Fault detection method and model training method for isolating switch in power equipment

Abstract

The embodiment of the invention provides a fault detection method and a model training method for an isolating switch in power equipment. The method comprises the following steps: detecting a current value of an isolating switch in power equipment, and obtaining a switch parameter set of the isolating switch in a preset time period; the switch parameter set comprises a current sequence, an angle sequence and a torque sequence; the preset time period is a corresponding time period when the current value of the isolating switch detected in real time is greater than or equal to a preset threshold value; the current sequence comprises a current value at each moment in a preset time period; the angle sequence comprises an angle value of each moment in a preset time period, and the torque sequence comprises a torque value of each moment in the preset time period; and inputting the current sequence, the angle sequence and the torque sequence in the switch parameter set into a preset fault identification model for fault identification processing to obtain a fault detection result of the isolation switch. The method is used for achieving the effect of accurately and quickly detecting the fault of the isolating switch of the power equipment.

Description

Technical Field

[0001] This application relates to the field of power equipment testing technology, and in particular to a fault detection method and model training method for disconnecting switches in power equipment. Background Technology

[0002] Disconnecting switches are key equipment in power systems responsible for opening and closing operations. Mechanical jamming of these switches can have serious consequences, potentially leading to equipment damage or even system failure.

[0003] In related technologies, fault detection mainly relies on manual inspection and fixed threshold alarms. However, manual inspection cannot capture transient anomalies in real time, and fixed threshold alarms are easily affected by environmental noise, leading to misjudgments.

[0004] Therefore, there is an urgent need for a solution that can accurately and quickly detect faults in disconnecting switches of power equipment. Summary of the Invention

[0005] This application provides a fault detection method and model training method for disconnecting switches in power equipment, so as to achieve accurate and rapid detection of faults in disconnecting switches of power equipment.

[0006] In a first aspect, embodiments of this application provide a fault detection method for disconnecting switches in power equipment, comprising:

[0007] The system detects the current value of a disconnector switch in a power equipment and obtains a set of switch parameters for the disconnector switch during a preset time period. The set of switch parameters includes a current sequence, an angle sequence, and a torque sequence. The preset time period is the period when the detected current value of the disconnector switch is greater than or equal to a preset threshold. The current sequence includes the current value at each moment within the preset time period. The angle sequence includes the angle value at each moment within the preset time period, representing the closing angle of the disconnector switch. The torque sequence includes the torque value at each moment within the preset time period, representing the rotational torque generated by the drive shaft of the disconnector switch during the opening and closing process.

[0008] The current sequence, angle sequence, and torque sequence from the switch parameter set are input into a preset fault identification model for fault identification processing to obtain the fault detection results of the disconnecting switch.

[0009] In one possible implementation, the current sequence, angle sequence, and torque sequence are input into a preset fault identification model for fault identification processing to obtain the fault detection results of the disconnecting switch, including:

[0010] Based on the preset fault identification model, feature extraction processing is performed on the current sequence, angle sequence, and torque sequence to obtain the current feature sequence, angle feature sequence, and torque feature sequence respectively;

[0011] Based on a preset fault identification model, the switch parameter set, current feature sequence, angle feature sequence, and torque feature sequence are processed to obtain the fault detection results of the disconnecting switch.

[0012] In one possible implementation, feature extraction processing is performed on the current sequence, angle sequence, and torque sequence based on a preset fault identification model to obtain current feature sequence, angle feature sequence, and torque feature sequence, including;

[0013] Based on the preset fault identification model, the current sequence, angle sequence, and torque sequence are processed by first-order difference to obtain first-order differential current sequence, first-order differential angle sequence, and first-order differential torque sequence.

[0014] Based on a pre-defined fault identification model, feature extraction processing is performed on the current sequence and the first-order differential current sequence to obtain the current feature sequence.

[0015] Based on the preset fault identification model, feature extraction processing is performed on the angle sequence and the first-order difference angle sequence to obtain the angle feature sequence;

[0016] Based on a preset fault identification model, feature extraction processing is performed on the torque sequence and the first-order differential torque sequence to obtain the torque feature sequence.

[0017] In one possible implementation, feature extraction processing is performed on the current sequence and the first-order differential current sequence based on a preset fault identification model to obtain a current feature sequence, including:

[0018] Based on the preset fault identification model, time-domain feature extraction processing is performed on the current sequence and the first-order differential current sequence to obtain the time-domain current sequence and the time-domain first-order differential current sequence.

[0019] Based on the preset fault identification model, frequency domain feature extraction processing is performed on the current sequence and the first-order differential current sequence to obtain the frequency domain current sequence and the frequency domain first-order differential current sequence.

[0020] Based on the time-domain current sequence, the first-order time-domain differential current sequence, the frequency-domain current sequence, and the first-order frequency-domain differential current sequence, the current characteristic sequence is obtained.

[0021] In one possible implementation, feature extraction processing is performed on the angle sequence and the first-order difference angle sequence based on a preset fault identification model to obtain an angle feature sequence, including:

[0022] Based on the preset fault identification model, time-domain feature extraction processing is performed on the angle sequence and the first-order difference angle sequence to obtain the time-domain angle sequence and the time-domain first-order difference angle sequence.

[0023] Based on the preset fault identification model, frequency domain feature extraction processing is performed on the angle sequence and the first-order difference angle sequence to obtain the frequency domain angle sequence and the frequency domain first-order difference angle sequence.

[0024] Angle feature sequences are obtained from the time-domain angle sequence, the time-domain first-order difference angle sequence, the frequency-domain angle sequence, and the frequency-domain first-order difference angle sequence.

[0025] In one possible implementation, feature extraction processing is performed on the torque sequence and the first-order differential torque sequence based on a preset fault identification model to obtain a torque feature sequence, including:

[0026] Based on the preset fault identification model, time-domain feature extraction processing is performed on the torque sequence and the first-order differential torque sequence to obtain the time-domain torque sequence and the time-domain first-order differential torque sequence.

[0027] Based on the preset fault identification model, frequency domain feature extraction processing is performed on the torque sequence and the first-order differential torque sequence to obtain the frequency domain torque sequence and the frequency domain first-order differential torque sequence.

[0028] The torque characteristic sequence is obtained from the time-domain torque sequence, the time-domain first-order difference torque sequence, the frequency-domain torque sequence, and the frequency-domain first-order difference torque sequence.

[0029] In one possible implementation, the set of switch parameters, current characteristic sequence, angle characteristic sequence, and torque characteristic sequence are processed based on a preset fault identification model to obtain the fault detection result of the disconnecting switch, including:

[0030] Based on a preset fault identification model, correlation coefficients are processed on the current sequence, angle sequence, and torque sequence to obtain a first correlation sequence, a second correlation sequence, and a third correlation sequence. The first correlation sequence includes the first correlation degree at each moment within a preset time period, representing the correlation between the current value and the angle value. The second correlation sequence includes the second correlation degree at each moment within the preset time period, representing the correlation between the current value and the torque value. The third correlation sequence includes the third correlation degree at each moment within the preset time period, representing the correlation between the angle value and the torque value.

[0031] Based on the preset fault identification model, the first correlation sequence, the second correlation sequence, the third correlation sequence, the current characteristic sequence, the angle characteristic sequence, and the torque characteristic sequence are processed to obtain the fault detection results of the disconnecting switch.

[0032] Secondly, embodiments of this application provide a model training method for detecting faults in disconnecting switches of power equipment, the method comprising:

[0033] The system acquires a set of current sequences, an angle sequence set, and a torque sequence set for the disconnecting switch during a preset time period. The preset time period is the period when the real-time detected current value of the disconnecting switch is greater than or equal to a preset threshold. The current sequence set includes at least one current sequence, which includes the current value at each moment during the preset time period. The angle sequence set includes at least one angle sequence, which includes the angle value at each moment during the preset time period, and the angle value represents the closing angle of the disconnecting switch. The torque sequence set includes at least one torque sequence, which includes the torque value at each moment during the preset time period, and the torque value represents the rotational torque generated by the drive shaft of the disconnecting switch during the opening and closing process.

[0034] The initial model is trained based on the current sequence set, angle sequence set, and torque sequence set to obtain the preset fault identification model;

[0035] The preset fault identification model is the preset fault identification model in the first aspect and / or in various possible implementations of the first aspect.

[0036] In one possible implementation, an initial model is trained based on a set of current sequences, a set of angle sequences, and a set of torque sequences to obtain a preset fault identification model, including:

[0037] Determine the first sequence median of the current sequence set, where the first sequence median represents the number of sequences in the current sequence set; determine the second sequence median of the angle sequence set, where the second sequence median represents the number of sequences in the angle sequence set; and determine the third sequence median of the torque sequence set, where the third sequence median represents the number of sequences in the torque sequence set.

[0038] Based on the median of the first sequence, the current sequences in the current sequence set are processed to unify the number of current values ​​in each current sequence, thus obtaining an initial current sequence set; and based on the median of the second sequence, the angle sequences in the angle sequence set are processed to unify the number of angle values ​​in each angle sequence, thus obtaining an initial angle sequence set; and based on the median of the third sequence, the torque sequences in the torque sequence set are processed to unify the number of torque values ​​in each torque sequence, thus obtaining an initial torque sequence set.

[0039] The initial model is trained based on the initial current sequence set, initial angle sequence set, and initial torque sequence set to obtain the preset fault identification model.

[0040] In one possible implementation, an initial model is trained based on an initial set of current sequences, an initial set of angle sequences, and an initial set of torque sequences to obtain a preset fault identification model, including:

[0041] Repeat the following steps until the preset conditions are met:

[0042] Based on the preset i-th parameter, the i-th initial model is updated to obtain the (i+1)-th initial model;

[0043] Based on the initial current sequence set, initial angle sequence set, and initial torque sequence set, the (i+1)th initial model is processed to obtain the (i+1)th evaluation index; where i is a positive integer greater than or equal to 1; and the value of i is determined by incrementing by 1.

[0044] Among them, the preset conditions are met, and the obtained evaluation indicators are used to obtain the preset fault identification model.

[0045] Thirdly, embodiments of this application provide a fault detection device for a disconnecting switch in a power equipment, comprising:

[0046] The acquisition module is used to detect the current value of the disconnecting switch in the power equipment and acquire the set of switching parameters of the disconnecting switch under a preset time period. The set of switching parameters includes a current sequence, an angle sequence, and a torque sequence. The preset time period is the time period corresponding to when the current value of the disconnecting switch detected in real time is greater than or equal to a preset threshold. The current sequence includes the current value at each moment under the preset time period. The angle sequence includes the angle value at each moment under the preset time period, and the angle value represents the closing angle of the disconnecting switch. The torque sequence includes the torque value at each moment under the preset time period, and the torque value represents the rotational torque value generated by the drive shaft of the disconnecting switch during the opening and closing process of the disconnecting switch.

[0047] The processing module is used to input the current sequence, angle sequence, and torque sequence from the switch parameter set into the preset fault identification model for fault identification processing, and obtain the fault detection results of the disconnecting switch.

[0048] In one possible implementation, the processing module includes:

[0049] The first processing module is used to perform feature extraction processing on the current sequence, angle sequence, and torque sequence based on the preset fault identification model to obtain the current feature sequence, angle feature sequence, and torque feature sequence respectively.

[0050] The second processing module is used to process the set of switch parameters, current feature sequence, angle feature sequence and torque feature sequence based on the preset fault identification model to obtain the fault detection result of the disconnecting switch.

[0051] In one possible implementation, the first processing module includes:

[0052] The third processing module is used to perform first-order differential processing on the current sequence, angle sequence, and torque sequence based on the preset fault identification model to obtain the first-order differential current sequence, the first-order differential angle sequence, and the first-order differential torque sequence.

[0053] The fourth processing module is used to perform feature extraction processing on the current sequence and the first-order differential current sequence based on the preset fault identification model to obtain the current feature sequence.

[0054] The fifth processing module is used to perform feature extraction processing on the angle sequence and the first-order difference angle sequence based on the preset fault identification model to obtain the angle feature sequence;

[0055] The sixth processing module is used to perform feature extraction processing on the torque sequence and the first-order differential torque sequence based on the preset fault identification model to obtain the torque feature sequence.

[0056] In one possible implementation, the fourth processing module includes:

[0057] Based on the preset fault identification model, time-domain feature extraction processing is performed on the current sequence and the first-order differential current sequence to obtain the time-domain current sequence and the time-domain first-order differential current sequence.

[0058] Based on the preset fault identification model, frequency domain feature extraction processing is performed on the current sequence and the first-order differential current sequence to obtain the frequency domain current sequence and the frequency domain first-order differential current sequence.

[0059] Based on the time-domain current sequence, the first-order time-domain differential current sequence, the frequency-domain current sequence, and the first-order frequency-domain differential current sequence, the current characteristic sequence is obtained.

[0060] In one possible implementation, the fifth processing module includes:

[0061] Based on the preset fault identification model, time-domain feature extraction processing is performed on the angle sequence and the first-order difference angle sequence to obtain the time-domain angle sequence and the time-domain first-order difference angle sequence.

[0062] Based on the preset fault identification model, frequency domain feature extraction processing is performed on the angle sequence and the first-order difference angle sequence to obtain the frequency domain angle sequence and the frequency domain first-order difference angle sequence.

[0063] Angle feature sequences are obtained from the time-domain angle sequence, the time-domain first-order difference angle sequence, the frequency-domain angle sequence, and the frequency-domain first-order difference angle sequence.

[0064] In one possible implementation, the sixth processing module includes:

[0065] Based on the preset fault identification model, time-domain feature extraction processing is performed on the torque sequence and the first-order differential torque sequence to obtain the time-domain torque sequence and the time-domain first-order differential torque sequence.

[0066] Based on the preset fault identification model, frequency domain feature extraction processing is performed on the torque sequence and the first-order differential torque sequence to obtain the frequency domain torque sequence and the frequency domain first-order differential torque sequence.

[0067] The torque characteristic sequence is obtained from the time-domain torque sequence, the time-domain first-order difference torque sequence, the frequency-domain torque sequence, and the frequency-domain first-order difference torque sequence.

[0068] In one possible implementation, the second processing module includes:

[0069] Based on a preset fault identification model, correlation coefficients are processed on the current sequence, angle sequence, and torque sequence to obtain a first correlation sequence, a second correlation sequence, and a third correlation sequence. The first correlation sequence includes the first correlation degree at each moment within a preset time period, representing the correlation between the current value and the angle value. The second correlation sequence includes the second correlation degree at each moment within the preset time period, representing the correlation between the current value and the torque value. The third correlation sequence includes the third correlation degree at each moment within the preset time period, representing the correlation between the angle value and the torque value.

[0070] Based on the preset fault identification model, the first correlation sequence, the second correlation sequence, the third correlation sequence, the current characteristic sequence, the angle characteristic sequence, and the torque characteristic sequence are processed to obtain the fault detection results of the disconnecting switch.

[0071] Fourthly, embodiments of this application provide a model training method for detecting faults in disconnecting switches of power equipment, the method comprising:

[0072] The acquisition module is used to acquire a set of current sequences, an angle sequence set, and a torque sequence set of disconnecting switches under a preset time period. The preset time period is the period when the real-time detected current value of the disconnecting switch is greater than or equal to a preset threshold. The current sequence set includes at least one current sequence, which includes the current value at each moment under the preset time period. The angle sequence set includes at least one angle sequence, which includes the angle value at each moment under the preset time period, and the angle value represents the closing angle of the disconnecting switch. The torque sequence set includes at least one torque sequence, which includes the torque value at each moment under the preset time period, and the torque value represents the rotational torque generated by the drive shaft of the disconnecting switch during the opening and closing process.

[0073] The training module is used to train the initial model based on the current sequence set, angle sequence set, and torque sequence set to obtain the preset fault identification model;

[0074] The preset fault identification model is the preset fault identification model in the first aspect and / or in various possible implementations of the first aspect.

[0075] In one possible implementation, the training module includes:

[0076] The confirmation module is used to determine the first sequence median value of the current sequence set, wherein the first sequence median represents the number of sequences in the current sequence set; and to determine the second sequence median value of the angle sequence set, wherein the second sequence median represents the number of sequences in the angle sequence set; and to determine the third sequence median value of the torque sequence set, wherein the third sequence median represents the number of sequences in the torque sequence set.

[0077] The processing module is used to process the current sequences in the current sequence set according to the median of the first sequence to unify the number of current values ​​in each current sequence in the current sequence set, thereby obtaining an initial current sequence set; and to process the angle sequences in the angle sequence set according to the median of the second sequence to unify the number of angle values ​​in each angle sequence in the angle sequence set, thereby obtaining an initial angle sequence set; and to process the torque sequences in the torque sequence set according to the median of the third sequence to unify the number of torque values ​​in each torque sequence in the torque sequence set, thereby obtaining an initial torque sequence set.

[0078] The training submodule is used to train the initial model based on the initial current sequence set, initial angle sequence set, and initial torque sequence set to obtain the preset fault identification model.

[0079] In one possible implementation, the training submodule includes:

[0080] Repeat the following steps until the preset conditions are met:

[0081] Based on the preset i-th parameter, the i-th initial model is updated to obtain the (i+1)-th initial model;

[0082] Based on the initial current sequence set, initial angle sequence set, and initial torque sequence set, the (i+1)th initial model is processed to obtain the (i+1)th evaluation index; where i is a positive integer greater than or equal to 1; and the value of i is determined by incrementing by 1.

[0083] Among them, the preset conditions are met, and the obtained evaluation indicators are used to obtain the preset fault identification model.

[0084] Fifthly, embodiments of this application provide an electronic device, including: a memory and a processor;

[0085] The memory stores instructions that the computer executes;

[0086] The processor executes computer execution instructions stored in memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0087] In a sixth aspect, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect.

[0088] In a seventh aspect, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0089] This application provides a fault detection method and model training method for disconnecting switches in power equipment. By real-time monitoring of the disconnecting switch current value and dynamically triggering multi-dimensional parameter acquisition, combined with intelligent model correlation analysis, it achieves accurate real-time detection and low false alarm rate diagnosis of transient faults such as mechanical jamming. Specifically, the method first detects the real-time current value of the disconnecting switch. When the current value reaches or exceeds a preset threshold (indicating that the equipment may be in an abnormal operating state), data acquisition for a preset time period is immediately initiated, synchronously recording the current value, closing angle value, and transmission shaft torque value at each moment within that time period, forming a set of switch parameters including current sequence, angle sequence, and torque sequence. Among them, the current sequence reflects the energy change characteristics during equipment operation, the angle sequence directly represents the mechanical action state of the disconnecting switch opening and closing, and the torque sequence reflects the load characteristics of the transmission mechanism. The three together constitute a multi-dimensional feature set for fault diagnosis. Subsequently, the above parameter set is input into a preset fault identification model. Through the model's learning and identification of the correlation patterns of multi-parameter time-series data, it can accurately capture fault feature combinations such as current mutations, angle deviations, and torque anomalies, avoiding misjudgments caused by environmental noise interference from a single parameter. This method ensures real-time data capture when a fault occurs through a dynamic threshold triggering mechanism, while using multi-parameter correlation analysis to improve feature reliability. It effectively solves the problems of lag in manual inspection and false alarms in fixed threshold alarms, and significantly improves the accuracy and response speed of disconnector switch fault detection. Attached Figure Description

[0090] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0091] Figure 1 A flowchart illustrating a fault detection method for disconnecting switches in power equipment provided in this application embodiment. Figure 1 ;

[0092] Figure 2 A flowchart illustrating a fault detection method for disconnecting switches in power equipment provided in this application embodiment. Figure 2 ;

[0093] Figure 3 A flowchart illustrating step S202 of a fault detection method for a disconnecting switch in power equipment provided in an embodiment of this application;

[0094] Figure 4 A flowchart illustrating a model training method for detecting faults in disconnecting switches of power equipment, provided in this application embodiment. Figure 1 ;

[0095] Figure 5 A flowchart illustrating a model training method for detecting faults in disconnecting switches of power equipment, provided in this application embodiment. Figure 2 ;

[0096] Figure 6a A graph showing the current values ​​of the disconnecting switch in different states as provided in the embodiments of this application;

[0097] Figure 6b A graph showing the angle values ​​of the disconnecting switch in different states as provided in the embodiments of this application;

[0098] Figure 6c A graph showing the torque values ​​of the disconnecting switch in different states as provided in the embodiments of this application;

[0099] Figure 7a A schematic diagram of the confusion matrix provided in the embodiments of this application. Figure 1 ;

[0100] Figure 7b A schematic diagram of the confusion matrix provided in the embodiments of this application. Figure 2 ;

[0101] Figure 8 A schematic diagram of the structure of a fault detection device for a disconnecting switch in a power equipment provided in an embodiment of this application;

[0102] Figure 9 A schematic diagram of the structure of a model training device for detecting faults in disconnecting switches of power equipment, provided in an embodiment of this application;

[0103] Figure 10 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0104] The accompanying drawings have illustrated specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to specific embodiments. Detailed Implementation

[0105] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0106] In power systems, disconnect switches are core devices for controlling circuit connection and disconnection, and their operational reliability directly affects the safety and stability of the power grid. These devices perform opening and closing operations through mechanical transmission mechanisms. During long-term operation, mechanical wear, lubrication failure, or the intrusion of foreign objects can easily lead to mechanical jamming faults. Initially, these faults manifest as minor abnormalities such as increased operating resistance and delayed action. If not addressed promptly, they can gradually escalate into serious consequences such as incomplete opening or closing, contact erosion, or even mechanism breakage, ultimately resulting in equipment damage or cascading faults, causing regional power outages.

[0107] Current fault detection methods used in the industry have significant limitations: traditional manual inspection relies on the experience and judgment of maintenance personnel, making it difficult to capture transient mechanical anomalies in real time and leaving blind spots in the inspection process; while alarm systems based on fixed thresholds can achieve automated monitoring, their criteria are too simplistic and easily affected by ambient temperature fluctuations, electromagnetic interference, and other noise factors, resulting in a high false alarm rate. Especially under complex operating conditions, the spectral characteristics of mechanical vibration signals and noise signals highly overlap, making it difficult for traditional detection methods to effectively distinguish fault characteristics from interference factors.

[0108] Therefore, this application provides a fault detection method for disconnecting switches in power equipment, which can solve the above-mentioned problems.

[0109] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0110] Figure 1 A flowchart illustrating a fault detection method for disconnecting switches in power equipment provided in this application embodiment. Figure 1 ,like Figure 1 As shown, the method includes:

[0111] S101. Detect the current value of the disconnecting switch in the power equipment and obtain the set of switching parameters of the disconnecting switch under a preset time period; wherein, the set of switching parameters includes a current sequence, an angle sequence, and a torque sequence; wherein, the preset time period is the time period corresponding to when the current value of the disconnecting switch detected in real time is greater than or equal to a preset threshold; the current sequence includes the current value at each moment under the preset time period; the angle sequence includes the angle value at each moment under the preset time period, the angle value representing the closing angle of the disconnecting switch; the torque sequence includes the torque value at each moment under the preset time period, the torque value representing the rotational torque value generated by the drive shaft of the disconnecting switch during the opening and closing process of the disconnecting switch.

[0112] For example, current sensors deployed in power equipment continuously monitor the current value of the line where the disconnecting switch is located. The current sensor converts the current signal into a measurable electrical signal and transmits it to a data acquisition system. The data acquisition system compares the received current value with a preset current threshold. When the real-time detected current value is greater than or equal to the preset threshold, the subsequent data recording process is triggered.

[0113] For example, the time period from when the current value reaches or exceeds a preset threshold to when the current value falls below the preset threshold again is defined as a preset time period. Within this preset time period, multiple parameters of the disconnector are simultaneously acquired: current sequence, angle sequence, and torque sequence. The current sequence is formed by continuously recording the current value at each moment using a current sensor. The angle sequence is formed by measuring the closing angle of the disconnector using an angle sensor (such as an encoder) and recording the angle value at each moment. The torque sequence is formed by monitoring the rotational torque generated by the disconnector's drive shaft during opening and closing using a torque sensor and recording the torque value at each moment.

[0114] In one possible embodiment, a preset threshold of 500A is set. When the current value is ≥500A, data recording is triggered. When the current value reaches 500A, the data acquisition system begins recording a timestamp and continuously acquires current, angle, and torque values.

[0115] For example, the following data was recorded during a certain period:

[0116] Current sequence: [500A, 520A, 510A, 490A, 480A]

[0117] Angle sequence: [30°,35°,40°,45°,50°]

[0118] Torque sequence: [10 N·m, 12 N·m, 11 N·m, 9 N·m, 8 N·m]

[0119] S102. Input the current sequence, angle sequence, and torque sequence from the switch parameter set into the preset fault identification model for fault identification processing to obtain the fault detection result of the disconnecting switch.

[0120] For example, the input current, angle, and torque sequences are standardized or normalized to eliminate the influence of differences in the dimensions and numerical ranges of different parameters, making the data more suitable for the model input requirements. Data integrity is checked, and missing or outlier values ​​are handled, for example, by using interpolation to fill in missing data and removing outlier data points that significantly deviate from the normal range.

[0121] The preprocessed current, angle, and torque sequences are integrated according to a preset format, typically forming a multidimensional array or data frame. Each row represents data at a given moment, and each column corresponds to a parameter (current, angle, torque). It is crucial to ensure that the dimensions and order of the input data match the requirements of the preset fault identification model. For example, the model might require the input data to have the shape of (number of samples, time step, number of parameters).

[0122] The prepared input data is fed into a pre-defined fault identification model. The model calculates and analyzes the input data based on the features and patterns it has learned internally. The model uses multi-layer neural networks or machine learning models to progressively extract high-level features from the data and classifies faults based on these features.

[0123] The model outputs fault detection results, typically in the form of probability values ​​or category labels. For example, it outputs a probability value between 0 and 1, indicating the likelihood of a fault in the disconnector switch; or it outputs a specific category label, such as "normal," "minor fault," or "serious fault." The output results are then parsed and interpreted. Based on preset thresholds or rules, the probability values ​​are converted into specific fault judgment conclusions, or the fault type and severity are determined based on the category label.

[0124] This application provides a fault detection method for disconnecting switches in power equipment. By real-time monitoring of the disconnecting switch current value and dynamically triggering multi-dimensional parameter acquisition, combined with intelligent model correlation analysis, it achieves accurate real-time detection and low false alarm rate diagnosis of transient faults such as mechanical jamming. Specifically, the method first detects the real-time current value of the disconnecting switch. When the current value reaches or exceeds a preset threshold (indicating that the equipment may be in an abnormal operating state), data acquisition for a preset time period is immediately initiated. The current value, closing angle value, and transmission shaft torque value at each moment within this time period are recorded simultaneously, forming a set of switch parameters including current sequence, angle sequence, and torque sequence. Among them, the current sequence reflects the energy change characteristics during equipment operation, the angle sequence directly represents the mechanical action state of the disconnecting switch opening and closing, and the torque sequence reflects the load characteristics of the transmission mechanism. The three together constitute a multi-dimensional feature set for fault diagnosis. Subsequently, the above parameter set is input into a preset fault identification model. Through the model's learning and identification of the correlation patterns of multi-parameter time-series data, it can accurately capture fault feature combinations such as current mutations, angle deviations, and torque anomalies, avoiding misjudgments caused by environmental noise interference from a single parameter. This method ensures real-time data capture when a fault occurs through a dynamic threshold triggering mechanism, while using multi-parameter correlation analysis to improve feature reliability. It effectively solves the problems of lag in manual inspection and false alarms in fixed threshold alarms, and significantly improves the accuracy and response speed of disconnector switch fault detection.

[0125] Figure 2 A flowchart illustrating a fault detection method for disconnecting switches in power equipment provided in this application embodiment. Figure 2 ,like Figure 2 As shown, in this embodiment... Figure 1 Based on the embodiments, a fault detection method for disconnecting switches in power equipment is described in detail. The method includes:

[0126] S201. Detect the current value of the disconnecting switch in the power equipment and obtain the set of switching parameters of the disconnecting switch under a preset time period; wherein, the set of switching parameters includes current sequence, angle sequence and torque sequence; wherein, the preset time period is the time period corresponding to when the current value of the disconnecting switch detected in real time is greater than or equal to a preset threshold.

[0127] For example, this step can refer to step S101 above, and will not be repeated here.

[0128] S202. Based on the preset fault identification model, feature extraction processing is performed on the current sequence, angle sequence, and torque sequence to obtain the current feature sequence, angle feature sequence, and torque feature sequence.

[0129] For example, data preprocessing is performed on current sequences, angle sequences, and torque sequences, including noise removal and handling missing values. Noise may originate from sensor errors, external environmental interference, or other factors; noise removal can improve data quality and accuracy. Missing values ​​may be caused by sensor malfunctions, data transmission interruptions, or other reasons; missing values ​​can be handled using methods such as interpolation or mean imputation.

[0130] Load a pre-trained fault identification model to extract fault-related features from the input sequence data. This model can be built based on machine learning algorithms, such as Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF), etc., or on deep learning algorithms, such as Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), etc. Different models have different characteristics and applicable scenarios. For example, CNN is good at processing data with local features, while RNN and its variants are suitable for processing sequence data.

[0131] Figure 3 This is a flowchart illustrating step S202 of a fault detection method for a disconnecting switch in power equipment provided in an embodiment of this application. Figure 3 As shown, step S202 includes:

[0132] S2021. Based on the preset fault identification model, the current sequence, angle sequence, and torque sequence are processed by first-order differential to obtain first-order differential current sequence, first-order differential angle sequence, and first-order differential torque sequence.

[0133] For example, the core idea of ​​first-order difference processing is to subtract adjacent data in the sequence. For a given sequence x = [x1, x2, x3, ..., x...] i The first-order difference sequence y is calculated as follows: y i =x i+1 -x i Where i = 1, 2, ..., n-1. This operation allows us to obtain the changes between adjacent data points in the original sequence, thus highlighting the trend of data change.

[0134] For a current sequence, following the first-order difference calculation method described above, starting from the second data point in the sequence, subtract the previous data point from each subsequent data point to obtain the first-order difference current sequence. For example, if the current sequence is [I1, I2, I3, ..., I...] t Then the first-order differential current sequence is [I2-I1,I3-I2,…,I…]. t -I t-1 ].

[0135] Similarly, the same operation is performed on the angle sequence and torque sequence. Angle sequence [A1, A2, A3, ..., A...] t After first-order difference processing, we get [A2-A1, A3-A2, ..., A t -A t-1 Torque sequence [T1,T2,T3,…,T] t After processing, we get [T2-T1,T3-T2,…,T] t -T t-1 ].

[0136] The obtained first-order differential current sequence, first-order differential angle sequence, and first-order differential torque sequence are organized and stored. These sequences can be saved to a database for use in subsequent fault identification models, or they can be directly passed to subsequent processing modules for further analysis and processing.

[0137] S2022. Based on the preset fault identification model, feature extraction processing is performed on the current sequence and the first-order differential current sequence to obtain the current feature sequence.

[0138] For example, for current sequences and first-order differential current sequences, time-domain characteristics are calculated, such as mean, variance, standard deviation, maximum value, minimum value, peak-to-peak value (the difference between the maximum and minimum values), skewness (measures the asymmetry of the data distribution), and kurtosis (measures the sharpness of the data distribution). These characteristics can reflect the central tendency, dispersion, and distribution pattern of the sequence from different perspectives.

[0139] A Fast Fourier Transform (FFT) is performed on the current sequence and the first-order differential current sequence to convert the time-domain signal into a frequency-domain signal. The FFT reveals the energy distribution of the signal at different frequencies. Features are extracted from the frequency-domain signal, such as the dominant frequency components (frequency with the highest energy), the frequency centroid (center of the frequency energy distribution), and the band energy (sum of energy across different frequency bands). These frequency-domain features reflect the frequency characteristics of the signal and are crucial for identifying frequency-related faults.

[0140] The features extracted from the time and frequency domains are integrated to form a comprehensive feature sequence. This feature sequence contains information from multiple aspects of the current sequence and the first-order differential current sequence, which can more comprehensively describe the characteristics of the sequence and provide richer input data for subsequent fault identification models.

[0141] In one example, based on a preset fault identification model, time-domain features are extracted from the current sequence and the first-order differential current sequence to obtain a time-domain current sequence and a time-domain first-order differential current sequence. Based on the preset fault identification model, frequency-domain features are extracted from the current sequence and the first-order differential current sequence to obtain a frequency-domain current sequence and a frequency-domain first-order differential current sequence. Based on the time-domain current sequence, the time-domain first-order differential current sequence, the frequency-domain current sequence, and the frequency-domain first-order differential current sequence, a current feature sequence is obtained.

[0142] For example, the collected current sequence and the first-order differential current sequence obtained through first-order difference operation are used as input data. These two sequences reflect the changes in current from different perspectives: the current sequence reflects the overall change of current over time, while the first-order differential current sequence highlights the rate of current change. A preset time-domain feature extraction algorithm is used to process the current sequence and the first-order differential current sequence. Common time-domain features include mean, variance, standard deviation, maximum value, minimum value, peak-to-peak value, skewness, and kurtosis. These features can describe the characteristics and distribution of the sequence from different dimensions. For example, the mean reflects the average level of the sequence, while the variance and standard deviation reflect the dispersion of the sequence.

[0143] In one possible implementation, the current sequence is I. i =[I1,I2,I3,…,I t ]; where i is a positive integer greater than or equal to 1 and less than or equal to t; the first-order differential current sequence is i i =[i1,i2,…,i t-1 ] = [I2-I1,I3-I2,…,I t -I t-1 ]; where i is a positive integer greater than or equal to 1 and less than or equal to t-1.

[0144] The current series and the first-order differential current series are processed by mean processing, maximum value processing, minimum value processing, standard deviation processing, skewness processing, kurtosis processing, interquartile range processing, median absolute deviation processing, area under the curve processing, and area under the square of the curve processing, respectively.

[0145] The formula for calculating the mean is:

[0146] The formula for calculating the maximum value is: max = max i xi

[0147] The formula for calculating the minimum value is: min = min i x i

[0148] The formula for calculating standard deviation is:

[0149] The formula for calculating skewness is:

[0150] The formula for calculating kurtosis is:

[0151] The formula for calculating the interquartile range (IQR) is: IQR = Q3 - Q1

[0152] The formula for calculating the median absolute deviation is: MAD = median(|x i -median(x)|)

[0153] The formula for calculating the area under the curve is:

[0154] The formula for calculating the area under the square of a curve is:

[0155] For example, an FFT is performed on a current sequence and a first-order differential current sequence to convert the time-domain signal into a frequency-domain signal. The FFT decomposes the signal into a superposition of sine and cosine components at different frequencies, thus revealing the energy distribution of the signal in the frequency domain. Features are extracted from the converted frequency-domain signal, such as the dominant frequency components (the frequencies with the highest energy), the frequency centroid (the central location of the frequency energy distribution), and the band energy (the sum of energy within different frequency bands). These frequency-domain features reflect the frequency characteristics of the signal and are of great significance for identifying frequency-related faults. For example, certain faults may cause an increase or decrease in the energy of specific frequency components.

[0156] In one possible implementation, the current sequence is I. i =[I1,I2,I3,…,I t ]; where i is a positive integer greater than or equal to 1 and less than or equal to t; the first-order differential current sequence is i i =[i1,i2,…,i t-1 ] = [I2-I1,I3-I2,…,I t -I t-1 ]; where i is a positive integer greater than or equal to 1 and less than or equal to t-1.

[0157] The current sequence and the first-order differential current sequence are processed by FFT to obtain frequency domain statistics, weighted frequency mean, the first five DFT coefficients, the first five maximum spectral peaks and their corresponding frequencies.

[0158] The input to the FFT is x n The sampling interval is T e The FFT formula is:

[0159]

[0160] Optionally, for X k Mean processing, variance processing, and kurtosis processing are performed separately, and the formulas are the same as those for time-domain processing, so they will not be repeated in this application.

[0161] The formula for calculating the weighted frequency mean is:

[0162] The first five FFT coefficients are: X1, X2, X3, X4, X5

[0163] S2023. Based on the preset fault identification model, feature extraction processing is performed on the angle sequence and the first-order difference angle sequence to obtain the angle feature sequence.

[0164] For example, the angle sequence records the angle changes of an object at different times, while the first-order differential torque sequence reflects the rate of change of torque between adjacent times.

[0165] For example, check the data for outliers, missing values, or noise. Outliers can be identified and processed using analytical methods (such as standard deviation-based methods), for example, replacing values ​​exceeding a certain standard deviation range with reasonable boundary values ​​or using interpolation methods to fill them in. Missing values ​​can be filled using mean interpolation, median interpolation, or model-based interpolation methods. Noise can be smoothed using filtering algorithms (such as moving average filtering, Kalman filtering, etc.).

[0166] The characteristics of an angle sequence, such as mean, variance, standard deviation, maximum value, minimum value, peak-to-peak value, skewness, and kurtosis, are calculated. These characteristics reflect the variation and distribution of angles over time. For example, the mean can represent the average position of the angle, while the variance and standard deviation can reflect the degree of fluctuation of the angle.

[0167] An FFT is performed on the angle sequence to transform it from the time domain to the frequency domain. Frequency domain features are then extracted, such as the dominant frequency components (frequency with the highest energy), the frequency centroid (the center of the frequency energy distribution), and the band energy (the sum of energy across different frequency bands). These frequency domain features can reveal the periodicity and frequency characteristics of the angle changes.

[0168] Similarly, the mean, variance, and standard deviation of the first-order difference torque sequence are calculated. These characteristics can reflect the properties of the torque change rate. For example, a larger variance indicates that the torque change rate fluctuates more.

[0169] In one example, based on a preset fault identification model, time-domain feature extraction is performed on the angle sequence and the first-order differential angle sequence to obtain a time-domain angle sequence and a time-domain first-order differential angle sequence; based on the preset fault identification model, frequency-domain feature extraction is performed on the angle sequence and the first-order differential angle sequence to obtain a frequency-domain angle sequence and a frequency-domain first-order differential angle sequence; based on the time-domain angle sequence, the time-domain first-order differential angle sequence, the frequency-domain angle sequence, and the frequency-domain first-order differential angle sequence, an angle feature sequence is obtained.

[0170] For example, based on a pre-defined fault identification model, time-domain features are extracted from the angle sequence. Common time-domain features include mean, variance, standard deviation, maximum value, minimum value, peak-to-peak value, skewness, and kurtosis. These features can describe the characteristics and distribution patterns of angles in the time domain. For instance, the mean reflects the average position of the angle, while the variance and standard deviation reflect the degree of fluctuation of the angle.

[0171] For example, similar time-domain features are extracted from the first-order differential angle sequence based on the same preset fault identification model. These features can reflect the characteristics of the angle change rate, which helps to describe the dynamic changes of the angle more comprehensively.

[0172] For example, FFT is used to transform the angle sequence from the time domain to the frequency domain. In the frequency domain, features such as the main frequency components (frequency with the highest energy), the frequency centroid (center of the frequency energy distribution), and the frequency band energy (sum of energy in different frequency bands) can be extracted. Frequency domain features can reveal the periodicity and frequency characteristics of angle changes, which is of great significance for identifying frequency-related fault modes.

[0173] For example, an FFT is also performed on the first-order difference angle sequence, and the corresponding frequency domain features are extracted. These features can reflect the frequency characteristics of the angle change rate, further enriching the feature information of the angle.

[0174] It is understandable that the formulas for time-domain feature processing and frequency-domain feature processing of the angle sequence and the first-order difference angle sequence are the same as those in step S2022, and will not be repeated here.

[0175] S2024. Based on the preset fault identification model, feature extraction processing is performed on the torque sequence and the first-order differential torque sequence to obtain the torque feature sequence.

[0176] For example, the torque sequence records the torque values ​​of electrical equipment at various times, serving as the foundational data for subsequent analysis. Based on a pre-defined fault identification model, the torque sequence and the first-order difference torque sequence undergo time-domain feature processing. Common time-domain features include mean, variance, standard deviation, maximum value, minimum value, peak-to-peak value, skewness, and kurtosis. These features describe the characteristics and distribution patterns of torque in the time domain. For instance, the mean reflects the average level of torque, while the variance and standard deviation reflect the degree of torque fluctuation.

[0177] For example, FFT is used to transform the torque sequence and the first-order differential torque sequence from the time domain to the frequency domain. In the frequency domain, features such as the main frequency components (frequency with the highest energy), the frequency centroid (center of the frequency energy distribution), and the band energy (sum of energy in different frequency bands) can be extracted. Frequency domain features can reveal the periodicity and frequency characteristics of torque changes, which is of great significance for identifying frequency-related fault modes.

[0178] In one example, based on a preset fault identification model, time-domain features are extracted from the torque sequence and the first-order differential torque sequence to obtain a time-domain torque sequence and a time-domain first-order differential torque sequence. Based on the preset fault identification model, frequency-domain features are extracted from the torque sequence and the first-order differential torque sequence to obtain a frequency-domain torque sequence and a frequency-domain first-order differential torque sequence. Based on the time-domain torque sequence, the time-domain first-order differential torque sequence, the frequency-domain torque sequence, and the frequency-domain first-order differential torque sequence, a torque feature sequence is obtained.

[0179] For example, based on a preset fault identification model, time-domain features are extracted from the torque sequence. Common time-domain features include mean (reflecting the average level of torque), variance (reflecting the degree of torque fluctuation), standard deviation (measuring the degree of torque dispersion), maximum value, minimum value, peak-to-peak value (the difference between the maximum and minimum values), skewness (describing the asymmetry of torque distribution), and kurtosis (describing the steepness of torque distribution), etc.

[0180] Based on the same pre-defined fault identification model, the same time-domain features are extracted from the first-order differential torque sequence. These features reflect the characteristics of the torque change rate.

[0181] The time-domain features of the extracted torque sequence and the first-order difference torque sequence are combined into sequence form to obtain the time-domain torque sequence and the time-domain first-order difference torque sequence.

[0182] The torque sequence was transformed from the time domain to the frequency domain using FFT. In the frequency domain, features such as the main frequency components (frequency with the highest energy), the frequency centroid (center of the frequency energy distribution), and the band energy (sum of energy across different frequency bands) were extracted. These frequency domain features reveal the periodicity and frequency characteristics of torque variations, which are important for identifying frequency-related fault modes.

[0183] The first-order differential torque sequence was also subjected to FFT, and the same frequency domain features were extracted. These features reflect the characteristics of the torque change rate in the frequency domain.

[0184] It is understandable that the formulas for time-domain feature processing and frequency-domain feature processing of the torque sequence and the first-order differential torque sequence are the same as those in step S2022, and will not be repeated here.

[0185] By employing multi-dimensional feature enhancement and intelligent fusion diagnostic techniques based on first-order differential analysis, a millisecond-level response and high-confidence detection effect are achieved for mechanical jamming faults in disconnecting switches. Specifically, this method first performs first-order differential processing on the acquired current, angle, and torque sequences to generate corresponding first-order differential current, angle, and torque sequences. First-order differential operation, by calculating the changes in data between adjacent time points, effectively filters out low-frequency drift caused by environmental noise while highlighting fault-related transient change characteristics.

[0186] When extracting features from current sequences and first-order differential current sequences, the model can capture both the absolute value change and the rate of change of the current; the processing of angle sequences and differential angle sequences can simultaneously reflect the actual position deviation of the contact and the abnormal operating speed; and the feature extraction of torque sequences and differential torque sequences can identify load anomalies.

[0187] By fusing features from raw and differential data, this method constructs a multi-dimensional feature space encompassing the time domain and the rate of change domain, enabling the fault identification model to distinguish between noise interference and genuine fault modes. Ultimately, this method significantly improves the real-time performance and accuracy of fault detection, effectively addressing the problems of high false alarm rates and weak transient fault detection capabilities in traditional methods, thus providing a reliable guarantee for the safe operation of power equipment.

[0188] S203. Based on the preset fault identification model, the switch parameter set, current characteristic sequence, angle characteristic sequence, and torque characteristic sequence are processed to obtain the fault detection result of the disconnecting switch.

[0189] For example, based on the preset fault identification task and data characteristics, a suitable fault identification model is selected, such as SVM, DT, or neural networks (such as CNN, RNN, LSTM, GRU, etc.). The selected model is initialized by setting its parameters, such as the number of layers, number of neurons, and learning rate of the neural network.

[0190] For example, the integrated feature matrix is ​​input into a pre-initialized fault identification model. The model will perform a series of calculations and transformations based on the input features to extract higher-level feature representations.

[0191] For example, the feature matrix of the disconnector switch to be detected is input into a preset fault identification model. The model will predict the input data based on the learned mapping relationship and output the probability of each sample belonging to different fault types or directly give the fault type label.

[0192] The model's predictions can be post-processed, for example, by setting a probability threshold. A sample is only considered to have a fault if the probability of it belonging to a certain fault type exceeds this threshold. Furthermore, the prediction results can be visualized to provide operators with a more intuitive understanding of the disconnector's fault status.

[0193] Based on the model prediction results and post-processing information, a detailed fault detection report is generated. The report should include basic information about the disconnecting switch, the detection time, the type of fault detected, and the severity of the fault.

[0194] In one example, correlation coefficient processing is performed on the current sequence, angle sequence, and torque sequence based on a preset fault identification model to obtain a first correlation sequence, a second correlation sequence, and a third correlation sequence. The first correlation sequence includes the first correlation degree at each moment within a preset time period, representing the correlation between the current value and the angle value. The second correlation sequence includes the second correlation degree at each moment within the preset time period, representing the correlation between the current value and the torque value. The third correlation sequence includes the third correlation degree at each moment within the preset time period, representing the correlation between the angle value and the torque value. The fault detection result of the disconnecting switch is obtained by processing the first correlation sequence, the second correlation sequence, the third correlation sequence, the current feature sequence, the angle feature sequence, and the torque feature sequence based on the preset fault identification model.

[0195] For example, a first correlation sequence is obtained by processing the correlation coefficient between the current sequence and the angle sequence based on a preset fault identification model. Specifically, for each current value in the current sequence and each angle value in the angle sequence, the correlation between the current value and the angle value is calculated to obtain the first correlation at each moment within a preset time period. The first correlation calculated at each moment is recorded to form the first correlation sequence. The correlation can be measured using methods such as the Pearson correlation coefficient, which reflects the degree of linear correlation between the two variables, current value and angle value.

[0196] Based on a preset fault identification model, the correlation coefficients of the current sequence and the torque sequence are processed to obtain a second correlation sequence. Similarly, at each moment within a preset time period, the correlation between the current value in the current sequence and the torque value in the torque sequence is calculated to obtain a second correlation degree. The second correlation degrees calculated at each moment are recorded to form the second correlation sequence.

[0197] Based on a preset fault identification model, the correlation coefficients of the angle sequence and torque sequence are processed to obtain a third correlation sequence. Similarly, at each moment within a preset time period, the correlation between the angle value in the angle sequence and the torque value in the torque sequence is calculated to obtain a third correlation degree. The third correlation degrees calculated at each moment are recorded to form the third correlation sequence.

[0198] The calculated first, second, and third correlation sequences are integrated with the pre-extracted current, angle, and torque feature sequences. Integration can be achieved by concatenating these sequences into a longer feature vector according to certain rules, or by inputting them as features from different channels into the model. Alternatively, a weighted concatenation method can be used for integration.

[0199] The integrated features are input into a pre-defined fault identification model. This model can be a machine learning model (such as SVM, DT, RF, etc.) or a deep learning model (such as CNN, RNN, LSTM, GRU, etc.). The model performs complex calculations and analyses on the input features to extract deeper feature representations.

[0200] The model outputs fault detection results for disconnecting switches based on the learned relationship between features and faults. The results can be binary (normal or faulty), multi-class (distinguishing between different types of faults), and can also provide information such as the probability of fault occurrence.

[0201] In one possible embodiment, the formula for calculating the first relevance in the first relevant sequence is:

[0202]

[0203] Where, ρ 1t The first relevance; cov(I) t A t () represents the covariance between the current value and the angle value; The standard deviation of the current series; denoted as the standard deviation of the angle sequence.

[0204] The formula for calculating the second relevance in the second relevance sequence is:

[0205]

[0206] Where, ρ 1t The second relevance; cov(I) t ,T t ) represents the covariance between the current value and the torque value; The standard deviation of the current series; denoted as the standard deviation of the torque series.

[0207] The formula for calculating the third relevance in a third-relevance sequence is:

[0208]

[0209] Where, ρ 1t The third relevance; cov(A) t ,T t ) represents the covariance between the angle value and the torque value; The standard deviation of the angle sequence; denoted as the standard deviation of the current series.

[0210] This application provides a fault detection method for disconnecting switches in power equipment. Through a phased feature extraction and multimodal data fusion intelligent diagnostic approach, it achieves high-precision identification and anti-interference detection of mechanical faults in disconnecting switches under complex operating conditions. Specifically, based on real-time monitoring of current values ​​and triggering data acquisition for a preset time period (starting when the current value reaches or exceeds a preset threshold), the method further performs layered processing on the acquired current sequence, angle sequence, and torque sequence: First, the feature extraction module in the preset fault identification model performs time-domain-frequency domain analysis on the three sequences respectively, generating current feature sequences, angle feature sequences, and torque feature sequences; then, the original switch parameter set and the extracted feature sequences are input into the model for correlation analysis, utilizing the complementarity of multi-dimensional data to enhance fault characterization capabilities. The derivation logic of this technique is that single parameters are easily affected by environmental noise, while phased feature extraction can filter noise and highlight the essential characteristics of the fault, while fusing the original data and feature sequences can retain temporal dynamic information. Ultimately, this method effectively solves the problems of high false alarm rate and weak transient fault detection capability in traditional methods, and achieves millisecond-level response and high-confidence diagnosis of faults such as mechanical jamming of disconnecting switches, significantly improving the reliability of power equipment operation and maintenance efficiency.

[0211] Figure 4 A flowchart illustrating a model training method for detecting faults in disconnecting switches of power equipment, provided in this application embodiment. Figure 1 ,like Figure 4 As shown,

[0212] S401. Obtain the current sequence set, angle sequence set, and torque sequence set of the disconnector switch under a preset time period; wherein, the preset time period is the time period corresponding to when the current value of the disconnector switch detected in real time is greater than or equal to a preset threshold; the current sequence set includes at least one current sequence, which includes the current value at each moment under the preset time period; the angle sequence set includes at least one angle sequence, which includes the angle value at each moment under the preset time period, and the angle value represents the closing angle of the disconnector switch; the torque sequence set includes at least one torque sequence, which includes the torque value at each moment under the preset time period, and the torque value represents the rotational torque value generated by the drive shaft of the disconnector switch during the opening and closing process of the disconnector switch.

[0213] For example, a dedicated current detection device (such as a current transformer) is used to monitor the current of the disconnecting switch in real time, continuously acquiring the current data of the disconnecting switch. The real-time detected current value is compared with a preset threshold. When the detected current value is greater than or equal to the preset threshold, time recording begins, and the current value is continuously monitored until the current value is lower than the preset threshold again, at which point time recording stops. This time period from when the current value reaches the threshold to when it falls below the threshold is the preset time period.

[0214] Within a preset time period, current values ​​are sampled and recorded at regular time intervals (e.g., every 0.1 seconds). The current values ​​at all sampling moments within the entire preset time period are arranged in chronological order to form a current sequence. If this sampling and recording process is performed multiple times within the preset time period, multiple current sequences will be obtained, which together constitute a current sequence set.

[0215] Within a preset time period, the closing angle of the disconnecting switch is measured in real time using devices such as angle sensors. The angle values ​​are sampled and recorded at time intervals that are the same as or different from those used for current sampling. The angle values ​​at each sampling moment are arranged in chronological order to form an angle sequence. Multiple sampling records will yield multiple angle sequences, which together constitute an angle sequence set.

[0216] Within a preset time period, the rotational torque generated by the drive shaft of the disconnecting switch during its opening and closing is measured using devices such as torque sensors. Samples are recorded at set time intervals, and the torque values ​​at each sampling moment are arranged chronologically to form a torque sequence. Multiple sampling records will yield multiple torque sequences, which together constitute a torque sequence set.

[0217] S402. Train the initial model based on the current sequence set, angle sequence set, and torque sequence set to obtain the preset fault identification model.

[0218] For example, data preprocessing is performed on the current sequence set, angle sequence set, and torque sequence set; optionally, the data preprocessing can be performed by processing the sequence set using the median of the sequence number, or by performing standardization or other methods.

[0219] Time-domain features, such as mean, variance, maximum, minimum, peak factor, and kurtosis, are extracted from the preprocessed sequence. For example, for a current sequence, the average current value and current fluctuation variance over a preset time period are calculated; these features reflect the overall change in current during that period. Frequency-domain feature processing uses methods such as FFT to convert the time-domain sequence into a frequency-domain sequence, and then extracts frequency-domain features, such as dominant frequency and frequency band energy. For example, for a torque sequence, analyzing its energy distribution at different frequencies may help identify certain specific fault modes.

[0220] Fault identification models include SVM, neural networks (such as CNN, RNN, LSTM, GRU, etc.), DT, RF, etc.

[0221] In one possible embodiment, the present application uses SVM as the initial model.

[0222] The processed sets of current sequence, angle sequence, and torque sequence are input into SVM for further processing to obtain a preset fault identification model.

[0223] This application provides a model training method for detecting faults in disconnecting switches of power equipment. By constructing a fault identification model with high generalization ability and anti-interference capabilities using multi-dimensional time-series data based on real fault scenarios and supervised learning, this method first monitors the current value in real time during the operation of the disconnecting switch. When the detected current value exceeds a preset threshold (indicating that the equipment may be in an abnormal operating state), a data acquisition mechanism for a preset time period is automatically triggered. The current value, closing angle value, and transmission shaft torque value are recorded synchronously at each moment within this time period, forming a set of switch parameters including current, angle, and torque sequences. To ensure the model can learn real fault characteristics, this method specifically collects sequence data including normal and various fault states, constructing a set of current, angle, and torque sequences covering all operating conditions. Subsequently, this multi-dimensional time-series data is input into the initial model for supervised learning training: by labeling the fault type in the data (such as jamming, offset, overload, etc.), the model can learn key features such as the correlation pattern between sudden current changes and abnormal torque, and the mapping relationship between angle trajectory deviation and mechanical action failure.

[0224] Traditional model training often relies on simulated data or single parameters, making it difficult to capture the complex characteristics of real faults. This application's embodiment, however, uses a dynamic threshold triggering mechanism to collect multi-parameter time-series data under actual fault scenarios, ensuring a high correlation between the training data and real faults. Simultaneously, cross-validation of multi-dimensional data effectively improves the model's feature discrimination capability. Ultimately, the preset fault identification model trained by this method possesses stronger environmental noise robustness and fault type identification accuracy, solving the problems of weak generalization ability and high false alarm rate of traditional models, and providing reliable core algorithm support for disconnector switch fault detection.

[0225] Figure 5 A flowchart illustrating a model training method for detecting faults in disconnecting switches of power equipment, provided in this application embodiment. Figure 2 ,like Figure 5 As shown,

[0226] S501. Obtain the current sequence set, angle sequence set, and torque sequence set of the disconnecting switch under a preset time period; wherein, the preset time period is the time period corresponding to when the current value of the disconnecting switch detected in real time is greater than or equal to a preset threshold.

[0227] For example, this step can refer to step 4201 above, and will not be repeated here.

[0228] S502. Determine the first median value of the current sequence set, where the first median value represents the number of sequences in the current sequence set; determine the second median value of the angle sequence set, where the second median value represents the number of sequences in the angle sequence set; determine the third median value of the torque sequence set, where the third median value represents the number of sequences in the torque sequence set; based on the first median value, process the current sequences in the current sequence set to unify the number of current values ​​in each current sequence, obtaining an initial current sequence set; based on the second median value, process the angle sequences in the angle sequence set to unify the number of angle values ​​in each angle sequence, obtaining an initial angle sequence set; based on the third median value, process the torque sequences in the torque sequence set to unify the number of torque values ​​in each torque sequence, obtaining an initial torque sequence set; train the initial model based on the initial current sequence set, initial angle sequence set, and initial torque sequence set to obtain a preset fault identification model.

[0229] For example, the first step involves calculating the median sequence length (i.e., the median of the first / second / third sequence) for each of the three sequence sets: current, angle, and torque. This value reflects the typical length of the data sequence for each physical quantity. The second step uses the median as a benchmark to adjust the length of all sequences in each sequence set: longer sequences are truncated to the median length, and shorter sequences are padded to the median length using linear interpolation or by repeating the last value. The third step inputs the three processed, standardized sequence sets into the initial model and trains the fault identification model using supervised learning.

[0230] For example, consider current sequence processing: Assume the original current sequence set contains 5 sequences with lengths [8, 12, 10, 15, 9], and the median of the first sequence is 10. Then, each sequence is processed as follows: for sequences of length 8, two duplicate values ​​are added to the end (e.g., the last value is copied); for sequences of length 12, the first 10 data points are truncated; and for sequences of length 15, the first 10 data points are truncated (to avoid outliers at the beginning and end). When processing angle sequences, if the median is 7, linear interpolation is used to generate intermediate values ​​for sequences of length 5 (e.g., 5 points interpolated to 7 points), and 7 points are extracted at equal intervals for sequences of length 9. The same principle applies to torque sequence processing; ultimately, the length of all sequences is unified to the corresponding median value.

[0231] For example, in the initial model training step, the processed initial set of current, angle, and torque sequences is used as feature input. Machine learning algorithms, such as SVM, are used to train the initial model. During training, the model learns the differences in current, angle, and torque characteristics between normal and fault states. For instance, when a specific fault occurs in the equipment, the current sequence may exhibit abnormal peaks, the rate of change in the angle sequence may accelerate, and the torque sequence may show irregular fluctuations. Through a large number of training samples, the model gradually develops the ability to recognize these feature differences, ultimately forming a model capable of effectively identifying preset faults. In practical applications, when new equipment operating data is input into the model, it can quickly and accurately determine whether the equipment has a fault and the type of fault based on the learned feature patterns, thereby achieving intelligent monitoring and fault early warning of equipment status.

[0232] S503. Repeat the following steps until the preset conditions are met: Based on the preset i-th parameter, update the i-th initial model to obtain the (i+1)-th initial model; Process the (i+1)-th initial model according to the initial current sequence set, the initial angle sequence set, and the initial torque sequence set to obtain the (i+1)-th evaluation index; where i is a positive integer greater than or equal to 1; and determine the value of i plus 1; where the preset conditions are met, the obtained evaluation indices are used to obtain the preset fault identification model.

[0233] For example, the preset parameters can be C and gamma parameters. The performance of the model is gradually improved by dynamically adjusting the C and gamma parameters of the SVM. The specific process is as follows: ① Based on the preset i-th parameter combination (e.g., C=1, gamma=0.1), the i-th initial model is adjusted and its structure optimized to generate the (i+1)-th improved model; ② Using the initial current, angle, and torque sequence set as input data, forward propagation calculation is performed on the new model to output the fault prediction result; ③ Based on the difference between the true label and the prediction result, the macro-average F1 score of the (i+1)-th iteration is calculated; ④ It is determined whether the preset stopping condition is met (e.g., the macro-average F1 score increases by less than 1% for three consecutive iterations or the maximum number of iterations of 200 is reached). If the condition is met, the iteration is terminated; otherwise, i is incremented by 1 and the above process is repeated.

[0234] For example, if the initial model is an SVM with parameters C=1 and gamma=0.1, in the first iteration, the parameters are adjusted to C=10 and gamma=0.01 to generate the second model. The initial current sequence set (normalized to 100 time steps) is input into this model, mapped to a high-dimensional space using the RBF kernel function, and the fault category probability distribution is output. When calculating the macro-average F1 score, precision, recall, and F1 value are calculated independently for each category, and the arithmetic mean is taken as 0.78. If the score improves by 0.15 compared to the initial model, and i=1 has not reached the maximum number of iterations (200), the parameters are further adjusted. In the second iteration, C=5 and gamma=0.05 are adjusted to generate the third model, and the macro-average F1 score improves to 0.82. If the score improvement is less than 0.05 for two consecutive iterations in subsequent iterations, the stopping condition is triggered, and the current model is used as the preset fault identification model.

[0235] For example, the grid search strategy in the parameter tuning stage is implemented as follows: Candidate values ​​for C are set to [0.1, 1, 10, 100], and candidate values ​​for gamma are [0.001, 0.01, 0.1, 1]. The macro-average F1 score for each parameter combination is calculated by iterating through all parameter combinations. For example, when C = 10 and gamma = 0.01, the macro-average F1 score of the model on the test set is 0.85; when C = 5 and gamma = 0.05, the score increases to 0.88. The parameter combination with the highest score (C = 5, gamma = 0.05) is selected as the optimal parameter. When calculating the macro-average F1 score, precision, recall, and F1 value are calculated independently for each fault category. For example, the F1 value for category A is 0.9, the F1 value for category B is 0.85, and the F1 value for category C is 0.78. The final macro-average F1 score is (0.9 + 0.85 + 0.78) / 3 = 0.843. Based on the optimal parameters as the parameters of the initial model, a preset fault identification model is obtained.

[0236] This application provides a model training method for detecting faults in disconnecting switches of power equipment. Through supervised learning methods involving dynamic sequence alignment and multi-round iterative optimization, it achieves the effect of constructing a high-precision and robust fault identification model. Specifically, the method first triggers data acquisition when the disconnecting switch current value exceeds a preset threshold, obtaining a set of current, angle, and torque sequences within a preset time period. To address the feature distortion problem caused by inconsistent sequence lengths, the median value of each sequence set is calculated, short sequences are padded with zeros / interpolated, and long sequences are truncated to obtain an initial set of current, angle, and torque sequences of uniform length, ensuring consistency in model input dimensions. Subsequently, the initial sequence set is input into the initial model for training, and performance is improved through multi-round iterative optimization: in each iteration, the model weights are updated based on preset parameters, and the current evaluation index is calculated. The process stops when the index reaches the target or the number of iterations saturates, ultimately obtaining a preset fault identification model with optimal parameters. This model training method eliminates the influence of data length differences through dynamic alignment, ensuring stable extraction of temporal features; multi-round iteration gradually approaches the boundary between fault and normal modes, significantly improving the model's ability to identify noise interference and multiple concurrent faults. The final trained model can accurately capture fault feature combinations such as sudden current changes, angle shifts, and abnormal torque, effectively solving the misjudgment problem caused by inconsistent data in traditional models, and providing reliable core algorithm support for disconnector switch fault detection.

[0237] Based on the above embodiments, Figure 6a A graph showing the current values ​​of the disconnecting switch in different states provided in the embodiments of this application; such as Figure 6a As shown, Figure 6a The blue line in the diagram represents the change in current value of the disconnector under normal conditions. Figure 6a The blue dots in the diagram represent the changes in the current value of the disconnecting switch when it is stuck. Figure 6a The blue line segments in the diagram represent the changes in the current value of the disconnecting switch under different conditions.

[0238] Figure 6b A graph showing the angle values ​​of the disconnecting switch in different states provided in the embodiments of this application; such as Figure 6b As shown, Figure 6b The blue line in the diagram represents the change in the angle value of the disconnector switch under normal conditions; Figure 6b The blue dots in the diagram represent the changes in the angle value of the disconnector switch under jammed conditions. Figure 6b The blue line segments in the diagram represent the changes in the angle value of the disconnector switch under different conditions.

[0239] Figure 6cA graph showing the torque values ​​of the disconnecting switch in different states provided in the embodiments of this application; as shown Figure 6c As shown, Figure 6c The blue line in the diagram represents the change in torque value of the disconnector under normal conditions. Figure 6c The blue dots in the diagram represent the change in torque value of the disconnector switch when it is stuck. Figure 6c The blue line segments in the diagram represent the changes in the torque value of the disconnector switch under different conditions.

[0240] Figure 7a A schematic diagram of the confusion matrix provided in the embodiments of this application. Figure 1 ;like Figure 7a As shown, the dataset obtained in the previous embodiment was divided into a training set and a test set in an 8:2 ratio. The fault identification model was then used to identify the data in the test set, resulting in a confusion matrix. This confusion matrix shows that the fault identification model achieved perfect classification on the training set: the normal class (4 samples), the out-of-phase class (10 samples), and the stuck class (17 samples) all achieved an exceptional performance of TP=100%, FP=0, and FN=0, with an overall accuracy of 100%.

[0241] Figure 7b A schematic diagram of the confusion matrix provided in the embodiments of this application. Figure 2 ;like Figure 7b As shown, the dataset obtained in the previous embodiment was divided into a training set and a test set in an 8:2 ratio. The fault identification model was used to identify the data in the test set, resulting in a confusion matrix. The confusion matrix analysis shows the classification results of the fault identification model on the test set: the normal class and the out-of-phase class achieved perfect classification (F1 = 100%), while the 'Kase' class had one false positive (misclassifying out-of-phase as 'Kase') and one false negative (missing 'Kase'), causing the F1 score to drop to 66.7%. The overall accuracy was 87.5%, with errors concentrated in the 'Kase' class (accounting for 2 / 8 of the samples).

[0242] Figure 8 This application provides a schematic diagram of the structure of a fault detection device for a disconnecting switch in a power equipment. Figure 8 As shown, the fault detection device 80 for disconnecting switches in power equipment provided in this embodiment includes:

[0243] The acquisition module 801 is used to detect the current value of the disconnecting switch in the power equipment and acquire the set of switching parameters of the disconnecting switch under a preset time period. The set of switching parameters includes a current sequence, an angle sequence, and a torque sequence. The preset time period is the time period when the current value of the disconnecting switch detected in real time is greater than or equal to a preset threshold. The current sequence includes the current value at each moment under the preset time period. The angle sequence includes the angle value at each moment under the preset time period, and the angle value represents the closing angle of the disconnecting switch. The torque sequence includes the torque value at each moment under the preset time period, and the torque value represents the rotational torque value generated by the drive shaft of the disconnecting switch during the opening and closing of the disconnecting switch.

[0244] The processing module 802 is used to input the current sequence, angle sequence and torque sequence in the switch parameter set into the preset fault identification model for fault identification processing, and obtain the fault detection result of the disconnecting switch.

[0245] In one possible implementation, the processing module 802 includes:

[0246] The first processing module 8021 is used to perform feature extraction processing on the current sequence, angle sequence, and torque sequence based on a preset fault identification model to obtain the current feature sequence, angle feature sequence, and torque feature sequence.

[0247] The second processing module 8022 is used to process the set of switch parameters, current feature sequence, angle feature sequence and torque feature sequence based on the preset fault identification model to obtain the fault detection result of the disconnecting switch.

[0248] In one possible implementation, the first processing module 8021 includes:

[0249] The third processing module 80211 is used to perform first-order differential processing on the current sequence, angle sequence, and torque sequence based on a preset fault identification model to obtain a first-order differential current sequence, a first-order differential angle sequence, and a first-order differential torque sequence.

[0250] The fourth processing module 80212 is used to perform feature extraction processing on the current sequence and the first-order differential current sequence based on the preset fault identification model to obtain the current feature sequence.

[0251] The fifth processing module 80213 is used to perform feature extraction processing on the angle sequence and the first-order difference angle sequence based on the preset fault identification model to obtain the angle feature sequence;

[0252] The sixth processing module 80214 is used to perform feature extraction processing on the torque sequence and the first-order differential torque sequence based on the preset fault identification model to obtain the torque feature sequence.

[0253] In one possible implementation, the fourth processing module 80212 includes:

[0254] Based on the preset fault identification model, time-domain feature extraction processing is performed on the current sequence and the first-order differential current sequence to obtain the time-domain current sequence and the time-domain first-order differential current sequence.

[0255] Based on the preset fault identification model, frequency domain feature extraction processing is performed on the current sequence and the first-order differential current sequence to obtain the frequency domain current sequence and the frequency domain first-order differential current sequence.

[0256] Based on the time-domain current sequence, the first-order time-domain differential current sequence, the frequency-domain current sequence, and the first-order frequency-domain differential current sequence, the current characteristic sequence is obtained.

[0257] In one possible implementation, the fifth processing module 80213 includes:

[0258] Based on the preset fault identification model, time-domain feature extraction processing is performed on the angle sequence and the first-order difference angle sequence to obtain the time-domain angle sequence and the time-domain first-order difference angle sequence.

[0259] Based on the preset fault identification model, frequency domain feature extraction processing is performed on the angle sequence and the first-order difference angle sequence to obtain the frequency domain angle sequence and the frequency domain first-order difference angle sequence.

[0260] Angle feature sequences are obtained from the time-domain angle sequence, the time-domain first-order difference angle sequence, the frequency-domain angle sequence, and the frequency-domain first-order difference angle sequence.

[0261] In one possible implementation, the sixth processing module 80214 includes:

[0262] Based on the preset fault identification model, time-domain feature extraction processing is performed on the torque sequence and the first-order differential torque sequence to obtain the time-domain torque sequence and the time-domain first-order differential torque sequence.

[0263] Based on the preset fault identification model, frequency domain feature extraction processing is performed on the torque sequence and the first-order differential torque sequence to obtain the frequency domain torque sequence and the frequency domain first-order differential torque sequence.

[0264] The torque characteristic sequence is obtained from the time-domain torque sequence, the time-domain first-order difference torque sequence, the frequency-domain torque sequence, and the frequency-domain first-order difference torque sequence.

[0265] In one possible implementation, the second processing module 8022 includes:

[0266] Based on a preset fault identification model, correlation coefficients are processed on the current sequence, angle sequence, and torque sequence to obtain a first correlation sequence, a second correlation sequence, and a third correlation sequence. The first correlation sequence includes the first correlation degree at each moment within a preset time period, representing the correlation between the current value and the angle value. The second correlation sequence includes the second correlation degree at each moment within the preset time period, representing the correlation between the current value and the torque value. The third correlation sequence includes the third correlation degree at each moment within the preset time period, representing the correlation between the angle value and the torque value.

[0267] Based on the preset fault identification model, the first correlation sequence, the second correlation sequence, the third correlation sequence, the current characteristic sequence, the angle characteristic sequence, and the torque characteristic sequence are processed to obtain the fault detection results of the disconnecting switch.

[0268] This embodiment provides a fault detection device for disconnecting switches in power equipment, which can execute the method provided in the above-described method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0269] Figure 9 A schematic diagram of a model training device for detecting faults in disconnecting switches of power equipment, provided in an embodiment of this application, is shown below. Figure 9 As shown, the model training device 90 for detecting faults in disconnecting switches of power equipment provided in this embodiment includes:

[0270] The acquisition module 901 is used to acquire a set of current sequences, an angle sequence set, and a torque sequence set of the disconnecting switch under a preset time period. The preset time period is the period when the real-time detected current value of the disconnecting switch is greater than or equal to a preset threshold. The current sequence set includes at least one current sequence, which includes the current value at each moment under the preset time period. The angle sequence set includes at least one angle sequence, which includes the angle value at each moment under the preset time period, and the angle value represents the closing angle of the disconnecting switch. The torque sequence set includes at least one torque sequence, which includes the torque value at each moment under the preset time period, and the torque value represents the rotational torque generated by the drive shaft of the disconnecting switch during the opening and closing process.

[0271] Training module 902 is used to train the initial model based on the current sequence set, angle sequence set, and torque sequence set to obtain a preset fault identification model;

[0272] The preset fault identification model is the preset fault identification model in the first aspect and / or in various possible implementations of the first aspect.

[0273] In one possible implementation, the training module 902 includes:

[0274] The confirmation module 9021 is used to determine the first sequence median value of the current sequence set, wherein the first sequence median represents the number of sequences in the current sequence set; and to determine the second sequence median value of the angle sequence set, wherein the second sequence median represents the number of sequences in the angle sequence set; and to determine the third sequence median value of the torque sequence set, wherein the third sequence median represents the number of sequences in the torque sequence set.

[0275] The processing module 9022 is used to process the current sequences in the current sequence set according to the median of the first sequence to unify the number of current values ​​in each current sequence in the current sequence set, thereby obtaining an initial current sequence set; and to process the angle sequences in the angle sequence set according to the median of the second sequence to unify the number of angle values ​​in each angle sequence in the angle sequence set, thereby obtaining an initial angle sequence set; and to process the torque sequences in the torque sequence set according to the median of the third sequence to unify the number of torque values ​​in each torque sequence in the torque sequence set, thereby obtaining an initial torque sequence set.

[0276] The training submodule 9023 is used to train the initial model based on the initial current sequence set, the initial angle sequence set, and the initial torque sequence set to obtain the preset fault identification model.

[0277] In one possible implementation, the training submodule 9023 includes:

[0278] Repeat the following steps until the preset conditions are met:

[0279] Based on the preset i-th parameter, the i-th initial model is updated to obtain the (i+1)-th initial model;

[0280] Based on the initial current sequence set, initial angle sequence set, and initial torque sequence set, the (i+1)th initial model is processed to obtain the (i+1)th evaluation index; where i is a positive integer greater than or equal to 1; and the value of i is determined by incrementing by 1.

[0281] Among them, the preset conditions are met, and the obtained evaluation indicators are used to obtain the preset fault identification model.

[0282] This embodiment provides a model training device for detecting faults in disconnecting switches of power equipment. It can execute the method provided in the above-described method embodiment, and its implementation principle and technical effect are similar. This embodiment will not elaborate further here.

[0283] Figure 10 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 10As shown, the electronic device 100 provided in this embodiment includes at least one processor 1001 and a memory 1002. Optionally, the device 100 further includes a communication component 1003. The processor 1001, memory 1002, and communication component 1003 are connected via a bus 1004.

[0284] In a specific implementation, at least one processor 1001 executes computer execution instructions stored in memory 1002, causing at least one processor 1001 to perform the above-described method.

[0285] The specific implementation process of processor 1001 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0286] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0287] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0288] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0289] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0290] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0291] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0292] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0293] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0294] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0295] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0296] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0297] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0298] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A fault detection method for disconnecting switches in power equipment, characterized in that, The method includes: The current value of a disconnector switch in a power equipment is detected, and a set of switching parameters of the disconnector switch under a preset time period is obtained. The set of switching parameters includes a current sequence, an angle sequence, and a torque sequence. The preset time period is the period corresponding to when the real-time detected current value of the disconnector switch is greater than or equal to a preset threshold. The current sequence includes the current value at each moment under the preset time period. The angle sequence includes the angle value at each moment under the preset time period, and the angle value represents the closing angle of the disconnector switch. The torque sequence includes the torque value at each moment under the preset time period, and the torque value represents the rotational torque generated by the drive shaft of the disconnector switch during the opening and closing process. The current sequence, angle sequence, and torque sequence from the switch parameter set are input into a preset fault identification model for fault identification processing to obtain the fault detection result of the disconnecting switch.

2. The method according to claim 1, characterized in that, The current sequence, angle sequence, and torque sequence are input into a preset fault identification model for fault identification processing to obtain the fault detection results of the disconnecting switch, including: Based on the preset fault identification model, feature extraction processing is performed on the current sequence, the angle sequence, and the torque sequence to obtain the current feature sequence, the angle feature sequence, and the torque feature sequence, respectively. Based on the preset fault identification model, the switch parameter set, the current feature sequence, the angle feature sequence, and the torque feature sequence are processed to obtain the fault detection result of the disconnecting switch.

3. The method according to claim 2, characterized in that, Based on the preset fault identification model, feature extraction processing is performed on the current sequence, the angle sequence, and the torque sequence to obtain current feature sequence, angle feature sequence, and torque feature sequence, including; Based on the preset fault identification model, the current sequence, the angle sequence, and the torque sequence are processed by first-order differential processing to obtain first-order differential current sequence, first-order differential angle sequence, and first-order differential torque sequence. Based on the preset fault identification model, feature extraction processing is performed on the current sequence and the first-order differential current sequence to obtain the current feature sequence; Based on the preset fault identification model, feature extraction processing is performed on the angle sequence and the first-order difference angle sequence to obtain the angle feature sequence; Based on the preset fault identification model, feature extraction processing is performed on the torque sequence and the first-order differential torque sequence to obtain the torque feature sequence.

4. The method according to claim 3, characterized in that, Based on the preset fault identification model, feature extraction processing is performed on the current sequence and the first-order differential current sequence to obtain the current feature sequence, including: Based on the preset fault identification model, time-domain feature extraction processing is performed on the current sequence and the first-order differential current sequence to obtain the time-domain current sequence and the time-domain first-order differential current sequence. Based on the preset fault identification model, frequency domain feature extraction processing is performed on the current sequence and the first-order differential current sequence to obtain the frequency domain current sequence and the frequency domain first-order differential current sequence. The current characteristic sequence is obtained based on the time-domain current sequence, the time-domain first-order differential current sequence, the frequency-domain current sequence, and the frequency-domain first-order differential current sequence.

5. The method according to claim 3, characterized in that, Based on the preset fault identification model, feature extraction processing is performed on the angle sequence and the first-order difference angle sequence to obtain the angle feature sequence, including: Based on the preset fault identification model, time-domain feature extraction processing is performed on the angle sequence and the first-order difference angle sequence to obtain the time-domain angle sequence and the time-domain first-order difference angle sequence. Based on the preset fault identification model, frequency domain feature extraction processing is performed on the angle sequence and the first-order difference angle sequence to obtain the frequency domain angle sequence and the frequency domain first-order difference angle sequence. The angle feature sequence is obtained based on the time-domain angle sequence, the time-domain first-order difference angle sequence, the frequency-domain angle sequence, and the frequency-domain first-order difference angle sequence.

6. The method according to claim 3, characterized in that, Based on the preset fault identification model, feature extraction processing is performed on the torque sequence and the first-order differential torque sequence to obtain the torque feature sequence, including: Based on the preset fault identification model, time-domain feature extraction processing is performed on the torque sequence and the first-order differential torque sequence to obtain the time-domain torque sequence and the time-domain first-order differential torque sequence. Based on the preset fault identification model, frequency domain feature extraction processing is performed on the torque sequence and the first-order differential torque sequence to obtain the frequency domain torque sequence and the frequency domain first-order differential torque sequence. The torque feature sequence is obtained based on the time-domain torque sequence, the time-domain first-order differential torque sequence, the frequency-domain torque sequence, and the frequency-domain first-order differential torque sequence.

7. The method according to any one of claims 1-6, characterized in that, Based on the preset fault identification model, the switch parameter set, the current feature sequence, the angle feature sequence, and the torque feature sequence are processed to obtain the fault detection result of the disconnecting switch, including: Based on the preset fault identification model, correlation coefficient processing is performed on the current sequence, the angle sequence, and the torque sequence to obtain a first correlation sequence, a second correlation sequence, and a third correlation sequence. The first correlation sequence includes a first correlation degree at each moment within the preset time period, representing the correlation between the current value and the angle value. The second correlation sequence includes a second correlation degree at each moment within the preset time period, representing the correlation between the current value and the torque value. The third correlation sequence includes a third correlation degree at each moment within the preset time period, representing the correlation between the angle value and the torque value. Based on the preset fault identification model, the first correlation sequence, the second correlation sequence, the third correlation sequence, the current feature sequence, the angle feature sequence, and the torque feature sequence are processed to obtain the fault detection result of the disconnecting switch.

8. A model training method for detecting faults in disconnecting switches of power equipment, characterized in that, The method includes: The system acquires a set of current sequences, an angle sequence set, and a torque sequence set for a disconnector switch within a preset time period. The preset time period is defined as the period when the real-time detected current value of the disconnector switch is greater than or equal to a preset threshold. The current sequence set includes at least one current sequence, which includes the current value at each moment within the preset time period. The angle sequence set includes at least one angle sequence, which includes the angle value at each moment within the preset time period, and the angle value represents the closing angle of the disconnector switch. The torque sequence set includes at least one torque sequence, which includes the torque value at each moment within the preset time period, and the torque value represents the rotational torque generated by the drive shaft of the disconnector switch during the opening and closing process. The initial model is trained based on the current sequence set, the angle sequence set, and the torque sequence set to obtain a preset fault identification model; The preset fault identification model is the preset fault identification model according to any one of claims 1-7.

9. The method according to claim 8, characterized in that, The initial model is trained based on the current sequence set, the angle sequence set, and the torque sequence set to obtain a preset fault identification model, including: Determine a first sequence median value for the current sequence set, wherein the first sequence median represents the number of sequences in the current sequence set; determine a second sequence median value for the angle sequence set, wherein the second sequence median represents the number of sequences in the angle sequence set; and determine a third sequence median value for the torque sequence set, wherein the third sequence median represents the number of sequences in the torque sequence set. Based on the median of the first sequence, the current sequences in the current sequence set are processed to unify the number of current values ​​in each current sequence in the current sequence set, thus obtaining an initial current sequence set; and based on the median of the second sequence, the angle sequences in the angle sequence set are processed to unify the number of angle values ​​in each angle sequence in the angle sequence set, thus obtaining an initial angle sequence set; and based on the median of the third sequence, the torque sequences in the torque sequence set are processed to unify the number of torque values ​​in each torque sequence in the torque sequence set, thus obtaining an initial torque sequence set; The initial model is trained based on the initial current sequence set, the initial angle sequence set, and the initial torque sequence set to obtain the preset fault identification model.

10. The method according to any one of claims 8-9, characterized in that, The initial model is trained based on the initial current sequence set, the initial angle sequence set, and the initial torque sequence set to obtain the preset fault identification model, including: Repeat the following steps until the preset conditions are met: Based on the preset i-th parameter, the i-th initial model is updated to obtain the (i+1)-th initial model; Based on the initial current sequence set, the initial angle sequence set, and the initial torque sequence set, the (i+1)th initial model is processed to obtain the (i+1)th evaluation index; where i is a positive integer greater than or equal to 1; and the value of i is determined to be incremented by 1. Wherein, the preset conditions are met, and the obtained evaluation indicators are used to obtain the preset fault identification model.

Images