Mechanical life prediction method and system for 10kv pole-mounted circuit breaker

Abstract

The application relates to the technical field of circuit breaker life prediction, and particularly discloses a mechanical life prediction method and system for a 10kV pole-mounted circuit breaker, which comprises the following steps: obtaining a multi-dimensional operation state signal of the circuit breaker, wherein the multi-dimensional operation state signal comprises an operating mechanism coil current sequence and a main circuit phase current sequence collected by a Hall current clamp, and an operating bus voltage and an environment temperature synchronously collected.The technical scheme of the application effectively reduces the interference caused by an outdoor harsh environment and power fluctuation on feature extraction by synchronously collecting the operating bus voltage and the environment temperature to benchmark convert the initial mechanical degradation parameters, so that the objective authenticity of the input model data is ensured.

Description

Technical Field

[0001] This invention relates to the field of circuit breaker life prediction technology, and in particular to a method and system for predicting the mechanical life of 10kV pole-mounted circuit breakers. Background Technology

[0002] Currently, 10kV pole-mounted circuit breakers are key control devices in distribution networks. Existing mechanical life assessment methods typically rely on periodic power outages for maintenance or statistical extrapolation based on a single operation. With the development of smart grids, some existing technologies attempt to extract current waveform characteristics by collecting operating parameters such as the circuit breaker's coil current to establish life assessment models, thereby achieving non-intrusive online condition monitoring.

[0003] However, in actual operating scenarios, large fluctuations in ambient temperature and instability in operating bus voltage can cause distortion of current waveforms, resulting in reference drift of extracted feature parameters. At the same time, after the evaluation model trained in the laboratory is put into the field, the model prediction accuracy will gradually decrease over time due to the difference between individual manufacturing tolerances and actual service conditions.

[0004] The resulting problem is that if environmental disturbances are not compensated for and the model is not updated, the lifetime prediction results will have serious deviations or even false alarms; on the other hand, if manual power outage calibration and model retraining are frequently performed in order to ensure accuracy, it directly violates the original intention of non-intrusive online monitoring to reduce operation and maintenance costs. Summary of the Invention

[0005] This invention aims to at least partially solve one of the technical problems in related technologies. Therefore, the objective of this invention is to propose a method and system for predicting the mechanical life of 10kV pole-mounted circuit breakers to ensure the reliability of power supply in distribution networks.

[0006] To achieve the above objectives, a first aspect of the present invention proposes a method for predicting the mechanical life of a 10kV pole-mounted circuit breaker, comprising:

[0007] The circuit breaker's multi-dimensional operating status signals are acquired, including the operating mechanism coil current sequence and main circuit phase current sequence acquired by Hall current clamp, as well as the operating bus voltage and ambient temperature acquired synchronously.

[0008] Event trigger detection is performed based on the current change rate of the coil current sequence of the operating mechanism. The triggered operating events are classified into opening operation, energy storage operation and closing operation. The closing operation is further subdivided into no-load closing state or load closing state based on the peak characteristics of the main circuit phase current sequence.

[0009] Based on the classified operation events, initial mechanical degradation parameters are extracted from the coil current sequence of the operating mechanism. The initial mechanical degradation parameters are then converted to a standard mechanical degradation feature vector by combining the operating bus voltage and the ambient temperature.

[0010] The standard mechanical degradation feature vector is input into the autoencoder feature mapping model, and the mechanical health index of the circuit breaker is mapped and output.

[0011] Obtain the field operation and maintenance feedback tag of the circuit breaker, construct the posterior error distribution based on Bayes' theorem, and dynamically and incrementally update the mapping parameters of the autoencoder feature mapping model to correct the mechanical health index. Then, calculate and output the mechanical predicted life based on the corrected mechanical health index and the preset life degradation model.

[0012] To achieve the above objectives, a second aspect of the present invention provides a mechanical life prediction system for 10kV pole-mounted circuit breakers, comprising:

[0013] A non-intrusive sensing module is used to acquire multi-dimensional operating status signals of the circuit breaker. The multi-dimensional operating status signals include the operating mechanism coil current sequence and the main circuit phase current sequence acquired by the Hall current clamp, as well as the operating bus voltage and ambient temperature acquired synchronously.

[0014] The operation event identification module is used to detect event triggering based on the current change rate of the coil current sequence of the operation mechanism, classify the triggered operation events into opening operation, energy storage operation and closing operation, and further subdivide the closing operation into no-load closing state or load closing state based on the peak characteristics of the main circuit phase current sequence.

[0015] The mechanical degradation assessment module is used to extract initial mechanical degradation parameters from the coil current sequence of the operating mechanism based on the classified operating events, and to perform benchmark conversion on the initial mechanical degradation parameters in combination with the operating bus voltage and the ambient temperature to generate a standard mechanical degradation feature vector.

[0016] The health index mapping module is used to input the standard mechanical degradation feature vector into the self-encoder feature mapping model and map and output the mechanical health index of the circuit breaker.

[0017] The self-calibration and life prediction module is used to obtain the field operation and maintenance feedback tags of the circuit breaker, construct the posterior error distribution based on Bayes' theorem, and dynamically and incrementally update the mapping parameters of the autoencoder feature mapping model to correct the mechanical health index. Based on the corrected mechanical health index and the preset life degradation model, the module calculates and outputs the predicted mechanical life.

[0018] To achieve the above objectives, a third aspect of the present invention provides an electronic device including a memory, a processor, and a computer program stored in the memory. When the computer program is executed by the processor, it implements the above-described method for predicting the mechanical life of a 10kV pole-mounted circuit breaker.

[0019] In practical power distribution network operation and maintenance applications, the technical solution of this invention effectively reduces the interference caused by harsh outdoor environments and power fluctuations on feature extraction by synchronously collecting the operating bus voltage and ambient temperature to perform benchmark conversion of initial mechanical degradation parameters, ensuring the objectivity and authenticity of the input model data. At the same time, combined with the real operation and maintenance feedback labels of the circuit breaker on site, the posterior error distribution is constructed based on Bayes' theorem to dynamically and incrementally update the autoencoder feature mapping model, which better balances the technical conflict between high-precision prediction and low operation and maintenance costs. This enables the model to have the ability to adapt to the environment and individual changes, thereby dynamically reducing the drift error caused by manufacturing tolerances and long-term service without the need for manual power outage intervention, and realizing high reliability and adaptive online prediction of the circuit breaker throughout its entire life cycle. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the method for predicting the mechanical life of 10kV pole-mounted circuit breakers provided by the present invention.

[0021] Figure 2 This is a schematic diagram of the operating mechanism coil current waveform and 8-dimensional mechanical degradation sensitive feature extraction and calibration in the mechanical life prediction method for 10kV pole-mounted circuit breakers provided by the present invention.

[0022] Figure 3 This is a fitting curve of the electrical wear degradation trajectory of the vacuum interrupter contact based on the double exponential degradation model in the mechanical life prediction method for 10kV pole-mounted circuit breakers provided by this invention.

[0023] Figure 4 This is an evolution diagram of the prior and posterior error distribution density of the feature mapping model parameters based on Bayes' theorem in the mechanical life prediction method for 10kV pole-mounted circuit breakers provided by this invention.

[0024] Figure 5 This is a least-squares identification scatter plot of the real-time dynamic resistance of the operating coil in the steady-state stage of electromagnetic engagement in the mechanical life prediction method for 10kV pole-mounted circuit breakers provided by this invention.

[0025] Figure 6 This is a t-SNE visualization scatter plot showing the difference in characteristic distribution between laboratory baseline latent variables and field individual initial latent variables in the mechanical life prediction method for 10kV pole-mounted circuit breakers provided by this invention.

[0026] Figure 7 This is a comparison curve of the mechanical health index of an individual circuit breaker before and after zero calibration of the initial health status benchmark of the field circuit breaker in the mechanical life prediction method for 10kV pole-mounted circuit breakers provided by the present invention.

[0027] Figure 8 This is a schematic diagram illustrating the implementation of the mechanical life prediction system for 10kV pole-mounted circuit breakers provided by the present invention.

[0028] Figure 9 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0029] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0030] The mechanical life prediction system, method, and electronic device for 10kV pole-mounted circuit breakers according to embodiments of the present invention are described below with reference to the accompanying drawings.

[0031] Example 1:

[0032] like Figure 1 As shown, this embodiment provides a method for predicting the mechanical life of 10kV pole-mounted circuit breakers. This method combines non-intrusive sensing of multi-source signals with edge intelligent algorithms to achieve accurate assessment of the status of critical switching equipment in the power system. Specifically, this embodiment includes the following:

[0033] In this embodiment, the sensing object is a 10kV pole-mounted circuit breaker, which is mainly responsible for load switching and fault isolation in the distribution network. In order to achieve status monitoring without changing the original secondary circuit wiring of the circuit breaker, this method first acquires multi-dimensional operating status signals of the circuit breaker through sensing hardware.

[0034] Specifically, the multi-dimensional operating status signals include the operating mechanism coil current sequence and the main circuit phase current sequence, which are non-intrusively acquired by four open-type Hall current clamps. The operating mechanism coil current sequence is further subdivided into the opening coil current, the closing coil current, and the energy storage motor current. These three types of current essentially cover the entire process of the circuit breaker from receiving commands and mechanism action to energy replenishment. Simultaneously, the system synchronously acquires the operating bus voltage and ambient temperature through a voltage divider circuit and a temperature sensor.

[0035] It should be noted that, in order to ensure the accuracy of subsequent arc energy calculation and feature fusion, the sampling frequency of all sensing channels is kept consistent, and the time synchronization error between signals is limited to within 0.1 milliseconds.

[0036] For example, the Hall current clamp is installed on the secondary leads inside the circuit breaker control box, converting the magnetic field generated by the induced current into a voltage signal. Since pole-mounted circuit breakers are exposed to complex outdoor electromagnetic environments year-round, the acquired raw current sequences are often superimposed with high-frequency noise. Therefore, in the signal preprocessing stage, the system uses a zero-phase Butterworth low-pass filter to denoise the current sequence of the operating mechanism coil, ensuring the clarity of the edges of current abrupt changes.

[0037] Subsequently, the system performs event trigger detection based on the rate of change of the current in the coil current sequence of the operating mechanism. Since the current will generate a sharp step jump at the moment the coil is energized, the moment of circuit breaker operation can be accurately captured by monitoring the first derivative of the current in real time.

[0038] Specifically, this embodiment classifies the triggered operation events into tripping operations, energy storage operations, and closing operations. For closing operations, the system does not perform a single evaluation, but further subdivides them into no-load closing state or load closing state based on the peak characteristics of the main circuit phase current sequence.

[0039] When the system detects that the rise in the trip coil current within the operating mechanism coil current sequence exceeds a first preset current threshold within a first preset time window, it determines that a tripping operation has been triggered. Here, the first preset time window is typically set to 1 millisecond, and the first preset current threshold is set to 0.5 amperes. Once triggered, the system automatically captures the complete current waveform from a first preset pre-trigger time (e.g., 20 milliseconds) before triggering to a first preset post-trigger time (e.g., 50 milliseconds) after the current returns to zero as the analysis object.

[0040] Similarly, for closing operations, the circuit is triggered when the closing coil current meets the above-mentioned sudden change condition, and the waveform is captured and extended to the second preset post-time (e.g., 100 milliseconds) after the current returns to zero. For energy storage operations, if the energy storage motor current rises by more than 1 ampere within 1 millisecond, an energy storage event detection is triggered. It should also be noted that the logic for distinguishing between no-load closing and load closing states is as follows: within the preset observation window (e.g., 50 milliseconds) after the closing command is issued, if the peak characteristic of the main circuit phase current sequence is less than the preset current reference value (e.g., 5 amperes), it is determined that the circuit breaker has only completed mechanical closing without any load current flowing through it, and is classified as no-load closing; otherwise, it is classified as load closing.

[0041] Based on the categorized operating events, the system extracts initial mechanical degradation parameters from the captured coil current sequence of the operating mechanism. These mechanical degradation parameters reflect physical changes within the circuit breaker operating mechanism, such as spring fatigue, connecting rod wear, and grease drying.

[0042] For example, this embodiment extracts 8-dimensional mechanical degradation sensitive features for different operational events. For the tripping operation, the first rising slope, the average value of the first plateau current, the duration of the first plateau, and the total pulse width of the first current are extracted from the tripping coil current waveform. Among them, the first rising slope reflects the current-receiving response speed of the coil inductance; the average value of the first plateau current is directly related to the armature movement resistance; and the duration of the first plateau characterizes the unlocking time of the tripping mechanism.

[0043] like Figure 2 This diagram illustrates the physical changes and feature extraction boundaries of multi-dimensional operational status signals. The horizontal axis represents time in milliseconds, and the vertical axis represents current in amperes.

[0044] The blue solid curve in the figure represents the original current waveform after filtering and preprocessing. It shows the waveform transformation law from zero point, through the intermediate concave plateau, to the steady state and finally cut off. This transformation law truly reflects the whole physical process of coil energization, armature overcoming resistance to generate motional electromotive force, and auxiliary switch cutting off the circuit.

[0045] To extract the key parameters of the tripping side that constitute the eight-dimensional mechanical degradation sensitive feature vector from the waveform, the rising slope fitting line of the initial stage of the waveform is marked by a red dashed line in the figure. This slope corresponds to the rising rate of the waveform in the range of 5 milliseconds to 12 milliseconds, which is used to characterize the current-receiving response speed of the coil inductance. The thick green solid line in the figure marks the platform feature extraction area. The average current on the vertical axis of this area is 1.84 amperes, spanning the time interval of 12 milliseconds to 28 milliseconds, which objectively quantifies the average resistance during the armature movement and the duration of tripping and unlocking. At the same time, the black total pulse width calibration line at the bottom of the figure accurately measures the span from the 5-millisecond command trigger to the current returning to zero in 48 milliseconds, thereby obtaining the total current pulse width parameter.

[0046] Through the precise calibration of the aforementioned multi-dimensional graphic features, this schematic diagram verifies the feasibility of the feature extraction steps in actual engineering waveforms, providing a reliable data foundation for subsequent evaluation of spring fatigue and connecting rod wear within mechanical mechanisms.

[0047] For the closing operation, the second rising slope, the average value of the second plateau current, the duration of the second plateau, and the total pulse width of the second current are extracted as closing-side features. For the energy storage operation, the focus is on extracting the motor starting peak current and the motor stable operating current. The starting peak current reflects the initial load condition of the energy storage spring, while the stable operating current characterizes the resistance state of the motor drive chain.

[0048] Since the current waveform of the circuit breaker coil is greatly affected by power supply voltage fluctuations and ambient temperature, the above-mentioned initial mechanical degradation parameters must be recalculated based on a benchmark.

[0049] Specifically, the system first constructs an equivalent resistance and inductance model of the coil, and normalizes the current amplitude features in the extracted characteristic parameters to the reference amplitude features under the preset rated voltage conditions. It should be noted that the equivalent resistance and inductance model of the coil follows Maxwell's electromagnetic induction principle, and its complete dynamic voltage balance equation including the motional electromotive force term is shown in the following formula (1):

[0050] (1)

[0051] in, This represents the measured operating bus voltage. This represents the equivalent resistance of the operating coil. This represents the instantaneous current value in the coil. This represents the equivalent inductance of the coil. This refers to the motional electromotive force generated by the change in air gap due to the movement of the armature.

[0052] It should be further explained that during the armature's movement, this motional electromotive force term is the core physical reason for the plateau and dip in the current waveform; however, once the mechanism completes its action and enters the electromagnetic engagement steady-state stage, since the armature is stationary, the inductance no longer changes and the first derivative of the current... Approximately 0, motional electromotive force This disappears. At this point, the model naturally degenerates into a pure resistance model, which provides a solid physical foundation for subsequent online parameter identification of the real-time dynamic resistance of the operating coil. Through this physical model, the system can accurately calculate the current amplitude that the mechanism should achieve under standard rated voltage.

[0053] Next, the system uses a pre-calibrated temperature correction coefficient to uniformly compensate and map the time-related features in the extracted feature parameters to the reference time features at a preset standard reference temperature.

[0054] For example, the preset standard reference temperature is set to 20 degrees Celsius. Since the viscosity of the lubricating grease increases as temperature decreases, leading to a longer mechanism operating time, this correction factor can eliminate the pseudo-degradation phenomenon caused by seasonal temperature fluctuations. Finally, the system integrates the reference amplitude feature and the reference time feature to generate the standard mechanical degradation feature vector.

[0055] Considering that the internal resistance of the coil may change due to insulation aging during the long service life of the circuit breaker, this embodiment also introduces a step of adaptive online calibration of the reference amplitude characteristics in the process of constructing the equivalent resistance and inductance model of the coil and normalizing the conversion.

[0056] Specifically, the system extracts the average current plateau segment from the coil current sequence of the operating mechanism during the electromagnetic engagement steady-state phase. Since the armature has completed its stroke at this point, the coil current reaches dynamic equilibrium, and its first derivative is approximately zero. Simultaneously, the system records the measured operating bus voltage at this moment.

[0057] The system calculates the real-time dynamic resistance of the operating coil at the current operating moment based on the ratio of the operating bus voltage to the average value of the current plateau segment. The calculation process follows the formula (2) below:

[0058] (2)

[0059] in, This indicates the real-time dynamic resistance of the operating coil. To synchronously acquire the operating bus voltage, This represents the steady-state average current value of the plateau segment in the current sequence.

[0060] By subtracting the real-time dynamic resistance from the preset static initial resistance in the coil's equivalent resistance-inductance model, the system generates a hardware characteristic drift correction operator. This operator can effectively identify current changes caused by hardware factors such as coil moisture and increased contact resistance, and perform residual compensation on the reference amplitude characteristics, thereby outputting a standard mechanical degradation feature vector after hardware self-calibration.

[0061] Then, the system inputs the standard mechanical degradation feature vector into the pre-trained autoencoder feature mapping model and outputs the mechanical health index of the circuit breaker.

[0062] Specifically, the autoencoder feature mapping model is constructed using data labeled with the percentage of remaining mechanical life from accelerated life tests of circuit breakers of the same model as the training set. The model consists of an encoder and a decoder, extracting the most representative degenerate manifold by compressing the dimension of the feature vector.

[0063] The mechanical health index is defined between a first set boundary value and a second set boundary value. For example, the first set boundary value is defined as 1, representing that the operating mechanism is in a brand new healthy state; the second set boundary value is defined as 0, representing that the mechanical life of the operating mechanism has ended. Based on the mechanical health index and a preset lifespan degradation trajectory model, the system ultimately calculates and outputs the predicted mechanical lifespan.

[0064] In addition to mechanical assessment, this method also includes a full-condition electrical life assessment step. For the aforementioned opening operation, the main loss of electrical life comes from the electric arc generated by contact separation.

[0065] Specifically, the system determines the arc initiation time based on the moment when the rate of change of the main circuit phase current sequence exceeds a first preset rate of change threshold after the tripping command is issued. This abrupt change signifies a sharp increase in the micro-contact resistance of the contacts and the generation of an arc. Similarly, the arc end time is determined based on the moment when the current crosses zero and does not rise further within the subsequent time window. Using the time difference between the arc initiation and arc end times, the system calculates the duration of a single arc.

[0066] The system matches the corresponding original arc-breaking energy in a pre-constructed arc-breaking time energy mapping library based on the peak current of the main circuit phase current sequence. Since the arc-breaking process is affected by the internal air pressure of the arc-extinguishing chamber and the operating bus voltage, the system uses the operating bus voltage to perform voltage coefficient normalization compensation on the original arc-breaking energy and uses the ambient temperature for environmental coefficient compensation to generate the single-time arc-breaking energy. The mathematical logic of the energy mapping library is shown in formula (3):

[0067] (3)

[0068] in, This represents the initial arcing energy during circuit breaker tripping. This represents the calculated single arcing time. and These represent the fitting coefficients for different current rating levels, with subscripts indicating the appropriate coefficients. Use natural numbers. This formula reveals the power function relationship between arcing time and ablation energy, and is the core physical basis for assessing the degree of contact burn-off.

[0069] In the full-condition electrical life assessment process, for the closing operation which is further subdivided into no-load closing state, the system focuses on assessing the contact damage caused by pre-breakdown events.

[0070] Specifically, the system extracts pre-breakdown events within a preset pre-breakdown time window after the closing coil is energized, such as 10 milliseconds to 100 milliseconds. The moment when the rising slope of the main circuit phase current sequence is greater than a second preset rate of change threshold is defined as the pre-breakdown start point, and the moment when the second derivative of the main circuit phase current sequence exhibits a negative extremum is defined as the pre-breakdown end point.

[0071] The system integrates the instantaneous current in the pre-breakdown interval using a preset equivalent arc voltage drop in the vacuum interrupter to generate the original pre-breakdown energy. Considering the real-time fluctuations in the distribution network voltage, the system uses synchronously acquired operating bus voltage (mapped here as a proportion of the system's primary side voltage) to compensate for bus voltage fluctuations, and combines this with temperature correction to generate the single-closing pre-breakdown energy. This method avoids directly measuring the contact gap electric field strength, achieving accurate estimation of the migration energy of molten metal droplets on the contact surface through secondary side signals.

[0072] The system adds the accumulated arcing energy of a single circuit breaker trip to the accumulated pre-breakdown energy of a single circuit breaker trip multiplied by a preset closing ablation efficiency coefficient, and calculates the total equivalent electrical wear energy.

[0073] Specifically, based on the bi-exponential degradation model, the system maps the total equivalent electrical wear energy to a full-condition electrical health index. The bi-exponential degradation model considers both the stable wear stage in the early stages of service and the accelerated ablation stage in the later stages. The model parameters include the contact ablation rate constant for the first stage and the contact deformation degradation rate constant for the second stage. Compared to a single linear model, the bi-exponential model can more objectively reflect the nonlinear evolution process of the vacuum interrupter contacts from slight oxidation to structural erosion.

[0074] like Figure 3 This figure illustrates the differences in the evolution of different assessment models when predicting contact electrical lifetime. The horizontal axis represents the cumulative number of operations, in units of times, and the vertical axis represents the electrical health index, in dimensionless units.

[0075] The blue linear degradation curve in the figure shows a uniform downward trend with a fixed slope, which is difficult to accurately reflect the complex physical losses. In contrast, the red double exponential degradation curve in the figure exhibits obvious two-stage nonlinear characteristics. In the early stage, when the cumulative number of operations increases from 0 to about 4000, the red curve declines relatively gently, indicating that the contact material is in a stable state of slight oxidation and minor wear. When the cumulative number of operations exceeds 5000 and enters the later stage, the downward slope of the red curve increases significantly, indicating that structural erosion and deformation occur on the contact surface, leading to accelerated health degradation.

[0076] The black actual wear sampling points in the figure are scattered around the red double-exponential degradation curve, and their fit with the red curve is significantly higher than that of the blue straight line throughout the entire life cycle.

[0077] Through the comparison and verification of the above multi-dimensional graphic features, the data graph confirms that the model used in this embodiment can objectively track the nonlinear electrical wear trajectory of the circuit breaker arc-extinguishing chamber under all operating conditions, thereby improving the accuracy of remaining electrical life prediction.

[0078] After outputting the predicted mechanical lifespan, the system generates an integrated early warning signal through feature fusion technology.

[0079] Specifically, the system introduces a first dynamic weighting coefficient and a second dynamic weighting coefficient to fuse the mechanical health index and the full-condition electrical health index to generate a comprehensive life-cycle health index. It should also be noted that the first and second dynamic weighting coefficients are not fixed, but are adaptively switched by the system based on pre-identified circuit breaker load types and operating frequency categories.

[0080] For typical feeder scenarios, the system tends to balance weights; however, for scenarios involving frequent operations such as switching capacitor banks and high inrush current, the system automatically increases the electrical life weighting coefficient to highlight the impact of contact burn-out on overall safety. When the mechanical predicted life, the all-condition electrical health index, or the all-life comprehensive health index exceeds the preset lower limit of the health threshold, the system generates a corresponding progressive early warning instruction, which is sent to the maintenance master station via the communication interface to guide technicians in conducting targeted repairs.

[0081] Optionally, in order to eliminate evaluation bias caused by manufacturing tolerances in different production batches of circuit breakers, this embodiment also includes a benchmark adaptive calibration branch based on individual differences in the field.

[0082] Specifically, the system acquires the operation data of the circuit breaker within a set lower limit of the number of operations during the initial stage of field commissioning (such as the first 10 operations) as initial calibration samples, and generates a corresponding set of individual field initial feature vectors. This set is then input into the autoencoder feature mapping model to extract the field initial latent variable matrix output by the model's hidden layer.

[0083] The system calculates the distributional divergence divergence between the initial latent variable matrix at the field and the pre-stored novel baseline latent variable matrix at the laboratory. This divergence reflects the inherent physical differences between the field individuals and the laboratory standard. The system maps this distributional divergence divergence to a compensation factor for the individual's initial health status.

[0084] Finally, the system superimposes the individual's initial health status compensation factor as a bias term onto the output layer of the autoencoder feature mapping model. This individual-differentiated calibration method enables the health index mapping to perform baseline zeroing calibration on initial deviations caused by manufacturing and assembly differences, ensuring the universality and accuracy of the evaluation model across different individuals in the entire network.

[0085] To ensure that the model's performance does not degrade during long-term service, the system executes a dynamic incremental update step based on Bayes' theorem.

[0086] Specifically, the edge terminal receives real-world observation datasets from the field, including mechanical failure feedback data, arc-extinguishing chamber replacement feedback data, or pre-breakdown anomaly feedback data. This feedback data supplements prior knowledge and triggers the correction of model parameters.

[0087] Specifically, the system calculates the posterior probability of the parameters according to Bayes' theorem, as shown in formula (4):

[0088] (4)

[0089] in, The mapping parameters to be updated for the autoencoder feature mapping model are... The posterior probability of the parameter is... This represents the likelihood distribution of the field feedback, i.e., the probability of observing this failure feedback given the known model parameters. For the prior probabilities of parameters based on laboratory data, This represents the marginal probability.

[0090] Specifically, the system continuously assimilates on-site feedback data to adjust the mapping parameters. The model gradually aligns with the actual service conditions of the circuit breaker. Furthermore, the system periodically calculates the model's prediction accuracy. When the accuracy falls below a preset trigger threshold, it automatically performs edge-increment retraining, thus ensuring the mechanical life prediction method has self-evolution capabilities.

[0091] like Figure 4 This diagram illustrates the physical process of dynamically incrementally updating the underlying mapping parameters of the autoencoder feature mapping model after incorporating on-site operational feedback data. The horizontal axis represents the model mapping parameter values ​​(dimensionless), and the vertical axis represents the probability density (dimensionless).

[0092] The blue initial prior distribution curve in the figure shows a normal distribution with a wide range and a low peak value. Its central extreme value is located around 0.4. This represents the general parameter distribution obtained by the model in the pre-training stage under the laboratory accelerated life test environment. Since there is a lack of actual field working condition data at this time, the uncertainty of the parameters is high.

[0093] After the edge terminal receives the first batch of real-world mechanical failure or maintenance feedback data, it performs preliminary corrections using Bayes' theorem, generating the green first-corrected posterior distribution curve shown in the figure. Compared to the blue curve, the extreme value of the distribution center of this green curve shifts to the right to around 0.55, and the curve shape becomes taller and thinner, indicating that the model has begun to assimilate the environmental differences of specific field individuals, and the certainty of the parameters has been initially improved.

[0094] With the equipment in long-term service and the continuous input of multiple batches of field anomaly feedback data, the model automatically performs multiple edge incremental retraining, eventually evolving into the multiple updated posterior distribution curve shown in red in the figure. The central extreme value of this red curve stably converges to around 0.65, and it has the highest probability density peak and the narrowest distribution span.

[0095] This convergence process, evolving from the blue curve to the green curve and then to the red curve, intuitively demonstrates that the edge self-correction mechanism provided in this embodiment can effectively assimilate real-world labels, effectively reduce drift errors caused by individual manufacturing tolerances and long-term service, and enable the life prediction model to have environmentally adaptive evolutionary capabilities, thereby ensuring the objective accuracy of online prediction results.

[0096] In summary, existing technologies often rely on fixed operation counts or single environmental compensation formulas, which are insufficient to cope with the challenges posed by drastic fluctuations in the outdoor environment of power distribution networks and individual manufacturing differences in circuit breakers.

[0097] This embodiment reduces the interference of voltage, temperature, and coil aging on feature extraction from the source by constructing a physical model of the coil's equivalent resistance and inductance and combining it with hardware characteristic drift adaptive calibration. Simultaneously, it achieves dual deep sensing of mechanical and electrical lifetimes by utilizing arc energy Weibull mapping and pre-breakdown energy assessment.

[0098] The overall solution transforms traditional extensive operation and maintenance into predictive maintenance based on precise data through a closed-loop logic of perception, feature extraction, environmental calibration, individual correction, and incremental learning. This not only significantly reduces the risk of distribution network accidents caused by mechanical failure of circuit breakers or failure of vacuum interrupters, but also avoids unnecessary over-maintenance, significantly improving the economy and reliability of power grid operation.

[0099] Example 2:

[0100] Building upon the aforementioned embodiments, this embodiment focuses on the long-term evolution of the operating mechanism coil hardware characteristics and its impact on the accuracy of mechanical life prediction. It provides a more environmentally adaptive and hardware self-calibrating method for predicting the mechanical life of 10kV pole-mounted circuit breakers. Specifically, it includes the following:

[0101] Phase 1: Feature extraction and environment mapping during the steady-state stage of electromagnetic attraction.

[0102] Specifically, when a 10kV pole-mounted circuit breaker executes closing or opening commands, the current evolution of its operating coil (opening coil or closing coil) follows complex electromagnetic transient laws. When the coil is energized, the current rises rapidly, and the generated electromagnetic force overcomes the spring prestress and the static friction of the mechanism, driving the armature to begin moving. As the armature moves to its limit position and completes mechanical impact engagement, the current enters a relatively stable equilibrium period. This physical process is defined in electrohydraulic dynamics as the electromagnetic engagement steady-state stage.

[0103] In this embodiment, during the process of constructing and normalizing the equivalent resistance and inductance model of the coil, high-frequency sampling data is first used to locate the current plateau region in the electromagnetic engagement steady-state stage of the operating mechanism coil current sequence. Since the armature has stopped displacing within this region, the back electromotive force generated by the coil inductive reactance tends to disappear, and the electrical characteristics of the entire circuit approximate a purely resistive load. At this time, the system extracts the average value of the current plateau segment within this region.

[0104] It should be noted that, in order to ensure the robustness of the calculation, the extraction of the mean value of the current plateau segment should avoid the oscillation noise generated at the moment of armature impact. Usually, the middle and later segments of the current waveform after it has stabilized are selected for arithmetic averaging.

[0105] Specifically, while extracting the average value of the current plateau segment, the system records the measured operating bus voltage that completely corresponds to the current sampling point through a synchronous triggering mechanism. For example, the measured operating bus voltage reflects the actual output level of the on-site battery pack or rectifier power supply at the moment of load application and is a fundamental physical quantity for subsequent electrical parameter identification.

[0106] The second stage: Real-time dynamic resistance calculation of the operating coil based on the least squares identification algorithm.

[0107] This embodiment introduces a real-time parameter identification mechanism to replace the fixed factory parameters in traditional solutions. Since pole-mounted circuit breakers operate outdoors for extended periods, their coil resistance will drift with the increase in service life. For example, microscopic carbonization of the winding insulation, oxidation and corrosion of the terminals, and the cumulative heat effect after multiple operations can all cause the coil internal resistance to deviate from the initial design value.

[0108] Specifically, in this embodiment, based on the measured operating bus voltage and the average current plateau segment, a least squares identification algorithm is used to find the resistance value that best approximates the actual physical state. The objective function in this process is shown in formula (5):

[0109] (5)

[0110] in, This indicates the identification and optimization objective function. This represents the resistance variable to be identified in the operating coil during the least squares iteration process. Indicates the first The measured instantaneous operating bus voltage values ​​at each sampling point Indicates the first Instantaneous value of the current plateau segment at each sampling point This represents the total number of sampling points during the steady-state phase. By finding the minimum value of this objective function, the system obtains the optimal solution that minimizes the residual, and this optimal solution is used as the real-time dynamic resistance of the operating coil at the current operating moment. .

[0111] For example, the calculation of the real-time dynamic resistance of the operating coil effectively transforms each circuit breaker operation into an online detection of the electrical state of the actuator. Compared to periodic manual power outage tests, this online identification can capture the dynamic response characteristics of the mechanism under actual operating loads, thus offering greater effectiveness.

[0112] like Figure 5 The figure demonstrates the mathematical optimization process by which the system identifies the true internal resistance of the coil online from fluctuating measured electrical quantities. The horizontal axis represents the instantaneous current during the plateau segment, in amperes, and the vertical axis represents the measured operating bus voltage, in volts.

[0113] The blue scatter points in the figure represent the instantaneous sampling discrete points acquired by the system at high frequency synchronously during the steady-state phase of electromagnetic engagement when the armature stops moving. Due to unavoidable measurement random errors in the field power distribution network and the slight fluctuations in the rectified power supply output, these blue scatter points do not fall on a strict straight line, but rather exhibit a discrete distribution around the center value, for example, forming relatively dense data clusters near the horizontal axis of 4 amperes and the vertical axis of 220 volts.

[0114] The solid red line in the figure represents the least-squares fitted regression line generated after the system uses the least-squares identification algorithm to minimize the sum of squared residuals of the discrete sampling point group. The slope of this solid red line physically represents the real-time dynamic resistance of the operating coil identified at the current operating moment. By showing the graphical relationship of the blue scatter points converging towards the solid red line, this figure verifies that the system can effectively suppress random jump interference from a single sampling point through a multi-point statistical algorithm, thereby objectively obtaining the true resistance value reflecting the aging or poor contact of the coil winding insulation.

[0115] This process decouples the electrical aging characteristics of the actuator from the purely mechanical wear characteristics at the underlying data processing level, providing reliable physical quantity support for the subsequent generation of hardware characteristic drift correction operators and the execution of residual compensation, and significantly reducing the probability of false alarms in the life prediction system under complex power grid environments.

[0116] Phase 3: Generation of hardware characteristic drift correction operators and deviation quantization.

[0117] The system calculates the difference and performs proportional operations between the real-time dynamic resistance of the operating coil and the preset static initial resistance in the coil's equivalent resistance-inductance model. The static initial resistance refers to the reference resistance value entered into the system through standard testing procedures when the circuit breaker is put into operation in a brand-new state.

[0118] Specifically, by comparing the two, the system generates a hardware characteristic drift correction operator. This operator is used to quantify the baseline deviation caused by hardware aging or drastic environmental changes. It should also be noted that the calculation logic of the hardware characteristic drift correction operator is shown in formula (6):

[0119] (6)

[0120] in, This represents the hardware characteristic drift correction operator. This indicates the preset static initial resistance.

[0121] The size of this operator directly reflects the aging degree of the hardware circuit. If A positive value that continues to increase usually indicates an increase in contact resistance in the wiring circuit or an evolution in the inter-turn insulation characteristics of the coil; if A sudden change may indicate an electrical fault in the secondary control circuit.

[0122] Specifically, in this embodiment, the hardware characteristic drift correction operator is defined as a dimensionless compensation coefficient, which establishes a mapping relationship between the current physical state of the hardware and its factory reference state. Through this relationship, the lifetime prediction system can separate its focus from the complex electrical drift and concentrate on pure mechanical wear and degradation.

[0123] Phase 4: Feature self-calibration and standard degradation vector output based on residual compensation.

[0124] After obtaining the hardware characteristic drift correction operator, the system uses this operator to perform residual compensation on the reference amplitude feature. Since the feature extraction module uses the factory reference voltage and reference resistance by default when calculating current amplitude features (such as rising slope, plateau current, etc.), if no compensation is performed, the increase in coil resistance will lead to a decrease in the measured current, thereby causing the model to make a misjudgment of increased mechanical friction or insufficient driving force.

[0125] Specifically, the residual compensation process aims to eliminate the feature mapping deviation caused by the aging drift of the operating coil's internal resistance. This is achieved by mathematically correcting the standard mechanically degraded feature vector, restoring it to the ideal coordinate system before hardware aging. It should also be noted that the calibrated feature calculation is shown in formula (7):

[0126] (7)

[0127] in, The components of the standard mechanical degradation eigenvector after hardware self-calibration are represented. This represents the initial reference amplitude characteristics without hardware calibration. This represents the preset characteristic sensitivity weighting coefficient, which is determined by electromagnetic characteristic experiments of different types of circuit breakers.

[0128] For example, through this residual compensation step, the system can output a standard mechanical degradation feature vector after hardware self-calibration. This means that even if the resistance of the circuit breaker's operating coil increases by 5% after ten years of service, the system can still accurately identify whether the wear of the mechanical linkage exceeds the safety threshold through the calibration algorithm, without being affected by changes in the coil's electrical parameters.

[0129] It is important to note that in actual operation and maintenance of power distribution networks, 10kV pole-mounted circuit breakers are widely distributed across a wide geographical range. For example, in coastal areas, salt spray corrosion can easily cause contact resistance drift in the terminal blocks inside the circuit breaker control box; in cold regions, long-term drastic temperature fluctuations can cause microscopic physical deformation of the coil windings, thereby affecting electrical parameters.

[0130] Traditional life prediction methods, lacking hardware self-calibration capabilities, often confuse aging phenomena in electrical circuits with mechanical wear of mechanisms. For example, a decrease in the peak closing current due to terminal oxidation is often misdiagnosed by existing algorithms as fatigue of the closing spring. The technical solution provided in this embodiment, by identifying the dynamic resistance of the coil online and introducing a correction operator, mathematically decouples the electrical and mechanical characteristics.

[0131] Specifically, the practical significance of this solution lies in significantly reducing the false alarm rate of the life prediction system. When maintenance personnel receive early warning information, they can clearly determine whether the warning stems from irreversible wear of mechanical mechanisms or from electrical drift in secondary circuits. This accurate diagnostic capability plays a crucial role in developing precise operation and maintenance strategies for the distribution network, extending equipment service life, and ensuring power supply reliability.

[0132] Therefore, this embodiment constructs a hardware calibration closed loop with self-sensing, self-identification, and self-compensation capabilities. Combining the technical solution of this embodiment with the multi-dimensional signal sensing and autoencoder model in the aforementioned embodiments forms a complete technology chain.

[0133] Compared to existing technologies, the advantages of this embodiment are reflected in the following aspects:

[0134] First, it solves the problem of dynamic drift of the measurement reference. Existing technologies mostly rely on static environmental compensation, while this solution achieves real-time monitoring of the aging of the actuator (coil) itself through least squares identification in Formula 5 and correction operator generation in Formula 6, ensuring that the reference baseline for life prediction is always in a state of dynamic updating.

[0135] Second, it improves the physical fidelity of the feature mapping. Through residual compensation in formula (7), the system transforms nonlinear interference in complex environments into linear compensation, so that the standard mechanical degradation feature vector input to the autoencoder model can eliminate noise from hardware aging and retain only the core degradation information related to mechanical friction, wear, and gaps.

[0136] Third, it achieves cost reduction and efficiency improvement at the operation and maintenance level. Through this solution, the health assessment of circuit breakers is no longer limited to ideal operating conditions set in the laboratory. The system can automatically adapt to the hardware evolution characteristics of each field circuit breaker, achieving accurate predictions tailored to each individual device. This not only avoids asset waste caused by premature retirement due to hardware drift but also prevents organizational refusal to operate accidents due to missed reporting.

[0137] In summary, this embodiment, by introducing a hardware self-calibration mechanism, provides a solid underlying data confidence level for predicting the mechanical life of 10kV pole-mounted circuit breakers, enabling the overall technical solution to exhibit extremely high robustness and industrial application value when facing real-world challenges such as long-term service, extreme environments, and manufacturing tolerances.

[0138] Example 3:

[0139] Based on the aforementioned embodiments, this embodiment focuses on how to eliminate the domain bias effect between the ideal laboratory model and the actual field conditions, and provides a method for predicting the mechanical life of 10kV pole-mounted circuit breakers with individual adaptive calibration capabilities.

[0140] Phase 1: Construction and limitation analysis of pre-trained laboratory models.

[0141] In this embodiment, the autoencoder feature mapping model is first pre-trained in a laboratory environment. Specifically, a deep autoencoder network is constructed using data labeled with the percentage of remaining mechanical life from accelerated life tests of the same type of circuit breaker. The autoencoder feature mapping model uses nonlinear dimensionality reduction techniques to compress high-dimensional standard mechanical degradation feature vectors into a low-dimensional manifold space, i.e., a hidden layer, to extract core features that can characterize the mechanical degradation of the mechanism.

[0142] For example, while laboratory accelerated life testing can simulate the service life of a circuit breaker for several years in a short period through high-frequency continuous operation, its test environment is typically under ideal conditions of constant temperature, constant humidity, and extremely stable power supply. It is also important to note that in actual production, even 10kV pole-mounted circuit breakers from the same batch may have slight manufacturing tolerances in terms of spring initial tension, connecting rod assembly clearance, and grease application thickness. When these individuals with slightly different physical properties are installed in high-altitude, extremely cold, or high-salt-spray field environments, their initial mechanical characteristic distribution often deviates from the laboratory baseline distribution. Without calibration, the model is highly susceptible to misinterpreting these inherent physical differences as acquired mechanical degradation, resulting in the mechanical health index of newly commissioned equipment being lower than the rated value in the initial stage.

[0143] Phase Two: Acquisition of individual health baseline data during the initial stage of on-site operation.

[0144] After the autoencoder feature mapping model completes pre-training, this method initiates a baseline adaptive calibration branch based on individual differences in the field. Specifically, it acquires the operational data of the circuit breaker within a set lower limit of the number of operations during the initial stage of field commissioning as the initial calibration sample.

[0145] Specifically, the lower limit of the set number of operations is defined as the frequency of operations sufficient to eliminate random sampling noise and reflect the initial stable state of the equipment. In this embodiment, it is recommended to select the operation data from the first 10 to 15 operations. For example, these first N operations typically occur during the equipment's commissioning phase. By sensing and extracting the corresponding multi-dimensional operating status signals through sensors, the system generates a corresponding set of initial on-site feature vectors for each individual unit installed in the field. It should also be noted that obtaining these samples is predicated on confirming that the circuit breaker is in a normally calibrated baseline state during the initial commissioning phase, i.e., there are no obvious installation defects or transportation damage.

[0146] The third stage: extraction of hidden layer feature distribution and construction of latent variable matrix.

[0147] The system inputs the initial feature vector set of each individual into the encoder part of the autoencoder feature mapping model one by one. Specifically, the encoder maps the input features to a low-dimensional feature space through nonlinear transformations of multiple layers of neurons.

[0148] Specifically, the initial latent variable matrix output by the hidden layer of the autoencoder feature mapping model is extracted. It is also important to note that the hidden layer, acting as the information bottleneck of the model, outputs latent variables that contain the most essential core features of the circuit breaker's mechanical mechanism. For a specific field instance, the distribution of its latent variable matrix reflects the initial mechanical response under the combined effects of manufacturing tolerances and the field installation environment. Simultaneously, the system pre-stores a corresponding novel laboratory baseline latent variable matrix, representing the initial response characteristics of the laboratory standard prototype under ideal conditions.

[0149] Fourth stage: Quantitative calculation of distribution difference divergence.

[0150] To quantify the distributional differences between field samples and laboratory standard prototypes, this embodiment introduces the concept of distributional divergence. Specifically, the system calculates the distributional divergence between the initial latent variable matrix in the field and the pre-stored novel baseline latent variable matrix in the laboratory.

[0151] Specifically, this embodiment uses a maximum mean difference algorithm based on kernel functions to measure the distance between two probability distributions. The calculation of the distribution divergence is shown in formula (8):

[0152] (8)

[0153] in, Indicates the divergence of the distribution; Indicates by The initial latent variable matrix of the field consists of sampling points; Indicates by A novel laboratory benchmark latent variable matrix composed of sampling points; This represents the total number of samples contained in the initial latent variable matrix at the site. This represents the total number of samples contained in the laboratory's new baseline latent variable matrix; Indicates the index of the field sample, corresponding to the first... First on-site operation; An index representing the laboratory reference sample; Represents the first hidden variable in the initial latent variable matrix of the site. One sample vector; Represents the first hidden variable in the new baseline latent variable matrix of the laboratory. One sample vector; This represents mapping the latent variables to a high-dimensional reproducing kernel Hilbert space. The feature mapping function.

[0154] For example, the distribution difference divergence physically represents the offset distance in mechanical performance between the field-mounted individual and the standard prototype. The greater this divergence, the greater the deviation between the inherent physical characteristics or field installation stress of the field circuit breaker and the laboratory environment.

[0155] like Figure 6 This diagram illustrates the spatial distribution of high-dimensional data from the hidden layer of the autoencoder feature mapping model after dimensionality reduction. The horizontal axis represents dimension one of the t-SNE model (dimensionless), and the vertical axis represents dimension two of the t-SNE model (dimensionless).

[0156] The blue scatter clusters in the figure represent sample points in a pre-stored new laboratory benchmark latent variable matrix. They are densely distributed around coordinates 15 on the horizontal axis and 20 on the vertical axis, representing the initial mechanical response benchmark characteristics of the same model standard circuit breaker under ideal test conditions such as constant temperature and humidity.

[0157] The red scatter clusters in the figure represent sample points corresponding to the initial latent variable matrices of individual circuit breakers extracted during the initial commissioning phase. They are mainly concentrated around the horizontal axis coordinate -10 and the vertical axis coordinate -5. The blue and red scatter clusters show obvious physical isolation and spatial distance in two-dimensional space. This significant distribution difference intuitively quantifies the inherent physical performance deviation of the same batch of circuit breakers caused by manufacturing tolerances such as initial spring tension and connecting rod assembly clearance, as well as the complex on-site climatic environment.

[0158] By revealing this spatial distribution separation phenomenon, the attached figure verifies the objective necessity of performing on-site individual health benchmark adaptive calibration steps in actual power distribution network operations. It demonstrates that directly applying the general pre-trained model in the laboratory can easily lead to misjudgments in the initial health score. This provides solid data evidence for calculating the distribution difference divergence and generating individual initial health state compensation factors, significantly reducing the probability of false alarms caused by individual differences in newly commissioned equipment.

[0159] Fifth stage: Mapping and generation of compensating factors for an individual's initial health status.

[0160] The system uses a preset nonlinear mapping function to map the distribution difference divergence into an individual's initial health state compensation factor.

[0161] Specifically, the individual's initial health status compensation factor is used to quantify the impact of the individual's innate performance bias on the final health score. It should also be noted that this compensation factor is essentially an adaptive bias vector, whose dimension matches the mechanical health index of the autoencoder output layer. Specifically, if the distributional divergence is within a preset tolerance range, the compensation factor approaches zero; if the divergence is significant, the compensation factor generates a corresponding correction based on the direction and magnitude of the bias to ensure that the calibrated initial health score is full.

[0162] Phase 6: Output layer bias superposition and reference zeroing calibration.

[0163] The individual's initial health status compensation factor is added as a bias term to the output layer of the autoencoder feature mapping model. Specifically, the addition operation occurs after the model calculates the original health score and before outputting the final prediction result.

[0164] Specifically, this embodiment uses this bias superposition to achieve benchmark zeroing calibration of the initial health index mapping deviation caused by manufacturing and assembly differences. The calibrated mechanical health index is calculated as shown in formula (9):

[0165] (9)

[0166] in, This indicates the mechanical health index after on-site benchmark zeroing calibration; This represents the nonlinear activation function of the output layer; This represents the weight matrix of the output layer; This represents the output vector of the hidden layer; This represents the original bias term that comes with the model; This represents the individual's initial health status compensation factor derived from Formula 8.

[0167] For example, through the calculation using formula (9), a newly commissioned circuit breaker that might have been mistakenly judged to have a health rating of 0.95 due to a slightly tight spring will have its score calibrated to 1.0. This means that the starting point of the life prediction system is precisely set at the moment the circuit breaker is commissioned. It should also be noted that this calibration process is only performed once during the initial commissioning phase, after which the compensation factor will be fixed in the model as a personalized parameter for evaluating the circuit breaker throughout its entire life cycle.

[0168] like Figure 7 This figure illustrates the longitudinal correction effect of initial assessment deviations caused by manufacturing and assembly differences after calibration with offset terms. The horizontal axis represents the cumulative number of operations (in times), and the vertical axis represents the mechanical health index (in dimensionless units).

[0169] The blue dashed line in the figure represents the original predicted trajectory before calibration. Due to inherent manufacturing tolerances and physical performance deviations of individual circuit breakers, such as excessive initial spring tension, the uncalibrated autoencoder feature mapping model misjudges these inherent physical differences as acquired mechanical degradation, causing the evaluation starting point of the blue dashed line to fall at 0.95, and gradually decrease with the increase of the number of operations.

[0170] The red solid line in the figure represents the baseline zeroing trajectory after calibration. The system calculates the distribution difference divergence and generates an individual's initial health status compensation factor. After superimposing this compensation factor onto the model output layer, the evaluation starting point of the red solid line is precisely pulled back and fixed at the full health position of 1.0.

[0171] The vertical translation comparison between the blue dashed line and the red solid line visually demonstrates that this calibration step effectively masks inherent individual differences that do not affect the safe operation of the equipment, ensuring that the starting point of the life prediction system is precisely set at the initial moment of the circuit breaker's on-site commissioning. This benchmark adaptive calibration technology guarantees that the subsequent descent trajectory only reflects mechanical wear and fatigue degradation caused by actual operation, significantly reducing the probability of false alarms in newly commissioned equipment during the early stages of its service life, and improving the objectivity and accuracy of the evaluation model applied to different circuit breakers across the entire network.

[0172] For example, in actual power distribution network operation and maintenance scenarios, the installation environments of 10kV pole-mounted circuit breakers vary greatly. For instance, the initial frictional characteristics of a circuit breaker installed in the arid, dusty northwest region will inevitably differ from those installed in the high-humidity coastal areas of southern China. If an uncalibrated uniform model is used, equipment in dusty environments may be on the verge of triggering warnings as soon as it goes online, resulting in numerous invalid false alarms and placing a heavy burden on the operation and maintenance master station for troubleshooting.

[0173] It should be noted that the impact of manufacturing tolerances is also significant. In precision mechanical assembly, even a 1% difference in torque will be significantly amplified in the high-dimensional features captured by the self-encoder. The benchmark zeroing calibration achieved by this technical solution through formulas (8) and (9) is essentially an in-situ digital twin modeling of each circuit breaker connected to the network.

[0174] Specifically, this calibration mechanism allows the model to ignore static feature offsets that do not affect safe operation and are merely due to individual differences, focusing instead on capturing dynamic degradation features that evolve over time. This enables the mechanical health index to accurately reflect the wear, fatigue, and performance degradation of circuit breakers after commissioning, greatly improving the accuracy and reliability of early warnings.

[0175] This embodiment establishes a complete chain of field adaptive calibration techniques. Compared to the overly radical model application methods in existing technologies, this technical solution exhibits the following overall advantages:

[0176] First, it solves the domain migration problem between laboratory data distribution and field data distribution. Existing technologies often cannot explain why initial monitoring data of the same model of equipment are inconsistent in different regions, while this solution scientifically quantifies this inconsistency by calculating the distribution difference divergence of the hidden layer.

[0177] Secondly, it achieves high-precision consistency in predicting the starting point. Through the bias superposition calibration of Formula 9, the system uniformly pulls the mechanical state of different individuals across the entire network back to the same healthy starting line, reducing initial spurious damage caused by manufacturing tolerances.

[0178] Finally, the robustness of the life prediction system in complex power grid environments was enhanced. This technical solution, through a progressive logic of pre-training, on-site calibration, and dynamic evaluation, ensures that the evaluation criteria for the mechanical health index of 10kV pole-mounted circuit breakers remain objective throughout the early, middle, and late stages of their service life.

[0179] By combining the multi-dimensional sensing in Example 1 and the hardware characteristic calibration in Example 2, this invention can not only effectively compensate for environmental voltage fluctuations but also adaptively correct coil aging drift and significantly reduce individual inherent differences. This closed-loop system makes the output of mechanical life prediction no longer a simple statistical inference but a precise judgment based on the mapping of physical entity characteristics, providing solid technical support for the digital transformation of power distribution networks.

[0180] Example 4:

[0181] like Figure 8 As shown, this embodiment provides a mechanical life prediction system for 10kV pole-mounted circuit breakers. In actual deployment, this system is usually integrated into a distribution automation terminal or a dedicated online monitoring device to perform real-time mechanical condition assessment and remaining life prediction of pole-mounted circuit breakers in operation.

[0182] The system specifically includes: a non-invasive sensing module, an operation event recognition module, a mechanical degradation assessment module, a health index mapping module, and a self-correction and lifespan prediction module.

[0183] Specifically, the non-invasive sensing module is configured to acquire multi-dimensional operating status signals of the circuit breaker. In practical engineering applications, this module includes multiple sensor units: in the operating mechanism coil circuit of the circuit breaker, an openable Hall current clamp (e.g., a clip-on Hall sensor with a range of 0~10A and an accuracy of ±1%) is non-invasively coupled to the coil power supply cable to collect the operating mechanism coil current sequence. The sampling frequency can be set to 10kHz~50kHz to fully capture the transient waveform characteristics of the current rising edge, peak value, and falling edge.

[0184] Meanwhile, in the main circuit on the incoming or outgoing side of the circuit breaker, another set of Hall current clamps (e.g., with a range of 0~630A) is used to synchronously collect the phase current sequence of the main circuit to determine the load status when the circuit is closed.

[0185] In addition, the module integrates a bus voltage sensor (e.g., a resistive voltage divider transformer connected to the circuit breaker operating bus) and a surface-mount temperature sensor (installed inside the mechanism housing or near the coil) to simultaneously collect the operating bus voltage and ambient temperature. These multi-dimensional signals are timestamped, converted from analog to digital, and packaged into structured data frames for subsequent module processing. The advantage of this non-intrusive design is that it eliminates the need to modify the circuit breaker's existing electrical circuits, ensuring the equipment's insulation performance and operational safety.

[0186] Specifically, the operation event recognition module receives a data stream from the non-intrusive sensing module and performs event trigger detection based on the current change rate of the operating mechanism coil current sequence. In practice, the module performs a differential operation on the coil current sequence, calculating the difference between every two adjacent sampling points divided by the sampling interval to obtain the instantaneous current change rate curve. When the absolute value of the current change rate exceeds a preset threshold (e.g., 0.5 A / ms) and its duration exceeds a set window (e.g., 0.2 ms), the start point of an operation event is determined to have been detected.

[0187] Subsequently, the module classifies the triggered operation events into tripping, energy storage, and closing operations based on the duration of the coil current waveform, peak plateau characteristics, and timing logic of the current cutoff point. For example, if the coil current waveform exhibits a short pulse (usually less than 15ms) with a back EMF spike at the trailing edge, it is classified as a tripping operation; if the waveform exhibits a relatively long flat-top segment (30ms~80ms) and the current amplitude is relatively stable, it is classified as an energy storage operation; if the waveform falls between the two, it is classified as a closing operation.

[0188] Furthermore, the module further subdivides the closing operation into no-load closing state or load closing state based on the peak characteristics of the main circuit phase current sequence. Specifically, after determining that a closing operation is to be performed, the module extracts the positive and negative peak values ​​of the main circuit phase current within half a cycle (i.e., 10ms) before and after the closing moment. If the peak values ​​of all three phase currents are lower than 5% of the rated current (for example, the rated current of a 10kV pole-mounted circuit breaker is 630A, so the threshold is approximately 31.5A), and there are no obvious load harmonic components in the current waveform, it is marked as no-load closing state; otherwise, it is marked as load closing state. This refined classification is of great significance for the subsequent normalization processing of degradation parameters, because the mechanical impact of load closing is often greater than that of the no-load state, and needs to be distinguished in the evaluation.

[0189] Specifically, the mechanical degradation assessment module extracts initial mechanical degradation parameters from the coil current sequence of the operating mechanism based on the categorized operational events. For each valid opening or closing operation, the module extracts the following parameters: coil energizing start current value, coil peak current value, peak time, maximum coil current rise slope, cutoff point time (i.e., the current drop point caused by auxiliary contact switching), and coil continuous energization time. These parameters constitute a set of initial mechanical degradation parameter vectors.

[0190] Considering that fluctuations in the operating bus voltage and changes in ambient temperature directly affect the coil current characteristics (e.g., a decrease in voltage leads to a decrease in peak current, and an increase in temperature leads to an increase in coil resistance), this module further combines the operating bus voltage and the ambient temperature to perform a benchmark conversion of the initial mechanical degradation parameters. The specific conversion method is as follows: using the rated operating bus voltage (e.g., DC220V or AC220V) and standard ambient temperature (e.g., 20℃) as benchmarks, the measured peak current, rise slope, and other parameters are linearly or piecewise linearly corrected.

[0191] For example, the corrected peak current = measured peak current × (rated voltage / measured bus voltage) × (1 + temperature correction factor × (measured temperature - 20℃)). The temperature correction factor is obtained through offline testing and is usually a positive value. This is because when the ambient temperature is higher than the reference temperature, the coil resistance increases, causing the measured current to be lower than the reference current, which needs to be compensated and amplified by a positive factor. After reference conversion, the module generates a standard mechanical degradation feature vector, which reduces the influence of the external environment and power supply conditions and truly reflects the mechanical state of the circuit breaker mechanism itself.

[0192] Specifically, the health index mapping module is used to input the standard mechanical degradation feature vector into the autoencoder feature mapping model and output the mechanical health index of the circuit breaker. This autoencoder feature mapping model is a three-layer neural network structure: the number of nodes in the input layer is consistent with the dimension of the degradation feature vector (e.g., 6-8 dimensions), the number of nodes in the hidden layer is set to half the number of nodes in the input layer (using a barrel structure), and the output layer is a scalar value, configurable from 0 to 100 or 0 to 1. The model is pre-trained using a large amount of historical data (including multiple sets of coil current samples from newly manufactured and decommissioned circuit breakers) in an unsupervised or semi-supervised manner, enabling the model to learn the intrinsic low-dimensional manifold of the feature vector under healthy conditions. During actual operation, the current standard feature vector is input into the model, and the model output is the mechanical health index. A higher index value indicates that the mechanical state is closer to the health benchmark, while a lower value indicates a more significant degree of degradation. To enhance robustness, the module can perform a moving average filter on the health index after multiple consecutive operations.

[0193] Specifically, the self-calibration and lifetime prediction module can be divided into two parts: online model calibration and remaining lifetime prediction.

[0194] In terms of online model calibration, the module acquires the field maintenance feedback tags of the circuit breaker. These feedback tags originate from actual maintenance records. For example, when maintenance personnel find out that the opening and closing time exceeds the tolerance, the contact wear exceeds the standard, or there is a jamming fault after performing mechanical characteristic tests on the circuit breaker, they enter the corresponding degradation level tag (such as "mild degradation", "severe degradation", "fault") into the system. The module constructs a posterior error distribution based on Bayes' theorem to dynamically and incrementally update the mapping parameters of the autoencoder feature mapping model. In specific implementation, the weight parameters of the autoencoder output layer are treated as random variables, and the prior distribution is set as the normal distribution after the last update. When a new maintenance feedback tag is obtained, the error between the current model output and the tag is calculated using the likelihood function, the posterior distribution is calculated according to Bayes' formula, and the expected value of the posterior distribution is taken as the updated parameter value, thereby realizing the online adjustment of the mapping parameters to correct the mechanical health index. This process avoids retraining the entire network and is suitable for lightweight updates in edge computing environments.

[0195] Regarding remaining service life prediction, the module calculates and outputs the predicted mechanical service life based on the corrected mechanical health index and a preset service life degradation model. The preset service life degradation model can adopt an exponential decay form, and its calculation logic is shown in the following formula: ;

[0196] in, Indicates the process Current mechanical health index after this operation. The initial mechanical health index is typically taken as the calibration value for new equipment, such as defined as 1. This is a parameter representing the rate of mechanical degradation. This is the cumulative number of machine operations.

[0197] This module utilizes a historically corrected mechanical health index sequence and fits it using the least squares method to obtain the individual current equipment data. The value is then set. A mechanical lifespan endpoint threshold is then established, for example, a mechanical health index below 30%. The remaining number of operations required to reach this threshold is calculated and output as the predicted mechanical lifespan. This prediction result can be uploaded to the main station system via a communication interface (such as IEC61850 or Modbus protocol) or displayed on a local HMI, providing a basis for decision-making in condition-based maintenance.

[0198] In summary, this embodiment constructs a closed-loop mechanical life prediction system for 10kV pole-mounted circuit breakers through non-invasive sensing, fine classification of operational events, environmental factor conversion, autoencoder health index mapping, and Bayesian online correction. This system can adaptively assess the degree of mechanical degradation and predict the remaining life under actual operating conditions.

[0199] Example 5:

[0200] Corresponding to the above embodiments, the present invention also proposes an electronic device.

[0201] like Figure 9 The diagram shows a structural schematic of an electronic device according to the present invention. The electronic device 100 includes a processor 101 and a memory 103. The processor 101 and the memory 103 are connected, for example, via a bus 102. Optionally, the electronic device 100 may further include a transceiver 104. It should be noted that in practical applications, the transceiver 104 is not limited to one unit, and the structure of this electronic device 100 does not constitute a limitation on the embodiments of the present invention.

[0202] Processor 101 may be a CPU, a general-purpose processor, a DSP, an ASIC, an FPGA, or other programmable logic device, transistor logic device, hardware component, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in connection with this disclosure. Processor 101 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0203] Bus 102 may include a pathway for transmitting information between the aforementioned components. Bus 102 may be a PCI bus or an EISA bus, etc. Bus 102 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 9 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0204] The memory 103 stores a computer program corresponding to the mechanical life prediction method for a 10kV pole-mounted circuit breaker according to the above embodiments of the present invention. This computer program is executed by the processor 101. The processor 101 executes the computer program stored in the memory 103 to implement the content shown in the aforementioned method embodiments.

[0205] Among them, electronic devices 100 include, but are not limited to: mobile terminals such as laptops and PADs (tablet computers) and fixed terminals such as desktop computers. Figure 9 The electronic device 100 shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of the present invention.

[0206] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for predicting the mechanical life of a 10kV pole-mounted circuit breaker, characterized in that, include: The circuit breaker's multi-dimensional operating status signals are acquired, including the operating mechanism coil current sequence and main circuit phase current sequence acquired by Hall current clamp, as well as the operating bus voltage and ambient temperature acquired synchronously. Event trigger detection is performed based on the current change rate of the coil current sequence of the operating mechanism. The triggered operating events are classified into opening operation, energy storage operation and closing operation. The closing operation is further subdivided into no-load closing state or load closing state based on the peak characteristics of the main circuit phase current sequence. Based on the classified operation events, initial mechanical degradation parameters are extracted from the coil current sequence of the operating mechanism. The initial mechanical degradation parameters are then converted to a standard mechanical degradation feature vector by combining the operating bus voltage and the ambient temperature. The standard mechanical degradation feature vector is input into the autoencoder feature mapping model, and the mechanical health index of the circuit breaker is mapped and output. Obtain the field operation and maintenance feedback tag of the circuit breaker, construct the posterior error distribution based on Bayes' theorem, and dynamically and incrementally update the mapping parameters of the autoencoder feature mapping model to correct the mechanical health index. Then, calculate and output the mechanical predicted life based on the corrected mechanical health index and the preset life degradation model.

2. The method according to claim 1, characterized in that, The event trigger detection based on the current change rate of the operating mechanism coil current sequence classifies the triggered operating events into opening operations, energy storage operations, and closing operations, including: When the rise of the tripping coil current in the coil current sequence of the operating mechanism is detected to be greater than the first preset current threshold within the first preset time window, the tripping operation is triggered, and the waveform from the first preset pre-time before triggering to the first preset post-time after the current returns to zero is captured. When the rise of the closing coil current in the coil current sequence of the operating mechanism is detected to be greater than the first preset current threshold within the first preset time window, the closing operation is triggered, and the waveform from the first preset pre-time before triggering to the second preset post-time after the current returns to zero is captured. When the rise of the energy storage motor current in the coil current sequence of the operating mechanism within the first preset time window is detected to be greater than the second preset current threshold, the energy storage operation is triggered, and the complete motor current waveform is captured.

3. The method according to claim 1, characterized in that, The extraction of initial mechanical degradation parameters from the coil current sequence of the operating mechanism includes extracting 8-dimensional mechanical degradation-sensitive features for different operating events, specifically including: For the circuit breaker tripping operation, the first rising slope, the average value of the first platform current, the duration of the first platform, and the total pulse width of the first current are extracted as features on the circuit breaker side. For the closing operation, the second rising slope, the average value of the second platform current, the duration of the second platform, and the total pulse width of the second current are extracted as closing-side features. For the energy storage operation, the peak current of the motor starting and the stable operating current of the motor are extracted as energy storage side features.

4. The method according to claim 1, characterized in that, The step of combining the operating bus voltage and the ambient temperature to perform a benchmark calculation on the initial mechanical degradation parameters to generate a standard mechanical degradation feature vector includes: Construct an equivalent resistance and inductance model for the coil, and normalize and convert the current amplitude features in the extracted feature parameters to the reference amplitude features under the preset rated voltage conditions. By using a pre-calibrated temperature correction coefficient, the time-related features in the extracted feature parameters are uniformly compensated and mapped to the reference time features at a preset standard reference temperature. The reference amplitude feature and the reference time feature are fused to generate the standard mechanical degradation feature vector.

5. The method according to claim 1, characterized in that, The method also includes a full-condition electrical life assessment step, specifically including: For the circuit breaker tripping operation, the arcing start time is determined based on the moment when the rate of change of the main circuit phase current sequence after the circuit breaker tripping command is issued is greater than the first preset rate of change threshold. The arcing end time is determined based on the moment when the current crosses zero and does not rise again in the subsequent time window. The arcing time is calculated using the time difference between the arcing start time and the arcing end time. Based on the peak current of the main circuit phase current sequence, the corresponding original arc-breaking energy is matched in the pre-constructed arc-breaking time energy mapping library. The original arc-breaking energy is compensated by voltage coefficient normalization using the operating bus voltage, and environmental coefficient compensation is performed using the ambient temperature to generate the single arc-breaking energy.

6. The method according to claim 5, characterized in that, The full-condition electrical life assessment steps also include: For the closing operation, which is further subdivided into no-load closing state, the pre-breakdown event within the preset pre-breakdown time window after the closing coil is energized is extracted. The moment when the rising slope of the main circuit phase current sequence is greater than the second preset rate of change threshold is defined as the pre-breakdown start point, and the moment when the second derivative of the main circuit phase current sequence has a negative extreme value is defined as the pre-breakdown end point. The instantaneous current in the pre-breakdown interval is integrated using the preset equivalent arc voltage drop of the vacuum interrupter to generate the original pre-breakdown energy. The bus voltage fluctuation is compensated using the operating bus voltage and the temperature is compensated using the ambient temperature to generate the single closing pre-breakdown energy.

7. The method according to claim 6, characterized in that, The full-condition electrical life assessment steps also include: The total equivalent electrical wear energy is calculated by adding the accumulated arcing energy of a single circuit breaker tripped to the accumulated pre-breakdown energy of a single circuit breaker tripped by multiplying it by the preset erosion efficiency coefficient. Based on a dual-exponential degradation model that includes the contact ablation rate constant in the first stage and the contact deformation degradation rate constant in the second stage, the total equivalent electrical wear energy is mapped to an electrical health index under all operating conditions.

8. The method according to claim 7, characterized in that, After outputting the predicted mechanical lifespan, the system also includes generating an integrated early warning signal, specifically including: A first dynamic weighting coefficient and a second dynamic weighting coefficient are introduced to fuse the mechanical health index and the full-condition electrical health index to generate a full-lifetime comprehensive health index; wherein, the first dynamic weighting coefficient and the second dynamic weighting coefficient are adaptively switched according to the pre-identified circuit breaker load type and operating frequency category. When the predicted mechanical lifespan, the full-condition electrical health index, or the full-lifespan comprehensive health index exceeds the preset lower limit of the health threshold, a progressive early warning instruction of the corresponding level is generated.

9. The method according to claim 1, characterized in that, The dynamic incremental update of the mapping parameters of the autoencoder feature mapping model based on the posterior error distribution constructed using Bayes' theorem specifically includes: We receive feedback data on mechanical failures, arc-extinguishing chamber replacements, or pre-breakdown anomalies from the field as the actual observation dataset. ; According to Bayes' theorem Calculate the posterior probability of the parameter, where The mapping parameters to be updated for the autoencoder feature mapping model are... The posterior probability of the parameter is... For the on-site feedback likelihood distribution, For the prior probability of the parameter, Marginal probability; The prediction accuracy of the autoencoder feature mapping model is calculated periodically. When the prediction accuracy is lower than the preset trigger threshold, edge incremental retraining is performed.

10. A mechanical life prediction system for 10kV pole-mounted circuit breakers, characterized in that, include: A non-intrusive sensing module is used to acquire multi-dimensional operating status signals of the circuit breaker. The multi-dimensional operating status signals include the operating mechanism coil current sequence and the main circuit phase current sequence acquired by the Hall current clamp, as well as the operating bus voltage and ambient temperature acquired synchronously. The operation event identification module is used to detect event triggering based on the current change rate of the coil current sequence of the operation mechanism, classify the triggered operation events into opening operation, energy storage operation and closing operation, and further subdivide the closing operation into no-load closing state or load closing state based on the peak characteristics of the main circuit phase current sequence. The mechanical degradation assessment module is used to extract initial mechanical degradation parameters from the coil current sequence of the operating mechanism based on the classified operating events, and to perform benchmark conversion on the initial mechanical degradation parameters in combination with the operating bus voltage and the ambient temperature to generate a standard mechanical degradation feature vector. The health index mapping module is used to input the standard mechanical degradation feature vector into the self-encoder feature mapping model and map and output the mechanical health index of the circuit breaker. The self-calibration and life prediction module is used to obtain the field operation and maintenance feedback tags of the circuit breaker, construct the posterior error distribution based on Bayes' theorem, and dynamically and incrementally update the mapping parameters of the autoencoder feature mapping model to correct the mechanical health index. Based on the corrected mechanical health index and the preset life degradation model, the module calculates and outputs the predicted mechanical life.

Images