the trigger sequence adjustment amount is superimposed on the trigger time corresponding to the initial graded protection trigger sequence to generate a corrected graded protection trigger sequence. Each trigger time in the corrected graded protection trigger sequence satisfies the minimum coordination time constraint and the correction magnitude is proportional to the inrush deviation.

[0042] Optionally,

[0043] When receiving a closing command, if the current time is the optimal closing time, the controller sends trigger signals sequentially to the multi-level protection components connected in parallel across the vacuum circuit breaker body according to the hierarchical protection trigger sequence, including the following steps:

[0044] The multi-level protection component includes a first-level fast response protection unit and a second-level high current suppression protection unit.

[0045] The controller calculates the optimal turn-on timing of the first-level fast response protection unit and the second-level high current suppression protection unit based on the graded protection trigger sequence, wherein the triggering time of the first-level fast response protection unit is earlier than the triggering time of the second-level high current suppression protection unit, and the time interval is inversely proportional to the expected inrush current rise rate.

[0046] The controller monitors the leakage current value of the first-level fast response protection unit in real time. When the leakage current value exceeds the preset current threshold, the amplitude of the second-level trigger pulse signal is increased.

[0047] The controller monitors the voltage difference across the multi-level protection component in real time. When the voltage difference is lower than a preset voltage difference threshold and continues for a preset period of time, it determines that the multi-level protection component has been turned on.

[0048] The controller collects the temperature parameters of the first-level fast response protection unit and the second-level high current suppression protection unit. When the temperature parameters exceed the corresponding temperature limit, the trigger time interval of subsequent protection levels is dynamically increased.

[0049] Optionally,

[0050] The controller controls the vacuum circuit breaker to close after all the multi-level protection components are turned on, and the steps to complete the inrush current suppression control include:

[0051] The controller calculates the optimal closing execution window based on the voltage phase angle change rate, the protection component current growth rate, and the system power angle stability index.

[0052] Within the optimal closing execution window, the controller predicts the response time of the closing mechanism based on the mechanical operating characteristics of the vacuum circuit breaker body and dynamically calculates the optimal moment for issuing the closing command.

[0053] Before sending the closing command, the controller calculates the transient power fluctuation of the system that will be caused by the closing operation, and performs the closing operation when the predicted power fluctuation is less than the preset power disturbance threshold.

[0054] During the closing process of the vacuum circuit breaker, the controller dynamically adjusts the conduction state of the multi-level protection components to achieve a smooth transition of the closing current. The smooth transition includes controlling the current rise rate to be less than the preset safe rise rate.

[0055] The controller collects the closing speed, rebound amplitude, and arc duration of the vacuum circuit breaker body contacts during this closing process, compares them with historical data, updates the mechanical characteristic parameter library, and optimizes the subsequent closing control strategy.

[0056] Secondly, a vacuum circuit breaker inrush current suppression system based on voltage phase angle is provided, including:

[0057] The first unit is used to acquire the voltage signals and historical closing data at both ends of the vacuum circuit breaker body and input them to the controller.

[0058] The second unit is used to calculate the voltage phase angle difference between the two ends of the vacuum circuit breaker body based on the voltage signal; the controller analyzes the historical closing data to obtain historical characteristic parameters.

[0059] The third unit is used to input the voltage phase angle difference, the voltage amplitude at both ends of the vacuum circuit breaker body, the system frequency, and the historical characteristic parameters into a double hidden layer neural network, and output the optimal closing time and the hierarchical protection trigger sequence.

[0060] The fourth unit is used to receive a closing command. If the current time is the optimal closing time, the controller sends trigger signals to the multi-level protection components connected in parallel at both ends of the vacuum circuit breaker body in sequence according to the hierarchical protection trigger sequence. If the current time is not the optimal closing time, the controller delays until the optimal closing time before sending the trigger signal.

[0061] The fifth unit is used to control the vacuum circuit breaker body to close after all the multi-level protection components are turned on, thereby completing the inrush current suppression control.

[0062] The sixth unit is used to add the actual inrush flow data of this closing operation to the historical closing data, in order to further optimize the dual hidden layer neural network.

[0063] Thirdly, a computer-readable storage medium is provided, having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0064] This invention utilizes a dual-hidden-layer neural network to process voltage phase angle differences and historical data, enabling precise calculation of the optimal closing time and intelligent generation of graded protection trigger sequences. This significantly improves the accuracy and adaptability of inrush current suppression. Compared to traditional methods, the peak inrush current is significantly reduced, effectively minimizing the impact on the power grid and equipment damage, and extending the service life of circuit breakers.

[0065] This invention constructs a complete closed-loop control system, which ensures the reliability and stability of the inrush current suppression process by monitoring the status of protection components in real time and dynamically adjusting the triggering strategy. The collaborative working mechanism of multi-level protection components makes the inrush current suppression process smoother, effectively reduces transient overvoltages, and improves power grid quality and system stability.

[0066] This invention continuously feeds historical closing data into a neural network model, forming a self-learning and optimization mechanism. This enables the system to adapt to different load characteristics and environmental conditions, and the inrush current suppression effect continuously improves over time. This method realizes a shift from passive protection to active predictive control, has broad engineering application value, and can significantly improve the safety and stability of power systems. Attached Figure Description

[0067] Figure 1 This is a schematic flowchart of the inrush current suppression method for vacuum circuit breakers based on voltage phase angle, according to an embodiment of the present invention.

[0068] Figure 2 This is a graph showing the relationship between time-varying weighting coefficients and system stability. Detailed Implementation

[0069] The technical solutions of the present invention will be described below with reference to the accompanying drawings. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0070] Figure 1 This is a schematic flowchart of the inrush current suppression method for vacuum circuit breakers based on voltage phase angle according to the present invention. Figure 1 As shown, the method includes:

[0071] The voltage signals and historical closing data at both ends of the vacuum circuit breaker body are acquired and input to the controller.

[0072] The controller calculates the voltage phase angle difference between the two ends of the vacuum circuit breaker body based on the voltage signal; the controller analyzes the historical closing data to obtain historical characteristic parameters;

[0073] The controller inputs the voltage phase angle difference, the voltage amplitude at both ends of the vacuum circuit breaker body, the system frequency, and the historical characteristic parameters into a double hidden layer neural network, and outputs the optimal closing time and the hierarchical protection trigger sequence.

[0074] When receiving a closing command, if the current time is the optimal closing time, the controller sends trigger signals to the multi-level protection components connected in parallel at both ends of the vacuum circuit breaker body in sequence according to the hierarchical protection trigger sequence; if the current time is not the optimal closing time, the controller delays until the optimal closing time before sending the trigger signal.

[0075] The controller closes the vacuum circuit breaker body after all the multi-level protection components are turned on, thus completing the inrush current suppression control.

[0076] The controller adds the actual inrush flow data of this closing operation to the historical closing data to further optimize the dual hidden layer neural network.

[0077] Optionally,

[0078] The steps by which the controller analyzes the historical closing data to obtain historical characteristic parameters include:

[0079] The historical closing data includes inrush waveform data, load parameter data, and environmental parameter data;

[0080] Wavelet transform is performed on the surge waveform data to obtain multi-scale decomposition coefficients, and the temporal distribution characteristics of the surge peak are extracted based on the multi-scale decomposition coefficients.

[0081] The environmental stress factor is obtained by calculating the temperature stress coefficient, humidity stress coefficient, and air pressure stress coefficient based on the environmental parameter data and multiplying them together.

[0082] The information entropy of the time distribution feature, the environmental stress factor, and the load parameter data are calculated respectively. The weight coefficient of each feature is determined according to the magnitude of the information entropy, and the weighted combination is used to generate a fusion feature matrix.

[0083] A radial basis function (RBF) neural network is constructed, comprising an input layer, a hidden layer, and an output layer. The number of nodes in the input layer is the same as the feature dimension of the fused feature matrix, and the hidden layer uses a Gaussian radial basis function as the activation function. The fused feature matrix and corresponding surge characteristic parameters from historical samples are input into the RBF neural network for training. The center position and width parameters of the radial basis function are optimized using the least squares method. The trained RBF neural network is then used to perform feature mapping on the fused feature matrix and output the historical feature parameters.

[0084] For example, this embodiment provides a method for obtaining feature parameters based on historical closing data. This method is executed by a controller and includes four main stages: data collection, feature extraction, feature fusion, and feature mapping.

[0085] During the data collection phase, the controller acquires historical closing data from the vacuum circuit breaker. This data includes three types of information: inrush current waveform data, load parameter data, and environmental parameter data. The inrush current waveform data records the complete waveform of current changes over time during each closing process, with a sampling frequency of 10kHz and a recording duration of 200ms after closing. The load parameter data includes information such as load type identification, rated power, power factor, and load impedance characteristics. The environmental parameter data includes information such as ambient temperature, relative humidity, and atmospheric pressure at the time of closing, with an ambient temperature range of -40℃ to 85℃, relative humidity range of 0% to 100%, and atmospheric pressure range of 80kPa to 110kPa.

[0086] In the feature extraction stage, the controller first performs wavelet transform processing on the surge waveform data. Specifically, the controller selects the db4 wavelet as the basis function and performs five-level wavelet decomposition on the surge waveform data to obtain wavelet coefficients for different frequency bands. From these wavelet coefficients, the controller extracts features such as energy distribution, the location of the maximum coefficient, and the coefficient decay rate to form the time distribution feature vector of the surge waveform. For example, for a typical closing surge waveform, the controller can obtain the energy distribution ratios of the low-frequency band (0-50Hz), the mid-frequency band (50-500Hz), and the high-frequency band (500-5000Hz) through wavelet decomposition as 65%, 25%, and 10%, respectively. The maximum coefficient appears at 8.5ms after closing, and the coefficient decay rate is 0.32 / ms.

[0087] The controller converts ambient temperature into a temperature stress coefficient. When the ambient temperature is 25℃, the temperature stress coefficient is 1.0; for every 10℃ increase in temperature, the temperature stress coefficient increases by 0.15; for every 10℃ decrease in temperature, the temperature stress coefficient increases by 0.2. Next, the controller calculates the humidity stress coefficient. When the relative humidity is 50%, the humidity stress coefficient is 1.0; for every 10% increase in humidity, the humidity stress coefficient increases by 0.1; for every 10% decrease in humidity, the humidity stress coefficient decreases by 0.05. Third, the controller calculates the barometric pressure stress coefficient. When the atmospheric pressure is 101.3 kPa, the barometric pressure stress coefficient is 1.0; for every 5 kPa increase in pressure, the barometric pressure stress coefficient decreases by 0.05; for every 5 kPa decrease in pressure, the barometric pressure stress coefficient increases by 0.08. Finally, the controller multiplies the temperature stress coefficient, humidity stress coefficient, and barometric pressure stress coefficient to obtain the environmental stress factor. For example, when the ambient temperature is 35℃, the relative humidity is 70%, and the atmospheric pressure is 95kPa, the calculated temperature stress coefficient is 1.15, the humidity stress coefficient is 1.2, the air pressure stress coefficient is 1.1, and the environmental stress factor is 1.518.

[0088] In the feature fusion stage, the controller first calculates the information entropy of each feature. For time distribution features, the controller divides them into several intervals, counts the frequency of feature values ​​in each interval, and calculates the information entropy value. For environmental stress factors, the controller similarly divides them into multiple intervals and calculates the information entropy. For load parameter data, the controller calculates the information entropy of each parameter separately. Then, the controller determines the weight coefficient of each feature based on the magnitude of the information entropy value. The larger the information entropy, the greater the amount of information contained in the feature, and the larger the weight coefficient assigned. The controller normalizes the information entropy values ​​of each feature to obtain the corresponding weight coefficients. For example, the information entropy of the time distribution feature is 2.35, the information entropy of the environmental stress factor is 1.86, and the information entropy of the load parameter is 2.08. The normalized weight coefficients are 0.37, 0.29, and 0.34, respectively. Subsequently, the controller uses these weight coefficients to weight and combine the features to generate a fused feature matrix. The number of rows in this matrix equals the number of historical samples, the number of columns equals the feature dimension, and each element in the matrix has been standardized, with a value ranging from -1 to 1.

[0089] In the feature mapping stage, the radial basis function neural network (RBF) consists of three layers: an input layer, hidden layers, and an output layer. The number of nodes in the input layer is the same as the feature dimension of the fused feature matrix, for example, 15 nodes. The number of nodes in the hidden layer is determined by the number of training samples, typically the square root of the number of training samples; for example, with 100 training samples, the hidden layer has 10 nodes. The number of nodes in the output layer is the same as the dimension of the historical feature parameters, for example, 8 nodes. The hidden layer uses the Gaussian radial basis function as the activation function. Its characteristic is that the closer the input data is to the center point, the larger the output value, and vice versa, effectively capturing local patterns in the feature space.

[0090] The controller inputs the fused feature matrix and corresponding inrush characteristic parameters from historical samples into the radial basis function neural network for training. The inrush characteristic parameters are key indicators extracted from historical closing records, including quantitative indicators such as peak inrush, duration, rate of rise, attenuation coefficient, harmonic content, and phase characteristics. These parameters directly reflect the inrush performance characteristics of the circuit breaker under different conditions. The training process consists of two stages: first, determining the center point and width parameter of the radial basis function; then, calculating the connection weights from the hidden layer to the output layer. The K-means clustering algorithm is used to determine the center point location, clustering the training samples into categories with the same number of nodes as the hidden layer. The center of each category is the center point of the radial basis function. The width parameter is calculated based on the average distance between the center points, typically set to 1.5 times the average distance. For example, when the average distance between the center points is 0.65, the width parameter is set to 0.975. After determining the center point and width parameter, the controller uses the least squares method to calculate the connection weights from the hidden layer to the output layer, minimizing the mean square error between the network output and the desired output.

[0091] After training, the controller uses the radial basis function neural network to perform feature mapping on the new fused feature matrix, outputting historical feature parameters. These historical feature parameters include statistical characteristics of historical inrush current peak values, rate of change characteristics of transient recovery voltage, and sensitivity coefficients for load types, which can comprehensively reflect the closing characteristics of the circuit breaker under different operating conditions. For example, for a new set of fused feature matrices, the mapped historical feature parameters include: the average historical inrush current peak value is 3.2 times the rated current, the standard deviation is 0.8 times the rated current, the rate of change of transient recovery voltage is 2.5 kV / ms, the sensitivity coefficient for inductive loads is 1.8, and the sensitivity coefficient for capacitive loads is 2.2, etc.

[0092] This invention utilizes multi-dimensional data analysis and advanced feature extraction techniques to achieve in-depth mining of historical closing data. It extracts high-value feature parameters from complex and ever-changing historical data, providing reliable data support for subsequent calculations of optimal closing times and generation of hierarchical protection trigger sequences. Compared to traditional methods, this invention more comprehensively considers the correlation between inrush current characteristics and environmental factors and load characteristics, improving the accuracy and representativeness of feature parameters. This significantly enhances the precision and adaptability of inrush current suppression, effectively reducing the impact of closing inrush current on the system, extending circuit breaker lifespan, and improving the safety and stability of power grid operation.

[0093] Optionally,

[0094] The steps by which the controller inputs the voltage phase angle difference, the voltage amplitude across the vacuum circuit breaker body, the system frequency, and the historical characteristic parameters into a dual-hidden-layer neural network, and outputs the optimal closing time and the graded protection trigger sequence, include:

[0095] The voltage phase angle difference is normalized to obtain a normalized phase angle difference. The deviation of the voltage amplitude is calculated based on the rated voltage to obtain the voltage deviation feature. The deviation of the system frequency is calculated based on the nominal frequency to obtain the frequency deviation feature. A historical feature vector containing inrush peak value and transient recovery voltage change rate is constructed based on the historical feature parameters.

[0096] A dual-hidden-layer neural network is constructed. The input layer of the dual-hidden-layer neural network receives the normalized phase angle difference, voltage deviation features, frequency deviation features, and historical feature vectors. The number of nodes in the first and second hidden layers of the dual-hidden-layer neural network is adaptively adjusted according to the square root of the product of the number of nodes in the input layer and the number of nodes in the output layer.

[0097] A combined optimization objective function is constructed based on minimizing the peak inrush current and minimizing the rate of change of transient recovery voltage. The dual-hidden-layer neural network is trained based on the combined optimization objective function, and the network weight parameters are updated using the backpropagation algorithm of the motive term.

[0098] The normalized phase angle difference, voltage deviation features, frequency deviation features, and historical feature vectors acquired in real time are input into the trained dual-hidden-layer neural network, and the optimal closing time is calculated based on the combined optimization objective function.

[0099] A graded protection trigger sequence is generated based on the optimal closing time, and the trigger time interval between adjacent protection levels in the graded protection trigger sequence is controlled to be greater than the preset minimum coordination time.

[0100] For example, the controller acquires the voltage phase angle difference across the vacuum circuit breaker body. This difference typically ranges from -180 degrees to +180 degrees. It is then normalized by dividing the original difference by 180 degrees to obtain the normalized phase angle difference, ensuring its range is controlled between -1 and +1. For instance, when the original phase angle difference is +75 degrees, the normalized value is +0.417; when the original phase angle difference is -120 degrees, the normalized value is -0.667. The controller also acquires the voltage amplitude across the vacuum circuit breaker body and compares it with the system's rated voltage to calculate the voltage deviation characteristic. Specifically, the calculation method involves subtracting the rated voltage value from the actual voltage amplitude and then dividing by the rated voltage value to obtain the relative deviation value. For example, when the system's rated voltage is 10kV and the actual measured voltage amplitude is 10.2kV, the calculated voltage deviation characteristic is +0.02; when the actual measured voltage is 9.8kV, the voltage deviation characteristic is -0.02. The controller acquires the current system frequency and compares it with the nominal grid frequency (usually 50Hz or 60Hz) to calculate the frequency deviation characteristic. Specifically, the actual frequency is subtracted from the nominal frequency, and then divided by the nominal frequency to obtain the relative deviation value. For example, when the nominal frequency is 50Hz and the actual measured frequency is 50.2Hz, the frequency deviation characteristic is +0.004; when the actual frequency is 49.8Hz, the frequency deviation characteristic is -0.004. Historical characteristic parameters contain inrush current characteristics of the system under different conditions. The controller extracts features related to the peak inrush current and the transient recovery voltage change rate from these parameters to form a historical feature vector. For example, the historical feature vector includes the average, standard deviation, maximum, and minimum values ​​of historical peak inrush current, as well as the average and coefficient of variation of the transient recovery voltage change rate. For a specific system, the historical feature vector is [3.2, 0.8, 5.1, 1.9, 2.5, 0.15], where the first four elements are related to the peak inrush current, and the last two elements are related to the transient recovery voltage change rate.

[0101] The dual-hidden-layer neural network structure includes an input layer, a first hidden layer, a second hidden layer, and an output layer. The number of nodes in the input layer is the same as the dimension of the input features, namely, normalized phase angle difference (1), voltage deviation feature (1), frequency deviation feature (1), and historical feature vector (6, depending on the actual situation), totaling 9 nodes. The output layer contains the optimal closing time (1) and the graded protection trigger sequence (3, depending on the number of protection levels), totaling 4 nodes. The number of hidden layer nodes is adaptively adjusted by calculating the square root of the product of the number of nodes in the input layer and the number of nodes in the output layer. For example, when the input layer has 9 nodes and the output layer has 4 nodes, the product is 36, and the square root is approximately 6. The controller can set the number of nodes in the first hidden layer to 6 and the number of nodes in the second hidden layer to 5, forming a 9-6-5-4 network structure. The connections between the input layer and the first hidden layer, the first hidden layer and the second hidden layer, and the second hidden layer and the output layer are all fully connected, with the initial connection weights randomly set within the range of [-0.5, +0.5]. Both the first and second hidden layers use the ReLU (Modified Linear Unit) function as the activation function. This function outputs the same value as the input when the input is positive and zero when the input is negative. It is characterized by its simple computation and stable gradient.

[0102] During the network training phase, the controller constructs a combined optimization objective function that simultaneously considers minimizing both the inrush peak current and the transient recovery voltage rate of change. The network is trained using a backpropagation algorithm with a driving term.

[0103] The normalized phase angle difference, voltage deviation features, frequency deviation features, and historical feature vectors acquired in real time are input into a trained dual-hidden-layer neural network. Through forward propagation calculation, the network outputs the optimal closing time, which is a comprehensive consideration based on the combined optimization objective function. For example, for a specific combination of input features, the network outputs the optimal closing time as 12.5 milliseconds after the current time, meaning that closing at this time achieves the best inrush current suppression effect.

[0104] During the trigger sequence generation phase, the controller generates a graded protection trigger sequence based on the calculated optimal closing time. The controller ensures that the trigger time interval between adjacent protection levels in the trigger sequence is greater than the preset minimum coordination time. If the generated initial trigger sequence does not meet this requirement, the controller will automatically adjust the trigger time of each protection level to ensure that the coordination time meets the requirements and is as close as possible to the optimal timing sequence.

[0105] This invention utilizes a dual-hidden-layer neural network to accurately calculate the optimal closing time of vacuum circuit breakers and intelligently generate graded protection trigger sequences. It organically combines voltage phase angle difference, voltage amplitude, system frequency, and historical characteristic parameters, fully considering the combined effects of inrush peak and transient recovery voltage change rate. Compared to traditional methods, this invention offers higher accuracy and adaptability, dynamically optimizing the closing strategy based on real-time system status and historical operating experience. This effectively reduces the impact of closing inrush current on the system, improves the safety and stability of power grid operation, and continuously enhances system performance through accumulated operating experience, providing strong support for the intelligent operation of power systems.

[0106] Optionally,

[0107] The steps of constructing a combined optimization objective function based on minimizing the peak inrush current and minimizing the transient recovery voltage rate of change, training the double-hidden-layer neural network based on the combined optimization objective function, and updating the network weight parameters using the backpropagation algorithm with motive terms include:

[0108] A transfer function for the vacuum circuit breaker system is constructed. Based on the transfer function, the standard deviation between the real-time state and the steady-state value of the system is calculated to obtain the transient process deviation. The rate of change of active power and reactive power of the load connected to the vacuum circuit breaker is calculated, and multiplied by the first sensitivity coefficient and the second sensitivity coefficient respectively, and then summed to obtain the weight adjustment amount.

[0109] A time-varying weight coefficient is constructed based on the transient process deviation. The time-varying weight coefficient decreases exponentially as the transient process deviation increases and increases exponentially over time. The time-varying weight coefficient is combined with the weight adjustment amount to obtain the inrush weight coefficient.

[0110] The inrush current characteristic is obtained by calculating the ratio of the peak inrush current to the rated current of the vacuum circuit breaker. The voltage characteristic is obtained by calculating the ratio of the transient recovery voltage change rate to the rated voltage. The product of the inrush current weighting coefficient and the peak inrush current characteristic is used as the short-term inrush current target item. The voltage weighting coefficient is obtained by subtracting the inrush current weighting coefficient from the unit value. The product of the voltage weighting coefficient and the transient recovery voltage characteristic is used as the medium-term voltage target item. The voltage deviation is calculated to obtain the long-term stability target item. Specifically, this includes collecting the frequency change rate and power angle change rate of the vacuum circuit breaker. When the frequency change rate is less than a first preset threshold and the power angle change rate is less than a second preset threshold, the product of the square of the voltage deviation and the long-term weighting coefficient is calculated to obtain the long-term stability target item.

[0111] The operating force and breaking current of the vacuum circuit breaker are collected. The mechanical life loss is obtained by calculating the exponential function of the operating force and the electrical life loss is obtained by calculating the power function of the breaking current. Life constraint terms are constructed based on the mechanical life loss and the electrical life loss.

[0112] The short-term inrush current target, medium-term voltage target, long-term stability target, and lifetime constraint are weighted and combined to construct an optimization objective function. The dual-hidden-layer neural network is trained using a backpropagation algorithm with momentum term, which is a weighted combination of the previous cycle update and the current gradient. The weight coefficients are dynamically adjusted based on the inner product of adjacent gradient vectors until the rate of change of the optimization objective function is less than a preset rate of change threshold for a consecutive preset number of cycles.

[0113] For example, the transfer function of a vacuum circuit breaker system is a second-order system, which includes the input signal (closing command), output response (system current / voltage), and system characteristic parameters (damping coefficient and natural frequency). For instance, for a typical 10kV vacuum circuit breaker system, the damping coefficient of its transfer function can be set to 0.35, and the natural frequency to 314 rad / s (corresponding to a 50Hz system). Based on this transfer function, the transient process deviation is obtained by sampling and calculating the difference between the real-time system state and the steady-state value, and then calculating the standard deviation of these differences. Specifically, the controller collects 20 equally spaced current sampling points within 100ms after closing, calculates the difference between these sampling points and the steady-state current value, and then calculates the standard deviation of these differences to obtain the transient process deviation value.

[0114] The active power P and reactive power Q are sampled every 5ms. The difference between two adjacent sampled values ​​is calculated and divided by the sampling time interval to obtain the rate of change. For example, if the active power measured at a certain moment is 8MW, and it is measured at 8.2MW after 5ms, then the rate of change of active power is 40MW / s; the rate of change of reactive power is calculated similarly. The controller multiplies the rate of change of active power by a first sensitivity coefficient (e.g., 0.6) and the rate of change of reactive power by a second sensitivity coefficient (e.g., 0.4), and then adds the two to obtain the weight adjustment amount. For example, when the rate of change of active power is 40MW / s and the rate of change of reactive power is 20Mvar / s, the weight adjustment amount is 0.6×40 + 0.4×20 = 32.

[0115] In the time-varying weight construction phase, the controller constructs a time-varying weight coefficient based on the transient process deviation. This coefficient has a dual characteristic: on the one hand, it decays exponentially with the increase of the transient process deviation; on the other hand, it increases exponentially with time. Specifically, the controller first calculates the basic weight value, which is the result of multiplying the negative transient process deviation of e by 0.5; then it calculates the time factor, which is the result of subtracting the negative time of e from 1 and dividing by the time constant, typically set to 50ms. The controller multiplies the basic weight value by the time factor to obtain the time-varying weight coefficient. For example, when the transient process deviation is 1.5 times the rated current and the closing time is 20ms, the basic weight value is 0.47, the time factor is 0.33, and the time-varying weight coefficient is 0.47 × 0.33 = 0.155. The time-varying weight coefficient is combined with the previously calculated weight adjustment amount to obtain the inrush current weight coefficient. The combination method is: multiply the time-varying weight coefficient by (1 plus the weight adjustment amount divided by 100). For example, when the time-varying weight coefficient is 0.155 and the weight adjustment amount is 32, the inrush weight coefficient is 0.155×(1+32 / 100)=0.205.

[0116] During the target construction phase, the controller first calculates the ratio of the peak inrush current to the rated current of the vacuum circuit breaker to obtain the inrush current characteristic. For example, when the peak inrush current is 250A and the rated current is 100A, the inrush current characteristic is 2.5. The controller also calculates the ratio of the transient recovery voltage change rate to the rated voltage to obtain the voltage characteristic. For example, when the transient recovery voltage change rate is 2kV / ms and the rated voltage is 10kV, the voltage characteristic is 0.2.

[0117] The controller multiplies the inrush current weighting coefficient by the inrush current characteristic to obtain the short-term inrush current target. For example, when the inrush current weighting coefficient is 0.205 and the inrush current characteristic is 2.5, the short-term inrush current target is 0.205 × 2.5 = 0.5125. The controller calculates the voltage weighting coefficient, which is 1 minus the inrush current weighting coefficient, for example, 0.795. The voltage weighting coefficient is multiplied by the voltage characteristic to obtain the medium-term voltage target. For example, when the voltage weighting coefficient is 0.795 and the voltage characteristic is 0.2, the medium-term voltage target is 0.795 × 0.2 = 0.159.

[0118] The controller collects the frequency change rate and power angle change rate of the vacuum circuit breaker. When the frequency change rate is less than a first preset threshold (e.g., 0.1 Hz / s) and the power angle change rate is less than a second preset threshold (e.g., 1 degree / s), the system is in a relatively stable state. The controller calculates the voltage deviation, which is the difference between the actual voltage and the rated voltage divided by the rated voltage. Then, the square of the voltage deviation is multiplied by a long-term weighting coefficient (e.g., 0.1) to obtain the long-term stability target term. For example, when the voltage deviation is 0.02 and the long-term weighting coefficient is 0.1, the long-term stability target term is 0.02. 2×0.1=0.00004.

[0119] During the constraint construction phase, the controller acquires the operating force and breaking current of the vacuum circuit breaker. The controller calculates an exponential function of the operating force, i.e., eoperating force divided by the rated operating force, to obtain the mechanical life loss. For example, when the operating force is 2000N and the rated operating force is 1500N, the mechanical life loss is e0. (2000 / 1500) =3.74. The controller calculates the breaking current using a power function, that is, the breaking current divided by the rated breaking current to the power of 2.5, to obtain the electrical life loss. For example, when the breaking current is 15kA and the rated breaking current is 25kA, the electrical life loss is (15 / 25). 2.5 =0.24. Multiply the mechanical life loss by 0.4, add the electrical life loss by 0.6, and then multiply by 0.05. For example, when the mechanical life loss is 3.74 and the electrical life loss is 0.24, the life constraint term is (3.74×0.4+0.24×0.6)×0.05=0.082.

[0120] During the optimization phase, the controller constructs an optimization objective function by weighting and combining short-term inrush current target terms, medium-term voltage target terms, long-term stability target terms, and lifetime constraint terms. The weighting method is as follows: the short-term inrush current target term is multiplied by 0.5, the medium-term voltage target term by 0.3, the long-term stability target term by 0.1, and the lifetime constraint term by 0.1, and then these are added together to obtain the final optimization objective function value. For example, when the short-term inrush current target term is 0.5125, the medium-term voltage target term is 0.159, the long-term stability target term is 0.00004, and the lifetime constraint term is 0.082, the optimization objective function value is 0.5125×0.5+0.159×0.3+0.00004×0.1+0.082×0.1=0.3108.

[0121] The controller trains the two-hidden-layer neural network using a backpropagation algorithm with a momentum term. In each iteration, the controller first calculates the network output through forward propagation, then calculates the objective function value, and finally calculates the gradient of each layer's weights through backpropagation. The controller adjusts the network weights using a weight update rule with a momentum term, where the momentum term is a weighted combination of the previous iteration's update and the current gradient. Specifically, the new weight update equals the learning rate (e.g., 0.01) multiplied by the current gradient, plus the momentum coefficient multiplied by the previous iteration's weight update.

[0122] The momentum coefficient is not a fixed value, but is dynamically adjusted based on the dot product of adjacent gradient vectors. When the dot product of adjacent gradient vectors is positive (similar directions), the momentum coefficient is increased, with a maximum of 0.95; when the dot product is negative (opposite directions), the momentum coefficient is decreased, with a minimum of 0.5. For example, when the dot product of adjacent gradient vectors is 0.8 (highly similar), the momentum coefficient can be set to 0.9; when the dot product is -0.3 (partially opposite), the momentum coefficient can be reduced to 0.6.

[0123] The controller repeats the above training process until the rate of change of the objective function is less than a preset rate of change threshold (e.g., 0.001) for a consecutive preset number of cycles (e.g., 10 cycles). The rate of change is calculated by dividing the absolute value of the difference between the current objective function value and the objective function value of the previous cycle by the objective function value of the previous cycle. For example, if the objective function value of the current cycle is 0.3108 and the previous cycle was 0.3110, then the rate of change is |0.3108 - 0.3110| / 0.3110 = 0.00064, which is less than the threshold of 0.001.

[0124] Figure 2 The graph shows the relationship between time-varying weight coefficients and system stability, illustrating the changing trends of these coefficients under different transient process deviations for three different weighting methods. The horizontal axis represents the transient process deviation (pu), ranging from 0 to 5; the vertical axis represents the time-varying weight coefficient, ranging from 0 to 1.0. The traditional fixed weighting method (represented by the dashed line) maintains a constant weight coefficient of 0.5 throughout the transient process, without adjusting to changes in system state, thus lacking adaptability. The linear time-varying weighting method (represented by the dotted line) shows a linear decreasing trend with increasing transient process deviation, gradually decreasing from an initial 0.9 to 0.4, demonstrating a certain degree of adaptability. The exponential time-varying weighting method (represented by the solid line) used in this invention exhibits significant nonlinear characteristics. When the transient process deviation is small, the weight coefficient remains at a high level (0.95-0.61), and decreases rapidly exponentially as the deviation increases, dropping to 0.08 when the deviation reaches 5. This characteristic allows the system to make precise weight adjustments under different operating conditions, more effectively suppressing inrush peaks. The figure also illustrates the effect of the time factor (represented by a thin dashed line), showing the trend of weight coefficient changes over time. At t = 20 ms, the time factor provides additional weight gain, ensuring that the system can dynamically adjust according to the time characteristics of the transient process.

[0125] This invention achieves comprehensive optimized control of the circuit breaker closing process by constructing a composite optimization objective function that integrates short-term inrush current suppression, medium-term voltage stability, long-term system stability, and equipment lifespan constraints. The innovation of this embodiment lies in the introduction of time-varying weight coefficients, enabling the control strategy to dynamically adjust with system state and time, and achieving adaptive weight adjustment through load characteristic sensing; the use of a backpropagation algorithm with dynamic momentum terms significantly improves the training efficiency of the neural network and the accuracy of the optimization results; and it realizes the transformation from single inrush current suppression to global system optimization, ensuring short-term inrush current suppression while also considering medium- and long-term system stability and equipment lifespan.

[0126] Optionally,

[0127] The steps of generating a graded protection trigger sequence based on the optimal closing time and controlling the trigger time interval between adjacent protection levels in the graded protection trigger sequence to be greater than the preset minimum coordination time include:

[0128] Construct a protection feature vector that includes voltage phase angle difference, inrush current peak characteristics, and transient recovery voltage change rate;

[0129] The ratio of the expected inrush amplitude to the preset benchmark inrush amplitude is calculated based on the optimal closing time and used as the graded weighting coefficient.

[0130] Using the Euclidean distance between the graded weight coefficient and the protection feature vector as the clustering criterion, a dynamic clustering method is adopted to divide the protection components into multiple protection levels, and the initial triggering time of the first protection level is determined based on the optimal closing time.

[0131] Calculate the trigger time interval for each protection level, minimize the sum of squared deviations between the trigger time interval and the preset minimum coordination time, and generate an initial graded protection trigger sequence, wherein the trigger time interval between adjacent protection levels in the initial graded protection trigger sequence is greater than the minimum coordination time;

[0132] Real-time inrush current data is collected during the closing process of the vacuum circuit breaker. The difference between the real-time inrush current data and the expected inrush current amplitude is calculated to obtain the inrush current deviation. The product of the square of the inrush current deviation and the adaptive coefficient is used as the trigger sequence adjustment amount.

[0133] When the inrush deviation exceeds the preset inrush threshold, the trigger sequence adjustment amount is superimposed on the trigger time corresponding to the initial graded protection trigger sequence to generate a corrected graded protection trigger sequence. Each trigger time in the corrected graded protection trigger sequence satisfies the minimum coordination time constraint and the correction magnitude is proportional to the inrush deviation.

[0134] For example, the protection characteristic vector comprises three main components: voltage phase angle difference, inrush current peak characteristic, and transient recovery voltage change rate. The voltage phase angle difference is directly derived from the previously calculated voltage phase angle difference across the vacuum circuit breaker body. For instance, when the voltage phase values ​​across the circuit breaker are 30 degrees and -40 degrees, the voltage phase angle difference is 70 degrees. The inrush current peak characteristic is the ratio of the expected inrush current peak to the rated current, obtained from historical data analysis. For example, for a certain type of 10kV vacuum circuit breaker with a rated current of 630A, historical data indicates that the expected inrush current peak under specific conditions is 1890A, then the inrush current peak characteristic is 3.0. The transient recovery voltage change rate refers to the rate of voltage rise during the voltage recovery process after the circuit breaker contacts have separated, typically expressed in kV / μs. For example, for a 10kV system, a typical transient recovery voltage change rate is 0.05 kV / μs. The controller combines these three parameters into a protection feature vector [70, 3.0, 0.05], which is used for subsequent protection level classification and trigger sequence generation.

[0135] During the weighting calculation phase, the controller calculates the ratio of the expected inrush current amplitude to the preset reference inrush current based on the optimal closing time, using this ratio as the weighting coefficient. The expected inrush current amplitude refers to the magnitude of the inrush current predicted at the optimal closing time. The preset reference inrush current is typically three times the rated current. For example, when the rated current is 630A, the preset reference inrush current is 1890A; if the expected inrush current amplitude at the optimal closing time is 1323A, then the weighting coefficient is 1323 / 1890 = 0.7. The smaller the weighting coefficient, the better the inrush current suppression effect, and the fewer protection levels are required; conversely, the larger the weighting coefficient, the more protection levels are needed to smoothly transition the closing current.

[0136] During the protection level classification stage, the controller uses the Euclidean distance between the classification weight coefficients and the protection feature vectors as the clustering criterion, employing a dynamic clustering method to divide the protection components into multiple protection levels. The Euclidean distance is calculated by summing the squares of the differences between the components of the protection feature vector and then taking the square root. For example, if there are two protection components with protection feature vectors [70, 3.0, 0.05] and [65, 2.8, 0.04] respectively, the Euclidean distance is 5.2. The controller uses the K-means clustering algorithm. The initial cluster center selection method is as follows: first, a protection component is randomly selected as the first cluster center; then, the minimum distance from the remaining protection components to the selected cluster center is calculated, and the next cluster center is selected according to the square of the distance. This process is repeated until K cluster centers are selected. During the clustering iteration process, the controller assigns each protection component to the nearest cluster center, then recalculates the center points of each class, repeating this process until the clustering results stabilize or the maximum number of iterations (usually 100) is reached.

[0137] The number of clusters K is dynamically determined based on the hierarchical weighting coefficient. Specifically: K is 2 when the weighting coefficient is less than 0.5; 3 when the weighting coefficient is greater than or equal to 0.5 and less than 0.8; and 4 when the weighting coefficient is greater than or equal to 0.8. For example, when the weighting coefficient is 0.7, the protection components will be divided into 3 levels. After the division, the controller determines the initial trigger time of the first protection level based on the optimal closing time. For a zero-phase-angle closing strategy, the trigger time of the first protection level is set to the optimal closing time; for a non-zero-phase-angle closing strategy, the trigger time of the first protection level may be earlier or later than the optimal closing time by a certain amount of time, depending on the phase angle difference and the expected inrush current characteristics.

[0138] During the trigger sequence generation phase, the controller calculates the trigger time interval for each protection level, minimizing the sum of squares of the deviations between the trigger time interval and the preset minimum coordination time. The minimum coordination time refers to the minimum time interval that must be maintained between adjacent protection levels to ensure that the preceding protection component has sufficient time to respond and achieve the expected protection effect. For vacuum circuit breaker systems, the minimum coordination time is typically set to 2 milliseconds. The controller employs a quadratic programming solution method, setting the constraint that the trigger time intervals of all adjacent protection levels are greater than or equal to the minimum coordination time. The objective function is the sum of squares of the deviations between the trigger time interval and the ideal interval (determined based on the characteristics of each protection level).

[0139] For example, for a protection system divided into three levels, the ideal triggering time interval is 2.5 milliseconds between the first and second levels, and 3 milliseconds between the second and third levels. Considering the constraint that the minimum coordination time is 2 milliseconds, the actual triggering time intervals obtained by the controller are: 2.3 milliseconds between the first and second levels, and 2.8 milliseconds between the second and third levels. Thus, the controller generates an initial hierarchical protection triggering sequence of [0, 2.3, 5.1] milliseconds, indicating that the first-level protection triggers at the optimal closing time, the second-level protection triggers after 2.3 milliseconds, and the third-level protection triggers after 5.1 milliseconds. This sequence satisfies the constraint that the triggering time interval between adjacent protection levels is greater than the minimum coordination time.

[0140] During the real-time monitoring and sequence correction phase, the controller collects real-time inrush current data during the vacuum circuit breaker closing process and compares it with the expected inrush current amplitude to calculate the inrush current deviation. For example, when the expected inrush current amplitude is 1323A and the measured inrush current is 1455A, the inrush current deviation is 132A, with a relative deviation of 10%. The controller multiplies the square of the inrush current deviation by an adaptive coefficient to obtain the trigger sequence adjustment. The adaptive coefficient is determined based on the system response characteristics and is typically set to 0.01 ms / (ampere squared). For example, when the inrush current deviation is 132A, the trigger sequence adjustment is 132A. 2×0.01=174.24 microseconds≈0.17 milliseconds.

[0141] When the inrush current deviation exceeds a preset inrush current threshold (usually set to 5% of the expected inrush current amplitude), the controller adds the trigger sequence adjustment amount to the trigger time corresponding to the initial graded protection trigger sequence to generate a corrected graded protection trigger sequence. The adjustment method is as follows: for excessively large inrush currents, the trigger time of subsequent protection levels is delayed; for excessively small inrush currents, the trigger time of subsequent protection levels is advanced. For example, when the measured inrush current is 10% greater than the expected inrush current, the trigger times of the second and third level protections are delayed by 0.17 milliseconds and 0.34 milliseconds, respectively, and the corrected graded protection trigger sequence becomes [0, 2.47, 5.44] milliseconds.

[0142] The controller ensures that each trigger moment in the corrected hierarchical protection trigger sequence still meets the minimum coordination time constraint. If the interval between adjacent trigger moments after correction is less than the minimum coordination time, the controller will perform a secondary adjustment to ensure that the trigger time interval of all adjacent protection levels is greater than or equal to the minimum coordination time. In addition, the correction amplitude is proportional to the inrush current deviation; the larger the inrush current deviation, the larger the correction amplitude. However, the maximum correction amplitude is usually no more than 2 milliseconds to avoid over-adjustment that could lead to system instability.

[0143] This invention achieves intelligent generation and dynamic correction of hierarchical protection trigger sequences based on optimal closing time. Through protection feature vector construction, dynamic clustering, and real-time monitoring and correction, it ensures the efficient and coordinated operation of the hierarchical protection system. This invention introduces hierarchical weight coefficients to dynamically determine the number of protection levels, uses Euclidean distance as a clustering criterion to achieve precise hierarchical classification of protection components, and achieves dynamic optimization of the trigger sequence through real-time inrush monitoring and feedback correction mechanisms. Compared with traditional fixed trigger sequence methods, this invention has higher adaptability and accuracy, and can dynamically adjust the protection strategy according to the real-time system status, effectively reducing the impact of closing inrush current on the system.

[0144] Optionally,

[0145] Upon receiving a closing command, if the current time is the optimal closing time, the controller sequentially sends trigger signals to the multi-level protection components connected in parallel across the vacuum circuit breaker body according to the hierarchical protection trigger sequence, including:

[0146] The multi-level protection component includes a first-level fast response protection unit and a second-level high-current suppression protection unit. The first-level fast response protection unit is made of an electric field-sensitive ceramic material whose response time is adapted to the voltage phase angle change rate. The second-level high-current suppression protection unit is made of a metal oxide varistor material whose varistor coefficient is proportional to the voltage phase angle difference.

[0147] The controller calculates the optimal turn-on timing of the first-level fast response protection unit and the second-level high current suppression protection unit based on the graded protection trigger sequence, wherein the triggering time of the first-level fast response protection unit is earlier than the triggering time of the second-level high current suppression protection unit, and the time interval is inversely proportional to the expected inrush current rise rate.

[0148] The controller generates a first-level trigger pulse signal and a second-level trigger pulse signal. The pulse width of the first-level trigger pulse signal matches the response time constant of the electric field-sensitive ceramic material, and the amplitude of the second-level trigger pulse signal is adapted to the conduction threshold voltage of the metal oxide varistor.

[0149] The controller monitors the leakage current value of the first-stage fast response protection unit in real time. When the leakage current value exceeds the preset current threshold, it increases the amplitude of the second-stage trigger pulse signal to ensure that the second-stage high current suppression protection unit can share the overload current in time.

[0150] The controller monitors the voltage difference across the multi-level protection component in real time. When the voltage difference is lower than a preset voltage difference threshold and continues for a preset period of time, it determines that the multi-level protection component has been fully turned on.

[0151] The controller collects the temperature parameters of the first-level fast response protection unit and the second-level high current suppression protection unit. When the temperature parameters exceed the corresponding temperature limit, the trigger time interval of subsequent protection levels is dynamically increased.

[0152] For example, the multi-stage protection assembly connected in parallel across both ends of the vacuum circuit breaker body mainly includes two core units: a first-stage fast-response protection unit and a second-stage high-current suppression protection unit. The first-stage fast-response protection unit is made of an electric field-sensitive ceramic material. This material is characterized by its ability to respond quickly to changes in the electric field, with the response time adapting to the voltage phase angle change rate. When the voltage phase angle change rate is 30 degrees / millisecond, the response time of the electric field-sensitive ceramic material is approximately 0.5 milliseconds; when the voltage phase angle change rate increases to 60 degrees / millisecond, the response time can be reduced to 0.3 milliseconds. The second-stage high-current suppression protection unit is made of a metal oxide varistor material. The varistor coefficient of this material is proportional to the voltage phase angle difference. For example, when the voltage phase angle difference is 30 degrees, the varistor coefficient is approximately 35; when the voltage phase angle difference increases to 60 degrees, the varistor coefficient can be increased to 70. This varistor material mainly uses zinc oxide (ZnO) as a matrix, with appropriate amounts of gadolinium oxide (Gd₂O₃), bismuth oxide (Bi₂O₃), etc., added as additives. The on-threshold voltage of this material can be controlled by adjusting the bismuth oxide content, typically set between 1.1 and 1.3 times the rated voltage.

[0153] Regarding the turn-on timing calculation, the controller calculates the optimal turn-on timing for the first-stage fast-response protection unit and the second-stage high-current suppression protection unit based on the hierarchical protection trigger sequence. The controller ensures that the triggering time of the first-stage fast-response protection unit is earlier than that of the second-stage high-current suppression protection unit, and the time interval is inversely proportional to the expected inrush current rise rate. When the expected inrush current rise rate is 100A / ms, the trigger time interval is set to 2.5 ms; when the expected inrush current rise rate increases to 200A / ms, the trigger time interval is reduced to 1.2 ms; and when the expected inrush current rise rate reaches 300A / ms, the trigger time interval is further shortened to 0.8 ms. This inverse relationship design ensures that the second-stage protection unit intervenes promptly when the inrush current rises rapidly, effectively suppressing the inrush current peak.

[0154] For example, in a typical 10kV vacuum circuit breaker system, when the detected voltage phase angle difference is 45 degrees and the expected inrush current rise rate is 150A / ms, the controller calculates the optimal conduction sequence as follows: the first-stage protection unit is triggered at the optimal closing time, and the second-stage protection unit is triggered 1.7 ms later. This tiered conduction method can achieve a smooth transition of inrush current and avoid the sudden changes that may be caused by single-stage protection.

[0155] In terms of trigger signal generation, the controller generates a first-level trigger pulse signal and a second-level trigger pulse signal. The pulse width of the first-level trigger pulse signal is matched to the response time constant of the electric field-sensitive ceramic material. For example, when the response time constant of the electric field-sensitive ceramic material is 0.4 milliseconds, the pulse width of the first-level trigger pulse signal is set to 0.8 milliseconds, twice the response time constant, ensuring that the electric field-sensitive ceramic material can fully respond and conduct. The signal amplitude is typically set to 1.2 times the rated control voltage; for example, when the control voltage is 24V, the signal amplitude is 28.8V. The amplitude of the second-level trigger pulse signal is adapted to the conduction threshold voltage of the metal oxide varistor. For example, when the conduction threshold voltage of the metal oxide varistor is 1.2 times the rated voltage, the amplitude of the second-level trigger pulse signal is set to 1.1 times the conduction threshold voltage. For a 10kV system, the conduction threshold voltage is 12kV, so the trigger pulse signal amplitude is 13.2kV. The signal pulse width is typically set to 1.5 milliseconds to ensure that the varistor has sufficient time to respond and conduct stably.

[0156] The controller generates these trigger signals via a pulse generation circuit. This circuit includes a high-speed digital-to-analog converter, a voltage amplifier, and an isolation drive module, capable of generating pulse signals with precise amplitude and timing control, and ensuring safe isolation between the control system and the high-voltage circuit.

[0157] For real-time monitoring and adjustment, the controller monitors the leakage current value of the first-stage fast-response protection unit in real time. Monitoring uses a Hall current sensor with a sampling frequency of 10kHz and a resolution of 0.1A. When the leakage current value exceeds a preset current threshold, the controller increases the amplitude of the second-stage trigger pulse signal to ensure that the second-stage high-current suppression protection unit can promptly share the overload current. The preset current threshold is typically set to 80% of the rated current of the first-stage protection unit. For example, when the rated current of the first-stage protection unit is 100A, and the monitored leakage current value reaches 85A, the controller increases the amplitude of the second-stage trigger pulse signal by 10%, from 13.2kV ​​to 14.52kV.

[0158] When the leakage current continues to rise and exceeds 90% of the rated current, the controller further increases the amplitude of the second-stage trigger pulse signal, up to a maximum of 30% of the original design value. This dynamic adjustment mechanism ensures the safe operation of the system under abnormal conditions and prevents overload damage to the first-stage protection unit.

[0159] Regarding conduction status determination, the controller monitors the voltage difference across the multi-stage protection components in real time. Monitoring employs a high-precision voltage divider and analog-to-digital converter, with a sampling frequency of 20kHz and a resolution of 0.1% of the rated voltage. When the voltage difference falls below a preset voltage difference threshold for a preset time period, the controller determines that the multi-stage protection components are fully conducting. The preset voltage difference threshold is typically set to 5% of the rated voltage, and the preset time period is 0.2 milliseconds. For example, for a 10kV system, when the detected voltage difference across the protection components is below 500V and lasts for more than 0.2 milliseconds, the controller determines that the protection components are fully conducting and the next step can proceed.

[0160] Once all multi-level protection components are determined to be in the conducting state, the controller sends a closing control signal to the vacuum circuit breaker body, completing the entire closing process. This voltage difference-based conduction determination method is more reliable than the traditional time-delay-based method, ensuring that the circuit breaker body is closed only after the protection components are truly conducting, effectively avoiding possible misoperation.

[0161] For temperature monitoring and protection, temperature acquisition uses thermocouples or PT100 temperature sensors with a measurement accuracy of ±1℃ and a sampling period of 100 milliseconds. When the temperature parameter exceeds the corresponding temperature limit, the controller dynamically increases the trigger time interval for subsequent protection levels. For electric field-sensitive ceramic materials, the temperature limit is typically set at 85℃; for metal oxide varistor materials, the temperature limit is typically set at 120℃.

[0162] For example, when the temperature of the first-level protection unit reaches 90℃, exceeding the temperature limit of 85℃, the controller increases the trigger interval of the second-level protection unit by 30%, from 1.7 milliseconds to 2.21 milliseconds. When the temperature further rises to 100℃, the trigger interval can be increased to twice the original design value. This temperature adaptive control mechanism can effectively prevent overheating damage to the protection components, extend the service life of the equipment, and ensure that the system can still operate normally under non-ideal temperature conditions.

[0163] This invention effectively suppresses inrush current during vacuum circuit breaker closing by precisely triggering and controlling multi-level protection components. Employing a design concept that matches material properties with electrical parameters, it organically combines the physical characteristics of electric field-sensitive ceramic materials and metal oxide varistor materials with the voltage phase angle change rate and difference, achieving precise response of the protection components to system conditions. Simultaneously, a real-time monitoring and dynamic adjustment mechanism ensures stable system operation under various conditions. This invention offers significant advantages such as fast response speed, good suppression effect, and strong adaptability, effectively reducing the impact of closing inrush current on the system.

[0164] Optionally,

[0165] The controller controls the vacuum circuit breaker to close after all the multi-level protection components are turned on, and the steps to complete the inrush current suppression control include:

[0166] The controller calculates the optimal closing execution window based on the voltage phase angle change rate, the protection component current growth rate, and the system power angle stability index.

[0167] Within the optimal closing execution window, the controller predicts the response time of the closing mechanism based on the mechanical operating characteristics of the vacuum circuit breaker body and dynamically calculates the optimal moment for issuing the closing command.

[0168] Before sending the closing command, the controller calculates the transient power fluctuation of the system that will be caused by the closing operation, and performs the closing operation when the predicted power fluctuation is less than the preset power disturbance threshold.

[0169] During the closing process of the vacuum circuit breaker, the controller dynamically adjusts the conduction state of the multi-level protection components to achieve a smooth transition of the closing current. The smooth transition includes controlling the current rise rate to be less than the preset safe rise rate.

[0170] The controller collects the closing speed, rebound amplitude, and arc duration of the vacuum circuit breaker body contacts during this closing process, compares them with historical data, updates the mechanical characteristic parameter library, and optimizes the subsequent closing control strategy.

[0171] For example, this embodiment provides a method for controlling the closing of a vacuum circuit breaker body after all multi-level protection components are turned on, including five main stages: calculation of the closing execution window, determination of the optimal closing time, system stability assessment, smooth control of the closing current, and updating of mechanical characteristic parameters.

[0172] During the closing execution window calculation phase, the controller calculates the optimal closing execution window based on the voltage phase angle change rate, the protection component current growth rate, and the system power angle stability index. The voltage phase angle change rate refers to the rate at which the voltage phase angle difference across the vacuum circuit breaker changes over time, typically measured in degrees per millisecond. The controller calculates the change rate by dividing the difference in phase angle between two adjacent samples by the sampling time interval. For example, if the phase angle difference at time t1 is 30 degrees and the phase angle difference at time t2 (t2-t1=1ms) is 31.5 degrees, then the voltage phase angle change rate is 1.5 degrees / millisecond. The protection component current growth rate refers to the rate at which the current in the multi-stage protection components increases over time, measured in A / millisecond. The controller collects the current values ​​in the protection components in real time using current sensors, with a sampling frequency of 10kHz and a resolution of 0.1A. The current growth rate is obtained by calculating the difference in current between two adjacent samples divided by the sampling time interval. For example, if the current is 50A at time t1 and 55A at time t2 (t2-t1 = 0.1ms), then the current growth rate is 50A / ms. The system power angle stability index is a parameter for evaluating the transient stability of a power system, typically characterized by the power angle change rate and power angle acceleration. The controller collects power angle data from the system's generator units at a sampling frequency of 100Hz and assesses system stability by calculating the trend of the power angle change rate. For example, when the power angle change rate is less than 0.2 degrees / second and the trend is stable, the system power angle stability index is "high"; when the power angle change rate is between 0.2 and 0.5 degrees / second and the trend fluctuates, the system power angle stability index is "medium"; when the power angle change rate is greater than 0.5 degrees / second or the trend fluctuates drastically, the system power angle stability index is "low".

[0173] Based on the above three parameters, the optimal closing execution window is calculated using fuzzy logic reasoning. When the voltage phase angle change rate is less than 0.5 degrees / millisecond, the protection component current growth rate is less than 20A / millisecond, and the system power angle stability index is "high," the optimal closing execution window width is set to 10 milliseconds. When the above parameters are within a medium range, the window width is set to 5 milliseconds. When the parameters are within an unfavorable range, the window width is reduced to 2 milliseconds. For example, in a certain closing operation, the voltage phase angle change rate is 0.3 degrees / millisecond, the protection component current growth rate is 15A / millisecond, and the system power angle stability index is "high." The optimal closing execution window calculated by the controller is within 10 milliseconds from the current moment.

[0174] During the optimal closing time determination phase, the mechanical operating characteristics of the vacuum circuit breaker body include parameters such as the operating mechanism's actuation time, contact travel time, and contact rebound time. The controller retrieves these parameters from the mechanical characteristic parameter library, makes corrections based on current ambient temperature, operating voltage, and other conditions, and predicts the response time of the closing mechanism from receiving the command to the complete closure of the contacts.

[0175] For example, for a certain type of vacuum circuit breaker, under standard conditions (ambient temperature 20℃, operating voltage 100% of rated value), the operating mechanism actuation time is 15 milliseconds, the contact travel time is 25 milliseconds, the contact bounce time is 5 milliseconds, and the total response time is 45 milliseconds. When the ambient temperature is -10℃, the operating mechanism actuation time increases by approximately 10%, i.e., 16.5 milliseconds; when the operating voltage is 90% of the rated value, the actuation time increases by approximately 5%, i.e., 17.3 milliseconds. Taking all these factors into account, the controller predicts a response time of 48 milliseconds under the current conditions.

[0176] Based on the predicted response time, the controller dynamically calculates the optimal time to issue the closing command. The calculation method involves extrapolating the response time backward from the end of the optimal closing execution window to obtain the command issuance time. For example, if the optimal closing execution window is within 10 milliseconds from the current time t0 (i.e., [t0, t0 + 10ms]), and the predicted response time is 48 milliseconds, then the optimal time to issue the closing command is t0 + 10ms - 48ms = t0 - 38ms. If the calculated result is earlier than the current time, then the window start time t0 is taken as the command issuance time.

[0177] During the system stability assessment phase, before sending the closing command, the controller calculates the transient power fluctuations that the closing operation will cause. Based on the topology, load characteristics, and system operating status of the line where the vacuum circuit breaker is located, the controller uses a small disturbance analysis method to estimate the power fluctuations after closing, obtains the voltage amplitude and phase angle at both ends of the line, and calculates the initial power transfer value after closing. Then, based on the system impedance parameters and load characteristics, it predicts the power change curve during the transient process. Finally, it calculates the maximum value of the transient power fluctuation, which is the predicted power fluctuation.

[0178] For example, for a transmission line with a rated power of 50MW, the current voltage amplitude is 110% of the rated value, the phase angle difference is 15 degrees, and the initial power transmission value after closing is approximately 50 × 1.1 × sin(15°) ≈ 14.2MW. Considering the dynamic characteristics of the system, the predicted maximum transient power fluctuation is 20% of the initial transmission value, i.e., 2.84MW. The controller compares this value with a preset power disturbance threshold (e.g., 5MW). If the predicted power fluctuation is less than the preset threshold, the closing operation is performed; otherwise, the controller will postpone closing and wait for the system conditions to improve.

[0179] During the smoothing control phase of the closing current, the core objective of smooth transition is to control the current rise rate to be less than the preset safe rise rate, typically set to 50A / ms. The controller monitors the rate of change of the closing current in real time. When the current rise rate is detected to be close to the preset safe rise rate, corresponding control measures are taken: the controller adjusts the conduction parameters of the multi-stage protection components based on the difference between the real-time current rise rate and the preset safe rise rate. When the current rise rate is below 80% of the preset value, the existing conduction state is maintained; when the current rise rate is between 80% and 95% of the preset value, the conduction parameters are fine-tuned, such as increasing the equivalent resistance of the second-stage protection component by 5%-10%; when the current rise rate is between 95% and 100% of the preset value, the conduction parameters are significantly adjusted, such as increasing the equivalent resistance of the second-stage protection component by 15%-20% and temporarily suspending the conduction process of the third-stage protection component.

[0180] For example, when the detected current rise rate is 45A / ms, which is close to 90% of the preset safe rise rate of 50A / ms, the controller adjusts the equivalent resistance of the second-stage protection component from the initially set 2 ohms to 2.2 ohms, an increase of 10%.

[0181] During the mechanical characteristic parameter update phase, the controller collects the closing speed, rebound amplitude, and arc duration of the vacuum circuit breaker body contacts during the current closing process, compares this data with historical data, updates the mechanical characteristic parameter library, and optimizes subsequent closing control strategies. The contact closing speed is measured by a stroke sensor, with a typical value of 1.0-1.5 m / s; the rebound amplitude is measured by a displacement sensor and should generally not exceed 5% of the contact stroke; the arc duration is obtained through current and voltage waveform analysis and should normally be less than 2 milliseconds.

[0182] The controller compares the collected parameters with reference values ​​in historical data and calculates the deviation rate. For example, if the reference closing speed is 1.2 m / s and the measured speed is 1.1 m / s, the deviation rate is -8.3%; if the reference rebound amplitude is 2 mm and the measured speed is 2.3 mm, the deviation rate is +15%. When the deviation rate exceeds a preset threshold (usually ±10%), the controller analyzes the cause of the deviation, which may be due to environmental factors, equipment aging, or abnormal operation. Based on the analysis results, the controller updates the relevant parameters in the mechanical characteristic parameter library. For example, if the contact closing speed of three consecutive closing operations is more than 8% lower than the reference value, the controller updates the reference closing speed to 95% of the original value.

[0183] Simultaneously, the controller optimizes subsequent closing control strategies based on updated mechanical characteristic parameters. For example, when the contact closing speed decreases, the closing command needs to be issued earlier; when the rebound amplitude increases, current smoothing control needs to be enhanced to reduce arc damage during the rebound process. This closed-loop optimization mechanism enables the closing control strategy to adapt to long-term changes in circuit breaker performance and maintain optimal control performance.

[0184] This invention constructs a complete closing control method for vacuum circuit breakers by accurately calculating the closing execution window, dynamically determining the optimal closing time, evaluating system stability, achieving smooth current transition, and continuously optimizing mechanical characteristic parameters. It organically combines electrical parameter analysis with mechanical characteristic prediction, while considering system stability and equipment lifespan factors, achieving high-precision control of the closing process. This invention has higher adaptability, reliability, and accuracy, and can dynamically optimize the closing strategy based on real-time system status and equipment characteristics, effectively reducing inrush current and transient disturbances.

[0185] Secondly, a vacuum circuit breaker inrush current suppression system based on voltage phase angle is provided, including:

[0186] The first unit is used to acquire the voltage signals and historical closing data at both ends of the vacuum circuit breaker body and input them to the controller.

[0187] The second unit is used to calculate the voltage phase angle difference between the two ends of the vacuum circuit breaker body based on the voltage signal; the controller analyzes the historical closing data to obtain historical characteristic parameters.

[0188] The third unit is used to input the voltage phase angle difference, the voltage amplitude at both ends of the vacuum circuit breaker body, the system frequency, and the historical characteristic parameters into a double hidden layer neural network, and output the optimal closing time and the hierarchical protection trigger sequence.

[0189] The fourth unit is used to receive a closing command. If the current time is the optimal closing time, the controller sends trigger signals to the multi-level protection components connected in parallel at both ends of the vacuum circuit breaker body in sequence according to the hierarchical protection trigger sequence. If the current time is not the optimal closing time, the controller delays until the optimal closing time before sending the trigger signal.

[0190] The fifth unit is used to control the vacuum circuit breaker body to close after all the multi-level protection components are turned on, thereby completing the inrush current suppression control.

[0191] The sixth unit is used to add the actual inrush flow data of this closing operation to the historical closing data, in order to further optimize the dual hidden layer neural network.

[0192] Thirdly, a computer-readable storage medium is provided, having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

Claims

1. A method for suppressing inrush current in vacuum circuit breakers based on voltage phase angle, characterized in that, include: The voltage signals and historical closing data at both ends of the vacuum circuit breaker body are acquired and input to the controller. The controller calculates the voltage phase angle difference between the two ends of the vacuum circuit breaker body based on the voltage signal; The controller analyzes the historical closing data to obtain historical characteristic parameters; The controller inputs the voltage phase angle difference, the voltage amplitude across the vacuum circuit breaker body, the system frequency, and the historical feature parameters into a dual-hidden-layer neural network, and outputs the optimal closing time and the graded protection trigger sequence. Specifically, this includes: normalizing the voltage phase angle difference to obtain a normalized phase angle difference; calculating the deviation of the voltage amplitude based on the rated voltage to obtain a voltage deviation feature; calculating the deviation of the system frequency based on the nominal frequency to obtain a frequency deviation feature; constructing a historical feature vector containing the inrush current peak value and the transient recovery voltage change rate based on the historical feature parameters; and constructing a dual-hidden-layer neural network. The input layer of the dual-hidden-layer neural network receives the normalized phase angle difference, voltage deviation feature, frequency deviation feature, and historical feature vector. The number of nodes in the first and second hidden layers of the dual-hidden-layer neural network is adaptively adjusted according to the square root of the product of the number of nodes in the input layer and the number of nodes in the output layer. A combined optimization objective function is constructed based on minimizing the inrush peak value and the transient recovery voltage change rate. The dual-hidden-layer neural network is trained based on the combined optimization objective function, and the network weight parameters are updated using a backpropagation algorithm with motive terms. The normalized phase angle difference, voltage deviation features, frequency deviation features, and historical feature vectors acquired in real time are input into the trained dual-hidden-layer neural network, and the optimal closing time is calculated based on the combined optimization objective function. A graded protection trigger sequence is generated according to the optimal closing time, and the trigger time interval between adjacent protection levels in the graded protection trigger sequence is controlled to be greater than the preset minimum coordination time. When receiving a closing command, if the current time is the optimal closing time, the controller sends trigger signals to the multi-level protection components connected in parallel at both ends of the vacuum circuit breaker body in sequence according to the hierarchical protection trigger sequence; if the current time is not the optimal closing time, the controller delays until the optimal closing time before sending the trigger signal. The controller closes the vacuum circuit breaker body after all the multi-level protection components are turned on, thus completing the inrush current suppression control. The controller adds the actual inrush flow data of this closing operation to the historical closing data to further optimize the dual hidden layer neural network.

2. The method according to claim 1, characterized in that, The steps by which the controller analyzes the historical closing data to obtain historical characteristic parameters include: The historical closing data includes inrush waveform data, load parameter data, and environmental parameter data; Wavelet transform is performed on the surge waveform data to obtain multi-scale decomposition coefficients, and the temporal distribution characteristics of the surge peak are extracted based on the multi-scale decomposition coefficients. The environmental stress factor is obtained by calculating the temperature stress coefficient, humidity stress coefficient, and air pressure stress coefficient based on the environmental parameter data and multiplying them together. The information entropy of the time distribution feature, the environmental stress factor, and the load parameter data are calculated respectively. The weight coefficient of each feature is determined according to the magnitude of the information entropy, and the weighted combination is used to generate a fusion feature matrix. A radial basis function (RBF) neural network is constructed, comprising an input layer, a hidden layer, and an output layer. The number of nodes in the input layer is the same as the feature dimension of the fused feature matrix, and the hidden layer uses a Gaussian radial basis function as the activation function. The fused feature matrix and corresponding surge characteristic parameters from historical samples are input into the RBF neural network for training. The center position and width parameters of the radial basis function are optimized using the least squares method. The trained RBF neural network is then used to perform feature mapping on the fused feature matrix and output the historical feature parameters.

3. The method according to claim 1, characterized in that, The steps of constructing a combined optimization objective function based on minimizing the peak inrush current and minimizing the transient recovery voltage rate of change, training the double-hidden-layer neural network based on the combined optimization objective function, and updating the network weight parameters using the backpropagation algorithm with motive terms include: A transfer function for the vacuum circuit breaker system is constructed. Based on the transfer function, the standard deviation between the real-time state and the steady-state value of the system is calculated to obtain the transient process deviation. The rate of change of active power and reactive power of the load connected to the vacuum circuit breaker is calculated, and multiplied by the first sensitivity coefficient and the second sensitivity coefficient respectively, and then summed to obtain the weight adjustment amount. A time-varying weight coefficient is constructed based on the transient process deviation. The time-varying weight coefficient decreases exponentially as the transient process deviation increases and increases exponentially over time. The time-varying weight coefficient is combined with the weight adjustment amount to obtain the inrush weight coefficient. The product of the inrush current weighting coefficient and the inrush current peak characteristic is used as the short-term inrush current target term; the voltage weighting coefficient is obtained by subtracting the inrush current weighting coefficient from the unit value, and the product of the voltage weighting coefficient and the transient recovery voltage characteristic is used as the medium-term voltage target term; the voltage deviation is calculated to obtain the long-term stability target term; The operating force and breaking current of the vacuum circuit breaker are collected. The mechanical life loss is obtained by calculating the exponential function of the operating force and the electrical life loss is obtained by calculating the power function of the breaking current. Life constraint terms are constructed based on the mechanical life loss and the electrical life loss. The short-term inrush current target, medium-term voltage target, long-term stability target, and lifetime constraint are weighted and combined to construct an optimization objective function. The dual-hidden-layer neural network is trained using a backpropagation algorithm with momentum term, which is a weighted combination of the previous cycle update and the current gradient. The weight coefficients are dynamically adjusted based on the inner product of adjacent gradient vectors until the rate of change of the optimization objective function is less than a preset rate of change threshold for a consecutive preset number of cycles.

4. The method according to claim 1, characterized in that, The steps of generating a graded protection trigger sequence based on the optimal closing time and controlling the trigger time interval between adjacent protection levels in the graded protection trigger sequence to be greater than the preset minimum coordination time include: Construct a protection feature vector that includes voltage phase angle difference, inrush current peak characteristics, and transient recovery voltage change rate; The ratio of the expected inrush amplitude to the preset benchmark inrush amplitude is calculated based on the optimal closing time and used as the graded weighting coefficient. Using the Euclidean distance between the graded weight coefficient and the protection feature vector as the clustering criterion, a dynamic clustering method is adopted to divide the protection components into multiple protection levels, and the initial triggering time of the first protection level is determined based on the optimal closing time. Calculate the trigger time interval for each protection level, minimize the sum of squared deviations between the trigger time interval and the preset minimum coordination time, and generate an initial graded protection trigger sequence, wherein the trigger time interval between adjacent protection levels in the initial graded protection trigger sequence is greater than the minimum coordination time; Real-time inrush current data is collected during the closing process of the vacuum circuit breaker. The difference between the real-time inrush current data and the expected inrush current amplitude is calculated to obtain the inrush current deviation. The product of the square of the inrush current deviation and the adaptive coefficient is used as the trigger sequence adjustment amount. When the inrush deviation exceeds the preset inrush threshold, the trigger sequence adjustment amount is superimposed on the trigger time corresponding to the initial graded protection trigger sequence to generate a corrected graded protection trigger sequence. Each trigger time in the corrected graded protection trigger sequence satisfies the minimum coordination time constraint and the correction magnitude is proportional to the inrush deviation.

5. The method according to claim 1, characterized in that, When receiving a closing command, if the current time is the optimal closing time, the controller sends trigger signals sequentially to the multi-level protection components connected in parallel across the vacuum circuit breaker body according to the hierarchical protection trigger sequence, including the following steps: The multi-level protection component includes a first-level fast response protection unit and a second-level high current suppression protection unit. The controller calculates the optimal turn-on timing of the first-level fast response protection unit and the second-level high current suppression protection unit based on the graded protection trigger sequence, wherein the triggering time of the first-level fast response protection unit is earlier than the triggering time of the second-level high current suppression protection unit, and the time interval is inversely proportional to the expected inrush current rise rate. The controller monitors the leakage current value of the first-level fast response protection unit in real time. When the leakage current value exceeds the preset current threshold, the amplitude of the second-level trigger pulse signal is increased. The controller monitors the voltage difference across the multi-level protection component in real time. When the voltage difference is lower than a preset voltage difference threshold and continues for a preset period of time, it determines that the multi-level protection component has been turned on. The controller collects the temperature parameters of the first-level fast response protection unit and the second-level high current suppression protection unit. When the temperature parameters exceed the corresponding temperature limit, the trigger time interval of subsequent protection levels is dynamically increased.

6. The method according to claim 1, characterized in that, The controller controls the vacuum circuit breaker to close after all the multi-level protection components are turned on, and the steps to complete the inrush current suppression control include: The controller calculates the optimal closing execution window based on the voltage phase angle change rate, the protection component current growth rate, and the system power angle stability index. Within the optimal closing execution window, the controller predicts the response time of the closing mechanism based on the mechanical operating characteristics of the vacuum circuit breaker body and dynamically calculates the optimal moment for issuing the closing command. Before sending the closing command, the controller calculates the transient power fluctuation of the system that will be caused by the closing operation, and performs the closing operation when the predicted power fluctuation is less than the preset power disturbance threshold. During the closing process of the vacuum circuit breaker, the controller dynamically adjusts the conduction state of the multi-level protection components to achieve a smooth transition of the closing current. The smooth transition includes controlling the current rise rate to be less than the preset safe rise rate. The controller collects the closing speed, rebound amplitude, and arc duration of the vacuum circuit breaker body contacts during this closing process, compares them with historical data, updates the mechanical characteristic parameter library, and optimizes the subsequent closing control strategy.

7. A vacuum circuit breaker inrush current suppression system based on voltage phase angle, used to implement the method of any one of claims 1-6, characterized in that, include: The first unit is used to acquire the voltage signals and historical closing data at both ends of the vacuum circuit breaker body and input them to the controller. The second unit is used to calculate the voltage phase angle difference between the two ends of the vacuum circuit breaker body based on the voltage signal; the controller analyzes the historical closing data to obtain historical characteristic parameters. The third unit is used to input the voltage phase angle difference, the voltage amplitude at both ends of the vacuum circuit breaker body, the system frequency, and the historical characteristic parameters into a double hidden layer neural network, and output the optimal closing time and the hierarchical protection trigger sequence. The fourth unit is used to send trigger signals to the multi-level protection components connected in parallel at both ends of the vacuum circuit breaker body in sequence according to the hierarchical protection trigger sequence when receiving the closing command if the current time is the optimal closing time. If the current time is not the optimal closing time, the controller will delay until the optimal closing time before sending the trigger signal; The fifth unit is used to control the vacuum circuit breaker body to close after all the multi-level protection components are turned on, thereby completing the inrush current suppression control. The sixth unit is used to add the actual inrush flow data of this closing operation to the historical closing data, in order to further optimize the dual hidden layer neural network.

8. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.

Images