A circuit breaking fault state diagnosis method and system based on a real-time sensing platform

Abstract

A circuit breaking fault state diagnosis method and system based on a real-time sensing platform. By collecting multiple types of operation data of a three-phase four-wire electric energy meter, time-frequency domain features, correlation features and electrical features are extracted and fused to construct a unified feature vector. A multi-task learning model is used to diagnose fault categories, locations and severity, and combined with time series data to predict the remaining service life of the electric energy meter. Based on the diagnosis results, a fault handling knowledge graph is constructed, combined with fault chain deduction and hierarchical trigger thresholds to automatically generate fault handling strategies. This method realizes intelligent diagnosis, life prediction and auxiliary decision of circuit breaking faults, and improves the intelligent level of electric energy meter operation and maintenance management.

Description

Technical Field

[0001] This invention belongs to the field of intelligent diagnostic technology for power systems, specifically relating to a method and system for diagnosing circuit fault status based on a real-time sensing platform. Background Technology

[0002] With the development of the energy internet and the digital and intelligent transformation of the power system, power grid companies have placed higher demands on the real-time monitoring and intelligent fault diagnosis of circuit breaker operation status. Due to the limitations of the secondary side access of the current transformer, three-phase four-wire energy meters usually do not have built-in remote opening and closing mechanisms, and still rely on external prepaid circuit breakers under high power loads; when communication or external switch failures occur, manual on-site operation is often required, resulting in a large workload for operation and maintenance.

[0003] Patent application CN121385772A, based on a feature extraction mechanism of orthogonal decoupling theory and a multi-level decision-making system of dynamic benchmark learning, overcomes individual differences in equipment and accurately reveals the fault coupling relationship of voltage / current loops, enabling fault diagnosis and location of energy meters. Patent application CN117907891A, by calculating the common-mode voltage error and setting an adaptive threshold, quickly and accurately locates the faulty switching transistor based on the diagnostic variable's endpoint, achieving open-circuit fault location of devices in a three-phase four-wire T-type inverter.

[0004] Existing circuit breakers generally focus on executing opening and closing commands, lacking built-in sensors to achieve active acquisition and remote interaction of operating status. Especially in high-power loads and scenarios with limited secondary side of the current transformer (CT), traditional monitoring methods relying on current loops fail, resulting in a blind spot in the perception of circuit breaker operating status. Furthermore, limited by a single data source and lacking multi-dimensional data fusion and analysis capabilities, existing methods are insufficient to meet the requirements of intelligent fault diagnosis for complex faults (closing / opening jams, power supply abnormalities, poor circuit contact, etc.). Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for diagnosing circuit fault conditions based on a real-time sensing platform. By collecting various types of operational data from a three-phase four-wire energy meter, time-frequency domain features, correlation features, and electrical features are extracted and fused to construct a unified feature vector. A multi-task learning model is then used to diagnose the fault type, location, and severity.

[0006] This application discloses a method for diagnosing open circuit fault status based on a real-time sensing platform, which adopts the following technical solution: The real-time sensing platform collects various types of data from three-phase four-wire energy meters through interfaces, extracts time-frequency domain features and constructs cross-signal features, and combines time-frequency domain features and cross-signal features to form basic signal features; based on numerical continuous signals in the various types of data, it calculates harmonic phase angle features, three-phase correlation features and nonlinear features, and integrates them with basic signal features to form a unified feature vector; The unified feature vector is dimensionality reduced to construct a multi-dimensional feature sample that includes dimensionality-reduced features, fault category, fault location, severity, and time series data. After adaptive attention enhancement of the dimensionality-reduced features at the channel dimension, multiple parallel task branches are constructed for joint training, and the fault diagnosis results of the three-phase four-wire energy meter are inferred and output. Based on the time series data, a lifespan prediction branch is constructed to predict the remaining lifespan of the energy meter based on the real-time dimensionality-reduced features, and the prediction is incorporated into the fault diagnosis results. Based on the fault diagnosis results, a fault handling knowledge graph is constructed by combining domain knowledge and expert experience; based on the nodes and edges defined in the graph, fault chain deduction is performed, and a candidate action set for fault handling strategies is output; combined with the hierarchical trigger threshold table embedded in the real-time perception platform, a deterministic handling strategy is generated.

[0007] Furthermore, the multiple data types include numerical continuous signals and event statistical signals. For the numerical continuous signal, time-frequency domain features and waveform features are acquired in the form of a sliding window. The event statistics signal includes event categories and statistical signals, and the event categories are encoded during the preprocessing stage. The event category code is mathematically combined with the corresponding numerical continuous signal to generate cross-signal features.

[0008] Furthermore, the harmonic phase angle characteristics and three-phase correlation characteristics are extracted, including the following steps: Harmonic distortion rate and phase jump characteristics are calculated based on the collected numerical continuous signals to form harmonic phase angle characteristics; Calculate the three-phase correlation coefficients and phase differences to form the three-phase correlation characteristics.

[0009] Furthermore, the severity level is a label indicating the continuity severity of the electricity meter fault, calculated using electrical feature mapping. The calculation process includes: When contact resistance is available, the resistance method is used; it is described as the difference between the measured contact resistance and the normal contact resistance, divided by the difference between the calibrated failure resistance and the normal contact resistance. When the contact resistance is not available, it is calculated by mapping electrical characteristic quantities. The description is as follows: the weighted values ​​of the harmonic distortion term, the temperature rise term, and the three mutual correlations are obtained, and then mapped to the range of 0 to 1 through the truncation function. The harmonic distortion term is the increment of harmonic distortion relative to the normal value, the temperature rise term is the temperature rise, and the three mutual correlations are 1 minus the minimum value of the three mutual correlation coefficients.

[0010] Furthermore, adaptive attention enhancement is applied to the channel dimension of the dimensionality-reduced features. The steps include: A fully connected mapping layer is used to project the dimensionality-reduced features. An SE attention module is then connected in series after the mapping layer to calculate and output the weight vector of the feature dimension after the projection transformation. The features after projection transformation are weighted using a weight vector, and an enhanced feature vector is output.

[0011] Furthermore, the parallel branches include: The fault location branch maps the enhanced feature vectors to a fault location probability distribution through a fully connected layer and an activation function, and outputs the predicted fault location with the maximum probability. At the same time, the three-phase correlation coefficient matrix is ​​calculated. If the three-phase correlation coefficient corresponding to the predicted fault location does not decrease, it is judged that a physical mechanism conflict has occurred, and a suspected misjudgment is output. The corresponding multi-dimensional feature samples are labeled, and the loss weight of the sample is increased in the next round of training and / or online learning.

[0012] Furthermore, the lifetime prediction branch includes: Read the time series data field from the current sample and map it to a single scalar of remaining lifetime through a time series network; Set a severity threshold for electricity meter failure, and then search for the moment when the severity of the first failure exceeds the severity threshold in subsequent multi-dimensional feature samples. The current sample's true remaining lifetime is... Subtract the current sample time index Then multiply by the sampling interval; The single scalar and the true remaining lifetime are optimized by backpropagation of the lifetime prediction branch using mean square error; the optimized lifetime prediction branch is then used to predict the remaining lifetime of the real-time dimensionality-reduced features.

[0013] Furthermore, the fault diagnosis results include: Fault type, fault location, fault severity, and remaining service life; define the fault risk level based on the fault severity.

[0014] Furthermore, the fault handling knowledge graph is a directed graph, including: Nodes include fault nodes, phenomenon nodes, consequence nodes, and action nodes; edge definitions include causal edges, concurrent edges, evolutionary edges, and inhibition edges. The causal edges are used to describe the phenomena caused by the fault, the concurrent edges are used to describe the symbiosis of phenomena, the evolution is used to describe the consequences caused by the phenomena, and the inhibition edges are used to describe the elimination of phenomena by actions.

[0015] The beneficial effects of this invention are that, compared with the prior art, 1. By fusing time-frequency domain features, cross-signal features, harmonic phase angle features, three-phase correlation features, and nonlinear features, a unified feature vector is constructed. An adaptive attention enhancement and multi-task joint learning mechanism are employed to simultaneously identify fault categories, locate fault positions, and assess fault severity. Compared to diagnostic methods that rely solely on single electrical parameters or threshold rules, this application can fully leverage the correlation information between multi-source heterogeneous data, improving the accuracy and robustness of three-phase four-wire open-circuit fault identification. Furthermore, by introducing a physical mechanism consistency verification mechanism, online correction and reinforcement learning are performed on suspected misdiagnosed samples, effectively reducing the misdiagnosis rate and improving the reliability of fault diagnosis under complex operating conditions.

[0016] 2. Building upon fault diagnosis, a lifespan prediction branch is further constructed. By combining time-series data, the remaining lifespan of the electricity meter is predicted, extending the focus from fault detection to fault prediction. Simultaneously, a fault handling knowledge graph is constructed by integrating domain knowledge and expert experience. Candidate handling actions are generated through fault chain deduction, and deterministic handling strategies are automatically matched based on tiered trigger thresholds. This not only enables early identification of potential failure risks and improves preventative maintenance capabilities but also achieves closed-loop linkage from fault detection, risk assessment, handling decisions to execution recommendations, reducing the decision-making burden on maintenance personnel and improving the intelligence and automation level of electricity meter operation and maintenance management. Attached Figure Description

[0017] Figure 1 The execution flowchart of the circuit breaker fault state diagnosis method provided in the embodiment is shown. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this application are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0019] In one embodiment, refer to Figure 1 This paper discloses a specific implementation method for diagnosing circuit fault status based on a real-time sensing platform.

[0020] S1: In one implementation, a real-time sensing platform is used to continuously collect operating data of three-phase four-wire energy meters and related circuit breakers.

[0021] The real-time sensing platform and the analog sampling channel and metering chip interface vector of the three-phase four-wire smart energy meter. The sensing platform includes at least a voltage / current transformer (for acquiring three-phase voltage and current), a temperature sensor (for acquiring the temperature of circuit breaker contacts and bus connection points), a switch quantity acquisition module (for acquiring the circuit breaker opening and closing position and tripping signal), and a clock synchronization module.

[0022] 1.1: For three-phase four-wire energy meters and related circuit breakers, the data collection objects and parameters include: Opening and closing status data: open / closed status, number of actions, action sequence, operation direction, etc., used to identify opening and closing jams, delays, or abnormal actions; Electrical quantity data: three-phase voltage, current, power, power factor, current imbalance, common-mode voltage, etc., used to identify undervoltage, overcurrent, circuit abnormalities and potential open circuit faults; Circuit characteristic data: operating power supply voltage, circuit contact resistance, relay coil resistance, etc., used to determine poor contact, loose wiring or abnormal drive; Mechanical / environmental data: vibration waveforms, temperature curves, changes in operating torque, etc., are used to determine abnormalities such as jamming during opening and closing, decreased arc extinguishing performance, and contact erosion; It can also synchronously acquire auxiliary time data, including operation commands, external fault alarms, abnormal load events, etc., which can be used to trigger high-frequency sampling or fault marking.

[0023] Furthermore, for numerical data, such as key parameters like current, voltage, temperature, and vibration, high-frequency continuous sampling is performed (e.g., 1kHz~10kHz, which can be adjusted according to the load); the opening and closing status and the number of actions are collected in real time to ensure that the entire process of the action is captured.

[0024] In one possible implementation, trigger conditions can be set to enter a high-frequency or multi-parameter acquisition mode after triggering, thereby improving the ability to capture key fault characteristics. Trigger conditions can be set as follows: sudden current change exceeding a threshold, abnormal action timing, vibration or temperature rise exceeding a set range, etc.

[0025] All acquisition modules use high-precision timestamps (such as milliseconds or microseconds) to ensure the consistency of time sequence of multi-source data, so as to support the time sequence alignment of subsequent feature extraction and model training.

[0026] 1.2: In a further implementation, the collected data is preprocessed. Numerical signal data and event / statistical signal data are standardized.

[0027] For continuous numerical signals, such as current, voltage, temperature, vibration amplitude, and circuit contact resistance, high-frequency interference and electromagnetic noise can be removed by using low-pass filtering or wavelet denoising. Different filtering parameters are set for signals in different frequency bands; for example, a 50~500Hz bandpass filter is used for vibration signals.

[0028] Z-score standardization maps data to a uniform range, such as 0~1 or -1~1, reducing the influence of different units.

[0029] For categorical data such as event / statistical signals, preprocessing is performed using one-hot encoding or integer encoding. Examples include circuit breaker status (open=0 / close=1), action direction (forward=1, reverse=0), and fault flag (present=1, absent=0). Statistical signals such as the number of opening and closing operations, average action interval, action delay time, and duration of the most recent action can form continuous numerical features to participate in the model input.

[0030] 1.3: In a further implementation, key features are extracted to form a unified fault feature vector.

[0031] Logarithmic signals are obtained using a sliding window approach to acquire time-domain features (mean, variance, etc.), frequency-domain features (Fourier transform to obtain frequency-related features), waveform features (rising edge slope, falling edge slope, peak-to-peak value, etc.), and cross-signal combination features.

[0032] Furthermore, cross-signal combination features are used to associate categorical events with corresponding numerical signals. For example: circuit breaker opening / closing status code + corresponding current peak value, circuit breaker opening / closing status code + corresponding vibration peak value. Formally, this can be represented as... ; This indicates cross-signal combination; This indicates the opening and closing status code, where "1" represents the closing status code and "0" represents the opening status code. This represents the peak value corresponding to a certain type of signal. For example, the tripping status code multiplied by the peak value of the operating current.

[0033] Temporal characteristics at the same sampling time point Frequency domain characteristics Waveform characteristics and cross-signal combination features By splicing the data, standardized basic signal characteristics are formed. , is represented as: ; Cross-signal combination features combine event / statistical signals and numerical electrical signals to simultaneously reflect the correlation patterns of event state, action patterns and electrical responses in a single feature vector.

[0034] 1.4: In one specific implementation, a special feature extraction method for complex electrical phenomena and three-phase system characteristics is further introduced.

[0035] The most common problems with three-phase four-wire energy meters are excessive neutral current, severe harmonic superposition, and / or abnormal three-phase correlation. Excessive neutral current may be caused by load imbalance, incorrect wiring, or grounding faults; third, fifth, and seventh harmonics are particularly critical, as they can affect measurement and protection operation. Abnormal three-phase correlation refers to the high correlation and stable phase difference between the three-phase voltage / current / power during normal operation; however, under abnormal conditions (such as single-phase grounding, open circuit, or load imbalance), the correlation drops sharply.

[0036] Based on the above considerations, this application considers standardizing the feature vector. Based on this, harmonic distortion rate and phase angle jump characteristics, three-phase correlation characteristics, and nonlinear characteristics are introduced. Specifically, the following steps are adopted: (1) Harmonic distortion rate and phase jump characteristics; Harmonic decomposition of voltage and current signals is performed using Fourier transform or digital filtering: ; , , These are the third, fifth, and seventh harmonic components, respectively, which have the most significant impact on the superposition of neutral current and measurement error. Further analysis using formulas... Calculate harmonic distortion rate .

[0037] Furthermore, for each harmonic component, the phase angle is calculated. Changes: ; in, Indicates the first The phase angle change of the harmonic components (such as the 3rd, 5th, 7th, etc.) is used to reflect the degree of change in the phase of the harmonic before and after the opening and closing of the circuit breaker. and Indicates the time before and after the opening and closing operation. Instantaneous phase angle of the subharmonic component.

[0038] when Exceeding the threshold means that there is a significant change in phase angle, which may indicate a wiring error, abnormal switch operation, or the presence of a harmonic source.

[0039] (2) Characteristics of triphase correlation; Three-phase correlation characteristics include cross-correlation coefficients and phase difference. Under normal operation, a three-phase four-wire energy meter exhibits a high correlation between three-phase voltage / current / power and a stable phase difference (approximately 120°). During abnormal conditions (single-phase grounding, open circuit, load imbalance), the correlation drops sharply. The cross-correlation coefficient is defined as: ; This represents the cross-correlation coefficient between phase A and phase B, under normal circumstances. It decreases significantly during abnormal conditions; Represents the current / voltage sequence of phase A and phase B. , Covariance between them; , Current / voltage sequences for phase A and phase B, respectively. , The standard deviation.

[0040] Phase difference: the instantaneous phase difference of three-phase voltage / current ; Small fluctuations indicate normalcy, while large fluctuations indicate abnormality.

[0041] (3) Nonlinear characteristics; Based on the high-frequency sampling sequence of current / voltage at the moment of closing. The Hilbert-Huang transform (HHT / EMD), higher-order spectral analysis (bispectral / trispectral), and time-frequency entropy calculation methods are used to extract the energy distribution, bispectral phase coupling characteristics, and time-frequency entropy values ​​of each intrinsic mode function (IMF) to characterize the degree of nonlinear distortion and the intensity of interharmonic interaction of the signal.

[0042] 1.5: Basic signal characteristics Harmonic phase angle characteristics Triphasic correlation characteristics and nonlinear characteristics Integrate them to form a unified high-dimensional feature vector. , is represented as: ; This feature vector comprehensively covers the signal's fundamental time-frequency properties, harmonic distortion, phase correlation, and nonlinear dynamic characteristics, providing high-quality input for subsequent fault diagnosis models.

[0043] S2: In one implementation, an intelligent diagnostic model is constructed and trained based on historical fault samples.

[0044] In one specific implementation, to eliminate redundant information and dimensionality effects between features and reduce the computational load in the early stages of training and inference, this application, before constructing the fault diagnosis model, processes the high-dimensional feature vector output by S1. Dimensionality reduction is performed.

[0045] In a further specific implementation of the reference, the aforementioned high-dimensional feature vectors are... Principal component analysis (PCA) was used for dimensionality reduction, and the top k principal components with a cumulative contribution rate of not less than a set threshold (85%) were selected to form a low-dimensional feature vector. .

[0046] In one reference implementation, an autoencoder can also be used to process high-dimensional feature vectors. Perform nonlinear feature compression.

[0047] The feature vector obtained after dimensionality reduction Based on this, an end-to-end deep neural network model is constructed to output the fault type, location, and severity, as well as predict the remaining life of a three-phase four-wire energy meter.

[0048] 2.1: In a further implementation, a multi-dimensional sample is established, including normal operation and various faults (such as poor contact, overcurrent, excessive harmonics, mechanical jamming, etc.).

[0049] The samples are constructed using a sliding window generation method; for the first... For each sample, its data structure is defined as follows: ; in, These are the sample features after dimensionality reduction in the preceding steps; This is a fault category label; 0 indicates no fault. ; For fault location labeling; The severity label indicates the continuity of the failure, with 0 representing no failure and 1 representing complete failure. This is time series data used for RUL (Remaining Useful Life) prediction.

[0050] Furthermore, in the sample data structure, the continuity severity label is calculated using a key electrical feature mapping. In this embodiment, the mapping is performed using harmonic distortion, temperature rise, and three-phase correlation. This is represented as: ; The above formula can be described as: when When the meter is in good working order, it indicates that the three phases are functioning normally, with no temperature rise and no electric arc; when... The time indicates that the electricity meter has completely failed, such as having tripped; if the trend of contact resistance change is known or can be directly monitored, the resistance method can be used. calculate, , and These are the measured contact resistance, normal contact resistance, and calibrated failure resistance of the electricity meter, respectively.

[0051] If the contact resistance cannot be measured, it can be calculated by mapping using the aforementioned characteristic quantities: ; In the formula, This represents the increase in harmonic distortion relative to the normal value. Indicates a temperature rise. It is the minimum value of the three cross-correlation coefficients; , as well as These are the weighting coefficients for each item; This is a truncation function used to restrict the calculated value to the range of 0 to 1.

[0052] It should be noted that, , and The three factors were normalized before being weighted.

[0053] Furthermore, in the sample data structure, the time series data is a sequence of dimensionality-reduced feature vectors within the sliding window of the sample, represented as: ; in, The index or timestamp of the current time window. This represents the length of the time window.

[0054] In one reference implementation, SMOTE (Synthetic Minority Oversampling) can be used to generate synthetic samples to address the problem of uneven sample distribution.

[0055] 2.2: Construct a deep feature mapping network that includes a channel attention mechanism to map input features to a high-dimensional space to uncover nonlinear coupling relationships and adaptively enhance the weights of key fault feature channels to suppress redundant noise interference.

[0056] Let the feature vector in the sample be The feature dimension is That is, the input tensor is First, design a fully connected mapping layer to... Projected to a high-dimensional feature space To enhance feature representation capabilities, the mapping formula is as follows: ;in, This is the weight matrix. For bias terms; This represents the activation function, preferably ReLU or GELU, used to introduce nonlinearity; This represents a high-dimensional feature representation.

[0057] Furthermore, after the mapping layer, a Squeeze-and-Excitation (SE) attention module is cascaded to explicitly model the importance of different feature channels (such as harmonic features, waveform features, and correlation features). This includes the following steps: For high-dimensional features Perform global information aggregation and generate channel descriptors through global average pooling (GAP). , This is used to characterize the global distribution of each channel. This is the feature channel index.

[0058] Furthermore, by learning the non-linear dependencies between channels through two fully connected layers (FC), a normalized channel weight vector is generated. , represented as ; To reduce the weights of the dimensionality reduction layer, This refers to the reduction ratio. To restore the layer weights; It is the ReLU activation function. This is the Sigmoid activation function, used to restrict the weights to the (0,1) interval.

[0059] Using weight vectors For high-dimensional features After weighted adjustment, the attention-enhanced feature vector is obtained. , ,in This indicates element-wise multiplication.

[0060] 2.3: Based on Enhanced Feature Vectors Three parallel task branches are constructed to achieve multi-dimensional joint reasoning on three-phase four-wire energy meters in terms of fault type, location, and severity.

[0061] The three parallel branches are as follows: (1) Fault classification branch: using a fully connected layer + Softmax activation function; Let the output feature dimension be First, a fully connected layer is used to map the dimension to the number of fault categories. For example, normal operation, wiring error, harmonic interference, mechanical failure, etc.

[0062] The output is then fed into a Softmax layer, which converts the output into a probability distribution; the fault classification branch outputs the predicted fault category probability distribution. .

[0063] In one reference implementation, the branch is trained using cross-entropy loss.

[0064] (2) Fault location branch: using a fully connected layer + Softmax activation function; A fully connected layer is used to map the dimension to the number of fault locations. For example, phase A, phase B, phase C, N line, and structural side.

[0065] The output is then fed into a Softmax layer, which converts the output into a probability distribution; the fault location branch outputs the predicted fault location probability distribution. .

[0066] Furthermore, in calculation Simultaneously, calculate the three-phase correlation coefficient matrix. , is represented as: ; According to physical principles, in a three-phase power system, the three-phase signals are highly symmetrical and strongly correlated during normal operation. When a fault occurs in one phase (such as a broken wire, short circuit, or harmonic distortion), its waveform distortion leads to a significant decrease in its correlation with the other two phases. This is reflected in the correlation coefficient matrix as follows: If the fault is in phase A, its coupling with phases B / C weakens. and / or The failure rate should decrease significantly; the same applies to failures in phases B and C. If all three phases fail, then... , , All three phases decreased. For reference, a significant decrease can be defined as a relative rate of change greater than a set value.

[0067] When the Softmax activation function of the fault location branch outputs the predicted fault location (such as phase A) based on the maximum probability, but the corresponding correlation coefficient does not decrease, it is judged that there is a physical mechanism conflict and outputs a suspected misjudgment.

[0068] Furthermore, the corresponding sample is labeled, and in the next round of training / online learning, it is assigned a higher loss weight. The loss function is updated as follows: ; For the updated loss function, The loss function for the previous training round; This is an indicator function; it is 1 when a sample exhibits a physical mechanism conflict, and 0 otherwise. This is the magnification factor.

[0069] (3) Fault severity branch: using a fully connected layer + Softmax activation function; Will Mapped to 1D output This indicates the severity of the fault, used to gauge the extent of its impact, rather than a simple binary judgment. For example, if... If so, a serious fault will be output.

[0070] The fault severity branch uses mean squared error as the loss function.

[0071] In one implementation, to ensure collaborative optimization among the branches, a weighted joint optimization of the loss functions of the three branches is designed.

[0072] 2.4: To predict the remaining service life of a three-phase four-wire energy meter, a RUL branch is constructed in addition to the three branches mentioned above; including the following steps: Read time series data fields from the sample Input LSTM or GRU, output the hidden state at the last time step. Calling a fully connected layer will Mapped to a single scalar .

[0073] Furthermore, a severity threshold for electricity meter failure is set, and the moment when the severity of the first failure exceeds this threshold is found in subsequent samples. Then the sample Corresponding true remaining lifespan for: ; For the sample The corresponding time window index, The sampling interval is denoted as . That is, a sample The corresponding RUL truth value.

[0074] The mapped single scalar With RUL truth value Backpropagation optimization is performed using mean squared error. The optimized RUL branch is then used for real-time dimensionality reduction features. Perform RUL prediction.

[0075] S3: In one implementation, an adaptive handling strategy is generated based on the fault diagnosis results.

[0076] After the aforementioned S2 step, the output diagnostic results include the fault type. Fault location Severity of the fault and remaining service life To form a quadruple .

[0077] 3.1: In a further implementation, the risk level of the three-phase four-wire energy meter is defined according to the severity of the fault, and it is encapsulated and output together with the quaternion.

[0078] Defining risk levels includes (Low risk, example corresponds) ), (Medium risk, example corresponding) ), (High risk, example provided) ), (Extreme risk, example corresponding) ).

[0079] Then the quadruple Updated to quintuple ; This is an index for risk levels. For example, .

[0080] 3.2: In a further embodiment, in Based on this, domain knowledge and expert experience are introduced to construct a fault handling knowledge graph.

[0081] First, construct the directed graph; this includes defining nodes and edges. The node definition includes: Fault Node : Combinations such as poor contact of phase A contacts; Phenomenon Node Electrical characteristics, such as excessive temperature rise, THD > 8%, three-phase imbalance, etc. Consequence Node Physical consequences, such as insulation breakdown, equipment meltdown, and upstream tripping; Action Node Executable commands, such as tripping the circuit breaker, limiting current, and starting the fan.

[0082] Edge definitions include: Causal edge Symptoms caused by the malfunction For example, poor contact can lead to increased temperature rise. Concurrent edge Symbiotic phenomena For example, THD exceeding the standard Three-phase imbalance; Evolutionary edge If this phenomenon is not eliminated, it will lead to consequences. For example, increased temperature rise leads to insulation breakdown; Suppress edge Actions can eliminate the phenomenon. For example, tripping the circuit breaker can eliminate temperature rise.

[0083] Furthermore, after constructing the directed graph, existing graph reasoning methods (such as graph traversal and path lookup) can be called to deduce the fault chain. The reasoning logic is as follows: and position along Find the connection , and then along Predicting potential Finally passed Reverse the candidate action set.

[0084] In one specific implementation, a hierarchical trigger threshold table is constructed and embedded in the sensing platform to determine the severity of the fault. and remaining service life Find a deterministic handling strategy for matching. Table 1 below provides an example.

[0085] Table 1. Tiered Trigger Residual Threshold Table

[0086] For example, This matches a three-tiered strategy. The corresponding action set for the handling strategy is "rate limiting 80%, bypass switching".

[0087] In one reference implementation, a virtual simulation is performed using a physical model or simulation platform before the disposal strategy is executed to avoid erroneous operations. For example, a simplified thermal balance / electrical lifetime model is used to simulate state transitions and predict the rate of temperature decrease and the severity of new faults after the disposal strategy is executed.

[0088] As an embodiment of this application, a circuit breaker fault state diagnosis system based on a real-time sensing platform is disclosed. Employing the specific implementation method described above for circuit breaker fault state diagnosis, the system includes: The energy meter sensing module is used by the real-time sensing platform to collect various types of data from three-phase four-wire energy meters through an interface, extract time-frequency domain features and construct cross-signal features, and combine the time-frequency domain features and cross-signal features to form basic signal features; based on the numerical continuous signals in the various types of data, it calculates harmonic phase angle features, three-phase correlation features and nonlinear features, and integrates them with the basic signal features to form a unified feature vector. The fault diagnosis module is used to reduce the dimensionality of the unified feature vector and construct a multi-dimensional feature sample including dimensionality-reduced features, fault category, fault location, severity, and time series data. After performing adaptive attention enhancement on the channel dimension of the dimensionality-reduced features, multiple parallel task branches are constructed for joint training, and the fault diagnosis results of the three-phase four-wire energy meter are inferred and output. Based on the time series data, a life prediction branch is constructed to predict the remaining life of the energy meter based on the real-time dimensionality-reduced features and incorporate it into the fault diagnosis results. The fault handling module is used to construct a fault handling knowledge graph based on fault diagnosis results, combined with domain knowledge and expert experience; to deduce the fault chain based on the nodes and edges defined in the graph, and to output a set of candidate actions for fault handling strategies; and to generate deterministic handling strategies by matching the hierarchical trigger threshold table embedded in the real-time perception platform.

[0089] As an embodiment of this application, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded into the processor, it employs the specific implementation described in the above-described open-circuit fault state diagnosis method.

[0090] As an embodiment of this application, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, employs the specific implementation method described above for diagnosing open circuit fault conditions.

[0091] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0092] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0093] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0094] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0095] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.

[0096] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A method for diagnosing open circuit fault status based on a real-time sensing platform, characterized in that, include: The real-time sensing platform collects various types of data from three-phase four-wire energy meters through interfaces, extracts time-frequency domain features and constructs cross-signal features, and combines time-frequency domain features and cross-signal features to form basic signal features; based on numerical continuous signals in the various types of data, it calculates harmonic phase angle features, three-phase correlation features and nonlinear features, and integrates them with basic signal features to form a unified feature vector; The unified feature vector is dimensionality reduced to construct a multi-dimensional feature sample that includes dimensionality-reduced features, fault category, fault location, severity, and time series data. After adaptive attention enhancement of the dimensionality-reduced features at the channel dimension, multiple parallel task branches are constructed for joint training, and the fault diagnosis results of the three-phase four-wire energy meter are inferred and output. Based on the time series data, a lifespan prediction branch is constructed to predict the remaining lifespan of the energy meter based on the real-time dimensionality-reduced features, and the prediction is incorporated into the fault diagnosis results. Based on the fault diagnosis results, a fault handling knowledge graph is constructed by combining domain knowledge and expert experience; based on the nodes and edges defined in the graph, fault chain deduction is performed, and a candidate action set for fault handling strategies is output; combined with the hierarchical trigger threshold table embedded in the real-time perception platform, a deterministic handling strategy is generated.

2. The method for diagnosing open circuit fault status based on a real-time sensing platform according to claim 1, characterized in that, The various data types include numerical continuous signals and event statistical signals. For the numerical continuous signal, time-frequency domain features and waveform features are acquired in the form of a sliding window. The event statistics signal includes event categories and statistical signals, and the event categories are encoded during the preprocessing stage. The event category code is mathematically combined with the corresponding numerical continuous signal to generate cross-signal features.

3. The method for diagnosing open circuit fault status based on a real-time sensing platform according to claim 1, characterized in that, The steps for extracting harmonic phase angle features and three-phase correlation features include: Harmonic distortion rate and phase jump characteristics are calculated based on the collected numerical continuous signals to form harmonic phase angle characteristics; Calculate the three-phase correlation coefficients and phase differences to form the three-phase correlation characteristics.

4. The method for diagnosing open circuit fault status based on a real-time sensing platform according to claim 1, characterized in that, The severity level is a label indicating the continuity severity of the electricity meter fault, calculated using electrical feature mapping. The calculation process includes: When contact resistance is available, the resistance method is used; it is described as the difference between the measured contact resistance and the normal contact resistance, divided by the difference between the calibrated failure resistance and the normal contact resistance. When the contact resistance is not available, it is calculated by mapping electrical characteristic quantities. The description is as follows: the weighted values ​​of the harmonic distortion term, the temperature rise term, and the three mutual correlations are obtained, and then mapped to the range of 0 to 1 through the truncation function. The harmonic distortion term is the increment of harmonic distortion relative to the normal value, the temperature rise term is the temperature rise, and the three mutual correlations are 1 minus the minimum value of the three mutual correlation coefficients.

5. The method for diagnosing open circuit fault status based on a real-time sensing platform according to claim 1, characterized in that, The steps for adaptive attention enhancement along the channel dimension of the dimensionality-reduced features include: A fully connected mapping layer is used to project the dimensionality-reduced features. An SE attention module is then connected in series after the mapping layer to calculate and output the weight vector of the feature dimension after the projection transformation. The features after projection transformation are weighted using a weight vector, and an enhanced feature vector is output.

6. The method for diagnosing open circuit fault status based on a real-time sensing platform according to claim 1, characterized in that, The parallel branches include: The fault location branch maps the enhanced feature vectors to a fault location probability distribution through a fully connected layer and an activation function, and outputs the predicted fault location with the maximum probability. At the same time, the three-phase correlation coefficient matrix is ​​calculated. If the three-phase correlation coefficient corresponding to the predicted fault location does not decrease, it is judged that a physical mechanism conflict has occurred, and a suspected misjudgment is output. The corresponding multi-dimensional feature samples are labeled, and the loss weight of the sample is increased in the next round of training and / or online learning.

7. The method for diagnosing open circuit fault status based on a real-time sensing platform according to claim 1, characterized in that, The lifetime prediction branch includes: Read the time series data field from the current sample and map it to a single scalar of remaining lifetime through a time series network; Set a severity threshold for electricity meter failure, and then search for the moment when the severity of the first failure exceeds the severity threshold in subsequent multi-dimensional feature samples. The current sample's true remaining lifetime is... Subtract the current sample time index Then multiply by the sampling interval; The single scalar and the true remaining lifetime are optimized by backpropagation of the lifetime prediction branch using mean square error; the optimized lifetime prediction branch is then used to predict the remaining lifetime of the real-time dimensionality-reduced features.

8. The method for diagnosing open circuit fault status based on a real-time sensing platform according to claim 1, characterized in that, The fault diagnosis results include: Fault type, fault location, fault severity, and remaining service life; define the fault risk level based on the fault severity.

9. The method for diagnosing open circuit fault status based on a real-time sensing platform according to claim 1, characterized in that, The fault handling knowledge graph is a directed graph, including: Nodes include fault nodes, phenomenon nodes, consequence nodes, and action nodes; edge definitions include causal edges, concurrent edges, evolutionary edges, and inhibition edges. The causal edges are used to describe the phenomena caused by the fault, the concurrent edges are used to describe the symbiosis of phenomena, the evolution is used to describe the consequences caused by the phenomena, and the inhibition edges are used to describe the elimination of phenomena by actions.

10. A circuit fault state diagnosis system based on a real-time sensing platform, executing the circuit fault state diagnosis method as described in any one of claims 1-9, characterized in that, The system includes: The energy meter sensing module is used by the real-time sensing platform to collect various types of data from three-phase four-wire energy meters through an interface, extract time-frequency domain features and construct cross-signal features, and combine the time-frequency domain features and cross-signal features to form basic signal features; based on the numerical continuous signals in the various types of data, it calculates harmonic phase angle features, three-phase correlation features and nonlinear features, and integrates them with the basic signal features to form a unified feature vector. The fault diagnosis module is used to reduce the dimensionality of the unified feature vector and construct a multi-dimensional feature sample including dimensionality-reduced features, fault category, fault location, severity, and time series data. After performing adaptive attention enhancement on the channel dimension of the dimensionality-reduced features, multiple parallel task branches are constructed for joint training, and the fault diagnosis results of the three-phase four-wire energy meter are inferred and output. Based on the time series data, a life prediction branch is constructed to predict the remaining life of the energy meter based on the real-time dimensionality-reduced features and incorporate it into the fault diagnosis results. The fault handling module is used to construct a fault handling knowledge graph based on fault diagnosis results, combined with domain knowledge and expert experience; to deduce the fault chain based on the nodes and edges defined in the graph, and to output a set of candidate actions for fault handling strategies; and to generate deterministic handling strategies by matching the hierarchical trigger threshold table embedded in the real-time perception platform.

11. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is loaded into the processor, it implements the open circuit fault state diagnosis method according to any one of claims 1-9.

12. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the circuit breaker fault state diagnosis method according to any one of claims 1-9.

Images