Primary and secondary fused circuit breaker state evaluation method, system, device and storage medium

Abstract

The present application relates to the technical field of circuit breaker state evaluation, and discloses a primary and secondary fusion circuit breaker state evaluation method, system, device and storage medium, which comprises the following steps: feature extraction is performed on a fault traveling wave signal to obtain a traveling wave feature; according to a tripping action time sequence, the tripping time is calculated; the traveling wave feature, the tripping time and a fault current parameter are spliced to obtain an enhanced feature vector; according to the difference between the enhanced feature vector and a reference vector, the health index at the current fault moment is calculated; according to the fault current parameter, the physically driven predicted residual life is calculated; the health index at the historical fault moment and the health index at the current fault moment are input into a residual life prediction model to obtain the data-driven predicted residual life; and the physically driven predicted residual life and the data-driven predicted residual life are weighted and fused to obtain the final predicted residual life. The present application can accurately predict the mechanical health state and the residual life of the circuit breaker, and improve the operation and maintenance efficiency of the power distribution system.

Description

Technical Field

[0001] This invention relates to the field of circuit breaker condition assessment technology, and in particular to a method, system, device and storage medium for condition assessment of integrated primary and secondary circuit breakers. Background Technology

[0002] The integrated primary and secondary pole-mounted circuit breaker is a core device in the distribution network for fault isolation and power restoration. The health of its mechanical operating mechanism directly determines the reliability of the circuit breaker's opening and closing, thus affecting the power supply safety and automation level of the distribution network. However, existing technologies for monitoring and evaluating the mechanical condition of circuit breakers have several significant shortcomings. Firstly, traditional mechanical characteristic monitoring methods require the additional installation of displacement sensors, accelerometers, Hall effect sensors, or angle encoders on the circuit breaker's operating mechanism. These sensors not only increase equipment cost and installation complexity but also raise concerns about their reliability under harsh outdoor environments (high temperature, high humidity, vibration) over long periods. Furthermore, sensor failures can introduce new monitoring blind spots. Secondly, conventional mechanical characteristic tests of circuit breakers are only conducted during power outages and maintenance, using dedicated testing equipment to simulate opening and closing operations. This offline testing cannot reflect the dynamic response of the circuit breaker during actual short-circuit fault opening. Therefore, offline test results are often overly optimistic and cannot accurately assess the true health level of the equipment.

[0003] Furthermore, most existing technologies can only achieve post-fault condition assessment or set a fixed number of tripping life indicators. This one-size-fits-all life management approach ignores individual equipment differences and cannot accurately predict the lifespan degradation trend of mechanical components, resulting in insufficient precision in the formulation of operation and maintenance strategies. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a method, system, device, and storage medium for assessing the condition of a primary and secondary integrated circuit breaker.

[0005] In a first aspect, the present invention provides a method for assessing the condition of a primary and secondary integrated circuit breaker, the method comprising: Acquire the fault traveling wave signal, tripping action timing, and fault current parameters of the primary and secondary integrated circuit breaker during fault tripping. Feature extraction is performed on the fault traveling wave signal to obtain traveling wave features. The tripping time is calculated according to the tripping action sequence. The traveling wave features, the tripping time, and the fault current parameters are concatenated to obtain an enhanced feature vector. The health index at the current fault moment is calculated based on the difference between the enhanced feature vector and the preset benchmark vector. Based on the fault current parameters, a physical model based on accumulated arc energy is used to calculate the predicted remaining life of the physical drive of the primary and secondary fused circuit breakers. The health index at the time of historical failure and the health index at the time of current failure are combined to form a health index sequence, which is then input into a preset remaining life prediction model to obtain the data-driven prediction of the remaining life of the primary and secondary fusion circuit breaker. The remaining life prediction model is constructed based on a convolutional neural network model. The predicted remaining lifetime of the physical-driven circuit breaker and the predicted remaining lifetime of the data-driven circuit breaker are weighted and fused to obtain the final predicted remaining lifetime of the primary and secondary fused circuit breaker.

[0006] Further, the step of extracting features from the fault traveling wave signal to obtain traveling wave features includes: The fault traveling wave signal is decomposed by wavelet packet to obtain high-frequency transient components, and features are extracted from the high-frequency transient components to obtain physical features, including high-frequency maximum features, high-frequency duration features, and energy percentage features. The fault traveling wave signal is normalized, and the processed sequence is input into a pre-trained feature extraction model to obtain deep features. The feature extraction model is constructed based on a one-dimensional convolutional neural network model. The physical features and the deep features are used as traveling wave features.

[0007] Further, the steps of performing wavelet packet decomposition on the fault traveling wave signal to obtain high-frequency transient components, and extracting features from the high-frequency transient components to obtain physical features include: The fault traveling wave signal is decomposed by wavelet packet decomposition to obtain multiple equal-width frequency band subspaces. In each frequency band subspace, the decomposition node coefficients with the center frequency located within a preset frequency band are selected for reconstruction to obtain high-frequency transient components. The maximum value among the absolute values ​​of the high-frequency transient components is taken as the high-frequency maximum feature; The time length of the high-frequency transient component from the starting point to the envelope decaying to the peak value by a preset proportion is used as the high-frequency duration feature. The proportion of the energy of the high-frequency transient component in the total energy of the fault traveling wave signal is used as the energy proportion feature. The high-frequency maximum value feature, the high-frequency duration feature, and the energy percentage feature are used as physical features.

[0008] Furthermore, the feature extraction model includes an input layer, several one-dimensional convolutional blocks, and a global average pooling layer connected in series. Each one-dimensional convolutional block contains a convolutional layer, a ReLU activation function, and a max pooling layer with a stride of 2.

[0009] Furthermore, the step of calculating the remaining physical drive prediction lifetime of the primary and secondary fused circuit breakers based on the fault current parameters and using a physical model based on accumulated arc energy includes: Calculate the energy of a single arc based on the fault current amplitude and arc duration; Based on the single arc energy, the historical cumulative arc energy is updated to obtain the updated cumulative arc energy; Based on the accumulated arc energy, a physical drive prediction of the remaining life of the primary and secondary fused circuit breakers is calculated using a physical model based on exponential decay.

[0010] Furthermore, the step of combining the health index at the historical fault time with the health index at the current fault time to form a health index sequence, and inputting it into a preset remaining life prediction model to obtain the data-driven prediction of the remaining life of the primary and secondary fusion circuit breaker includes: The health index at the time of the historical failure and the health index at the time of the current failure are combined to form a health index sequence, which is then input into a preset remaining life prediction model to obtain the predicted value of the health index at the next time step. The remaining life prediction model is constructed based on a long short-term memory neural network model. Based on the preset lifespan mapping function, the predicted value of the health index at the next moment is mapped to obtain the data-driven prediction of the remaining lifespan of the primary and secondary integrated circuit breakers.

[0011] Furthermore, after the step of obtaining the final predicted remaining life of the primary and secondary fusion circuit breakers, the method further includes: Based on the health index at the current time of failure and the final predicted remaining lifespan, a graded early warning is generated according to a preset grading rule. The graded early warning includes a mild warning, a moderate warning, and an emergency warning. Based on the aforementioned tiered early warning system, maintenance recommendations for the primary and secondary integrated circuit breakers are generated.

[0012] Secondly, the present invention provides a primary and secondary integrated circuit breaker condition assessment system, the system comprising: The data processing module is used to acquire the fault traveling wave signal, tripping action timing and fault current parameters of the primary and secondary integrated circuit breaker during fault tripping. Feature extraction is performed on the fault traveling wave signal to obtain traveling wave features. The tripping time is calculated according to the tripping action sequence. The traveling wave features, the tripping time, and the fault current parameters are concatenated to obtain an enhanced feature vector. The health index at the current fault moment is calculated based on the difference between the enhanced feature vector and the preset benchmark vector. The physical drive prediction module is used to calculate the remaining physical drive prediction life of the primary and secondary fusion circuit breakers based on the fault current parameters and a physical model based on accumulated arc energy. The data-driven prediction module is used to combine the health index at the time of historical failure with the health index at the time of current failure to form a health index sequence, and input it into a preset remaining life prediction model to obtain the data-driven predicted remaining life of the primary and secondary fusion circuit breaker. The remaining life prediction model is constructed based on a convolutional neural network model. The final life prediction module is used to perform weighted fusion of the physical-driven predicted remaining life and the data-driven predicted remaining life to obtain the final predicted remaining life of the primary and secondary fusion circuit breaker.

[0013] Thirdly, embodiments of the present invention also provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method.

[0014] Fourthly, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described method.

[0015] This invention provides a method, system, device, and storage medium for primary and secondary integrated circuit breaker condition assessment. It automatically learns the deep implicit features of traveling wave signals through a one-dimensional convolutional network model, overcoming the limitations of manual feature design relying on expert experience and improving assessment accuracy. By combining physical and data-driven approaches for remaining lifetime prediction, it combines physical interpretability with the high precision of data-driven methods, thereby improving prediction accuracy. Through accurate prediction of the mechanical health status and remaining lifetime of circuit breakers, this invention can improve the operation and maintenance efficiency of power distribution automation systems. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the primary and secondary fusion circuit breaker status assessment method in an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of the primary and secondary integrated circuit breaker condition assessment system in an embodiment of the present invention; Figure 3 This is an internal structural diagram of the computer device in an embodiment of the present invention.

[0017] Figure label: 10. Data processing module; 20. Physics-driven prediction module; 30. Data-driven prediction module; 40. Final life prediction module. Detailed Implementation

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

[0019] Please see Figure 1 The first embodiment of the present invention proposes a method for assessing the condition of a primary and secondary integrated circuit breaker, comprising steps S10 to S60: Step S10: Obtain the fault traveling wave signal, tripping action sequence and fault current parameters of the primary and secondary fusion circuit breaker during fault tripping. Step S20: Extract features from the fault traveling wave signal to obtain traveling wave features; calculate the tripping time according to the tripping action sequence; and concatenate the traveling wave features, the tripping time, and the fault current parameters to obtain an enhanced feature vector. Step S30: Calculate the health index at the current fault moment based on the difference between the enhanced feature vector and the preset benchmark vector; Step S40: Based on the fault current parameters, use a physical model based on accumulated arc energy to calculate the predicted remaining life of the physical drive of the primary and secondary fusion circuit breakers. Step S50: Combine the health index at the time of the historical fault with the health index at the time of the current fault to form a health index sequence, and input it into a preset remaining life prediction model to obtain the data-driven prediction of the remaining life of the primary and secondary fusion circuit breaker. The remaining life prediction model is constructed based on a convolutional neural network model. Step S60: The predicted remaining lifetime of the physical drive and the predicted remaining lifetime of the data drive are weighted and fused to obtain the final predicted remaining lifetime of the primary and secondary fused circuit breaker.

[0020] In this embodiment, the integrated primary and secondary circuit breaker incorporates a wideband current sensor (bandwidth covering power frequency to megahertz) and is equipped with a feeder terminal unit (FTU). This sensor is originally used for traveling wave fault location. The FTU, as the edge computing core, possesses synchronous data acquisition, digital signal processing, and lightweight deep learning inference capabilities. When a short-circuit fault in the distribution network causes the integrated primary and secondary circuit breaker to trip, multi-source signals are acquired through the feeder terminal unit. These multi-source signals include the fault traveling wave signal, the tripping action timing, and fault current parameters. Here, the fault traveling wave signal refers to the traveling wave signal under fault tripping conditions in the distribution network, and the fault current parameters include the fault current amplitude and arc duration.

[0021] Specifically, the FTU achieves microsecond-level time synchronization between the traveling wave acquisition board and the main control CPU through an in-station clock synchronization protocol (such as IEEE 1588 or IRIG-B), ensuring that the traveling wave signal, tripping command, and auxiliary contact status are aligned under the same time base. When a fault trip occurs, the feeder terminal unit continuously acquires the three-phase current at a preset high sampling rate (e.g., not less than 5MHz) using a wideband current sensor (e.g., bandwidth of 0.1Hz to 1MHz) built into the primary and secondary fusion circuit breaker. When a sudden change in the current traveling wave is detected (e.g., the rate of change of current exceeds a threshold or the instantaneous value change exceeds the set value), a signal segment is extracted from the period before the trigger (covering the power frequency half-wave) to the period after the trigger (covering the entire arc extinction process). The length of each phase signal is fixed to the preset number of sampling points (corresponding to a millisecond-level time window). This segment fully contains the high-frequency arc transient component generated at the moment of contact separation (center frequency in the tens to hundreds of kilohertz band), which contains rich mechanical information about the contact separation process; the feeder terminal unit records the moment the circuit breaker tripping command is issued and the moment the auxiliary contacts separate through the digital input channel, forming the tripping action sequence; at the same time, the feeder terminal unit extracts the fault current amplitude I from the power frequency half-cycle data before the traveling wave triggering. fault (RMS or peak value) Simultaneously, taking the starting point of the traveling wave (arc ignition) as the starting point and the attenuation of the high-frequency components to the background noise level as the ending point, the arc duration is obtained. .

[0022] Then, the tripping time is calculated based on the tripping action sequence. Specifically, the tripping time is obtained by subtracting the time when the circuit breaker tripping command is issued from the time when the auxiliary contacts separate. This parameter directly reflects the response speed of the operating mechanism (such as the trip unit, transmission link, and spring mechanism), and its gradual increase indicates mechanical degradation such as spring fatigue and transmission jamming.

[0023] In this embodiment, parallel feature extraction is performed on the fault traveling wave signal, and the specific steps include: The fault traveling wave signal is decomposed by wavelet packet to obtain high-frequency transient components, and features are extracted from the high-frequency transient components to obtain physical features, including high-frequency maximum features, high-frequency duration features, and energy percentage features. The fault traveling wave signal is normalized, and the processed sequence is input into a pre-trained feature extraction model to obtain deep features. The feature extraction model is constructed based on a one-dimensional convolutional neural network model. The physical features and the deep features are used as traveling wave features.

[0024] This embodiment employs two parallel feature extraction paths to extract features from the fault traveling wave signal. These parallel paths are physical feature extraction based on wavelet packet decomposition and deep feature extraction based on a one-dimensional convolutional neural network model (1D-CNN). The specific steps of physical feature extraction based on wavelet packet decomposition include: The fault traveling wave signal is decomposed by wavelet packet decomposition to obtain multiple equal-width frequency band subspaces. In each frequency band subspace, the decomposition node coefficients with the center frequency located within a preset frequency band are selected for reconstruction to obtain high-frequency transient components. The maximum value among the absolute values ​​of the high-frequency transient components is taken as the high-frequency maximum feature; The time length of the high-frequency transient component from the starting point to the envelope decaying to the peak value by a preset proportion is used as the high-frequency duration feature. The proportion of the energy of the high-frequency transient component in the total energy of the fault traveling wave signal is used as the energy proportion feature. The high-frequency maximum value feature, the high-frequency duration feature, and the energy percentage feature are used as physical features.

[0025] In this embodiment, wavelet packet decomposition is performed on each phase traveling wave signal segment in the fault traveling wave signal. A wavelet basis function with a suitable vanishing moment (such as the Daubechies 4 wavelet) is selected, and the number of decomposition layers (usually 5 to 8 layers) is determined based on the sampling frequency and the high-frequency band of interest (e.g., 50kHz to 500kHz). Wavelet packet decomposition divides the fault traveling wave signal into multiple equal-width frequency band subspaces. Then, the center frequency of each subband is calculated, and all nodes falling within the high-frequency band are selected. Inverse wavelet packet transform is performed on the selected subband coefficients to obtain the time-domain reconstructed signal, i.e., the high-frequency transient component.

[0026] Then, multiple features related to physical properties are extracted from the time series of high-frequency transient components. Specifically, the maximum absolute value of each element in the sequence is taken as the high-frequency maximum feature. This characteristic reflects the concentration of arc energy at the moment of contact separation; after contact ablation, the arc disperses. The frequency exhibits a decreasing trend; the duration of the high-frequency duration is taken as the time from the starting point (when the amplitude first exceeds the noise threshold) to the point where the envelope decays to a certain proportion (e.g., 1 / e) of the peak value. This characteristic, the effective duration of the high-frequency component, reflects the contact separation speed. When the opening speed decreases due to mechanism jamming or spring fatigue... Extend; calculate the percentage of the energy of the high-frequency transient components (amplitude squared over time) to the total energy of the original fault traveling wave, as the energy proportion characteristic. This characteristic reflects the efficiency of energy transfer within the mechanism; aging leads to a decrease in energy transfer efficiency and an increase in high-frequency energy loss. The frequency decreases. Finally, the high-frequency maximum characteristics, high-frequency duration characteristics, and energy percentage characteristics obtained above are used as physical characteristics.

[0027] The deep feature extraction, which is performed in parallel with the physical feature extraction mentioned above, adopts a feature extraction model based on a one-dimensional convolutional neural network model (1D-CNN). Specifically, the fault traveling wave signal is normalized to obtain an input sequence with zero mean and unit variance, and then the normalized sequence is input into the pre-trained feature extraction model. This model is constructed using a one-dimensional convolutional neural network (1D-CNN). Its network structure consists of an input layer, multiple one-dimensional convolutional blocks, and a global average pooling layer. The input layer receives a fixed-length one-dimensional signal, i.e., the preprocessed sequence. Multiple one-dimensional convolutional blocks are connected in series. Each convolutional block contains a convolutional layer, a ReLU activation function, and a max pooling layer with a stride of 2. The kernel size of each convolutional layer decreases and the number of channels increases layer by layer. Preferably, the feature extraction model uses three one-dimensional convolutional blocks. The kernel sizes of the convolutional layers in each convolutional block are 64, 32, and 16, and the number of channels are 32, 64, and 128, respectively. Each layer is followed by a ReLU activation function and a max pooling layer with a stride of 2. Finally, a 128-dimensional deep feature vector is output after global average pooling. In this embodiment, the feature extraction model is pre-trained using data constructed based on historical fault traveling wave signals. The samples are historical fault traveling wave signals, and the labels are the actual mechanical state of the circuit breaker, including healthy state and unhealthy state caused by mechanical degradation. Labels such as slight degradation, moderate degradation, and severe degradation can be set according to the degree of degradation. The loss function is cross-entropy loss. Through the pre-trained feature extraction model, deep features related to mechanical degradation can be extracted from the fault traveling wave signals. These deep features contain mechanical degradation information, such as subtle morphological changes in contact bounce, transient characteristics of arc reignition, and waveform distortion caused by nonlinearity of the transmission mechanism. This information is difficult to fully describe using manually designed wavelet packet features.

[0028] The physical and deep features extracted by the parallel branch are used as traveling wave features and concatenated with the opening time, fault current amplitude, and arc duration to obtain an enhanced feature vector. For ease of subsequent health index calculation, the enhanced feature vector in this embodiment is a normalized feature vector. Then, based on the difference between the enhanced feature vector and the preset benchmark vector, the health index of the circuit breaker at the current fault moment is calculated. Specifically, the traveling wave data collected under fault opening scenarios is used as the benchmark sample. The circuit breaker at the corresponding moment of the benchmark sample is a circuit breaker in a non-degradable state, belonging to the initial stage of equipment commissioning or after the replacement of core components (contacts, opening springs, transmission mechanisms). Its mechanical characteristics are all within the factory nominal qualified range. The number of benchmark samples is between 3 and 10. The larger the number of benchmark samples, the more stable the benchmark vector. Generally, 5 measurements are sufficient to meet the accuracy requirements.

[0029] For the reference vector corresponding to the deep features in the enhanced feature vector, the reference samples are input into the pre-trained feature extraction model to obtain the deep features of each reference sample. The median of all reference samples is then taken as the reference value for each dimension. For other non-deep features, the physical features, tripping time, and fault current parameters corresponding to the reference samples are extracted following the same feature extraction steps. Similarly, the median of all reference samples is taken as the reference value for each feature dimension. The median is far more robust to accidental signal interference and outliers than the arithmetic mean. Furthermore, to account for the dimensional differences in the features, the feature values ​​of the reference vector need to be normalized.

[0030] Then, the weighted Euclidean distance is used to calculate the similarity between the enhanced feature vector and the baseline vector. The greater the distance, the less similar the current state of the circuit breaker is to the healthy state, that is, the less healthy the mechanical state of the circuit breaker is. Therefore, in this embodiment, the similarity is used as the health index. The health index can be expressed as 1 minus the calculated weighted Euclidean distance. The value of the health index is between 0 and 1. The closer it is to 1, the healthier the mechanical state is. It should be noted that the feature weights used in calculating the weighted Euclidean distance for the enhanced feature vector are pre-set weights. Since deep features have many dimensions, the weights are first divided into deep and non-deep features. For example, the total proportion of deep features is 0.65, and the total proportion of other features is 0.35. Then, the total proportion is allocated to each feature dimension. Since non-deep features only have 6 dimensions, weights can be manually set using expert experience. For deep features, the initial weights of the multi-dimensional features obtained during the pre-training phase of the feature extraction model are used. These initial weights are then normalized to obtain the relative weights within each deep feature. Finally, each internal relative weight is multiplied by the total proportion of the deep feature to obtain the final global weight for each deep feature dimension.

[0031] In this embodiment, in addition to feature extraction and calculation of the circuit breaker's health index at the current moment, the collected multi-source data also needs to calculate the circuit breaker's remaining lifetime. Furthermore, to accurately predict the circuit breaker's remaining lifetime, this embodiment employs a physical data hybrid-driven approach for remaining lifetime prediction. The specific steps of the physical-driven remaining lifetime prediction include: Calculate the energy of a single arc based on the fault current amplitude and arc duration; Based on the single arc energy, the historical cumulative arc energy is updated to obtain the updated cumulative arc energy; Based on the accumulated arc energy, a physical drive prediction of the remaining life of the primary and secondary fused circuit breakers is calculated using a physical model based on exponential decay.

[0032] In this embodiment, a physical model based on accumulated arc energy is used for remaining lifetime prediction. In this physical model, the energy of a single arc is first calculated based on the amplitude of the fault current and the duration of the arc collected in this study. Specifically, in the prediction of the remaining lifetime of the circuit breaker, the arc energy is usually defined by the integral of voltage, current, and time, i.e., arc energy. Where u(t) is the arc voltage at time t, and i(t) is the arc current at time t. However, in practical engineering, the arc voltage waveform is complex, and its amplitude is affected by the contact spacing, the arc-extinguishing medium, and the arc morphology, making it difficult to measure directly and accurately; while the fault current amplitude... and arc duration This value can then be reliably obtained through a current transformer. Therefore, this embodiment employs a simplified physical model based on the square of the current, the principle of which is as follows: During short-circuit tripping, the arc equivalent resistance Although exhibiting nonlinear characteristics, numerous experiments have shown that for the same type of circuit breaker under similar fault conditions, there is an approximate proportional relationship between the arc voltage and current during the arcing process: ,in, Let be the experimental fitting constant. The instantaneous arc power at time t is... Since the fault current waveform is approximately sinusoidal and its amplitude is relatively stable, the integral yields... Considering the product of the square of the effective current value or amplitude and time, in practical engineering, the proportionality coefficient is further combined into k. .in Assuming the waveform factor, the simplified formula is obtained: In the formula, The value represents the energy of a single electric arc, and k is a proportionality coefficient, obtained by fitting short-circuit tripping test data of the same type of circuit breaker. The fault current amplitude, This represents the duration of the electric arc.

[0033] Then the energy of a single electric arc is added to the previously accumulated electric arc energy: In the formula, To accumulate arc energy, the initial value is zero if no fault trip has occurred before; otherwise, the arc energy of each trip is recorded and added together to obtain the accumulated arc energy before the current fault time.

[0034] Since arc energy and circuit breaker life are strongly correlated, this embodiment uses arc energy to predict remaining life. Furthermore, because the relationship between arc energy and circuit breaker life is not linear, this embodiment uses an exponential function to predict remaining life. In the formula, Predict the remaining lifetime for the physical drive at time t. For the initial rated mechanical life, λ The attenuation coefficient is determined through accelerated aging tests in the laboratory.

[0035] In this embodiment, the data-driven remaining lifetime prediction uses a neural network model, preferably a long short-term memory neural network model, to construct the remaining lifetime prediction model. The specific prediction steps include: The health index at the time of the historical failure and the health index at the time of the current failure are combined to form a health index sequence, which is then input into a preset remaining life prediction model to obtain the predicted value of the health index at the next time step. The remaining life prediction model is constructed based on a long short-term memory neural network model. Based on the preset lifespan mapping function, the predicted value of the health index at the next moment is mapped to obtain the data-driven prediction of the remaining lifespan of the primary and secondary integrated circuit breakers.

[0036] In this embodiment, a two-layer Long Short-Term Memory (LSTM) neural network is used to construct the remaining lifetime prediction model. The input data of this model is a health index sequence. The model uses two LSTM layers to capture long-term dependencies in the sequence data. The first LSTM layer returns the output of all time steps, providing complete sequence information for the second LSTM layer. The final fully connected layer is used to generate the final output of the model, namely the health index at the next time step. Specifically, the health index calculated at the current time step is combined with the historical health indices at previously stored historical fault times to form a health index sequence, which is then input into the pre-trained remaining lifetime prediction model to obtain the predicted value of the health index at the next time step. Then, the predicted value of the health index at the next time step is converted into the data-driven remaining lifetime prediction value L of the circuit breaker at the current fault time step through a pre-fitted mapping function.LSTM (t). It should be noted that the remaining life prediction model adopts a typical two-layer LSTM model architecture. The training dataset consists of health indices obtained from actual operation of the same type of circuit breaker and health indices generated by experimental simulation. The model structure and training process will not be described in detail here.

[0037] Furthermore, since the input data for the remaining lifetime prediction model uses historical health indices, there may be situations where historical health indices are not yet stored or the stored historical health indices do not meet the sequence window requirements. Therefore, in this embodiment, when there are not enough historical health indices, such as when the number of fault trips is less than 3, the remaining lifetime calculation result of the physical model is directly used as the final predicted remaining lifetime. If the stored historical health indices still do not meet the sequence window requirements, a variable-length sequence input is used, and gaps are masked through a masking layer, or time-series sample enhancement is used to slightly perturb and enhance the stored historical health indices, generating multiple simulated samples that conform to the degradation trend to fill the window. In addition, the mapping function is established offline based on the relationship between the health index and the corresponding remaining lifetime in the historical samples, using methods such as piecewise linear interpolation or support vector regression. The construction steps of the mapping function will not be described in detail here.

[0038] The reason this embodiment uses a predicted health index instead of a health index calculated at the current moment for remaining life mapping is that the mechanical performance degradation of circuit breakers is a continuous, trend-driven, and non-linear temporal evolution process. A single-moment health index only reflects the current static state and cannot reflect the rate of degradation or future trends. Direct mapping would result in a large error in lifespan. For example, two circuit breakers of the same model may have the same current health index, but different operating times. Therefore, the remaining lifespan predicted using the current health index cannot reflect the difference in their degradation rates. Furthermore, circuit breakers degrade slowly during the initial commissioning and stable operation periods, but accelerate towards the end of their lifespan, with the aging rate increasing exponentially. A single-moment health index cannot determine whether the current stage is slow degradation or has entered an accelerated degradation stage, leading to generally optimistic lifespan mappings and potential false alarms. Therefore, this embodiment uses an LSTM model to automatically extract implicit features such as degradation rate and acceleration from the historical trajectory of the health index, thereby accurately predicting the health index at the next moment and achieving accurate assessment of remaining lifespan.

[0039] After obtaining the physical-driven predicted remaining lifetime and the data-driven predicted remaining lifetime, the two are weighted and summed to obtain the final predicted remaining lifetime of the primary and secondary fused circuit breaker: In the formula, The final predicted remaining lifetime at time t, This is the fusion coefficient.

[0040] It also outputs the prediction uncertainty interval (such as the 95% confidence interval calculated based on the error variance of the LSTM model) for operation and maintenance personnel to refer to.

[0041] In a preferred embodiment, an adaptive weighted fusion method is used to predict the remaining lifetime of both physical-driven and data-driven predictions, defining an adaptive fusion coefficient. Its initial value is set to 0.6 (biased towards the physical model), and it is dynamically adjusted through Bayesian online learning as the number of actual fault trips increases. When the amount of data is sufficient and the LSTM prediction error converges, The weight can be gradually reduced to 0.3–0.4, giving the data-driven model a higher weight. Specifically, in the dynamic adjustment phase of Bayesian online learning, the Beta distribution is selected as the prior / posterior distribution of α. First, an initial prior distribution is set. After each fault trip, the true label of the current prediction is obtained to judge the prediction accuracy of the two models. The true label can be obtained by mapping the actual health index calculated by multi-feature fusion after the current trip through a mapping function. Then, the prediction error of the two models is calculated based on the true label. Based on the error comparison results of the two models, the parameters of the Beta distribution are incrementally updated. If the physical model predicts more accurately in the current observation, the confidence parameter of the physical model is increased by 1. If the LSTM model predicts more accurately in the current observation, the confidence parameter of the data-driven model is increased by 1. If the error difference between the two is less than 5%, the prediction accuracy of the two is considered to be equivalent, and the parameters are not updated to avoid meaningless fluctuations. After the update, the new fusion coefficient α is the mean of the posterior distribution, and a 95% confidence interval is output synchronously with the weight update for operation and maintenance reference.

[0042] Furthermore, to allow the model to gradually adapt to the unique degradation trajectory of the current circuit breaker, lightweight incremental learning is performed after each fault trip. The acquired multi-source signals, enhanced feature vectors, and health index are stored in a circular buffer. Model fine-tuning is triggered every preset number of new samples or when the health index changes beyond a certain range. For 1D-CNN, all parameters of the first two convolutional blocks are frozen (these parameters learn general traveling wave features), and only the parameters of the last convolutional block and the fully connected layer after the global average pooling layer are updated. For LSTM, the parameters of the first LSTM layer are frozen, and only the parameters of the second LSTM layer and the output fully connected layer are updated. A small learning rate (e.g., 0.001) and a single iteration (1 epoch) are used, and the batch size is dynamically set according to the number of cached samples. The fine-tuned new model parameters overwrite the old model file, while retaining the most recent 2-3 versions for rollback.

[0043] Finally, based on the health index and the final predicted remaining lifespan, a tiered early warning and maintenance recommendation are generated. The specific steps include: Based on the health index at the current time of failure and the final predicted remaining lifespan, a graded early warning is generated according to a preset grading rule. The graded early warning includes a mild warning, a moderate warning, and an emergency warning. Based on the aforementioned tiered early warning system, maintenance recommendations for the primary and secondary integrated circuit breakers are generated.

[0044] In this embodiment, tiered early warnings and maintenance recommendations are issued based on the health index and the threshold range to which the final predicted remaining lifespan falls. These tiered early warnings include mild, moderate, and emergency warnings. A mild warning is issued when the health index is within the first index threshold range (e.g., 0.5-0.6), and it is recommended to add mechanical characteristic testing to the next power outage plan. A moderate warning is issued when the health index is below the minimum value of the first index threshold range (e.g., below 0.5) and the final predicted remaining lifespan is above the lifespan threshold, and it is recommended to schedule maintenance within the next month, focusing on checking the trip spring, transmission linkage, and contacts. An emergency warning is issued when the final predicted remaining lifespan is below the lifespan threshold, and it is recommended to immediately replace the circuit breaker or perform a major overhaul. Furthermore, early warnings can also be issued in conjunction with accumulated arc energy. For example, when the accumulated arc energy is close to the energy threshold (obtained from factory tests or historical data) and the health index is below the minimum value of the first index threshold range, a moderate warning is issued, indicating that both the electrical and mechanical lifespans of the contacts are nearing the end, and recommending complete replacement.

[0045] Finally, the early warning information, health index, final predicted remaining lifespan, fault location results (synchronously output based on the traveling wave ranging principle), and model version number are sent to the distribution automation master station through standard communication protocols. At the same time, all raw data and intermediate results are stored locally in compressed format for subsequent offline analysis.

[0046] This embodiment provides a condition assessment method for integrated primary and secondary circuit breakers. This embodiment fully reuses the existing traveling wave ranging broadband current sensor and FTU of the integrated primary and secondary circuit breaker, eliminating the need for any additional displacement, acceleration, or other sensors, significantly reducing the deployment cost of condition monitoring. Based on the traditional wavelet packet decomposition for extracting high-frequency component features, this embodiment automatically learns the deep implicit features of the traveling wave signal through a one-dimensional convolutional network model, overcoming the limitations of manual feature design relying on expert experience and improving assessment accuracy. The remaining lifetime prediction is performed using a combination of physical and data-driven methods, combining physical interpretability with the high precision of data-driven methods, thus improving prediction accuracy. This embodiment, through accurate prediction of the mechanical health status and remaining lifetime of the circuit breaker, can improve the operation and maintenance efficiency of the power distribution automation system.

[0047] Please see Figure 2 Based on the same inventive concept, the second embodiment of this invention proposes a primary and secondary integrated circuit breaker condition assessment system, comprising: Data processing module 10 is used to acquire the fault traveling wave signal, tripping action timing and fault current parameters of the primary and secondary integrated circuit breaker during fault tripping. Feature extraction is performed on the fault traveling wave signal to obtain traveling wave features. The tripping time is calculated according to the tripping action sequence. The traveling wave features, the tripping time, and the fault current parameters are concatenated to obtain an enhanced feature vector. The health index at the current fault moment is calculated based on the difference between the enhanced feature vector and the preset benchmark vector. The physical drive prediction module 20 is used to calculate the remaining physical drive prediction life of the primary and secondary fusion circuit breakers based on the fault current parameters and a physical model based on accumulated arc energy. The data-driven prediction module 30 is used to combine the health index at the time of historical failure with the health index at the time of current failure to form a health index sequence, and input it into a preset remaining life prediction model to obtain the data-driven predicted remaining life of the primary and secondary fusion circuit breaker. The remaining life prediction model is constructed based on a convolutional neural network model. The final life prediction module 40 is used to perform weighted fusion of the physical-driven predicted remaining life and the data-driven predicted remaining life to obtain the final predicted remaining life of the primary and secondary fusion circuit breaker.

[0048] The technical features and effects of the primary and secondary integrated circuit breaker condition assessment system proposed in this invention are the same as those of the method proposed in this invention, and will not be repeated here. Each module in the above-mentioned primary and secondary integrated circuit breaker condition assessment system can be implemented entirely or partially through software, hardware, or a combination thereof. Each module can be embedded in or independent of the processor in a computer device in hardware form, or it can be stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0049] Furthermore, embodiments of the present invention also propose a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method.

[0050] Please see Figure 3The diagram illustrates the internal structure of a computer device in one embodiment. This computer device can specifically be a terminal or a server. The computer device includes a processor, memory, network interface, display, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface of the computer device is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for assessing the state of a primary and secondary fusion circuit breaker. The display screen of the computer device can be a liquid crystal display (LCD) or an e-ink display. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0051] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computing devices may include more or fewer components than those shown in the figure, or combine certain components, or have the same component arrangement.

[0052] Furthermore, embodiments of the present invention also propose a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described method.

[0053] In summary, the embodiments of this invention propose a method, system, device, and storage medium for assessing the condition of a primary and secondary integrated circuit breaker. The method acquires the fault traveling wave signal, tripping action sequence, and fault current parameters of the primary and secondary integrated circuit breaker during fault tripping. It extracts features from the fault traveling wave signal to obtain traveling wave features, calculates the tripping time based on the tripping action sequence, and concatenates the traveling wave features, the tripping time, and the fault current parameters to obtain an enhanced feature vector. Based on the difference between the enhanced feature vector and a preset benchmark vector, it calculates the health index at the current fault moment. Based on the fault current parameters, it uses a physical model based on accumulated arc energy to calculate the physical drive predicted remaining lifetime of the primary and secondary integrated circuit breaker. It combines the health indices at historical fault moments with the health index at the current fault moment to form a health index sequence, which is then input into a preset remaining lifetime prediction model to obtain the data-driven predicted remaining lifetime of the primary and secondary integrated circuit breaker. The remaining lifetime prediction model is constructed based on a convolutional neural network model. Finally, it performs a weighted fusion of the physical drive predicted remaining lifetime and the data-driven predicted remaining lifetime to obtain the final predicted remaining lifetime of the primary and secondary integrated circuit breaker. This invention automatically learns the deep implicit features of traveling wave signals through a one-dimensional convolutional network model, overcoming the limitations of relying on expert experience in manual feature design and improving the evaluation accuracy. It uses a combination of physical and data-driven approaches to predict remaining lifetime, combining physical interpretability with the high precision of data-driven methods, thus improving prediction accuracy. This invention can improve the operation and maintenance efficiency of power distribution automation systems by accurately predicting the mechanical health status and remaining lifetime of circuit breakers.

[0054] The various embodiments in this specification are described in a progressive manner. For directly identical or similar parts of the embodiments, refer to each other. Each embodiment focuses on its differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. It should be noted that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0055] The embodiments described above are merely preferred embodiments of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various improvements and substitutions without departing from the technical principles of this invention, and these improvements and substitutions should also be considered within the scope of protection of this application. Therefore, the scope of protection of this patent application should be determined by the scope of the claims.

Claims

1. A method for condition assessment of a primary and secondary integrated circuit breaker, characterized in that, include: Acquire the fault traveling wave signal, tripping action timing, and fault current parameters of the primary and secondary integrated circuit breaker during fault tripping. Feature extraction is performed on the fault traveling wave signal to obtain traveling wave features. The tripping time is calculated according to the tripping action sequence. The traveling wave features, the tripping time, and the fault current parameters are concatenated to obtain an enhanced feature vector. The health index at the current fault moment is calculated based on the difference between the enhanced feature vector and the preset benchmark vector. Based on the fault current parameters, a physical model based on accumulated arc energy is used to calculate the remaining life of the physical drive prediction of the primary and secondary fused circuit breakers. The health index at the time of historical failure and the health index at the time of current failure are combined to form a health index sequence, which is then input into a preset remaining life prediction model to obtain the data-driven prediction of the remaining life of the primary and secondary fusion circuit breaker. The remaining life prediction model is constructed based on a convolutional neural network model. The predicted remaining lifetime of the physical-driven circuit breaker and the predicted remaining lifetime of the data-driven circuit breaker are weighted and fused to obtain the final predicted remaining lifetime of the primary and secondary fused circuit breaker.

2. The method for assessing the condition of a primary and secondary integrated circuit breaker according to claim 1, characterized in that, The step of extracting features from the fault traveling wave signal to obtain traveling wave features includes: The fault traveling wave signal is decomposed by wavelet packet to obtain high-frequency transient components, and features are extracted from the high-frequency transient components to obtain physical features, including high-frequency maximum features, high-frequency duration features, and energy percentage features. The fault traveling wave signal is normalized, and the processed sequence is input into a pre-trained feature extraction model to obtain deep features. The feature extraction model is constructed based on a one-dimensional convolutional neural network model. The physical features and the deep features are used as traveling wave features.

3. The method for assessing the condition of a primary and secondary integrated circuit breaker according to claim 2, characterized in that, The steps of performing wavelet packet decomposition on the fault traveling wave signal to obtain high-frequency transient components, and extracting features from the high-frequency transient components to obtain physical features include: The fault traveling wave signal is decomposed by wavelet packet decomposition to obtain multiple equal-width frequency band subspaces. In each frequency band subspace, the decomposition node coefficients with the center frequency located within a preset frequency band are selected for reconstruction to obtain high-frequency transient components. The maximum value among the absolute values ​​of the high-frequency transient components is taken as the high-frequency maximum feature; The time length of the high-frequency transient component from the starting point to the envelope decaying to the peak value by a preset proportion is used as the high-frequency duration feature. The proportion of the energy of the high-frequency transient component in the total energy of the fault traveling wave signal is used as the energy proportion feature. The high-frequency maximum value feature, the high-frequency duration feature, and the energy percentage feature are used as physical features.

4. The method for assessing the condition of a primary and secondary integrated circuit breaker according to claim 2, characterized in that, The feature extraction model includes an input layer, several one-dimensional convolutional blocks, and a global average pooling layer, which are connected in series. Each one-dimensional convolutional block contains a convolutional layer, a ReLU activation function, and a max pooling layer with a stride of 2.

5. The method for assessing the condition of a primary and secondary integrated circuit breaker according to claim 1, characterized in that, The step of calculating the remaining life of the physical drive prediction of the primary and secondary fused circuit breakers based on the fault current parameters and using a physical model based on accumulated arc energy includes: Calculate the energy of a single arc based on the fault current amplitude and arc duration; Based on the single arc energy, the historical cumulative arc energy is updated to obtain the updated cumulative arc energy; Based on the accumulated arc energy, a physical drive prediction of the remaining life of the primary and secondary fused circuit breakers is calculated using a physical model based on exponential decay.

6. The condition assessment method for a primary and secondary integrated circuit breaker according to claim 1, characterized in that, The step of combining the health index at the historical fault time and the health index at the current fault time to form a health index sequence, and inputting it into a preset remaining life prediction model to obtain the data-driven prediction of the remaining life of the primary and secondary fusion circuit breaker includes: The health index at the time of the historical failure and the health index at the time of the current failure are combined to form a health index sequence, which is then input into a preset remaining life prediction model to obtain the predicted value of the health index at the next time step. The remaining life prediction model is constructed based on a long short-term memory neural network model. Based on the preset lifespan mapping function, the predicted value of the health index at the next moment is mapped to obtain the data-driven prediction of the remaining lifespan of the primary and secondary integrated circuit breakers.

7. The method for assessing the condition of a primary and secondary integrated circuit breaker according to claim 1, characterized in that, Following the step of obtaining the final predicted remaining lifetime of the primary and secondary fusion circuit breakers, the method further includes: Based on the health index at the current time of failure and the final predicted remaining lifespan, a graded early warning is generated according to a preset grading rule. The graded early warning includes a mild warning, a moderate warning, and an emergency warning. Based on the aforementioned tiered early warning system, maintenance recommendations for the primary and secondary integrated circuit breakers are generated.

8. A condition assessment system for integrated primary and secondary circuit breakers, characterized in that, include: The data processing module is used to acquire the fault traveling wave signal, tripping action timing and fault current parameters of the primary and secondary integrated circuit breaker during fault tripping. Feature extraction is performed on the fault traveling wave signal to obtain traveling wave features. The tripping time is calculated according to the tripping action sequence. The traveling wave features, the tripping time, and the fault current parameters are concatenated to obtain an enhanced feature vector. The health index at the current fault moment is calculated based on the difference between the enhanced feature vector and the preset benchmark vector. The physical drive prediction module is used to calculate the remaining physical drive prediction life of the primary and secondary fusion circuit breakers based on the fault current parameters and a physical model based on accumulated arc energy. The data-driven prediction module is used to combine the health index at the time of historical failure with the health index at the time of current failure to form a health index sequence, and input it into a preset remaining life prediction model to obtain the data-driven predicted remaining life of the primary and secondary fusion circuit breaker. The remaining life prediction model is constructed based on a convolutional neural network model. The final life prediction module is used to perform weighted fusion of the physical-driven predicted remaining life and the data-driven predicted remaining life to obtain the final predicted remaining life of the primary and secondary fusion circuit breaker.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images