A method and system for early warning of electrical assembly failure

Abstract

This application relates to the field of electrical equipment condition monitoring and fault early warning technology, and in particular to a method and system for fault early warning of electrical components. It includes: collecting temperature, voltage, load, and environmental parameters of electrical components; combining thermal inertia conversion, operating cycle division, and a health benchmark model; extracting degradation-related differences and generating a comprehensive degradation state characterization value; then determining the incubation period starting point based on historical evolution sequences, constructing a degradation trajectory, and predicting the remaining service life. This solves the problem in existing technologies where it is difficult to accurately distinguish between the true degradation starting point and transient operational anomalies under conditions of frequent load fluctuations, leading to inaccurate determination of the fault incubation period starting point.

Description

Technical Field

[0001] This application relates to the field of electrical equipment condition monitoring and fault early warning technology, and in particular to a method and system for early warning of electrical component faults. Background Technology

[0002] In power supply and distribution and industrial control scenarios, electrical components often operate under complex conditions of continuous load and frequent load fluctuations. Existing monitoring methods typically rely on continuous observation of operating parameters such as temperature and voltage to determine the presence of abnormal signs and trigger warnings. However, the parameters of electrical components themselves fluctuate normally with load changes, which inherently limits the method of judging anomalies by simply relying on parameter thresholds. When the conductive connections of components show slight aging or unstable contact, localized heating and instantaneous voltage drops are more likely to occur during high load periods, but these anomalies are partially masked during low load periods, resulting in intermittent anomalies that show significant declines after each occurrence. Since the true starting point of the fault latency period and the instantaneous anomalies caused by operating condition fluctuations highly overlap in time, existing conventional monitoring methods often struggle to accurately distinguish between the true degradation starting point and operating condition-related instantaneous anomalies. If the first short-term anomaly is directly regarded as the starting point of the latency period, subsequent warnings will be triggered prematurely, prolonging the risk process and causing unnecessary maintenance costs. If these anomalies are all treated as load disturbances and ignored, historical anomalies cannot be continuously inherited, the latency period will be confirmed later, and the optimal maintenance opportunity will be missed. This misjudgment of the starting point of the fault latency period under conditions of frequent load fluctuations seriously affects the continuous reception of fault warnings and the accuracy of trend judgment, posing a hidden danger to the safe and stable operation of the electrical system.

[0003] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0004] This application discloses an electrical component fault early warning method and system, which aims to solve the problem that the existing technology is unable to accurately distinguish between the true degradation start point and the conditional instantaneous anomaly under the condition of frequent load fluctuation, resulting in inaccurate determination of the fault latency start point.

[0005] The technical solution of this application is as follows:

[0006] In a first aspect, this application discloses a method for early warning of electrical component faults, including:

[0007] The actual temperature parameters, actual voltage parameters, load parameters, and environmental parameters of the electrical components are obtained. Based on the physical thermal inertia of the electrical components, the load parameters are converted into time response characteristics to obtain an equivalent thermal effect sequence that is consistent with the time response characteristics of the actual temperature parameters.

[0008] Based on the load parameters and the rate of change determined by the load parameters, the operating cycle of the electrical components is divided, and the context information within each operating cycle is obtained. A benchmark model is established with equivalent thermal action sequence and environmental parameters as input, and reflects the operating parameter response of the electrical components under healthy conditions. Based on the benchmark model, the theoretical normal temperature parameters and theoretical normal voltage parameters are derived under the current operating conditions.

[0009] The difference between the actual temperature parameters and the theoretical normal temperature parameters is calculated to obtain the temperature difference sequence. The difference between the actual voltage parameters and the theoretical normal voltage parameters is calculated to obtain the voltage difference sequence. The temperature difference sequence and the voltage difference sequence together constitute the parameter difference sequence reflecting the additional effects of degradation.

[0010] The parameter difference sequence is aggregated to obtain multiple characteristic indicators reflecting the degree of degradation of electrical components. The multiple characteristic indicators are then fused, and based on the synergistic change relationship between the multiple characteristic indicators, a comprehensive degradation state characterization value of the electrical components is generated. The synergistic change relationship includes the unidirectional change relationship of multiple characteristic indicators within the same operating condition cycle. The comprehensive degradation state characterization value generated within each operating condition cycle is stored to form a historical evolution sequence.

[0011] Based on the comprehensive degradation status characterization value, the fault latency determination process is executed, which includes normal status, suspected latency period, and confirmed latency period;

[0012] When the process is suspected to be in the latent period, the evolution trend of the comprehensive degradation state characterization value is analyzed based on the historical evolution sequence, and the cross-operational cycle comparison and verification are carried out to obtain the evolution trend analysis results.

[0013] When the evolution trend analysis results indicate that the comprehensive degradation state value meets the preset continuous deterioration condition, or when the comparison and verification of the comprehensive degradation state value across operating cycle is abnormally increased, the judgment process will be switched to the diagnosis latency period, and the starting point of the fault latency period will be confirmed.

[0014] Based on the starting point of the fault latency period, an evolution trajectory of the comprehensive degradation state characterization value is constructed, and based on the evolution trajectory, the remaining service life of electrical components is predicted.

[0015] Furthermore, the parameter difference sequences are aggregated to obtain multiple characteristic indicators reflecting the degree of degradation of electrical components, including:

[0016] Based on the context information within each operating cycle, determine the low load duration between the current operating cycle and the previous operating cycle, and obtain the temperature difference corresponding to the start time of the high load window of the current operating cycle from the temperature difference sequence.

[0017] Based on the physical thermal inertia of the electrical components, determine the thermal time constant of the electrical components, and judge whether the duration of low load is sufficient to completely dissipate the additional heat generated in the previous operating cycle based on the thermal time constant.

[0018] When the duration of low load is insufficient to completely dissipate the additional heat generated in the previous operating cycle, a virtual difference baseline function that decays exponentially over time is constructed. The virtual difference baseline function starts with the temperature difference corresponding to the beginning of the high load window of the current operating cycle and decays according to the thermal time constant.

[0019] Within the high-load window of the current operating cycle determined based on context information, the virtual difference baseline value at the corresponding moment is subtracted from the temperature difference in the temperature difference sequence to obtain the corrected dynamic degradation difference.

[0020] By integrating the corrected dynamic degradation difference over time, the cumulative additional temperature rise characteristics after stripping away the historical residual effects are obtained.

[0021] When the low load duration is sufficient to completely dissipate the additional heat generated in the previous operating cycle, the temperature difference sequence is integrated over time during the high load window of the current operating cycle to obtain the cumulative additional temperature rise characteristics.

[0022] The voltage difference sequence is averaged during the high-load window of the current operating cycle to obtain the average additional voltage drop characteristic, and the cumulative additional temperature rise characteristic and the average additional voltage drop characteristic are used as multiple characteristic indicators.

[0023] Secondly, this application also discloses an electrical component fault early warning system, comprising:

[0024] The equivalent thermal effect sequence generation module is used to obtain the actual temperature parameters, actual voltage parameters, load parameters and environmental parameters of electrical components. Based on the physical thermal inertia of electrical components, the load parameters are converted into time response characteristics to obtain an equivalent thermal effect sequence that is consistent with the time response characteristics of the actual temperature parameters.

[0025] The operating cycle and benchmark model processing module is used to divide the operating cycle of electrical components based on load parameters and the rate of change determined by the load parameters, and to obtain the context information within each operating cycle. It establishes a benchmark model with equivalent thermal action sequence and environmental parameters as input, and reflects the operating parameter response of electrical components under healthy conditions. Based on the benchmark model, the theoretical normal temperature parameters and theoretical normal voltage parameters are derived under the current operating conditions.

[0026] The parameter difference sequence generation module is used to calculate the difference between the actual temperature parameter and the theoretical normal temperature parameter to obtain the temperature difference sequence, and to calculate the difference between the actual voltage parameter and the theoretical normal voltage parameter to obtain the voltage difference sequence. The temperature difference sequence and the voltage difference sequence together constitute the parameter difference sequence reflecting the degradation additional effect.

[0027] The comprehensive characterization and historical evolution sequence generation module is used to aggregate parameter difference sequences to obtain multiple feature indicators reflecting the degree of degradation of electrical components. It also integrates multiple feature indicators and generates a comprehensive degradation state characterization value of electrical components based on the synergistic change relationship between multiple feature indicators. The synergistic change relationship includes the unidirectional change relationship of multiple feature indicators within the same operating condition cycle. The module stores the comprehensive degradation state characterization value generated within each operating condition cycle to form a historical evolution sequence.

[0028] The incubation period determination process execution module is used to execute the fault incubation period determination process based on the comprehensive degradation state characterization value. The determination process includes normal state, suspected incubation period and confirmed incubation period.

[0029] The evolution trend analysis module is used to analyze the evolution trend of the comprehensive degradation state characterization value based on the historical evolution sequence when the judgment process is in the suspected latent period, and to compare and verify it across operating cycle to obtain the evolution trend analysis results.

[0030] The diagnostic switching module is used to switch the judgment process to the diagnostic latency period when the comprehensive degradation status characterization value of the evolution trend analysis results meets the preset continuous deterioration condition, or when the comprehensive degradation status characterization value of the cross-operational cycle comparison verification characterization increases abnormally, and to confirm the starting point of the fault latency period.

[0031] The life prediction module is used to construct the evolution trajectory of the comprehensive degradation state characterization value based on the starting point of the fault latency, and predict the remaining operating life of electrical components based on the evolution trajectory.

[0032] Beneficial Effects: This application isolates the impact of load fluctuations on temperature response by constructing an equivalent thermal effect sequence, establishes a benchmark model reflecting the health status, and calculates the difference sequence between actual parameters and theoretical normal parameters, thereby effectively isolating the interference of operating condition fluctuations and accurately extracting the additional effects caused by component degradation. By dividing the operating condition cycle and obtaining contextual information, combined with the fusion of multiple feature indicators and the analysis of synergistic changes, a comprehensive degradation state characterization value is generated, realizing a global quantitative assessment of the degradation state. Furthermore, by establishing a three-level judgment process including normal state, suspected latent period, and confirmed latent period, and performing cross-cycle comparison verification based on historical evolution sequences, the true degradation starting point and operating condition instantaneous anomalies are effectively distinguished, avoiding premature or late confirmation of the latent period starting point. Finally, by constructing an evolution trajectory and predicting the remaining operating life, accurate prediction of fault development trends is achieved. Compared with existing technologies, this application significantly improves the accuracy, continuity, and reliability of fault early warning under complex operating conditions with frequent load fluctuations, effectively solves the technical problem of difficult fault latent period initiation determination, and provides a scientific basis for preventive maintenance of electrical components. Attached Figure Description

[0033] Figure 1 This is a schematic diagram of a fault early warning method for electrical components provided in this application.

[0034] Figure 2 A schematic flowchart of an electrical component fault early warning method is provided in another embodiment of this application.

[0035] Figure 3 This is a schematic diagram of the structure of an electrical component fault early warning system provided in this application.

[0036] In the diagram: 1. Equivalent thermal effect sequence generation module; 2. Operating condition cycle and benchmark model processing module; 3. Parameter difference sequence generation module; 4. Comprehensive characterization and historical evolution sequence generation module; 5. Latency period determination process execution module; 6. Evolution trend analysis module; 7. Diagnostic switching module; 8. Lifetime prediction module. Detailed Implementation

[0037] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0038] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0039] In industrial power distribution systems or large-scale industrial control scenarios, electrical components such as circuit breakers, contactors, or power relays typically undertake frequent load switching tasks. Taking the main circuit breaker in a distribution cabinet serving a group of CNC machine tools as an example, it undergoes dozens or even hundreds of load switching cycles daily, with the load current fluctuating between a few amperes under no-load conditions and hundreds of amperes under full load. Under such operating conditions, when the conductive connections of the circuit breaker experience slight contact defects due to long-term mechanical vibration or electro-corrosion, the contact resistance increases during high-load periods, leading to localized Joule heating and measurable temperature rise and voltage drop. When switching to a low-load state, as the current decreases and heat gradually dissipates, these abnormal characteristics subside and approach normal levels. The intermittent and periodic abnormal behaviors mentioned above highly overlap with the normal operating condition responses caused by load fluctuations in the time domain, making it difficult for the monitoring system to determine the starting point of the fault latency period: if the first short-term abnormality is directly identified as the starting point of degradation, the subsequent warning timeline will be unnecessarily lengthened, leading to over-maintenance; if it is regarded as a load disturbance and ignored, the real degradation process is interrupted, and the alarm is only triggered after the abnormality continues to worsen, which may easily lead to missing the maintenance window.

[0040] To address the problem of latency determination in scenarios with frequent load fluctuations, this application proposes a fault early warning method based on physical thermal inertia modeling and multi-cycle collaborative verification. Before detailing the specific technical implementation, several key concepts need to be explained. Physical thermal inertia refers to the physical property of electrical components where the temperature response lags behind the instantaneous power change due to their own thermal capacity and thermal resistance characteristics. This lag characteristic causes the temperature change during sudden load changes to exhibit an exponential stabilization rather than an instantaneous jump. Taking a molded case circuit breaker with a rated current of 400A as an example, the thermal capacity of its contact system is mainly composed of the superposition of the thermal capacities of the copper alloy contacts, conductive plates, and insulating shell. In a set of example parameters consistent with subsequent calibration results, the total thermal capacity C is approximately 136.4 J / K, and the steady-state thermal resistance R_th from the contacts to the environment is approximately 0.88 K / W. Therefore, the thermal time constant τ = R_th × C ≈ 120 seconds is calculated, meaning that when the load suddenly increases from 0 to the rated value, the contact temperature needs approximately 360 seconds (3τ) to reach 95% of the steady-state temperature rise. The operating cycle refers to a continuous operating period divided according to the amplitude and rate of change of load parameters. Within each cycle, the load maintains a relatively stable trend, such as a high-load steady-state period, a load-reducing transition period, and a low-load steady-state period. Contextual information encompasses situational parameters such as load level, ambient temperature, and duration within each operating cycle, used to characterize the operating conditions of that cycle. The benchmark model is a mapping relationship constructed based on historical operating data of healthy electrical components, capable of extrapolating theoretical normal temperature and voltage responses based on thermal inputs and environmental conditions. The parameter difference sequence is the deviation sequence between the actual acquired parameters and the parameters extrapolated from the benchmark model, including both measurement noise and systemic additional effects caused by component degradation. The comprehensive degradation state characterization value is a single quantitative index obtained by fusing multi-dimensional difference features, used to characterize the degradation state of electrical components during operating condition fluctuations.

[0041] Reference Figure 1 Based on the above conceptual framework, the implementation process of the technical solution in this application is as follows:

[0042] S1000: Acquires the actual temperature parameters, actual voltage parameters, load parameters, and environmental parameters of electrical components. Based on the physical thermal inertia of the electrical components, it performs time response characteristic conversion on the load parameters to obtain an equivalent thermal action sequence that is consistent with the time response characteristics of the actual temperature parameters.

[0043] S2000: Based on load parameters and the rate of change determined by the load parameters, the operating cycle of electrical components is divided, and the context information within each operating cycle is obtained. A benchmark model is established with equivalent thermal action sequence and environmental parameters as input, and reflects the operating parameter response of electrical components under healthy conditions. Based on the benchmark model, theoretical normal temperature parameters and theoretical normal voltage parameters are derived under the current operating conditions.

[0044] S3000: Calculate the difference between actual temperature parameters and theoretical normal temperature parameters to obtain a temperature difference sequence, and calculate the difference between actual voltage parameters and theoretical normal voltage parameters to obtain a voltage difference sequence. The temperature difference sequence and the voltage difference sequence together constitute a parameter difference sequence reflecting the additional effects of degradation.

[0045] S4000: Aggregates the parameter difference sequence to obtain multiple characteristic indicators reflecting the degree of degradation of electrical components, merges multiple characteristic indicators, and generates a comprehensive degradation state characterization value of electrical components based on the synergistic change relationship between multiple characteristic indicators. The synergistic change relationship includes the same-direction change relationship of multiple characteristic indicators in the same operating condition cycle. It stores the comprehensive degradation state characterization value generated in each operating condition cycle to form a historical evolution sequence.

[0046] S5000: Based on the comprehensive degradation state characterization value, execute the fault latency determination process, which includes normal state, suspected latency period and confirmed latency period;

[0047] S6000: When the process is suspected to be in the latent period, the evolution trend of the comprehensive degradation state characterization value is analyzed based on the historical evolution sequence, and the cross-operational cycle comparison and verification are carried out to obtain the evolution trend analysis results.

[0048] S7000: When the comprehensive degradation state characterization value of the evolution trend analysis result meets the preset continuous deterioration condition, or when the comprehensive degradation state characterization value of the cross-operational cycle comparison verification characterization increases abnormally, the judgment process will be switched to the diagnosis latency period, and the starting point of the fault latency period will be confirmed.

[0049] S8000: Based on the starting point of the fault latency, construct the evolution trajectory of the comprehensive degradation state characterization value, and predict the remaining service life of electrical components based on the evolution trajectory.

[0050] In specific implementation, load parameters include load current, load power, or load power calculated from load current and voltage; preset continuous deterioration conditions include at least the rate of change of the comprehensive degradation state characterization value continuously increasing over multiple consecutive operating cycles, and the rate of change exceeding a preset acceleration threshold; abnormal increase includes at least the current operating cycle exceeding a preset statistical range relative to historical operating cycles with similar context information. A multi-sensor collaborative acquisition network needs to be established for parameter acquisition. Temperature monitoring can be performed by arranging a PT100 platinum resistance temperature sensor at each of the three-phase contacts of the circuit breaker, using a three-wire connection to eliminate the influence of lead resistance, converting it into a 4-20mA standard current signal through a signal conditioning circuit, and acquiring it at a 1Hz sampling frequency using a data acquisition card. The signal is then converted into the actual temperature value using the calibration formula T=(I_T-4)×(150-(-20)) / (20-4)+(-20), where I_T is the current signal corresponding to the temperature sensor, in mA, and the temperature range covers -20℃ to 150℃. Voltage parameters are obtained by installing 0.5-accuracy voltage transformers at the three-phase input terminals, with a transformation ratio of 400V / 100V. The secondary side is processed by an AD536A true RMS converter chip to obtain a DC voltage signal proportional to the actual RMS voltage value. After sampling by a 16-bit ADC, assuming the ADC count value is N_V and the full scale corresponds to 400V on the primary side, the actual voltage value can be restored using V_actual = N_V × 400 / 65535. Load parameters are obtained by using a through-type current transformer (600A / 5A transformation ratio) with a precision sampling resistor to obtain the current signal. Combined with the real-time voltage, the apparent power S = √3 × U × I_L is calculated as the load parameter, where U is the RMS value of the three-phase line voltage and I_L is the RMS value of the load current. The sampling frequency is 1Hz, synchronized with the temperature. The ambient temperature is measured by installing a DS18B20 digital temperature sensor at the air inlet of the distribution cabinet, with an accuracy of ±0.5℃, and the reading is updated every 30 seconds.

[0051] For the physical thermal inertia conversion, a first-order RC equivalent thermal circuit model is used to simulate the thermal dynamic response of the component. The establishment of this model requires experimental determination of the heat capacity C and thermal resistance R_th parameters. The specific calibration method is as follows: at an ambient temperature of 25℃, a step load (suddenly increasing from 0 to 80% of the rated current) is applied to the circuit breaker, and the response curve of the contact temperature rising from the initial value to the steady-state value is recorded. The least squares method is used to fit the exponential curve T(t) = T_steady_state × (1 - exp(-t / τ)) + T_initial, where T(t) is the contact temperature at time t, T_steady_state is the steady-state temperature, T_initial is the initial temperature, and τ is the thermal time constant. The thermal time constant τ for this type of circuit breaker is found to be 120 seconds. The thermal resistance R_th is calculated as the ratio of the steady-state temperature rise ΔT to the power consumption P_loss, R_th = ΔT / P_loss. The steady-state temperature rise at a current of 320A is measured to be 45K, and the power consumption is P_loss = I_L. 2×R_contact≈320 2 ×0.0005=51.2W, wherein R_contact is contact resistance, therefore R_th≈45 / 51.2≈0.88K / W, and further the heat capacity C=τ / R_th≈136.4J / K. The instantaneous load power P(t) is input into the equivalent thermal circuit, and the output Q(t) is the equivalent thermal action sequence. The mathematical expression is τ·dQ / dt+Q=P(t). The differential equation is discretely solved by the forward Euler method: Q(t+Δt)=Q(t)+(P(t)-Q(t))×Δt / τ, wherein P(t) is the instantaneous load power at time t, Q(t) is the value of the equivalent thermal action sequence at time t, Δt is the sampling interval and is set to 1 second in this embodiment. This sequence smoothes the instantaneous fluctuation of the load and remains consistent with the inertial response of the temperature sensor.

[0052] Working condition cycle division can implement state machine switching based on the load change rate. In specific implementation, the power change rate dP / dt between consecutive sampling points is calculated, a positive threshold θ_up=5kW / s (for load increase determination), a negative threshold θ_down=-5kW / s (for load decrease determination), a high load threshold P_high=240kW (60% of rated power) and a low load threshold P_low=40kW (10% of rated power) are set. When dP / dt>θ_up, it is determined as load increase transition; when dP / dt<θ_down, it is determined as load decrease transition; when dP / dt is between the thresholds and P>P_high, it is determined as high-load steady state; when P<P_low, it is determined as low-load steady state, and the remaining cases belong to the medium-load transition region. An operating condition cycle is a complete load fluctuation process starting from the end of one low-load period, successively passing through load increase, high-load window, load decrease and returning to the low-load steady state. The average load P_avg, ambient temperature T_env, and duration Δt_cycle within the cycle are recorded as context information, with the storage format {P_avg, T_env, Δt_cycle, starting timestamp}, wherein Δt_cycle is used to represent the duration of the operating condition cycle to distinguish it from the sampling interval Δt.

[0053] The baseline model was established using a hybrid approach combining data-driven and physical mechanism methods. First, under healthy component conditions (contact resistance <50μΩ confirmed by DC resistance testing), at least three months of historical operating data were collected, covering combinations of different load levels (20%, 40%, 60%, 80%, 100% rated load) and ambient temperatures (15℃, 25℃, 35℃, 45℃). After preprocessing, the data was used as input features, with the equivalent thermal effect sequence Q(t) and ambient temperature T_env as input features, and the actual temperature T_actual and voltage drop ΔV as output labels, to train a Long Short-Term Memory (LSTM) network. The network structure was configured with two neurons in the input layer (Q, T_env), two LSTM layers with 64 neurons each in the hidden layers, and two neurons in the output layer (T_theory, ΔV_theory). The Adam optimizer was used with a learning rate of 0.001, and the loss function was the mean squared error (MSE). When the validation set MSE < 0.5 and no longer decreased for 10 consecutive epochs, the model parameters were saved, resulting in the baseline model. In online applications, the real-time calculated Q(t) and T_env are input into the model to perform forward calculations, obtaining the theoretical normal temperature parameters and theoretical normal voltage parameters. The parameter difference sequence is obtained by point-by-point subtraction, such as temperature difference ΔT_diff(t) = T_actual(t) - T_theory(t), voltage difference ΔV_diff(t) = V_actual(t) - V_theory(t), where T_actual(t) is the actual temperature parameter at time t, T_theory(t) is the theoretical normal temperature parameter at time t, V_actual(t) is the actual voltage parameter at time t, and V_theory(t) is the theoretical normal voltage parameter at time t.

[0054] In the feature aggregation stage, a sliding window integration is performed on the temperature difference sequence. The window length is set to the duration of the high-load window period T_window in the current operating cycle, and the integration formula is A_T=∫ΔT_diff(t)dt, using the trapezoidal rule for numerical integration: A_T=Σ(ΔT_diff,i+ΔT_diff,i+1)×Δt / 2. The voltage difference sequence is averaged to obtain the average additional voltage drop feature: A_V=ΣΔV_diff,i / N, where A_T is the cumulative additional temperature rise feature, A_V is the average additional voltage drop feature, and N is the number of sampling points within the current high-load window period. In the fusion stage, a two-dimensional feature vector X=[A_T,A_V] is constructed, and its Mahalanobis distance with the healthy baseline vector is calculated as the comprehensive degradation state characterization value D. The historical evolution sequence stores this characterization value in chronological order, forming a degradation trajectory, which is stored in a circular buffer, retaining data from the most recent 200 cycles. The judgment process is set up with three levels of status: under normal conditions, D < 1.0; under suspected incubation period, 1.0 ≤ D < 2.5, at which point trend analysis is initiated; under confirmed incubation period, D ≥ 2.5 or the condition of continuous deterioration is met. Cross-cycle comparison verification is performed by conducting a two-sample t-test between the current cycle's characteristic value and the characteristic value of historical similar operating condition cycles (|P_avg_diff| < 10kW and |T_env_diff| < 3℃), calculating the p-value, and if p < 0.05, the difference is considered significant, where P_avg_diff is the difference between the average load of the current cycle and the average load of historical similar operating condition cycles, and T_env_diff is the difference between the ambient temperature of the current cycle and the ambient temperature of historical similar operating condition cycles.

[0055] In another embodiment of this application, it is further proposed to aggregate the parameter difference sequence to obtain multiple characteristic indicators reflecting the degree of degradation of electrical components, including:

[0056] S4100: Based on the context information within each operating cycle, determine the low load duration between the current operating cycle and the previous operating cycle, and obtain the temperature difference corresponding to the start time of the high load window of the current operating cycle from the temperature difference sequence.

[0057] S4200: Based on the physical thermal inertia of the electrical components, determine the thermal time constant of the electrical components, and determine whether the duration of low load is sufficient to completely dissipate the additional heat generated in the previous operating cycle based on the thermal time constant.

[0058] S4300: When the duration of low load is insufficient to completely dissipate the additional heat generated in the previous operating cycle, a virtual difference baseline function that decays exponentially over time is constructed. The virtual difference baseline function starts with the temperature difference corresponding to the beginning of the high load window of the current operating cycle and decays according to the thermal time constant.

[0059] S4400: During the high-load window period of the current operating cycle determined based on context information, the virtual difference baseline value at the corresponding moment is subtracted from the temperature difference in the temperature difference sequence to obtain the corrected dynamic degradation difference.

[0060] S4500: Time integration of the corrected dynamic degradation difference yields the cumulative additional temperature rise characteristics after stripping away historical residual effects;

[0061] S4600: When the low load duration is sufficient to completely dissipate the additional heat generated in the previous operating cycle, the temperature difference sequence is integrated over time within the high load window of the current operating cycle determined based on context information to obtain the cumulative additional temperature rise characteristics.

[0062] S4700: The voltage difference sequence is averaged within the high-load window of the current operating condition cycle determined based on context information to obtain the average additional voltage drop characteristic, and the cumulative additional temperature rise characteristic and the average additional voltage drop characteristic are used as multiple characteristic indicators.

[0063] In practical implementation, the determination of the thermal time constant τ needs to be combined with the physical characteristics of the specific components. For a molded case circuit breaker with a rated current of 400A, it can be determined experimentally: at an ambient temperature of 25℃, first operate at 80% of the rated current (320A) until thermal equilibrium is reached, and record the steady-state temperature rise ΔT_steady; then disconnect the load and record the time required for the temperature to drop from ΔT_steady to 36.8%ΔT_steady. This time is the thermal time constant τ, where ΔT_steady is the steady-state temperature rise of the contact temperature relative to the ambient temperature during stable operation. The measured value for this type of circuit breaker is τ = 120 seconds. The threshold for judging the low load duration is set to 3τ = 360 seconds. If the low load duration t_rest < 360 seconds, it is considered that the additional heat generated in the previous operating cycle has not been completely dissipated; if the low load duration t_rest ≥ 360 seconds, it is considered that the additional heat generated in the previous operating cycle has been basically dissipated.

[0064] When the additional heat generated in the previous operating cycle has not been completely dissipated, the virtual difference baseline function is constructed as ΔT_base(t) = ΔT_start·exp(-(t-t_start) / τ), where ΔT_start is the temperature difference corresponding to the start time t_start of the high-load window of the current operating cycle, t is the current time, τ is the thermal time constant, and ΔT_base(t) is the virtual difference baseline value at time t. This exponential function is used to simulate the natural decay trend that the residual heat from the previous operating cycle should exhibit in the current operating cycle when there is no new degradation increment. By subtracting the virtual difference baseline value from the measured temperature difference, the corrected dynamic degradation difference is obtained, thereby eliminating the influence of historical residual heat on the degradation characteristics of the current operating cycle.

[0065] For example, if the low load lasts only 60 seconds after the previous high load cycle ends, which is less than 3τ = 360 seconds and does not meet the complete dissipation condition, and the temperature difference at the beginning of the current cycle's high load window is 5℃, then the virtual difference baseline function decays exponentially from 5℃. At the 30th second of the current cycle, the virtual difference baseline value is 5·exp(-30 / 120)≈3.89℃. If the measured temperature difference at this moment is 6℃, then the corrected dynamic degradation difference is 6-3.89=2.11℃, which reflects the new degradation increment in the current operating cycle. The corrected dynamic degradation difference is accumulated using the trapezoidal integral method to obtain the cumulative additional temperature rise feature ΔT_cum=∫(ΔT_measured(t)-ΔT_base(t))dt, where ΔT_cum is the cumulative additional temperature rise feature, ΔT_measured(t) is the measured temperature difference in the temperature difference sequence at time t, and ΔT_base(t) is the virtual difference baseline value at time t.

[0066] When the duration of low load is sufficient to completely dissipate the additional heat generated in the previous operating cycle, there is no need to construct a virtual difference baseline function. The temperature difference sequence can be directly integrated over time within the high-load window of the current operating cycle determined by context information to obtain the cumulative additional temperature rise characteristic. Simultaneously, the voltage difference sequence is averaged within the high-load window of the current operating cycle determined by context information to obtain the average additional voltage drop characteristic. The cumulative additional temperature rise characteristic and the average additional voltage drop characteristic are used as multiple feature indicators, thereby enabling temperature-side features and voltage-side features to jointly characterize the degree of degradation of electrical components.

[0067] This technical solution can eliminate the interference of residual thermal history on the current degradation assessment, avoid misjudging the heat accumulated in the previous operating cycle as the degradation aggravation in the current operating cycle, and improve the accuracy and physical consistency of degradation feature extraction in scenarios with frequent load switching.

[0068] In another embodiment of this application, it is further proposed to integrate multiple characteristic indicators and generate a comprehensive degradation state characterization value of electrical components based on the synergistic change relationship between the multiple characteristic indicators, including:

[0069] S4800: Determines the load level and ambient temperature corresponding to the current operating cycle based on the context information within the current operating cycle;

[0070] S4810: Based on the load level and ambient temperature, select correlation parameters that match the current operating cycle. Correlation parameters are used to characterize the synergistic change relationship between multiple characteristic indicators.

[0071] S4820: Based on multiple characteristic indicators within the current operating cycle, construct the current degradation feature vector, and based on the values ​​of multiple characteristic indicators in the health state of electrical components, construct the health baseline feature vector.

[0072] S4830: Calculate the distance between the current degraded feature vector and the healthy baseline feature vector based on the correlation parameter;

[0073] S4840: Generate a comprehensive degradation status characterization value for electrical components based on distance.

[0074] In practical implementation, the establishment of correlation parameters requires multi-condition calibration under healthy conditions. The specific steps are as follows: Assuming the components are confirmed to be healthy, collect operating data covering different load levels (30%, 50%, 70%, 90% of rated load) and ambient temperatures (20℃, 30℃, 40℃). Calculate the statistical characteristics of the cumulative additional temperature rise characteristic x1 and the average additional voltage drop characteristic x2 under each operating condition, where x1 represents the cumulative additional temperature rise characteristic and x2 represents the average additional voltage drop characteristic. Construct an operating condition-covariance mapping table. The following table illustrates the covariance parameters corresponding to some operating condition combinations:

[0075] Table 1: Covariance parameters for some operating condition combinations

[0076]

[0077] When applied online, based on the current load level P_avg (e.g., 85% of rated load) and ambient temperature T_env (e.g., 38℃), the covariance matrix Σ=[[σ11,σ12],[σ12,σ22]] of the current operating condition is obtained from the operating condition-covariance mapping table through bilinear interpolation. Here, P_avg is the average load level corresponding to the current operating condition cycle, T_env is the ambient temperature corresponding to the current operating condition cycle, σ11 is the variance of the cumulative additional temperature rise feature x1, σ22 is the variance of the average additional voltage drop feature x2, σ12 is the covariance between the cumulative additional temperature rise feature x1 and the average additional voltage drop feature x2, and Σ is the covariance matrix composed of σ11, σ12 and σ22. The current degradation feature vector is X = [x1, x2], and the healthy baseline vector is μ = [μ1, μ2], where μ1 and μ2 are the mean values ​​of the features corresponding to the current operating condition under the healthy state, and are stored together with the covariance parameter in the operating condition-covariance mapping table. The distance is calculated using Mahalanobis distance D = sqrt((X-μ)^T·Σ^(-1)·(X-μ)), where D is the comprehensive degradation state characterization value, (X-μ)^T is the transpose of the difference between the current degradation feature vector and the healthy baseline vector, and Σ^(-1) is the inverse of the covariance matrix Σ. This distance metric considers the synergistic changes among multiple feature indicators, avoiding the problem of repeated measurement of the deviation degree of related features when using Euclidean distance.

[0078] In another embodiment of this application, a fault latency determination process is further proposed based on the comprehensive degradation state characterization value, including:

[0079] S5100: Obtain the comprehensive degradation status characterization value within the current operating cycle, obtain the current status of the judgment process, and determine the load level and ambient temperature corresponding to the current operating cycle based on the context information within the current operating cycle.

[0080] S5200: Based on the load level and ambient temperature, selects a state switching rule that matches the current operating cycle from a preset set of operating condition-related state switching rules;

[0081] S5300: Based on the state switching rules, determine whether the comprehensive degradation state characterization value meets the switching conditions for switching from the current state to the next state. The switching conditions include the state duration requirement and the state regression inhibition condition. The next state includes switching from the normal state to the suspected incubation period, or switching from the suspected incubation period to the confirmed incubation period.

[0082] S5400: When the comprehensive degradation state characterization value meets the switching conditions, the judgment process will be switched to the next state;

[0083] S5500: When the comprehensive degradation state characterization value does not meet the switching conditions, maintain the current state of the judgment process.

[0084] In practical implementation, the state switching rule set is constructed using a condition-based threshold strategy. State switching rules can be set according to load ranges and temperature ranges. The following example lists the rules corresponding to three typical operating condition combinations:

[0085] Table 2: Rules corresponding to three typical operating condition combinations

[0086]

[0087] During online operation, the system determines the threshold by looking up values ​​in a table based on the current P_avg and T_env, where P_avg is the average load level corresponding to the current operating cycle, and T_env is the ambient temperature corresponding to the current operating cycle. When the current operating condition does not directly fall into one of the typical combinations listed in the table, a stricter state switching rule can be selected based on adjacent operating cycle intervals, or the corresponding threshold can be determined through a preset interpolation method. The state duration requirement refers to the requirement that the comprehensive degradation state characterization value D must exceed the corresponding threshold for N consecutive operating cycle periods before a state switch can be triggered, preventing instantaneous spike interference. Here, D is the comprehensive degradation state characterization value, and N is the number of consecutive operating cycle periods required to trigger a state switch. The state rollback suppression condition refers to the confirmation period of M operating cycle after the judgment process enters the suspected latent period or confirmed latent period. During this period, even if the comprehensive degradation state characterization value D drops briefly, it will not immediately roll back to the lower risk state. The judgment process is only allowed to roll back when the comprehensive degradation state characterization value D is lower than the maintenance threshold corresponding to the current state for M consecutive operating cycle, and the switching conditions for the next state are not met. M is the number of consecutive confirmation cycles that must be met before the state rollback is allowed.

[0088] For example, under medium load and medium temperature conditions, when D > 1.0 for three consecutive operating cycles, the system switches from the normal state to the suspected incubation period. After entering the suspected incubation period, even if D drops to 0.9 in the fourth operating cycle, the system still maintains the suspected incubation period state. Only when the subsequent five consecutive operating cycles are all below 1.0 and the conditions for maintaining the suspected incubation period or switching from suspected to confirmed cases are not met again, will the system consider reverting from the suspected incubation period to the normal state.

[0089] In another embodiment of this application, it is further proposed that, based on the state transition rules, the determination of whether the comprehensive degradation state characterization value meets the transition condition from the current state to the next state includes:

[0090] S5310: Determine whether the load level exceeds the preset instantaneous overload threshold, or whether the rate of change of ambient temperature determined based on the ambient temperature of the current operating cycle and the previous operating cycle exceeds the preset sudden change threshold of the environment.

[0091] S5320: When the load level exceeds the preset instantaneous overload threshold, or the rate of change of ambient temperature exceeds the preset sudden change threshold of environment, the current operating condition cycle is determined as the extreme operating condition cycle.

[0092] S5330: When the current operating condition cycle is an extreme operating condition cycle, determine the historical operating condition cycles with similar contextual information from the historical evolution sequence, and compare the comprehensive degradation state characterization value in the current operating condition cycle with the comprehensive degradation state characterization value in the historical operating condition cycle to obtain the extreme operating condition comparison result.

[0093] S5340: When the extreme operating condition comparison result indicates that the deviation between the comprehensive degradation state characterization value in the current operating condition cycle and the comprehensive degradation state characterization value in the historical operating condition cycle is within a preset range, the current state of the judgment process is maintained.

[0094] S5350: When the extreme operating condition comparison result indicates that the deviation between the comprehensive degradation state characterization value in the current operating condition cycle and the comprehensive degradation state characterization value in the historical operating condition cycle exceeds the preset range, the decision-making process should be switched to the next state based on the switching conditions.

[0095] S5360: When the current operating cycle is not an extreme operating cycle, determine whether the process should switch to the next state based on the switching conditions.

[0096] In practical implementation, the instantaneous overload threshold is set to 120% of the rated load (i.e., the power corresponding to 480A), and the environmental change threshold is set to a temperature change exceeding 5℃ per minute (i.e., |ΔT_env / Δt|>5℃ / min). When a sudden increase in current is detected (such as a motor starting impact, with the current suddenly increasing from 100A to 500A) or a sudden rise in ambient temperature (such as an air conditioner malfunction, with the temperature rising from 25℃ to 35℃ within 5 minutes), the system marks that cycle as an extreme operating condition. At this time, historical cycles with similar overload amplitude (±10%) or similar temperature change rate (±1℃ / min) are retrieved from the historical database. These historical cycles are required to occur during the period when the components are confirmed to be healthy (within the first 3 months after installation and no abnormalities detected). The comprehensive degradation state characterization values ​​of these historical cycles under healthy conditions are retrieved as a benchmark, and the statistics are calculated as: mean μ_hist and standard deviation σ_hist. If the current characteristic value D_current falls within the range of [μ_hist-2σ_hist, μ_hist+2σ_hist], it is determined to be a normal response under extreme conditions, and the current state is maintained. If D_current>μ_hist+2σ_hist, it indicates that even under comparable extreme stress, the current degradation has exceeded the historical healthy level, and the state is determined according to the conventional switching conditions. For example, when the current overload is 150%, search for historical periods with overloads of 140%-160% that are confirmed to be healthy. If the historical D value is distributed between 0.5 and 1.0 (μ=0.75, σ=0.125), and the current D=1.5, then 1.5>0.75+2×0.125=1.0, which is determined to be abnormal, and the suspected latency period determination process is initiated.

[0097] This technical solution effectively identifies and isolates the interference of extreme working conditions on degradation judgment. By comparing with similar historical working conditions, it distinguishes between "normal stress response under extreme working conditions" and "abnormal degradation aggravated by extreme working conditions", avoiding misjudgment or omission under extreme conditions and significantly enhancing the accuracy and robustness of the system in abnormal working conditions.

[0098] During the latent phase of diagnosis, the system has confirmed the true starting point of the fault evolution process. However, to reliably predict the remaining service life of electrical components, a deep analysis and verification of the historical evolution trend of the comprehensive degradation state characterization value is necessary. In industrial scenarios with frequent load fluctuations, the degradation process may exhibit nonlinear acceleration or intermittent plateau characteristics, and simple threshold judgment is insufficient to distinguish between temporary fluctuations and continuous deterioration. More importantly, the degradation rate within different operating condition cycles varies significantly due to load levels and ambient temperature. Directly using a single time series analysis method can easily misjudge differences in operating conditions as differences in degradation, or mask the true degradation trend under operating condition noise. Therefore, a verification mechanism is needed to identify accelerated degradation characteristics and perform cross-cycle comparisons of similar operating conditions to confirm the authenticity and persistence of the degradation trend.

[0099] Reference Figure 2 In another embodiment of this application, it is further proposed that when the determination process is in a suspected latent period, the evolution trend of the comprehensive degradation state characterization value is analyzed based on the historical evolution sequence, and a comparison and verification is performed across operating cycle to obtain the evolution trend analysis results, including:

[0100] S6100: Determine the rate of change of the comprehensive degradation state characterization value between adjacent operating condition cycles based on the historical evolution sequence;

[0101] S6200: Based on whether the rate of change of the comprehensive degradation status characterization value continues to increase in multiple consecutive operating condition cycles, and whether the increase in the rate of change of the comprehensive degradation status characterization value exceeds the preset acceleration threshold, it is determined whether the comprehensive degradation status characterization value shows an accelerating deterioration trend, and the acceleration deterioration judgment result is obtained.

[0102] S6300: When performing cross-operational cycle comparison verification, the rate of change of the comprehensive degradation state characterization value of the current operating cycle is compared with the rate of change of the comprehensive degradation state characterization value of the historical operating cycle with similar contextual information to obtain the comparison result.

[0103] S6400: Based on the results of the accelerated deterioration judgment and the comparison results, the evolution trend analysis results are obtained.

[0104] To better understand the implementation details of the above technical solution, the key concepts and implementation process are explained below. The rate of change reflects the speed at which the comprehensive degradation state characterization value evolves per unit time, and is a fundamental indicator for judging whether degradation is accelerating. The acceleration threshold is the critical boundary distinguishing normal fluctuations from accelerated deterioration, and can be preset based on the degradation characteristics of specific components. Similar contextual information mainly refers to operating conditions where the load level and ambient temperature are within a preset deviation range, used to ensure that cross-cycle comparisons are conducted under similar operating conditions, avoiding misjudging differences in operating conditions as differences in degradation.

[0105] In practical applications, the rate of change can be obtained as follows: The system reads continuous sampling point data from the historical evolution sequence in chronological order, denoted as D_i and D_{i-1}, corresponding to the comprehensive degradation state characterization values ​​of the i-th and (i-1)-th operating condition cycles, respectively, and records the time interval Δt_i between the two sampling points. The rate of change v_i is calculated using the difference method, i.e., v_i = (D_i - D_{i-1}) / Δt_i, where D_i is the comprehensive degradation state characterization value of the i-th operating condition cycle, D_{i-1} is the comprehensive degradation state characterization value of the (i-1)-th operating condition cycle, Δt_i is the time interval between the corresponding sampling points of the i-th and (i-1)-th operating condition cycles, and v_i is the rate of change corresponding to the i-th operating condition cycle. To identify the accelerating deterioration trend, the system establishes a rate of change sequence {v_i} and further calculates the rate of change increment a_i = (v_i - v_{i-1}) / Δt_v,i, where v_{i-1} is the rate of change corresponding to the (i-1)th operating cycle, Δt_v,i is the time interval between the calculation times of the rates of change corresponding to v_i and v_{i-1}, and a_i is used to characterize the rate of change increment of the comprehensive degradation state characterization value. The preset acceleration threshold a_threshold can be determined based on the historical degradation data of this component model. For example, by statistically analyzing historical cases confirmed to have entered the accelerated degradation stage, the 95th percentile of the rate of change increment distribution can be taken as the threshold, which can be set to 0.01 / hour for example. 2 When the system detects that a_i is greater than a_threshold for three consecutive operating cycles, it determines that the comprehensive degradation status characterization value is showing an accelerated deterioration trend.

[0106] The implementation process of cross-cycle comparison verification is as follows. The system extracts the load level P_current and ambient temperature T_current from the context information of the current operating cycle. Then, it retrieves historical operating cycles from the historical evolution sequence that satisfy |P_hist - P_current| < 5% of rated load and |T_hist - T_current| < 2℃, where P_current is the load level of the current operating cycle, T_current is the ambient temperature of the current operating cycle, P_hist is the load level of the historical operating cycle, and T_hist is the ambient temperature of the historical operating cycle. The rate of change {v_hist} corresponding to these similar operating cycles is extracted, and its statistical mean μ_v and standard deviation σ_v are calculated, where v_hist is the rate of change corresponding to historical similar operating cycles, μ_v is the mean of the rate of change of historical similar operating cycles, σ_v is the standard deviation of the rate of change of historical similar operating cycles, and v_current is the rate of change corresponding to the current operating cycle. If the current rate of change v_current > μ_v + 2σ_v, then the rate of change of the comprehensive degradation state characterization value of the current operating cycle is determined to be abnormally large compared to similar historical operating cycles. This comparison process can eliminate degradation rate changes caused by differences in operating conditions, making the detected acceleration trend more reflective of the continuous deterioration of the component's own health status.

[0107] This technical solution can identify the accelerated deterioration characteristics of the degradation process, and eliminate the interference of operating condition fluctuations by comparing similar operating conditions across cycles. It can determine whether the evolution trend of the comprehensive degradation status characterization value truly reflects the continuous deterioration of the component's health status, and provide trend criteria for subsequent remaining lifetime prediction.

[0108] After identifying the inception point of the fault latency period and analyzing its evolution trend, constructing a complete and continuous evolution trajectory becomes the foundation for predicting the remaining service life. In actual industrial monitoring environments, the collection of comprehensive degradation state characterization values ​​may be intermittent due to communication interruptions, temporary sensor failures, or system maintenance. Directly ignoring these interruptions during trajectory construction will lead to fragmented degradation processes and information loss, thus affecting the accuracy of life prediction. Furthermore, different operating conditions result in varying confidence levels for degradation assessments; high-load, long-cycle data reflects the true degradation level better than short, transitional data. Therefore, this difference needs to be reflected in trajectory construction. Consequently, a trajectory construction method is required that can handle data gaps, ensure trajectory continuity, and reflect data reliability.

[0109] In another embodiment of this application, it is further proposed that the evolution trajectory of the comprehensive degradation state characterization value be constructed based on the starting point of the fault latency, including:

[0110] S8100: Takes each comprehensive degradation state characterization value formed from the beginning of the fault latency period as a sampling point and obtains the time interval between adjacent sampling points;

[0111] S8200: Determine whether there are data gaps in the comprehensive degradation status characterization values ​​based on the time interval;

[0112] S8300: When there is a data interruption, the comprehensive degradation state characterization value is completed based on the theoretical normal temperature parameters, theoretical normal voltage parameters and environmental parameters in the corresponding time period of the data interruption, and a virtual comprehensive degradation state characterization value is generated.

[0113] S8400: Based on the context information within the operating cycle corresponding to each sampling point, the sampling points are weighted to obtain the weighted sampling points;

[0114] S8500: When there is a data discontinuity, the weighted sampling points and the virtual comprehensive degradation state characterization value are smoothed to construct the evolution trajectory of the comprehensive degradation state characterization value; when there is no data discontinuity, the weighted sampling points are smoothed to construct the evolution trajectory of the comprehensive degradation state characterization value.

[0115] The following example illustrates the process in detail. Data discontinuity identification is based on monitoring the sampling time interval. Under normal circumstances, the system collects data according to a fixed sampling period T_cycle (e.g., one operating cycle every 30 minutes). The system calculates the time interval Δt_sample between adjacent sampling points. When Δt_sample > 1.5 × T_cycle (i.e., 45 minutes), a data discontinuity is determined to exist.

[0116] For data gaps, the generation of the virtual integrated degradation state characterization value needs to be completed in conjunction with the physical model. Specifically, the system retrieves the equivalent thermal effect sequence and environmental parameters for the corresponding time period of the data gap and inputs this data into a pre-established health state baseline model. This baseline model is trained using historical data from healthy components and can output the theoretical normal temperature T_theory and the theoretical normal voltage V_theory. Assuming that during the data gap, the degradation state of the component remains unchanged from the level at the last moment before the gap, i.e., there is no new development of degradation-related effects, the virtual integrated degradation state characterization value D_virtual is calculated as D_virtual = D_last + k × (T_theory - T_env), where D_last is the characterization value of the last valid sampling point before the gap, k is the temperature correction coefficient, and T_env is the ambient temperature. This formula takes into account the residual temperature difference caused by thermal inertia.

[0117] The weighted processing step reflects the varying degrees of importance of data across different operating conditions. Based on the context information corresponding to each sampling point, the system extracts the average load level P_avg and duration Δt_duration for that operating condition cycle, and calculates the weight w_i = (P_avg / P_rated) × (Δt_duration / T_ref), where P_rated is the rated load and T_ref is the reference time (e.g., 1 hour). This means that operating conditions with high load and long duration receive higher weights because their contribution to degradation accumulation is more significant.

[0118] The smoothing process employs cubic spline interpolation to fit curves to the discretely weighted sampling points (including virtual sampling points), generating a continuously differentiable evolutionary trajectory function D(t). This method ensures smooth transitions at data discontinuities and avoids sharp points or unreasonable jumps that may occur with linear connections.

[0119] This technical solution can effectively fill the information gap caused by missing monitoring data. By generating virtual values ​​based on physical models, it ensures the physical consistency of the trajectory. Combined with working condition correlation weighting, it highlights high-contribution data points and ultimately constructs a complete, continuous evolution trajectory that reflects the real degradation process, providing a high-quality data foundation for remaining life prediction.

[0120] In another embodiment of this application, it is further proposed to predict the remaining operational life of electrical components based on their evolution trajectory, including:

[0121] S8600: Based on the latest preset number of sampling points arranged in chronological order in the evolution trajectory, determine the set of terminal sampling points, and based on the change in the comprehensive degradation state characterization value corresponding to adjacent sampling points in the set of terminal sampling points and the time interval between adjacent sampling points, determine the local curvature change of the evolution trajectory and the degradation rate corresponding to the set of terminal sampling points.

[0122] S8700: Determines the current load level and ambient temperature based on context information within the current operating cycle;

[0123] S8800: Adjusts the weight of each sampling point in the end sampling point set in the remaining service life prediction based on local curvature changes, current load level and ambient temperature;

[0124] S8900: Based on the weighted set of end sampling points, a preliminary prediction of the remaining service life of electrical components is made, and a preliminary prediction result is obtained;

[0125] S8910: When the rate of degradation of the comprehensive degradation state characterization value increases due to the change in local curvature, and the difference between the current load level and the load level corresponding to the previous operating cycle exceeds the preset load fluctuation threshold, or the difference between the ambient temperature and the ambient temperature corresponding to the previous operating cycle exceeds the preset ambient temperature fluctuation threshold, a prediction and correction process based on physical failure boundaries is triggered. The physical failure boundaries include the preset critical failure characterization value corresponding to the comprehensive degradation state characterization value.

[0126] S8920: In the prediction and correction process, the preliminary prediction results are constrained and corrected based on the preset failure critical characterization value and the degradation rate corresponding to the end sampling point set to obtain the prediction result;

[0127] S8930: When the prediction correction process is not triggered, the preliminary prediction result is used as the prediction result;

[0128] S8940: Based on the weights of the end sampling points of a preset number of groups, the remaining service life of electrical components is predicted to obtain multiple candidate remaining service life prediction values; the fluctuation range of the prediction result is determined according to the difference between the prediction result and the multiple candidate remaining service life prediction values.

[0129] S8950: When the fluctuation range exceeds the preset robustness threshold, the multi-predictor fusion mechanism is activated, and the remaining operating life prediction value is generated based on the multi-predictor fusion mechanism.

[0130] S8990: When the fluctuation range does not exceed the preset robustness threshold, the prediction result will be used as the predicted value of the remaining operating life.

[0131] The following explains the specific implementation details of the above prediction process. First, the system extracts the latest 5 sampling points arranged in chronological order from the evolution trajectory, forming the terminal sampling point set {D_{n-4}, D_{n-3}, D_{n-2}, D_{n-1}, D_n}, where D_n is the comprehensive degradation state characterization value corresponding to the latest sampling point, and D_{n-4} to D_{n-1} are the comprehensive degradation state characterization values ​​corresponding to sampling points arranged in chronological order that are earlier than D_n. The local curvature change can be determined based on the degradation rate change of adjacent sampling points: first calculate v_i=(D_i-D_{i-1}) / Δt_i, then calculate κ_i=|v_i-v_{i-1}| / Δt_{v,i}, where v_i is the degradation rate corresponding to the i-th sampling point, Δt_i is the time interval between D_i and D_{i-1}, κ_i is the local curvature change corresponding to the i-th sampling point, and Δt_{v,i} is the time interval between the calculation times corresponding to v_i and v_{i-1}. The degradation rate v_degrad corresponding to the set of terminal sampling points is obtained by dividing the difference between the first and last sampling points at the end by the total time: v_degrad=(D_n-D_{n-4}) / ΣΔt_i, where ΣΔt_i is the sum of the time intervals of all adjacent sampling points from D_{n-4} to D_n.

[0132] The weighting strategy dynamically changes based on the degree of degradation nonlinearity. When the calculated local curvature change κ_i > κ_threshold (e.g., 0.05), it indicates an increase in the degradation rate of the comprehensive degradation state characterization value. In this case, an exponential weighting method is adopted, w_i = exp(α × (in)), and the weights are normalized to obtain w_i' = w_i / Σw_i, where κ_threshold is the local curvature change threshold, α is the sensitivity coefficient, w_i is the initial weight of the i-th sampling point, w_i' is the normalized weight of the i-th sampling point, and Σw_i is the sum of the initial weights of all sampling points in the final sampling point set. Under this exponential weighting method, the i-th value corresponding to the sampling point closer to the current time is closer to n, thus obtaining a higher weight, while the weights of earlier sampling points decay exponentially.

[0133] The preset load fluctuation threshold and ambient temperature fluctuation threshold are set based on operating condition stability requirements. For example, the load fluctuation threshold is set to 20% of the rated load, and the ambient temperature fluctuation threshold is set to 5℃. When |P_current - P_prev| > 20% of the rated load or |T_env_current - T_env_prev| > 5℃ is detected, a predictive correction process is triggered, where P_current is the current load level corresponding to the current operating condition cycle, P_prev is the load level corresponding to the previous operating condition cycle, T_env_current is the ambient temperature corresponding to the current operating condition cycle, and T_env_prev is the ambient temperature corresponding to the previous operating condition cycle.

[0134] The preset failure threshold characterization value D_failure can be obtained through accelerated life testing of the modules. Specifically, a progressive loading test is performed on modules of the same model, and parameters such as contact resistance and temperature rise are monitored. When the contact resistance increases to three times the initial value or the temperature rise exceeds the limit operating temperature of the insulation material (e.g., 150°C), the corresponding comprehensive degradation state characterization value is recorded, which is D_failure. D_failure is the preset failure threshold characterization value corresponding to the comprehensive degradation state characterization value.

[0135] Preliminary predictions can employ an exponential extrapolation model. To avoid inconsistencies in dimensions, the exponential growth coefficient λ is first determined based on the degradation rate corresponding to the set of final sampling points: λ = v_degrad / max(D_n, ε), where λ is the exponential growth coefficient and ε is a preset positive number to prevent D_n from being too small and causing division by zero. Then, a prediction model D(t) = D_n × exp(λ × (t - t_n)) is constructed, where t_n is the sampling time corresponding to the latest sampling point, t is the prediction time, and D(t) is the comprehensive degradation state characterization value corresponding to the prediction time t. Let D(t) = D_failure, and the preliminary remaining lifetime RUL_preliminary is obtained, i.e., RUL_preliminary = (ln(D_failure / D_n)) / λ. In the prediction and correction process, if the prediction endpoint corresponding to the initial remaining lifetime exceeds D_failure, or if the degradation rate corrected based on the current operating conditions causes the prediction curve to reach D_failure ahead of time, then the prediction endpoint is constrained according to D_failure so that the failure time corresponding to the prediction result falls at D(t)=D_failure.

[0136] The fluctuation range can be calculated using the Bootstrap resampling method. The system constructs a preset number of end-sampling point weights based on the last 5 sampling points. For example, 100 sets of end-sampling point weights are formed by random sampling with replacement 100 times. Each set of weights is used to recalculate the remaining lifetime, resulting in 100 candidate remaining lifetime predictions. Based on the difference between the prediction result and the multiple candidate remaining lifetime predictions, the 95% confidence interval width is calculated, and this width is used as the fluctuation range of the prediction result. If the ratio of this width to the predicted median exceeds 20% (a preset robustness threshold), the fluctuation range is determined to exceed the preset robustness threshold, and the multi-predictor fusion mechanism is activated; otherwise, the prediction result is used as the remaining lifetime prediction.

[0137] This technical solution can sense local curvature changes in the degradation trajectory and adaptively adjust prediction weights. When operating conditions change abruptly, prediction correction is performed through physical failure boundaries to avoid overestimation of life caused by nonlinear extrapolation. At the same time, prediction uncertainty is assessed through fluctuation range, and a fusion mechanism is triggered when uncertainty is too high to improve the accuracy, robustness and reliability of remaining life prediction.

[0138] In another embodiment of this application, it is further proposed that when the fluctuation range exceeds a preset robustness threshold, a multi-predictor fusion mechanism is activated, and a predicted value of the remaining operating lifetime is generated based on the multi-predictor fusion mechanism, including:

[0139] S8951: Inputting evolutionary trajectories into a prediction system that integrates physical degradation mechanisms with data-driven approaches;

[0140] S8952: By using a prediction system to perform pattern recognition on the evolutionary trajectory and multiple feature indicators that form the evolutionary trajectory, the degradation pattern recognition results are obtained. The degradation pattern recognition results include single degradation patterns or multi-mode coupled degradation.

[0141] S8953: When the degradation pattern identification result is a single degradation pattern, the prediction system generates a predicted value of the remaining service life based on the evolution trajectory;

[0142] S8954: When the degradation pattern recognition result is multi-mode coupled degradation, the prediction system decomposes the degradation trend corresponding to each degradation mode according to the evolution trajectory and multiple feature indicators that form the evolution trajectory, and adjusts the prediction weight of each degradation mode according to the contribution of each degradation mode to the comprehensive degradation state characterization value at the current stage.

[0143] S8955: The prediction system performs weighted fusion of the degradation trends corresponding to each degradation mode based on the adjusted prediction weights of each degradation mode, and generates the predicted value of the remaining service life.

[0144] The following describes the specific implementation of the multi-predictor fusion mechanism. The fusion prediction system comprises three sub-predictors: a physical degradation model based on the Arrhenius equation (suitable for thermally activated degradation, such as oxidation), a contact fatigue model based on the Paris formula (suitable for mechanical wear), and a data-driven model based on a Long Short-Term Memory (LSTM) network. The construction process of these sub-predictors is as follows: The Arrhenius model obtains the activation energy E_a and the pre-exponential factor A_arr by performing accelerated aging tests at different temperatures, where E_a is the activation energy and A_arr is the pre-exponential factor in the Arrhenius model; the Paris model obtains the crack propagation coefficient C_P and the exponent m through cyclic loading tests, where C_P is the crack propagation coefficient in the Paris formula and m is the crack propagation exponent, and C_P is used to avoid confusion with the heat capacity C; the LSTM model uses historical degradation trajectory data to obtain network weights through time series prediction training.

[0145] Degradation pattern recognition is achieved through feature extraction and cluster analysis. The system extracts the following feature indicators from the evolutionary trajectory: the variance of the curvature sequence, the correlation coefficient ρ between temperature and voltage differences, and the trend of temperature rise rate. A three-dimensional feature vector X_mode=[σ_κ,ρ,dT_rise / dt] is constructed, where X_mode is the feature vector used for degradation pattern recognition, σ_κ is the variance of the curvature sequence, ρ is the correlation coefficient between temperature and voltage differences, dT_rise / dt is the temperature rise rate, and T_rise represents the temperature rise value. The K-means clustering algorithm (preset K=2) is used to cluster these feature vectors. If the current feature vector is closer to the cluster center marked as multi-mode coupling, it is determined to be multi-mode coupling degradation; otherwise, it is determined to be a single degradation pattern.

[0146] The contribution of each degradation mode is calculated using the projection pursuit method. It is assumed that the comprehensive degradation state characterization value D can be decomposed into a linear superposition of the contributions of each mode: D = β_1 × D_oxidation + β_2 × D_wear + β_3 × D_residual, where D_oxidation, D_wear, and D_residual represent the theoretical degradation components corresponding to oxidation, wear, and other residual factors, respectively, and β_1, β_2, and β_3 are the contribution coefficients of the corresponding degradation modes. In the actual solution, a fitting objective minΣ_j|D_j-(β_1 × D_oxidation,j + β_2 × D_wear,j + β_3 × D_residual,j)| can be established based on multiple sampling points within the end sampling point set. 2Where D_j is the comprehensive degradation state characterization value corresponding to the j-th sampling point, and D_oxidation,j, D_wear,j, and D_residual,j are the theoretical degradation components corresponding to oxidation, wear, and other residual factors at the j-th sampling point, respectively. The contribution coefficient β_i is solved by least squares fitting, and the prediction weight is determined based on the contribution coefficient. The prediction weight can be expressed as w_i=max(β_i,0) / Σmax(β_i,0) to avoid negative weights due to negative contribution coefficients, where w_i is the prediction weight corresponding to the i-th degradation mode.

[0147] Each sub-predictor outputs a remaining lifetime estimate RUL_i based on its own mechanism or data foundation. The final fusion result is obtained by weighted averaging: RUL_fusion = Σ(w_i × RUL_i), where RUL_i is the remaining lifetime estimate output by the i-th sub-predictor, and RUL_fusion is the fused remaining lifetime prediction. For example, if the identification results show that thermal oxidation contributes 60%, mechanical wear contributes 40%, and other factors contribute negligibly, and the LSTM model is assigned a small weight (e.g., 0.1) as a correction term, then after normalization, the Arrhenius model weight is approximately 0.55, the Paris model weight is approximately 0.36, and the LSTM model weight is approximately 0.09. The final prediction value comprehensively reflects multiple degradation mechanisms and data-driven correction results.

[0148] This technical solution can identify complex combinations of degradation modes and dynamically adjust the prediction weights based on the real-time contribution of each degradation mechanism to the current degradation state. It achieves the complementarity of physical mechanisms and data-driven advantages, effectively addresses the prediction uncertainty in multi-mode coupled degradation scenarios, and generates more robust and reliable remaining service life predictions.

[0149] Reference Figure 3 This application also proposes an electrical component fault early warning system, comprising:

[0150] The equivalent thermal effect sequence generation module 1 is used to obtain the actual temperature parameters, actual voltage parameters, load parameters and environmental parameters of the electrical components. Based on the physical thermal inertia of the electrical components, the load parameters are converted into time response characteristics to obtain an equivalent thermal effect sequence that is consistent with the time response characteristics of the actual temperature parameters.

[0151] The operating cycle and benchmark model processing module 2 is used to divide the operating cycle of electrical components based on load parameters and the rate of change determined by the load parameters, and to obtain the context information within each operating cycle. It establishes a benchmark model with equivalent thermal action sequence and environmental parameters as input, and reflects the operating parameter response of electrical components under healthy conditions. Based on the benchmark model, it derives the theoretical normal temperature parameters and theoretical normal voltage parameters under the current operating conditions.

[0152] The parameter difference sequence generation module 3 is used to calculate the difference between the actual temperature parameter and the theoretical normal temperature parameter to obtain the temperature difference sequence, and to calculate the difference between the actual voltage parameter and the theoretical normal voltage parameter to obtain the voltage difference sequence. The temperature difference sequence and the voltage difference sequence together constitute the parameter difference sequence reflecting the degradation additional effect.

[0153] The comprehensive characterization and historical evolution sequence generation module 4 is used to aggregate the parameter difference sequence to obtain multiple feature indicators reflecting the degree of degradation of electrical components, and to fuse multiple feature indicators. Based on the synergistic change relationship between multiple feature indicators, a comprehensive degradation state characterization value of electrical components is generated. The synergistic change relationship includes the same-direction change relationship of multiple feature indicators in the same operating condition cycle. The comprehensive degradation state characterization value generated in each operating condition cycle is stored to form a historical evolution sequence.

[0154] The incubation period determination process execution module 5 is used to execute the fault incubation period determination process based on the comprehensive degradation state characterization value. The determination process includes normal state, suspected incubation period and confirmed incubation period.

[0155] The evolution trend analysis module 6 is used to analyze the evolution trend of the comprehensive degradation state characterization value based on the historical evolution sequence when the judgment process is in the suspected latent period, and to perform cross-operational cycle comparison and verification to obtain the evolution trend analysis results.

[0156] The diagnostic switching module 7 is used to switch the judgment process to the diagnostic latency period and confirm the starting point of the fault latency period when the comprehensive degradation state characterization value of the evolution trend analysis result meets the preset continuous deterioration condition, or when the comprehensive degradation state characterization value of the cross-operational cycle comparison verification characterization increases abnormally.

[0157] The life prediction module 8 is used to construct the evolution trajectory of the comprehensive degradation state characterization value based on the starting point of the fault latency, and predict the remaining operating life of the electrical components based on the evolution trajectory.

[0158] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for early warning of electrical component faults, characterized in that, include: The actual temperature parameters, actual voltage parameters, load parameters, and environmental parameters of the electrical components are obtained. Based on the physical thermal inertia of the electrical components, the load parameters are converted to time response characteristics to obtain an equivalent thermal action sequence that is consistent with the time response characteristics of the actual temperature parameters. Based on the load parameters and the rate of change determined by the load parameters, the operating cycle of the electrical components is divided, and the context information within each operating cycle is obtained. A benchmark model is established with the equivalent thermal effect sequence and the environmental parameters as input, and reflects the operating parameter response of the electrical components under healthy conditions. Based on the benchmark model, the theoretical normal temperature parameters and theoretical normal voltage parameters are derived under the current operating conditions. The difference between the actual temperature parameter and the theoretical normal temperature parameter is calculated to obtain a temperature difference sequence, and the difference between the actual voltage parameter and the theoretical normal voltage parameter is calculated to obtain a voltage difference sequence. The temperature difference sequence and the voltage difference sequence together constitute a parameter difference sequence reflecting the additional effects of degradation. The parameter difference sequence is aggregated to obtain multiple feature indicators reflecting the degree of degradation of the electrical component. The multiple feature indicators are then fused together, and a comprehensive degradation state characterization value of the electrical component is generated based on the synergistic change relationship between the multiple feature indicators. The synergistic change relationship includes the unidirectional change relationship of the multiple feature indicators within the same operating condition cycle. The comprehensive degradation state characterization value generated within each operating condition cycle is stored to form a historical evolution sequence. Based on the comprehensive degradation state characterization value, a fault latency determination process is executed, which includes normal state, suspected latency period, and confirmed latency period; When the determination process is in the suspected latent period, the evolution trend of the comprehensive degradation state characterization value is analyzed based on the historical evolution sequence, and a comparison and verification is performed across operating cycle to obtain the evolution trend analysis results. When the evolution trend analysis result indicates that the comprehensive degradation state characterization value meets the preset continuous deterioration condition, or when the cross-operational cycle comparison verification indicates that the comprehensive degradation state characterization value increases abnormally, the judgment process is switched to the diagnosis latency period, and the starting point of the fault latency period is confirmed. Based on the starting point of the fault latency period, the evolution trajectory of the comprehensive degradation state characterization value is constructed, and based on the evolution trajectory, the remaining service life of the electrical component is predicted.

2. The electrical component fault early warning method according to claim 1, characterized in that, The aggregation of the parameter difference sequence yields multiple characteristic indicators reflecting the degree of degradation of the electrical components, including: Based on the context information within each operating cycle, determine the low load duration between the current operating cycle and the previous operating cycle, and obtain the temperature difference corresponding to the start time of the high load window of the current operating cycle from the temperature difference sequence. Based on the physical thermal inertia of the electrical component, the thermal time constant of the electrical component is determined, and based on the thermal time constant, it is determined whether the duration of the low load is sufficient to completely dissipate the additional heat generated in the previous operating cycle. When the duration of the low load is insufficient to completely dissipate the additional heat generated in the previous operating cycle, a virtual difference baseline function that decays exponentially over time is constructed. The virtual difference baseline function takes the temperature difference corresponding to the start of the high load window of the current operating cycle as its starting point and decays according to the thermal time constant. Within the high-load window period of the current operating cycle determined based on the context information, the virtual difference baseline value at the corresponding moment is subtracted from the temperature difference in the temperature difference sequence to obtain the corrected dynamic degradation difference. By integrating the corrected dynamic degradation difference over time, the cumulative additional temperature rise characteristics after removing the effects of historical residues are obtained. When the duration of the low load is sufficient to completely dissipate the additional heat generated in the previous operating cycle, the temperature difference sequence is integrated over time during the high load window of the current operating cycle to obtain the cumulative additional temperature rise characteristics. The voltage difference sequence is averaged over the high-load window period of the current operating cycle to obtain the average additional voltage drop characteristic, and the cumulative additional temperature rise characteristic and the average additional voltage drop characteristic are used as the multiple characteristic indicators.

3. The electrical component fault early warning method according to claim 2, characterized in that, The process of integrating the multiple feature indicators and generating a comprehensive degradation status characterization value for the electrical component based on the synergistic change relationship among the multiple feature indicators includes: Based on the context information within the current operating cycle, determine the load level and ambient temperature corresponding to the current operating cycle; Based on the load level and the ambient temperature, a correlation parameter matching the current operating cycle is selected. The correlation parameter is used to characterize the coordinated change relationship between the multiple characteristic indicators. Based on the multiple characteristic indicators within the current operating cycle, a current degradation feature vector is constructed, and based on the multiple characteristic indicators of the electrical components in a healthy state, a health baseline feature vector is constructed. Based on the correlation parameter, calculate the distance between the current degraded feature vector and the health baseline feature vector; Based on the distance, a comprehensive degradation status characterization value for the electrical component is generated.

4. The electrical component fault early warning method according to claim 1, characterized in that, The process involves analyzing the evolution trend of the comprehensive degradation state characterization value based on the historical evolution sequence, and performing comparative verification across operating cycle conditions to obtain the evolution trend analysis results, including: Based on the historical evolution sequence, determine the rate of change of the comprehensive degradation state characterization value between adjacent operating condition cycles; Based on whether the rate of change of the comprehensive degradation state characterization value continues to increase in multiple consecutive operating condition cycles, and whether the increase in the rate of change of the comprehensive degradation state characterization value exceeds a preset acceleration threshold, it is determined whether the comprehensive degradation state characterization value shows an accelerating deterioration trend, and an acceleration deterioration judgment result is obtained. When performing cross-operational cycle comparison verification, the rate of change of the comprehensive degradation state characterization value of the current operating cycle is compared with the rate of change of the comprehensive degradation state characterization value of the historical operating cycle with similar context information to obtain the comparison result. Based on the accelerated deterioration judgment results and the comparison results, the evolutionary trend analysis results are obtained.

5. The electrical component fault early warning method according to claim 1, characterized in that, The process of constructing the evolution trajectory of the comprehensive degradation state characterization value based on the starting point of the fault latency includes: Each of the comprehensive degradation state characterization values ​​formed from the starting point of the fault latency period is used as a sampling point, and the time interval between adjacent sampling points is obtained. Based on the time interval, determine whether there is a data gap in the comprehensive degradation state characterization value; When the data discontinuity exists, the comprehensive degradation state characterization value is completed based on the theoretical normal temperature parameter, theoretical normal voltage parameter and environmental parameter within the corresponding time period of the data discontinuity, and a virtual comprehensive degradation state characterization value is generated. Based on the context information within the operating cycle corresponding to each sampling point, the sampling points are weighted to obtain weighted sampling points; When the data discontinuity exists, the weighted sampling points and the virtual comprehensive degradation state characterization value are smoothed to construct the evolution trajectory of the comprehensive degradation state characterization value; when the data discontinuity does not exist, the weighted sampling points are smoothed to construct the evolution trajectory of the comprehensive degradation state characterization value.

6. The electrical component fault early warning method according to claim 5, characterized in that, Predicting the remaining operational life of the electrical components based on the evolution trajectory includes: Based on the latest preset number of sampling points arranged in chronological order in the evolution trajectory, a set of terminal sampling points is determined. Based on the change in the comprehensive degradation state characterization value corresponding to adjacent sampling points in the set of terminal sampling points and the time interval between adjacent sampling points, the local curvature change of the evolution trajectory and the degradation rate corresponding to the set of terminal sampling points are determined. Determine the current load level and ambient temperature based on the context information within the current operating cycle; Based on the local curvature change, the current load level, and the ambient temperature, the weight of each sampling point in the end sampling point set in the remaining service life prediction is adjusted. Based on the weighted set of end sampling points, a preliminary prediction of the remaining service life of the electrical components is made, and a preliminary prediction result is obtained. When the local curvature change indicates an increase in the degradation rate of the comprehensive degradation state characterization value, and the difference between the current load level and the load level corresponding to the previous operating cycle exceeds a preset load fluctuation threshold, or the difference between the ambient temperature and the ambient temperature corresponding to the previous operating cycle exceeds a preset ambient temperature fluctuation threshold, a prediction and correction process based on physical failure boundaries is triggered. The physical failure boundaries include the preset failure critical characterization value corresponding to the comprehensive degradation state characterization value. In the prediction and correction process, the preliminary prediction results are constrained and corrected based on the preset failure critical characterization value and the degradation rate corresponding to the end sampling point set to obtain the prediction result; When the prediction correction process is not triggered, the preliminary prediction result is used as the prediction result; Based on the weights of the end sampling points of a preset number of groups, the remaining service life of the electrical components is predicted to obtain multiple candidate remaining service life prediction values; the fluctuation range of the prediction result is determined according to the difference between the prediction result and the multiple candidate remaining service life prediction values. When the fluctuation range exceeds the preset robustness threshold, the multi-predictor fusion mechanism is activated, and the remaining operating life prediction value is generated according to the multi-predictor fusion mechanism. When the fluctuation range does not exceed the preset robustness threshold, the prediction result is used as the predicted value of the remaining operating life.

7. The electrical component fault early warning method according to claim 6, characterized in that, The step of generating the remaining lifetime prediction value based on the multi-predictor fusion mechanism includes: The evolutionary trajectory is input into a prediction system that integrates physical degradation mechanisms with data-driven approaches; The prediction system performs pattern recognition on the evolutionary trajectory and the multiple feature indicators that form the evolutionary trajectory to obtain degradation pattern recognition results, which include single degradation patterns or multi-mode coupled degradation. When the degradation pattern identification result is a single degradation pattern, the prediction system generates the remaining service life prediction value based on the evolution trajectory; When the degradation pattern identification result is multi-mode coupled degradation, the prediction weight of each degradation pattern is adjusted according to the contribution of each degradation pattern to the comprehensive degradation state characterization value at the current stage. The prediction system performs weighted fusion of the degradation trends corresponding to each degradation mode based on the adjusted prediction weights of each degradation mode, and generates the predicted value of the remaining service life.

8. The electrical component fault early warning method according to claim 1, characterized in that, The process of determining the fault latency based on the comprehensive degradation state characterization value includes: Obtain the comprehensive degradation state characterization value within the current operating condition cycle, and determine the load level and ambient temperature corresponding to the current operating condition cycle based on the context information within the current operating condition cycle. Based on the load level and the ambient temperature, select a state switching rule that matches the current operating cycle from a preset set of operating condition-related state switching rules; Based on the state switching rules, it is determined whether the comprehensive degradation state characterization value meets the switching conditions for switching from the current state to the next state. The switching conditions include state duration requirements and state regression inhibition conditions. The next state includes switching from the normal state to the suspected incubation period, or switching from the suspected incubation period to the confirmed incubation period. When the comprehensive degradation state characterization value meets the switching condition, the determination process is switched to the next state; When the comprehensive degradation state characterization value does not meet the switching condition, the current state of the determination process is maintained.

9. The electrical component fault early warning method according to claim 8, characterized in that, The step of determining whether the comprehensive degradation state representation value meets the switching conditions for transitioning from the current state to the next state based on the state switching rules includes: Determine whether the load level exceeds a preset instantaneous overload threshold, or whether the rate of change of ambient temperature determined based on the ambient temperature of the current operating cycle and the previous operating cycle exceeds a preset sudden environmental change threshold. When the load level exceeds the preset instantaneous overload threshold, or the rate of change of the ambient temperature exceeds the preset sudden change threshold of the environment, the current operating condition cycle is determined as an extreme operating condition cycle. When the current operating condition cycle is an extreme operating condition cycle, historical operating condition cycles with similar contextual information are determined from the historical evolution sequence, and the comprehensive degradation state characterization value in the current operating condition cycle is compared with the comprehensive degradation state characterization value in the historical operating condition cycle to obtain the extreme operating condition comparison result. When the deviation between the comprehensive degradation state characterization value in the current operating condition cycle and the comprehensive degradation state characterization value in the historical operating condition cycle, as indicated by the extreme operating condition comparison result, is within a preset range, the current state of the determination process is maintained. When the extreme operating condition comparison result indicates that the deviation between the comprehensive degradation state characterization value in the current operating condition cycle and the comprehensive degradation state characterization value in the historical operating condition cycle exceeds a preset range, the determination process is judged to switch to the next state based on the switching conditions. When the current operating cycle is not an extreme operating cycle, the determination process is used to switch to the next state based on the switching conditions.

10. An electrical component fault early warning system, characterized in that, include: The equivalent thermal effect sequence generation module is used to obtain the actual temperature parameters, actual voltage parameters, load parameters and environmental parameters of the electrical components, and to perform time response characteristic conversion on the load parameters based on the physical thermal inertia of the electrical components to obtain an equivalent thermal effect sequence that is consistent with the time response characteristics of the actual temperature parameters. The operating condition cycle and benchmark model processing module is used to divide the operating condition cycle of the electrical component according to the load parameters and the rate of change determined according to the load parameters, and to obtain the context information within each operating condition cycle. It establishes a benchmark model with the equivalent thermal effect sequence and the environmental parameters as input and reflects the operating parameter response of the electrical component under healthy conditions. Based on the benchmark model, it derives the theoretical normal temperature parameters and theoretical normal voltage parameters under the current operating condition. The parameter difference sequence generation module is used to calculate the difference between the actual temperature parameter and the theoretical normal temperature parameter to obtain a temperature difference sequence, and to calculate the difference between the actual voltage parameter and the theoretical normal voltage parameter to obtain a voltage difference sequence. The temperature difference sequence and the voltage difference sequence together constitute a parameter difference sequence reflecting the additional effects of degradation. The comprehensive characterization and historical evolution sequence generation module is used to aggregate the parameter difference sequence to obtain multiple feature indicators reflecting the degree of degradation of the electrical component, fuse the multiple feature indicators, and generate a comprehensive degradation state characterization value of the electrical component based on the synergistic change relationship between the multiple feature indicators. The synergistic change relationship includes the same-direction change relationship of the multiple feature indicators within the same operating condition cycle. The module stores the comprehensive degradation state characterization value generated within each operating condition cycle to form a historical evolution sequence. The incubation period determination process execution module is used to execute the fault incubation period determination process based on the comprehensive degradation state characterization value. The determination process includes normal state, suspected incubation period and confirmed incubation period. The evolution trend analysis module is used to analyze the evolution trend of the comprehensive degradation state characterization value based on the historical evolution sequence when the judgment process is in the suspected latent period, and to perform cross-operational condition cycle comparison and verification to obtain the evolution trend analysis results. The diagnostic switching module is used to switch the judgment process to the diagnostic latency period and confirm the starting point of the fault latency period when the evolution trend analysis result indicates that the comprehensive degradation state characterization value meets the preset continuous deterioration condition, or when the cross-operational cycle comparison verification indicates that the comprehensive degradation state characterization value has increased abnormally. The life prediction module is used to construct the evolution trajectory of the comprehensive degradation state characterization value based on the starting point of the fault latency, and predict the remaining operating life of the electrical component based on the evolution trajectory.

Images