A self-adaptive phase selection control system and method for a pneumatic main circuit breaker of a motor train unit

Abstract

This invention discloses an adaptive phase selection control system and method for a pneumatic main circuit breaker in a high-speed train, relating to the field of rail transit technology. The invention includes a microprocessor, an I / O drive module, a sensor module, an ADC sampling module, and a VCB signal monitoring module. The microprocessor stores the stable opening time, target opening phase, and target closing phase of the pneumatic circuit breaker, and loads a closing action time prediction model constructed based on a gradient boosting decision tree. The sensor module monitors and uploads data on characteristic factors affecting the closing action time. The ADC sampling module collects voltage and current signals from the train's network in real time. The VCB signal monitoring module receives and intercepts circuit breaker action commands transmitted from the train and determines whether the command type is an opening command or a closing command. This invention solves the incompatibility problem between pneumatic circuit breakers and phase selection control technology by constructing a closing action time prediction model.

Description

Technical Field

[0001] This invention belongs to the field of rail transit technology, and in particular relates to an adaptive phase selection control system and method for a pneumatic main circuit breaker of a high-speed train. Background Technology

[0002] When a high-speed train crosses a phase-splitting section, frequent tripping of the main circuit breaker can generate switching overvoltages. If the amplitude of these overvoltages is too large, it can affect the safe and stable operation of the train's electrical equipment. Furthermore, the accompanying arcing can burn the pantograph and other contact network components, causing traction substation tripping, and further leading to partial power outages and locomotive shutdowns. Phase selection control technology can effectively suppress switching overvoltages. However, many high-speed trains in my country are equipped with pneumatic vacuum main circuit breakers, and currently there are no mature phase control solutions or products specifically designed for these pneumatic vacuum circuit breakers.

[0003] Currently, the suppression of operational overvoltages caused by circuit breakers is mainly achieved by adding protective measures to mitigate the damage. However, the current phase selection control technology, if applied to circuit breakers, requires providing accurate operating times for the circuit breaker. Therefore, it is not suitable for the pneumatic circuit breakers with highly dispersed closing operating times that are widely used on high-speed trains. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes an adaptive phase selection control system and method for a pneumatic main circuit breaker of a high-speed train.

[0005] To achieve the above objectives, the present invention provides the following technical solution: An adaptive phase selection control system for a pneumatic main circuit breaker of a high-speed train includes: The microprocessor is used to store the stable opening action time, target opening phase, and target closing phase of the pneumatic circuit breaker, and loads the closing action time prediction model built based on gradient boosting decision tree. The sensor module is used to monitor and upload data on characteristic factors affecting the closing action time to the microprocessor. The characteristic factors include the internal air pressure at the moment of closing, the external air source pressure of the circuit breaker, the duration of the last closing state, the last closing action time, the ambient temperature, and the temperature of the internal air storage tank of the circuit breaker. The ADC sampling module is used to collect voltage and current signals from the train's network in real time and transmit them to the microprocessor. The VCB signal monitoring module is used to receive and intercept circuit breaker operation commands transmitted from the vehicle, determine whether the command type is a tripping command or a closing command, and transmit the determination result to the microprocessor. The I / O driver module is used to send opening or closing signals to the circuit breaker according to the control instructions of the microprocessor.

[0006] Furthermore, the closing action time prediction model stored in the microprocessor is a gradient boosting decision tree regression model with six feature factors as input and the closing action time of the pneumatic circuit breaker as output; the training set of the prediction model can be adaptively supplemented and updated.

[0007] An adaptive phase selection control method for a pneumatic main circuit breaker of a high-speed train is implemented based on the aforementioned adaptive phase selection control system for a pneumatic main circuit breaker of a high-speed train, including a tripping control process and a closing control process. The tripping control process includes: when the VCB signal monitoring module receives the tripping command, the ADC sampling module monitors the vehicle-to-grid current and identifies the next current phase zero-crossing point as the reference zero-crossing point. The microprocessor calculates the first delay time based on the stored tripping action time, target tripping phase, and current cycle. After the first delay time starts at the reference zero-crossing point, the I / O drive module releases the tripping signal to the circuit breaker. The closing control process includes: when the VCB signal monitoring module receives the closing command, the ADC sampling module monitors the vehicle-to-grid voltage and identifies the next voltage phase zero-crossing point as the reference zero-crossing point; the sensor module collects the values ​​of six characteristic factors at the current moment; the microprocessor determines whether the amount of data in the current prediction model training set has reached a set threshold; if the threshold has not been reached, the I / O drive module directly releases the closing signal and records the closing action time and corresponding characteristic factors as new data to supplement the training set; if the threshold has been reached, the values ​​of the six characteristic factors are input into the prediction model to obtain the predicted value of the closing action time; the microprocessor calculates the second delay time based on the predicted value, the target closing phase, and the voltage period; after the second delay time begins at the reference zero-crossing point, the I / O drive module releases the closing signal and records the actual closing action time and corresponding characteristic factors to update the training set.

[0008] Furthermore, the first delay time is calculated as follows: select the smallest integer such that the first delay time is positive. The first delay time is equal to the integer multiplied by the current period minus the tripping action time plus the time required from the zero phase to the target tripping phase.

[0009] Furthermore, the second delay time is calculated as follows: select the smallest integer such that the second delay time is positive. The second delay time is equal to the integer multiplied by the voltage period minus the predicted closing action time plus the time required from the zero phase to the target closing phase.

[0010] Furthermore, during the closing control process, when the amount of training set data has not reached the set threshold, a direct closing operation is performed, and after the closing is completed, the closing action time and the corresponding six characteristic factor values ​​are added to the training set as a new set of data; when the amount of training set data has reached the set threshold, a phase-selective closing operation based on the prediction model is performed, and after the closing is completed, the closing action time and the corresponding six characteristic factor values ​​are updated in sequence to a set of training set data.

[0011] Furthermore, the six feature factors were determined by filtering the candidate feature factors based on their importance using a gradient boosting decision tree algorithm. The candidate feature factors include the internal air pressure at the moment of closing, the external air source pressure of the circuit breaker, the duration of the last closing state, the internal air pressure of the last closing, the internal air pressure difference between the two closing operations, the time of the last closing operation, the ambient temperature, and the temperature of the internal air storage tank of the circuit breaker.

[0012] Furthermore, the importance screening includes: collecting closing action time data and candidate feature factor data of pneumatic circuit breakers, performing data cleaning and missing value processing, mapping the data to a unified interval using a standardization method, calculating the cumulative contribution value of each feature to the reduction of the loss function during all weak decision tree splits using a gradient boosting decision tree, obtaining the importance score of each feature after normalization, and screening out the six feature factors with the highest importance scores.

[0013] Furthermore, the method for constructing the closing action time prediction model includes: using the six feature factors as input and the closing action time of the pneumatic circuit breaker as output, constructing a set of weak regression trees, fitting the residual between the current predicted value and the true value for each weak regression tree, using the mean squared error loss function, and optimizing the parameters such as the number of decision trees, learning rate, maximum depth, minimum number of samples for internal node splitting, and minimum number of samples for leaf nodes through a grid search method, and accumulating the prediction results of all weak regression trees to obtain a strong regression model.

[0014] Furthermore, during the tripping control process, the target tripping phase is the current phase; during the closing control process, the target closing phase is the voltage phase, and the target closing phase is set to zero phase with the goal of suppressing operational overvoltage.

[0015] The present invention has the following beneficial effects: This invention provides a theoretical analysis of the closing process of a pneumatic circuit breaker and, with the help of test data, filters the importance of features to identify the main characteristic factors affecting the closing action time.

[0016] This invention solves the problem of incompatibility between pneumatic circuit breakers and phase selection control technology by constructing a closing action time prediction model.

[0017] This invention features an adaptive supplementation and update function for the prediction model training set, resolving the issue of individual differences in circuit breakers and adapting it to each vehicle and each circuit breaker. It also prevents a decrease in model prediction accuracy due to factors such as aging of the circuit breaker's mechanical structure and climate change, thereby improving its engineering applicability.

[0018] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the feature importance screening results of the present invention; Figure 2 This is a hardware schematic diagram of the phase selection control system of the present invention; Figure 3 This is a schematic diagram of the phase selection tripping method of the present invention; Figure 4 This is a schematic diagram of the phase selection closing of the present invention; Figure 5 This is a schematic diagram of the overall process of the adaptive phase selection control of the pneumatic circuit breaker of the present invention; Figure 6 This is a scatter plot of the prediction model results. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Please see Figure 2 As shown, the present invention is an adaptive phase selection control system for a pneumatic main circuit breaker of a high-speed train, comprising: The microprocessor is used to store the stable opening action time, target opening phase, and target closing phase of the pneumatic circuit breaker, and loads the closing action time prediction model built based on gradient boosting decision tree. The sensor module is used to monitor and upload data on characteristic factors affecting the closing action time to the microprocessor. The characteristic factors include the internal air pressure at the moment of closing, the external air source pressure of the circuit breaker, the duration of the last closing state, the last closing action time, the ambient temperature, and the temperature of the internal air storage tank of the circuit breaker. The ADC sampling module is used to collect voltage and current signals from the train's network in real time and transmit them to the microprocessor. The VCB signal monitoring module is used to receive and intercept circuit breaker operation commands transmitted from the vehicle, determine whether the command type is a tripping command or a closing command, and transmit the determination result to the microprocessor. The I / O driver module is used to send opening or closing signals to the circuit breaker according to the control instructions of the microprocessor.

[0023] The closing action time prediction model stored in the microprocessor is a gradient boosting decision tree regression model with six feature factors as input and the closing action time of the pneumatic circuit breaker as output; the training set of the prediction model can be adaptively supplemented and updated.

[0024] Please see Figure 5 As shown, the present invention is an adaptive phase selection control method for a pneumatic main circuit breaker of a high-speed train, which is based on the aforementioned adaptive phase selection control system for a pneumatic main circuit breaker of a high-speed train, and includes a tripping control process and a closing control process. The tripping control process includes: when the VCB signal monitoring module receives the tripping command, the ADC sampling module monitors the vehicle-to-grid current and identifies the next current phase zero-crossing point as the reference zero-crossing point. The microprocessor calculates the first delay time based on the stored tripping action time, target tripping phase, and current cycle. After the first delay time starts from the reference zero-crossing point, the I / O drive module releases the tripping signal to the circuit breaker. The closing control process includes: when the VCB signal monitoring module receives the closing command, the ADC sampling module monitors the vehicle-to-grid voltage and identifies the next voltage phase zero-crossing point as the reference zero-crossing point. The sensor module collects the values ​​of six characteristic factors at the current moment. The microprocessor determines whether the amount of data in the current prediction model training set has reached the set threshold. If the threshold has not been reached, the I / O drive module directly releases the closing signal and records the closing action time and corresponding characteristic factors as new data to supplement the training set. If the threshold has been reached, the values ​​of the six characteristic factors are input into the prediction model to obtain the predicted value of the closing action time. The microprocessor calculates the second delay time based on the predicted value, the target closing phase, and the voltage period. After the second delay time starts from the reference zero-crossing point, the I / O drive module releases the closing signal and records the actual closing action time and corresponding characteristic factors to update the training set.

[0025] The first delay time is calculated as follows: select the smallest integer so that the first delay time is positive. The first delay time is equal to the integer multiplied by the current period minus the tripping action time plus the time required from the zero phase to the target tripping phase.

[0026] The second delay time is calculated as follows: select the smallest integer so that the second delay time is positive. The second delay time is equal to the integer multiplied by the voltage period minus the predicted closing action time plus the time required from the zero phase to the target closing phase.

[0027] During the closing control process, if the amount of training set data has not reached the set threshold, a direct closing operation is performed, and after the closing is completed, the closing action time and the corresponding six characteristic factor values ​​are added to the training set as a new set of data; if the amount of training set data has reached the set threshold, a phase-selective closing operation based on the prediction model is performed, and after the closing is completed, the closing action time and the corresponding six characteristic factor values ​​are updated in sequence to a set of training set data.

[0028] The six feature factors were determined by filtering the candidate feature factors based on their importance using the gradient boosting decision tree algorithm. The candidate feature factors include the internal air pressure at the moment of closing, the external air source pressure of the circuit breaker, the duration of the last closing state, the internal air pressure of the last closing, the internal air pressure difference between the two closings, the time of the last closing action, the ambient temperature, and the temperature of the internal air storage tank of the circuit breaker.

[0029] Feature importance screening includes: collecting closing action time data of pneumatic circuit breakers and candidate feature factor data, cleaning and handling missing values, mapping the data to a unified interval using a standardization method, calculating the cumulative contribution of each feature to the reduction of the loss function during all weak decision tree splits using gradient boosting decision trees, obtaining the importance score of each feature after normalization, and selecting the six feature factors with the highest importance scores.

[0030] The method for constructing the closing action time prediction model includes: taking six feature factors as input and the closing action time of the pneumatic circuit breaker as output, constructing a set of weak regression trees, fitting the residual between the current predicted value and the true value for each weak regression tree, using the mean squared error loss function, and optimizing the parameters such as the number of decision trees, learning rate, maximum depth, minimum number of samples for internal node splitting, and minimum number of samples for leaf nodes through a grid search method, and summing the prediction results of all weak regression trees to obtain a strong regression model.

[0031] During the tripping control process, the target tripping phase is the current phase; during the closing control process, the target closing phase is the voltage phase. To suppress operational overvoltage, the target closing phase is set to zero phase.

[0032] The specific application of this embodiment is as follows: I. Preliminary Algorithm Preparation 1. Screening of key characteristic factors Analyzing the opening and closing principles and specific data of pneumatic circuit breakers, it is found that their opening time is stable and suitable for phase selection control technology, while their closing time is dispersed and not suitable. A theoretical analysis of the closing principle of pneumatic circuit breakers reveals that closing is achieved by external high-pressure air driving a cylinder piston. When the solenoid valve is energized, the inlet valve opens, and high-pressure air enters the cylinder through the solenoid valve and the switching valve, driving the piston inside the cylinder. The piston pushes the insulating rod, which in turn pushes the moving contact of the vacuum tube, ultimately closing and pressing it against the stationary contact. Simultaneously, the return spring mounted on the moving contact is compressed and stores energy. Factors contributing to the strong dispersion in the closing action time include: (1) Air pressure factor When the solenoid valve is energized, compressed air enters the transmission cylinder, pushing the transmission rod to close the main contacts. The air pressure inside the transmission cylinder directly determines the thrust pushing the transmission rod, thus affecting the time required for the main contacts to close after the solenoid valve is energized. Characteristic factors to be identified: internal air pressure at the moment of closing (A1), and external air pressure of the circuit breaker (A2).

[0033] (2) Reset Spring Mechanism When the circuit breaker closes, the drive rod compresses the return spring to ensure tight contact between the contacts. The total compression stroke of the return spring remains constant each time the circuit breaker closes. During normal train operation, the circuit breaker is in the closed state, and the return spring is in a compressed state. In this state, the spring experiences stress relaxation. When the circuit breaker closes again after leaving the phase separation zone, the return spring needs to be compressed again. However, due to the stress loss caused by the previous stress relaxation, the spring force that the insulating rod needs to overcome to push the moving contact upward changes, leading to increased dispersion in the closing action time of the pneumatic circuit breaker. Furthermore, considering the different air pressures within the drive cylinder each time the circuit breaker closes, the different forces and speeds required to compress the return spring to the same deformation may also affect the spring stress. Proposed characteristic factors: duration of the previous closing state (A3), internal air pressure of the previous closing (A4), internal air pressure difference between two closings (A5), and operating time of the previous closing (A6).

[0034] (3) Ambient temperature The mechanical structure of pneumatic vacuum circuit breakers is complex, and changes in ambient temperature may affect the materials and mechanical properties, leading to pressure fluctuations in the internal compressed air or deviations in control signals. Proposed characteristic factors: ambient temperature (A7), temperature of the internal air reservoir of the circuit breaker (A8).

[0035] In summary, based on the analysis of the closing principle of the pneumatic vacuum circuit breaker, eight characteristic factors affecting the closing action time are identified: internal air pressure at the moment of closing (A1), external air source pressure of the circuit breaker (A2), duration of the previous closing state (A3), internal air pressure of the previous closing (A4), internal air pressure difference between the two closing operations (A5), time of the previous closing action (A6), ambient temperature (A7), and temperature of the internal air storage tank of the circuit breaker (A8).

[0036] A test platform was built to collect closing action time data and influencing factor data, and the GBDT algorithm was used for feature importance screening. First, the dataset underwent data cleaning and missing value processing. Invalid samples resulting from test anomalies, equipment failures, or recording errors required data cleaning. A pneumatic circuit breaker is essentially a specialized power switch integrating a mechanical actuator, arc-extinguishing system, and intelligent control unit. The closing action time and its dispersion can be represented by the switch's delay and bounce. Therefore, the closing action time data from the dataset can be used... average The delay of the circuit breaker is represented by its root mean square error or standard deviation, while the chatter of the circuit breaker is represented by its root mean square error or standard deviation.

[0037] Therefore, the average delay of the pneumatic circuit breaker is: In the formula, For the first Second closing operation time; This represents the total number of samples. The vibration of the pneumatic circuit breaker is: Using the normal distribution as the mathematical model for the closing action time of a pneumatic circuit breaker, the mathematical expectation in the probability density formula is... Replace with average closing action time Standard deviation Replace with the closing bounce time of a pneumatic circuit breaker The formula is as follows: In the formula, The value of the closing action time is a random variable; therefore, 3 can be selected. Outlier identification is performed based on the principle of centralizing and organizing the closing action time data under similar characteristic factors in the dataset. For samples that deviate from the average delay by more than 3 times the jitter time, the data is reviewed in conjunction with the test records, and the original data that is confirmed to be test error is removed. For values ​​that are physically impossible, such as air pressure exceeding the sensor's range or negative action time, the original data of that group is directly removed.

[0038] In addition to data cleaning, missing value processing was performed on the dataset to ensure its integrity and meet the training set requirements of the prediction model. For abnormal data where the feature factors were collected correctly, random numbers were taken from the average delay of the pneumatic circuit breaker under the same feature factors within a jitter time to supplement it. This ensured that the number of valid samples for each type of pneumatic circuit breaker remained above 900. Analysis was performed on the original dataset obtained from the tests. Since all feature factors affecting the closing action time of the pneumatic circuit breaker, except for some ambient temperature values, were positive, and there were no abnormal extreme values ​​after the first step of data cleaning, Min-Max standardization was used for data processing. The maximum value of the same type of feature was set as the upper limit and mapped to 1, the minimum value was set as the lower limit and mapped to 0, and other feature values ​​were mapped proportionally to the interval, thus linearly stretching or compressing the value range of the original data to the [0,1] interval. The specific formula is as follows: in, These are the eigenvalues ​​whose standardized values ​​are fixed in the interval [0,1]. It is the minimum value within the sample of this characteristic factor; This is the maximum value within the sample of this feature factor.

[0039] GBDT is an ensemble learning method based on decision trees. It calculates the cumulative contribution of a feature to the reduction in the loss function during the splitting of all weak decision trees. By summing the local importance of a feature at all splitting nodes of all weak decision trees and then normalizing the sum, the global importance score of that feature is obtained. Assuming the model consists of K weak decision trees, the loss function value before splitting at node t of the k-th tree is... The loss function values ​​of the left and right leaf points after splitting are respectively and Therefore, the reduction in loss from this split is: So features In the The importance of a local feature on a tree is the sum of the loss reductions brought by that feature at all split nodes in the entire tree: In the formula, For the first All internal split nodes of the tree; For nodes Adopting the first Split based on each feature; For nodes The splitting gain resulting from the splitting process. Based on the first... The importance of local features in each decision tree can be further calculated. Throughout the entire Global importance within a GBDT model of decision trees: To facilitate comparison of feature importance among various features, the global importance is normalized so that its value ranges between [0,1], resulting in a normalized importance score for each feature: The main influencing factors ultimately include: internal air pressure at the moment of closing (A1), external air source pressure of the circuit breaker (A2), duration of the last closing state (A3), time of the last closing action (A6), ambient temperature (A7), and temperature of the internal air reservoir of the circuit breaker (A8). Figure 1 As shown.

[0040] 2. Prediction Model Construction Based on these six main feature factors as input and the closing action time of the pneumatic circuit breaker as output, a set of weak learners (trees) is constructed. Each time, the current prediction is used as the benchmark, and the next weak classifier is used to fit the residual of the error function with respect to the predicted value. The results of multiple decision trees are accumulated as the final prediction output. The initialization model formula is as follows: In the formula, For input samples; These are model parameters; For classification and regression trees; The weight of each tree; For the first A decision tree; Let be the total number of decision trees. Using the negative gradient of the loss function as the residual fitting method, in regression tasks, a predicted value is generated for each sample in each iteration. The loss function is the mean squared error loss function. : In the formula, For the first The true value of each sample; For the first Predicted values ​​for each sample; This represents the residual between the actual value and the predicted value. This iterative process is repeated until the preset number of weak trees or the loss no longer decreases. Finally, the prediction results of the initial model and all weighted weak trees are summed to form a strong regression model of the closing action time of the pneumatic circuit breaker. The key parameters of the GBDT model are shown in Table 1 below: Table 1 Important parameters of the GBDT model The grid search method was used to optimize the important parameters of the GBDT model. First, the candidate values ​​of the important parameters were determined: n_estimators[100,200,300,400,500], learning_rate[0.01,0.05,0.1,0.2], max_depth[3,5,7,9], min_samples_split[2,4,6,8,10], and min_samples_leaf[1,2,3,4,5]. These were defined as the important parameter space of the model. K-fold cross-validation was performed on each combination of important parameters in the parameter grid. After comparing the average cross-validation scores of all parameter combinations, the combination with the highest score was selected as the optimal parameters, as shown in Table 2 below.

[0041] Table 2 Important parameters of the GBDT model II. Introduction to Phase Selection Control System The phase selection control system consists of a microprocessor, I / O driver module, sensor module, ADC sampling module, and VCB signal monitoring module, and is connected to the central control unit (CCU) of the EMU. The microprocessor is responsible for storing, calculating, and implementing predictive models; the ADC sampling module is responsible for real-time acquisition and feedback of voltage and current signals from the EMU's network; the VCB signal detection module is responsible for monitoring each circuit breaker operation command received from the vehicle. The circuit breaker control voltage is DC110V; if the DC110V signal changes from low to high, it is a closing signal; otherwise, it is a opening signal; the I / O driver module controls relays to disconnect or send DC110V signals to the circuit breaker, enabling it to complete the opening and closing actions; the sensor module is responsible for monitoring and uploading air pressure and temperature data, among other characteristic factors. The phase selection control system stores the stable opening time of the pneumatic circuit breaker. The target tripping phase α and the target closing phase β are loaded into the GBDT prediction model framework.

[0042] III. Specific Phase Selection Control Methods The specific phase selection control method is as follows: The VCB signal monitoring module receives and intercepts the circuit breaker operation command transmitted from the vehicle and determines the command type. If it is a tripping command, the ADC sampling module monitors the vehicle-to-grid current and identifies the next current phase zero-crossing point as a reference zero-crossing point, combined with the circuit breaker tripping operation time. Phase of the target circuit breaker The microprocessor calculates the first delay time. , The calculation method is as follows: In the formula, To ensure that the first delay time in the formula The smallest positive integer; The current cycle time; From zero phase to Phase delay. The internal clock delay of the phase selection control system at the current phase zero-crossing point. Afterwards, the I / O drive module releases the trip signal to the circuit breaker. The timing logic of the trip control process is as follows: Figure 3 As shown. The circuit breaker receives the trip signal, and after its own tripping action time... Then, in the target current phase The tripping operation is completed. If the command is to close, the ADC sampling module monitors the vehicle-to-grid voltage and identifies the next voltage phase zero-crossing point as a reference zero-crossing point. The sensor module and internal clock collect the values ​​of six characteristic factors at the current moment. Then, it is determined whether the amount of training data in the prediction model has reached the set threshold (the threshold for the dataset is set by comprehensively considering the closing frequency of the circuit breaker during the actual operation of the EMU and the amount of dataset required for model prediction; the threshold can be manually entered). If the threshold is reached, the I / O drive module directly releases the closing signal to the circuit breaker and records this moment. Record the moment when the circuit breaker completes the closing action. The final closing time will be... - and corresponding characteristic factor values This data is added to the training set as a supplement. If the threshold has been reached, the feature factor values ​​are... As input, the predicted closing time is output using the prediction model. Calculate the second delay time . The calculation method is as follows: In the formula, To ensure that the second delay time in the formula The smallest positive integer; The voltage cycle time; From zero phase to Phase delay. The internal clock delay of the phase selection control system at the voltage phase zero-crossing point. Afterwards, the I / O drive module releases a closing signal to the circuit breaker and records the time. The circuit breaker receives the closing signal and, after its own closing action time... Then, at the target voltage phase Complete the closing operation and record this moment. The timing relationship of the closing control process is as follows: Figure 4As shown. The final closing operation time will be... - and corresponding characteristic factor values As a set of data, update a set of training data sequentially.

[0043] Furthermore, with the goal of suppressing operational overvoltage, the target closing phase is set to 0° (at which point the theoretical overvoltage value is 0). Combining the phase selection closing control process and delay time calculation formula described above, the deviation between the final closing phase and the target phase is the deviation between the model prediction and the actual value. The prediction model results are as follows... Figure 6 As shown, the statistical difference between the predicted and actual values ​​from the model prediction results is within ±4ms, with most within ±2.5ms. Taking the reference overvoltage standard for AC25kV trains as an example, actual measurements and simulations show that overvoltage generated during closing within ±60° will not affect the train. Using 0° as the target phase angle and combining the prediction model for phase selection closing control, the probability of the actual closing phase being within ±60° is 93%.

[0044] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0045] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. An adaptive phase selection control system for a pneumatic main circuit breaker of a high-speed train, characterized in that, include: The microprocessor is used to store the stable opening action time, target opening phase, and target closing phase of the pneumatic circuit breaker, and loads the closing action time prediction model built based on gradient boosting decision tree. The sensor module is used to monitor and upload data on characteristic factors affecting the closing action time to the microprocessor. The characteristic factors include the internal air pressure at the moment of closing, the external air source pressure of the circuit breaker, the duration of the last closing state, the last closing action time, the ambient temperature, and the temperature of the internal air storage tank of the circuit breaker. The ADC sampling module is used to collect voltage and current signals from the train's network in real time and transmit them to the microprocessor. The VCB signal monitoring module is used to receive and intercept circuit breaker operation commands transmitted from the vehicle, determine whether the command type is a tripping command or a closing command, and transmit the determination result to the microprocessor. The I / O driver module is used to send opening or closing signals to the circuit breaker according to the control instructions of the microprocessor.

2. The adaptive phase selection control system for a pneumatic main circuit breaker of a high-speed train according to claim 1, characterized in that, The closing action time prediction model stored in the microprocessor is a gradient boosting decision tree regression model with six feature factors as input and the closing action time of the pneumatic circuit breaker as output; the training set of the prediction model can be adaptively supplemented and updated.

3. An adaptive phase selection control method for a pneumatic main circuit breaker of a high-speed train, implemented based on the adaptive phase selection control system for the pneumatic main circuit breaker of a high-speed train as described in any one of claims 1-2, characterized in that... This includes the opening control process and the closing control process; The tripping control process includes: when the VCB signal monitoring module receives the tripping command, the ADC sampling module monitors the vehicle-to-grid current and identifies the next current phase zero-crossing point as the reference zero-crossing point. The microprocessor calculates the first delay time based on the stored tripping action time, target tripping phase, and current cycle. After the first delay time begins at the reference zero-crossing point, the I / O drive module releases the tripping signal to the circuit breaker. The closing control process includes: when the VCB signal monitoring module receives the closing command, the ADC sampling module monitors the vehicle-to-grid voltage and identifies the next voltage phase zero-crossing point as the reference zero-crossing point; the sensor module collects the values ​​of six characteristic factors at the current moment; the microprocessor determines whether the amount of data in the current prediction model training set has reached a set threshold; if the threshold has not been reached, the I / O drive module directly releases the closing signal and records the closing action time and corresponding characteristic factors as new data to supplement the training set; if the threshold has been reached, the values ​​of the six characteristic factors are input into the prediction model to obtain the predicted value of the closing action time; the microprocessor calculates the second delay time based on the predicted value, the target closing phase, and the voltage period; after the second delay time begins at the reference zero-crossing point, the I / O drive module releases the closing signal and records the actual closing action time and corresponding characteristic factors to update the training set.

4. The adaptive phase selection control method for a pneumatic main circuit breaker of a high-speed train according to claim 3, characterized in that, The first delay time is calculated as follows: select the smallest integer so that the first delay time is positive. The first delay time is equal to the integer multiplied by the current period minus the tripping action time plus the time required from the zero phase to the target tripping phase.

5. The adaptive phase selection control method for a pneumatic main circuit breaker of a high-speed train according to claim 3, characterized in that, The second delay time is calculated as follows: select the smallest integer so that the second delay time is positive. The second delay time is equal to the integer multiplied by the voltage period minus the predicted closing action time plus the time required from the zero phase to the target closing phase.

6. The adaptive phase selection control method for a pneumatic main circuit breaker of a high-speed train according to claim 3, characterized in that, During the closing control process, when the amount of training set data has not reached the set threshold, a direct closing operation is performed, and after the closing is completed, the closing action time and the corresponding six characteristic factor values ​​are added to the training set as a new set of data; when the amount of training set data has reached the set threshold, a phase-selective closing operation based on the prediction model is performed, and after the closing is completed, the closing action time and the corresponding six characteristic factor values ​​are updated in sequence to a set of training set data.

7. The adaptive phase selection control method for a pneumatic main circuit breaker of a high-speed train according to claim 3, characterized in that, The six characteristic factors were determined by filtering candidate characteristic factors based on their importance using a gradient boosting decision tree algorithm. The candidate characteristic factors include the internal air pressure at the moment of closing, the external air source pressure of the circuit breaker, the duration of the last closing state, the internal air pressure of the last closing, the internal air pressure difference between the two closing operations, the time of the last closing operation, the ambient temperature, and the temperature of the internal air storage tank of the circuit breaker.

8. The adaptive phase selection control method for a pneumatic main circuit breaker of a high-speed train according to claim 7, characterized in that, The importance screening includes: collecting closing action time data of pneumatic circuit breakers and candidate feature factor data, cleaning and handling missing values, mapping the data to a unified interval using a standardization method, calculating the cumulative contribution of each feature to the reduction of the loss function during all weak decision tree splits using a gradient boosting decision tree, obtaining the importance score of each feature after normalization, and screening out the six feature factors with the highest importance scores.

9. The adaptive phase selection control method for a pneumatic main circuit breaker of a high-speed train according to claim 3, characterized in that, The method for constructing the closing action time prediction model includes: using the six feature factors as input and the closing action time of the pneumatic circuit breaker as output, constructing a set of weak regression trees, fitting the residual between the current predicted value and the true value for each weak regression tree, using the mean squared error loss function, and optimizing the parameters such as the number of decision trees, learning rate, maximum depth, minimum number of samples for internal node splitting, and minimum number of samples for leaf nodes through a grid search method, and accumulating the prediction results of all weak regression trees to obtain a strong regression model.

10. The adaptive phase selection control method for a pneumatic main circuit breaker of a high-speed train according to claim 3, characterized in that, During the tripping control process, the target tripping phase is the current phase; during the closing control process, the target closing phase is the voltage phase, and the target closing phase is set to zero phase with the goal of suppressing operational overvoltage.

Images