Intelligent electric appliance cabinet state evaluation method based on multi-dimensional time sequence data

Abstract

The invention relates to the technical field of electric appliance cabinet control, in particular to an intelligent electric appliance cabinet state evaluation method based on multi-dimensional time sequence data, and the method comprises the steps: collecting operation and environment data when an electric appliance cabinet acts; dynamically correcting the historical sample by using the temperature difference and the waveform change rate to generate a self-adaptive reference waveform; calculating a topology distortion index and an energy efficiency deviation index based on the current time sequence data and the adaptive reference waveform; and fusing the topological distortion index and the energy efficiency deviation index to calculate a mahalanobis distance to obtain a comprehensive abnormal score, and evaluating the state of the electric appliance cabinet according to the comprehensive abnormal score. By generating the environment adaptive reference waveform and fusing the multi-dimensional features, environment interference can be effectively stripped, macroscopic and microscopic fault features are captured, and the accuracy and robustness of fault diagnosis are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of electrical cabinet control, and particularly to an intelligent electrical cabinet state evaluation method based on multi-dimensional time-series data. Background Art

[0002] As a core control node of modern power distribution networks, intelligent electrical cabinets integrate key components such as vacuum circuit breakers, and the opening and closing operations of these components mainly rely on spring operating mechanisms. Due to the complex and variable operating environment of the power grid, the mechanical properties of the operating mechanisms are directly related to the action reliability of the equipment and the continuity of power supply in the power grid. Therefore, online monitoring and state evaluation of the mechanical characteristics of electrical cabinets are important measures to maintain the safe and stable operation of the power system.

[0003] The existing mechanical characteristic detection methods for such equipment mainly rely on type test standards and generally use static threshold methods based on scalar data for evaluation. This method collects key discrete feature data during the equipment operation process by deploying sensors, mainly including closing time, opening time, speed at the instant of separation and instant of contact, and peak coil current. Subsequently, each collected value is compared with a preset standard range one by one. If all the index values are within the specified range, it is determined that the current equipment state is normal, and this is used as the basis for the equipment to continue to be put into operation.

[0004] The existing technologies have limitations in dealing with the early evolution of mechanical failures. Early hidden dangers such as wear, lubrication failure, or spring fatigue generated during the long-term operation of the operating mechanism are first reflected in the microscopic topological distortion of the action waveform. Given the power redundancy usually existing in mechanical system designs, such local changes at the waveform level are difficult to cause macroscopic scalar values such as closing and opening times to exceed the qualified range, resulting in a pseudo-healthy state of the equipment in terms of numerical indicators. The existing methods focus mainly on the final discrete numerical results and ignore the waveform morphological evolution information during the entire action process, making it difficult to identify potential risks of internal mechanical state deterioration with qualified numerical indicators and difficult to meet the needs of maintenance personnel to formulate accurate maintenance strategies based on the actual deterioration trend of the equipment. Summary of the Invention

[0005] To solve the above technical problem that the existing detection technology only focuses on macroscopic scalar values and ignores the microscopic distortion of the action waveform, resulting in the inability to identify the early deterioration of the mechanical state with qualified numerical indicators, the present invention provides an intelligent electrical cabinet state evaluation method based on multi-dimensional time-series data, and this method includes the following steps: The system acquires operational and environmental data during a single operation of the electrical cabinet. The operational data includes current timing data and instantaneous operating voltage values, while the environmental data includes temperature data. Based on the temperature difference between the current operation's temperature data and the historical health sample temperature data of the electrical cabinet, a correction amount is calculated using the waveform change rate of the historical health sample. This correction amount is then used to compensate for the amplitude of the historical health sample, resulting in an adaptive reference waveform for the current operation. The system calculates the dynamic time warping distance and waveform slope difference using the current timing data and the adaptive reference waveform. A weighted fusion of the dynamic time warping distance and waveform slope difference yields the topology distortion index for the current operation. The system calculates the current integral difference and the operation duration difference between the current timing data and the adaptive reference waveform. Based on the current integral difference and the operation duration difference, it calculates the energy efficiency deviation index for the current operation. The topology distortion index and the energy efficiency deviation index are merged into a feature vector. The Mahalanobis distance between the feature vector and the center of the historical health feature vector is calculated to obtain a comprehensive anomaly score. The system evaluates the state of the electrical cabinet based on the comprehensive anomaly score.

[0006] This invention first acquires operational and environmental data of the electrical cabinet during operation. Based on the temperature difference between the current operation's temperature data and historical healthy sample temperature data, a correction amount is calculated using the waveform change rate to compensate for historical samples, thereby obtaining an adaptive reference waveform adapted to the current operating conditions. This method reduces the interference of non-fault drift of mechanical characteristics caused by changes in lubricating oil viscosity due to ambient temperature variations on the evaluation results. Furthermore, this invention obtains a topology distortion index by calculating the dynamic time warping distance and waveform slope difference between the current timing data and the adaptive reference waveform. This can identify abnormalities in waveform geometry. Simultaneously, it calculates the current integral difference and the operation duration difference to obtain an energy efficiency deviation index, thereby assessing whether the mechanism has a problem of decreased work efficiency due to damping aging or jamming. Finally, this invention merges the topology distortion index and energy efficiency deviation index into a feature vector and calculates its Mahalanobis distance to the center of the historical healthy feature vector. Utilizing the Mahalanobis distance's ability to consider the correlation between multi-dimensional features, a comprehensive evaluation of the electrical cabinet's condition is achieved, reducing potential misjudgments from single-dimensional index evaluations and improving the reliability of electrical cabinet mechanical fault diagnosis.

[0007] Preferably, the step of weightedly fusing the dynamic time warp distance and the waveform slope difference to obtain the topology distortion index for this action includes: dividing the dynamic time warp distance by the total number of sampling points to obtain a first component; calculating the product of the square of the second derivative of the adaptive reference waveform at each time step and a preset balance coefficient, and adding 1 to the product to obtain the weight corresponding to each time step; calculating the product of the absolute value of the waveform slope difference and the corresponding weight at each time step to obtain the weighted slope difference at each time step; integrating the weighted slope difference over a preset sampling duration to obtain a second component; and weightedly summing the first component and the second component to obtain the topology distortion index for this action.

[0008] When acquiring the topological distortion index, this invention divides the dynamic time warping distance by the total number of sampling points as the first component to evaluate the overall contour similarity of the waveform and reduce the impact of sampling time jitter. At the same time, it introduces the square plus one of the second derivative of the adaptive reference waveform as a dynamic weight, and calculates the weighted integral of the waveform slope difference as the second component. This weighting mechanism uses the waveform curvature characterized by the second derivative to increase the proportion of waveform inflection points or drastic change regions in the difference evaluation, so that the algorithm can capture the waveform micro-morphological distortion caused by early mechanical wear, thus taking into account the detection capabilities of macro-contour anomalies and micro-detail faults.

[0009] Preferably, the second component satisfies the following relation: ; in, This is the second component of this action; It is the sampling duration; The adaptive reference waveform is in the 1st... The second derivative at time t; , These are the current timing data and the adaptive reference waveform at the [number]th [time series]. The first derivative at time t; These are weighting coefficients. It is the absolute value symbol; It is the balance coefficient.

[0010] This invention utilizes integral calculation to determine the second component. It obtains the instantaneous velocity deviation by calculating the absolute value of the difference between the first derivative of the current timing data and the adaptive reference waveform, and constructs a dynamic gain coefficient using the second derivative of the adaptive reference waveform. This calculation logic physically simulates the sensitivity distribution of a mechanical mechanism to changes in resistance. Specifically, in the unsteady-state region where the mechanism accelerates or decelerates drastically, the second derivative value is larger, thus amplifying the impact of small velocity differences on the final score within this region. Conversely, in the uniform or stationary region, the weight is reduced, thereby suppressing background noise interference in non-critical areas. This waveform abrupt change attention-based calculation mechanism enhances the ability to capture minute waveform distortions caused by wear on mechanical mating surfaces or sudden changes in local resistance.

[0011] Preferably, the adaptive reference waveform for this action is in the 1st... Amplitude at time Satisfying the relation: ; in, It is the first Time decay weights for each historical healthy sample; It is the weight normalization factor, whose value is all weights. sum; It is the total number of historical healthy samples; It is the sensitivity coefficient; It is the average waveform of all historical healthy samples at the 1st Amplitude at any given moment; The average waveform is at the 1st The first derivative of a time point with respect to time; This is the temperature data for this action; It is the first Temperature data from a historical healthy sample; It is the first The historical health sample in the first Amplitude at any given moment; It is the absolute value symbol.

[0012] This invention constructs the calculation logic of an adaptive reference waveform by introducing a time decay weight and a sensitivity coefficient. The time decay weight gives higher importance to recent historical healthy samples, so that the generated reference waveform can follow the natural aging trend of the equipment's mechanical characteristics and prevent parameter drift caused by normal aging from being misjudged as sudden failure. At the same time, the product of the waveform's derivative with respect to time and the temperature difference is used as a correction term to simulate the non-uniform effect of temperature changes on the dynamic damping of the mechanical mechanism. That is, a larger compensation amount is applied during the high-speed movement phase of the mechanism, while a smaller compensation amount is applied during the steady phase. This makes the generated adaptive reference waveform more consistent with the actual physical conditions at the current temperature and improves the accuracy of subsequent comparative evaluation.

[0013] Preferably, the step of calculating the energy efficiency deviation index of the current action based on the current integral difference and the action duration difference includes: exponentializing the action duration difference to obtain a time suppression factor; and multiplying the current integral difference by the time suppression factor to obtain the energy efficiency deviation index.

[0014] Preferably, the energy efficiency deviation index satisfies the following relationship: ; in, This is the energy efficiency deviation index for this action; It is the first The amplitude of the current data at any given time; The adaptive reference waveform for this action is at the [number]th [number]th [number]. Amplitude at any given moment; This is the actual travel time of this action; This is the base travel time; It is the time sensitivity coefficient; It is a natural exponential function; It is the absolute value symbol; It refers to the sampling duration.

[0015] This invention utilizes the current integral ratio to reflect energy consumption deviation and introduces a natural exponential function containing a time-sensitive coefficient to construct a nonlinear constraint term for the difference in action time. Multiplying the two yields the energy efficiency deviation index. This calculation method takes into account the high sensitivity of high-voltage switchgear to action time. That is, when there is a slight delay in the action time of the equipment, the exponential function will cause the energy efficiency deviation index to increase sharply, thereby amplifying the slow-action fault characteristics caused by increased mechanical damping or lubrication failure. This allows the evaluation results to highlight potential hidden dangers that, although the waveform shape does not change much, seriously affect the opening and closing speed, thus improving the early warning capability for the risk of failure to operate or slow operation.

[0016] Preferably, the calculation of the Mahalanobis distance between the feature vector and the center of the historical healthy feature vector includes: obtaining the mean vector of the two-dimensional feature vector of the historical healthy sample and the inverse matrix of the covariance matrix of the two-dimensional feature vector of the historical healthy sample; using the mean vector and the inverse matrix of the covariance matrix, calculating the Mahalanobis distance of the two-dimensional feature vector of the current action from the health center, and using the Mahalanobis distance as the comprehensive anomaly score of the current action.

[0017] Preferably, acquiring the operating data and environmental data during a single operation of the electrical cabinet includes: collecting the operating data and environmental data during a single operation of the electrical cabinet; calculating a correction coefficient based on Ohm's law using the ratio of the rated voltage of the electrical cabinet to the instantaneous value of the operating voltage; and linearly scaling the collected initial current data using the correction coefficient to obtain current timing data.

[0018] Preferably, the step of assessing the status of the appliance cabinet based on the comprehensive anomaly score includes: calculating the natural logarithm of the comprehensive anomaly score; dividing a preset health score benchmark value by the sum of the natural logarithm of the comprehensive anomaly score and 1 to obtain a health score; and assessing the status of the appliance cabinet based on the health score.

[0019] Preferably, the step of evaluating the status of the appliance cabinet based on the health score includes: setting a pre-warning status threshold and a health status threshold, wherein the health status threshold is greater than the pre-warning status threshold; if the health score is greater than or equal to the health status threshold, the appliance cabinet is determined to be in a healthy state; if the health score is less than the health status threshold but greater than or equal to the pre-warning status threshold, the appliance cabinet is determined to be in a pre-warning state; if the health score is less than the pre-warning status threshold, the appliance cabinet is determined to be in an abnormal state.

[0020] The beneficial effects of this invention are as follows: By collecting environmental parameters during the operation process, this invention dynamically corrects historical health samples using temperature difference and waveform change rate, generating an adaptive benchmark waveform that adapts to the current operating conditions. This reduces the interference of environmental temperature fluctuations on mechanical characteristic evaluation and ensures the accuracy of the benchmark data. Simultaneously, this invention integrates a topological distortion index based on dynamic time warping and second derivative weighting, and an energy efficiency deviation index based on energy and time exponent coupling. The former focuses on identifying waveform micromorphological anomalies caused by mechanical wear, while the latter focuses on capturing the decrease in work efficiency and action delay caused by damping aging, achieving a multi-dimensional complementary evaluation of the electrical cabinet's mechanical state. Furthermore, this invention utilizes the Mahalanobis distance algorithm to fuse multi-dimensional feature vectors, automatically measuring the correlation and dispersion between various feature dimensions based on the statistical distribution patterns of historical health data. This achieves a comprehensive evaluation of the equipment's health status and improves the robustness of online monitoring and fault diagnosis for intelligent electrical cabinets. Attached Figure Description

[0021] Figure 1 A flowchart of a method for assessing the status of an intelligent electrical cabinet based on multidimensional time-series data, provided in an embodiment of the present invention; Figure 2 A comparison diagram of motion waveform features provided in an embodiment of the present invention; Figure 3 This is a multidimensional feature space distribution map provided for an embodiment of the present invention. Detailed Implementation

[0022] This invention provides a method for assessing the status of intelligent electrical cabinets based on multi-dimensional time-series data, such as... Figure 1 As shown, this method is an evaluation process for each opening or closing action of the electrical cabinet, including steps S100-S500: Step S100: Obtain the operating data and environmental data during a single operation of the electrical cabinet; the operating data includes current timing data and instantaneous operating voltage values, and the environmental data includes temperature data.

[0023] It should be noted that the single-action characteristics of the electrical cabinet's operating mechanism depend not only on the mechanical state but also on the combined effects of external power supply voltage and temperature data. On the one hand, fluctuations in the mains voltage can cause non-fault-related deviations in the coil drive current and action time; on the other hand, physical parameters such as lubricant viscosity and metal-metal fit clearance are highly sensitive to temperature changes. For example, increased viscosity at low temperatures can lead to increased action resistance. Without environmental parameters, subsequent evaluations will struggle to distinguish whether the distortion of the action waveform is a normal characteristic drift caused by environmental factors or a genuine fault caused by mechanical jamming. Therefore, this invention simultaneously records voltage and temperature parameters while collecting the action data, providing a data foundation for subsequently separating the effects of electrical fluctuations and temperature data.

[0024] Specifically, when the electrical cabinet is detected to perform a tripping or closing action, the built-in acquisition terminal is activated. Within a preset sampling time, the analog signals output by the current transformer and displacement sensor are continuously acquired at a preset sampling frequency to obtain the original current data and the moving contact travel data of this action.

[0025] Simultaneously, the temperature data for this action is acquired through the temperature sensor integrated inside the cabinet, and the instantaneous value of the operating voltage at the moment of action is recorded. Subsequently, based on the physical characteristic that the coil impedance can be regarded as constant within the sampling time, and according to Ohm's law that current and voltage are linearly proportional, the ratio of the rated voltage to the actual acquired voltage is used as a correction coefficient to linearly scale the original current data, thereby obtaining the current timing data for this action.

[0026] At this point, the operational and environmental data of the electrical cabinet during a single operation were obtained.

[0027] Step S200: Based on the temperature difference between the temperature data of the current action and the temperature data of the historical health samples of the electrical cabinet, calculate the correction amount by combining the waveform change rate of the historical health samples, and use the correction amount to compensate for the amplitude of the historical health samples to obtain the adaptive reference waveform of the current action.

[0028] It should be noted that the frictional characteristics of moving parts in mechanical operating mechanisms are affected by temperature. If a fixed waveform is used as the evaluation standard, temperature changes may be misjudged as mechanical failures. Therefore, this step aims to generate a reference waveform adapted to the environmental conditions during this operation. Considering the non-uniformity of the effect of lubricating oil viscosity changes on mechanical damping—that is, it is more affected by temperature in the high-dynamic operating range of the mechanical mechanism and less affected in the steady-state range—this invention introduces temperature difference to assess the magnitude of environmental changes and waveform change rate to assess the sensitivity to temperature at different time stages. The correction amount is determined through the coupling of these two factors, achieving dynamic compensation for historical health data.

[0029] Specifically, the first step is to obtain historical health samples. These historical health samples are action waveform data and their corresponding historical temperature data collected when the electrical cabinet is in a confirmed normal operating state, such as during the factory acceptance phase or after the equipment has been restored to normal after maintenance and repair.

[0030] Then, taking any moment within the sampling period as the current moment, perform the following operations: calculate the difference between the temperature data of the current action and the historical temperature data of each historical healthy sample at the corresponding moment; at the same time, based on the difference principle of discrete data, calculate the first derivative of the average waveform of the historical healthy sample with respect to time at the current moment, so as to obtain the rate of change of the waveform in the time dimension.

[0031] Finally, the amplitude at the current moment is calculated. It should be noted that while historical health samples reflect the equipment's baseline health state, the temperature data recorded at that time differs from the current moment, making direct averaging insufficient to reflect the true current operating conditions. Therefore, this invention uses the waveform of historical health samples as a benchmark, employing the coupled product of real-time temperature difference and waveform change rate as a gain term to dynamically fine-tune the amplitude of each historical health sample. Simultaneously, to eliminate random noise from single sampling and reflect the evolution trend of the equipment's state, a weighted average is used to fuse the corrected samples, ensuring that the generated benchmark waveform possesses both the statistical stability of historical data and adaptability to the current environment.

[0032] Based on the above logic, the adaptive reference waveform for this action is at the [number]th [number]. Amplitude at time Satisfying the relation: ; in, It is the first Time decay weights for each historical healthy sample; It is the weight normalization factor, whose value is all weights. The sum is used to ensure that the magnitude of the amplitude remains consistent with historical healthy samples; It is the total number of historical healthy samples; It is a sensitivity coefficient used to balance physical dimensions and map the product of the rate of change and temperature difference into a dimensionless correction gain; It is the average waveform of all historical healthy samples at the 1st Amplitude at any given moment; The average waveform is at the 1st The first derivative of a time point with respect to time; This is the temperature data for this action; It is the first Temperature data from a historical healthy sample; It is the first The historical health sample in the first Amplitude at any given moment; It is the absolute value symbol.

[0033] In this relation, For temperature compensation, It reflects the temperature difference between the current temperature data and historical temperature data. The larger the temperature difference, the larger the base amount that needs to be compensated. It is the waveform rate of change magnitude, reflecting the intensity of mechanical action. When the machine is in a high dynamic range, such as during startup or contact, the value of this derivative term is large, which increases the weight of the compensation term, thereby simulating the effect of viscosity change on damping under high dynamics. Conversely, in the stable phase of the waveform, the derivative term approaches zero, which causes the compensation term to automatically decrease and maintain waveform stability. Partially constitutes the first A correction factor for each historical health sample is used to dynamically reshape the data based on real-time operating conditions.

[0034] It should be noted that the sensitivity coefficient The damping setting should be determined based on the equipment's lubrication characteristics and mechanical tolerances. For equipment using high viscosity index lubricating oil and with tight mechanical fits, temperature has a more significant impact on damping; in this case, the setting should be adjusted accordingly. A larger value, such as 0.005, enhances the algorithm's ability to compensate for temperature changes; conversely, for equipment using low-viscosity lubricating oil or with a looser mechanical structure, the effect of temperature is relatively small, and the algorithm can be set to a smaller value. Setting it to a small value, such as 0.001, reduces noise introduced by overcompensation. In this embodiment, it is preferably set to 0.003.

[0035] It should also be noted that the time decay weight... The setting is to balance the statistical stability and timeliness of historical data. This invention preferably uses the exponential decay method found in existing technologies. Considering that the mechanical characteristics of equipment will undergo slight irreversible drift with the years of operation, such as spring fatigue, historical healthy samples closer to the current moment more accurately reflect the current basic state of the equipment. Therefore, by introducing a forgetting factor to construct an exponential decay model, more weight is given to recent samples, while reducing the influence of older samples. If the equipment is in its break-in or aging period, its characteristics change rapidly, and the forgetting factor should be set smaller to accelerate weight decay; if the equipment is in a stable period, it can be set larger to enhance robustness.

[0036] like Figure 2 The figure shows a comparison of the waveform characteristics of the action, with the horizontal axis representing time and the vertical axis representing the current amplitude. The dashed line corresponds to the adaptive reference waveform generated based on historical health samples, while the solid line corresponds to the waveform of the field-acquired data for the current action. The shaded area enclosed by both represents the region of difference. From the waveform morphology in the figure, the field-acquired data waveform shows a significant time lag and pulse broadening at its peak compared to the adaptive reference waveform. This morphological difference is consistent with the sluggish action characteristics caused by increased damping or lubrication aging in the mechanical transmission mechanism. The presence of the shaded area visually assesses the deviation between the two in terms of energy integration and geometric topology, proving that this invention can effectively capture minute mechanical fault characteristics.

[0037] Thus, by traversing all moments within all sampling durations, an adaptive reference waveform composed of the amplitude values ​​at each moment was obtained to evaluate the current action.

[0038] Step S300: Calculate the dynamic time warping distance and waveform slope difference using the current timing data and the adaptive reference waveform. Then, perform weighted fusion of the dynamic time warping distance and waveform slope difference to obtain the topology distortion index for this action.

[0039] It should be noted that early faults in mechanical operating mechanisms mainly manifest as local distortions in waveform shape. However, sampling time jitter or random delays in mechanical transmission can cause an overall time axis offset between the measured waveform and the reference waveform. This non-fault-related offset can lead to misjudgments in rigidity measures such as Euclidean distance. This step aims to evaluate the geometric differences in waveforms. This invention introduces a Dynamic Time Warping (DTW) algorithm, which can nonlinearly stretch the time axis, thereby focusing on comparing the shape similarity of two waveforms while ignoring time translation differences. Simultaneously, by combining the second derivative weighting of the waveform, it focuses on capturing subtle deformations at waveform inflection points.

[0040] Specifically, the DTW algorithm is first applied to calculate the minimum matching path distance between the current time series data and the adaptive reference waveform for this action, which serves as a measure of the overall profile difference. The first derivative of the current time series data and the adaptive reference waveform with respect to time at each moment is calculated to obtain their slope data. At the same time, the second derivative of the adaptive reference waveform is calculated to determine the curvature weight at each moment. The path distance, slope difference, and second derivative weight are substituted into the calculation model, and the topology distortion index for this action is obtained by weighted summation.

[0041] Based on the above logic, the topological distortion index of this action satisfies the following relationship: ; in, It is the topology distortion index of this action, used to characterize the degree of geometric anomaly of the current timing data of this action relative to the adaptive reference waveform; This is the current timing data for this action. With adaptive reference waveform The minimum matching path distance between them; This is the total number of sampling points; It is the sampling duration; It is an adaptive reference waveform In the The second derivative at time t; , These are current timing data. Adaptive reference waveform In the The first derivative at time t; , These are the weighting coefficients for the macroscopic topological term and the integral term, respectively. It is the absolute value symbol; It is the balance coefficient, and its unit is... The units of these quantities are reciprocals of each other, used to eliminate differences in physical dimensions.

[0042] In this relation, This is a macroscopic topological term that uses DTW distance to assess overall geometric similarity. Its function is to eliminate interference from nonlinear scaling of the time axis and to determine whether the general outline of the waveform is normal; within the integral... This is a micro-dynamic term, which calculates the slope difference between the measured data and the reference data at the same moment. Physically, it reflects the instantaneous deviation of the mechanism's motion speed and can sensitively detect minute jamming or speed lag caused by sudden changes in local friction. The integral term... This is the feature attention weight term, which uses the square of the second derivative of the reference waveform to construct a dynamic weight. At the inflection point or drastic change of the waveform, the weight increases significantly, thereby amplifying the influence of the micro-dynamic term in the key area; in the flat segment, the weight approaches 1, avoiding the interference of background noise in the stable area on the final score.

[0043] It should be noted that the balancing weight coefficient and It is not only used to adjust the proportion of macroscopic and microscopic features, but also implicitly plays a role in dimensional balance, mapping distance and slope integral measures to the same numerical order of magnitude. Its setting should be determined based on the fault prevention strategy and detection focus of the electrical cabinet. If the application scenario focuses on preventing severe functional failures, such as severe deformation of connecting rods or breakage of energy storage springs, which cause macroscopic faults that alter the overall waveform profile, then it should be set to... Set it to a large value, such as set to , Set it to a smaller value, such as set to This enhances the algorithm's tolerance to the overall waveform trend and focuses on structural integrity. Conversely, if the application scenario emphasizes early and precise diagnosis, aiming to detect hidden faults such as the initial stage of grease drying or slight wear on mating surfaces that cause minor burrs or abnormal slopes in local waveforms, then it is necessary to... Set it to a large value, such as set to , Set it to a smaller value, such as set to This enhances the algorithm's sensitivity to changes in microscopic details. In this embodiment, to balance the safety of the macroscopic structure with potential early-stage microscopic hazards, it is preferable to... and All are set to 0.5. In this embodiment, the balance coefficient... The value of can be set based on the statistical characteristics of the square of the second derivative of the reference waveform. Preferred... ,in, It is an adaptive reference waveform In the The second derivative at time t; It is the maximum value function. This setting normalizes the value of the square of the second derivative of the reference waveform to the interval [0,1], ensuring that it is comparable to the constant term 1 in terms of magnitude.

[0044] Thus, the topological distortion index, which characterizes the degree of geometric anomaly in the waveform of this action, was obtained.

[0045] Step S400: Calculate the current integral difference and action duration difference between the current timing data and the adaptive reference waveform. Based on the current integral difference and action duration difference, calculate the energy efficiency deviation index of this action.

[0046] It should be noted that focusing solely on waveform shape may miss high-energy-consuming and slow-moving damping aging faults. For example, when lubricating grease ages and becomes sticky, the mechanism may maintain a similar waveform shape to complete the same stroke, but it will consume more electrical energy and the action time will be longer. Therefore, this step aims to evaluate the physical work efficiency of the operating mechanism.

[0047] Furthermore, to accurately assess this efficiency decline, this invention constructs an evaluation logic based on exponential coupling. Considering that the increase in mechanical damping often exhibits a nonlinear deterioration trend, and that high-voltage switchgear has extremely low tolerance for delays in operating time—meaning that even a small time delay can imply a serious risk of arc-extinguishing failure—simple linear weighting cannot reflect the sharp increase in this risk. Therefore, this invention utilizes an exponential function to nonlinearly amplify the time delay and couples the energy consumption deviation with the time delay through a product. The advantage of this design is that it can not only capture single anomalies in energy or time but also produce an extremely sensitive response to complex degradation characteristics that are both power-intensive and slow, thereby improving the ability to identify early damping aging faults.

[0048] Specifically, firstly, the current timing data and the adaptive reference waveform of this action are integrated within the sampling time to obtain the current-time integral value; then, the measured stroke time is extracted from the moving contact stroke timing data of this action, and the reference stroke time is obtained from the historical health samples; finally, the energy consumption ratio and the relative value of time delay are substituted into the above-constructed exponential coupling model to calculate the energy efficiency deviation index of this action.

[0049] Based on the above logic, the energy efficiency deviation index of this action satisfies the following relationship: ; in, It is the energy efficiency deviation index of this action, used to characterize the degree of abnormality in the physical efficiency of this action; It is the first The amplitude of the current data at any given time; The adaptive reference waveform for this action is at the [number]th [number]th [number]. Amplitude at any given moment; This is the actual travel time of this action; This is the baseline travel time, and its value is greater than 0. It is the time sensitivity coefficient; It is a natural exponential function; It is the absolute value symbol; It refers to the sampling duration.

[0050] In this relation, The deviation rate of energy consumption was evaluated, and the increase in mechanical resistance led to an increase in the current waveform area, causing this value to rise. It is a suppression term based on time delay. Slower action will cause this term to grow nonlinearly. The two are multiplied together, and the time delay is used as a gain factor to amplify the energy deviation. This allows the algorithm to give a high anomaly score to the case of increased energy consumption and slower action, which accurately reflects the characteristics of mechanical damping aging or lubrication failure.

[0051] It should be noted that the time sensitivity coefficient The settings should be determined based on the stringency of the equipment's timeliness requirements for opening and closing actions. For core equipment involved in power grid protection, such as high-voltage circuit breakers, even a slight delay in their opening and closing time can lead to arc-extinguishing failure or cause system stability problems. In such cases, [the settings should be adjusted accordingly]. Set a larger value, such as 10 to 15, to impose strict time constraints; for equipment that does not undertake the task of interrupting fault current and has relatively relaxed time requirements, such as disconnecting switches or grounding switches, the time constraint can be adjusted accordingly. The value is set to a relatively small value, such as 5 to 8, to allow for normal time fluctuations within a certain range and avoid triggering erroneous alarms due to non-faulty mechanical discreteness. In this embodiment, for the circuit breaker scenario, a value of 10 is preferred.

[0052] Thus, the energy efficiency deviation index, which characterizes the abnormal physical work efficiency of this action, was obtained.

[0053] Step S500: Combine the topological distortion index and energy efficiency deviation index into a feature vector, calculate the Mahalanobis distance between the feature vector and the center of the historical healthy feature vector, and obtain a comprehensive anomaly score; evaluate the status of the appliance cabinet based on the comprehensive anomaly score.

[0054] It should be noted that single-dimensional indicators often have limitations when dealing with complex coupled fault modes. For example, a micro-short circuit between coil turns mainly causes an abnormal increase in current amplitude, which may lead to a significant increase in the operational efficiency deviation index, but the overall geometry of the waveform may not be drastically distorted. Conversely, early minor mechanical jamming may cause small spikes or abrupt slope changes in the waveform, increasing the operational topology distortion index, but its impact on total energy consumption may not be significant. Using a fixed-weight manual weighting method is difficult to adapt to such variable fault characteristics. Considering that the Mahalanobis distance algorithm has the ability to automatically measure the correlation and dispersion between various feature dimensions using the covariance matrix of historical data, it can automatically assign dynamic weights to indicators based on their stability under historical health conditions—that is, the smaller the historical variance of an indicator, the greater the weight its current small fluctuations will be given in the final evaluation, thus achieving an adaptive evaluation logic of "whoever is stable is sensitive." Based on this, this invention uses the Mahalanobis distance algorithm, based on the statistical distribution pattern of historical health data, to calculate the statistical distance of the current state from the health center, thereby achieving a comprehensive and objective evaluation of the equipment status.

[0055] Specifically, a two-dimensional feature vector is constructed, consisting of the topological distortion index of this action and the energy efficiency deviation index of this action.

[0056] Subsequently, the mean vector and the inverse matrix of the covariance matrix of the two-dimensional feature vectors of the historical healthy samples are obtained. It should be noted that the mean vector and the inverse matrix of the covariance matrix of the two-dimensional feature vectors are pre-calculated based on all historical healthy samples stored in the database.

[0057] Specifically, for each historical healthy sample, it is treated as a single action data point. The topological distortion index and energy efficiency deviation index of that sample relative to the adaptive baseline waveform are calculated to construct a two-dimensional feature vector for that sample. After traversing all historical healthy samples, the arithmetic mean of all sample two-dimensional feature vectors is calculated to obtain the mean vector of the two-dimensional feature vectors. The covariance matrix and its inverse matrix of these two-dimensional feature vectors are then calculated. How to calculate the covariance matrix is ​​existing technology and will not be elaborated here.

[0058] Then, using the mean vector and the inverse of the covariance matrix of the feature vectors of historical healthy samples, the Mahalanobis distance of the current state vector from the healthy center is calculated, which is used as the comprehensive abnormality score of this action.

[0059] Based on the above logic, the overall anomaly score for this action satisfies the following relationship: ; in, This is the overall abnormal score for this action; It is the two-dimensional feature vector of this action. ,in, This is the energy efficiency deviation index for this action. It is the topological distortion index of this action; It is the mean vector of the two-dimensional feature vectors of historical healthy samples; It is the inverse matrix of the covariance matrix of the two-dimensional eigenvectors of historical health samples; It is the vector transpose symbol; It is the dot product operator.

[0060] In this relationship, the inverse of the covariance matrix acts as an automatic weighting matrix. If a certain indicator is very stable in historical health samples, its... The corresponding element is larger, and small fluctuations in this indicator will be given higher weight, resulting in a significant increase in the overall abnormal score of this action. This mechanism utilizes the statistical characteristics of data to improve the sensitivity and accuracy of the assessment.

[0061] like Figure 3 The figure shows the multidimensional feature space distribution map, with the horizontal axis representing the topological distortion index and the vertical axis representing the energy efficiency deviation index. The circular markers clustered below and to the left of the origin correspond to historical healthy samples, while the star-shaped markers at the far upper right corner correspond to the current detection result. From the spatial distribution in the figure, the historical healthy samples exhibit high-density clustering, indicating minimal feature fluctuations under normal equipment conditions. In contrast, the current detection result deviates from the distribution center of the healthy samples in both the horizontal and vertical dimensions, leading to an increase in the calculated Mahalanobis distance. This significant statistical distance separation confirms that the feature vector of this action belongs to an anomalous outlier, validating the sensitivity and accuracy of the multidimensional feature fusion method in fault identification.

[0062] Finally, in order to provide a valid basis for maintenance strategies, it is necessary to map the comprehensive anomaly score to obtain the health score of this action, and evaluate the status of the electrical cabinet based on the health score.

[0063] In one feasible implementation, the health score for this action satisfies the following relationship: ; in, This is the health score for this action; This is the overall abnormal score for this action; It is the baseline value for the health score, which can be set to 100, and the implementers can also set it according to their needs; It is the natural logarithm function.

[0064] In this formula, the denominator utilizes the logarithmic function's property of suppressing large numerical fluctuations: when the equipment is in a healthy or slightly faulty state, the slope of the logarithmic function is large, making the health score sensitive to changes in the overall anomaly score, thus ensuring high-resolution identification of early faults; when the equipment experiences a serious fault causing an explosive increase in the overall anomaly score, the slope of the logarithmic function gradually slows down, compressing the value into an effective linear range and preventing the indicator signal from bottoming out and saturating. This design allows the system to maintain an effective data gradient even under extreme conditions such as severe faults, providing accurate data support for maintenance.

[0065] After calculating the health score, a multi-level status assessment mechanism based on thresholds is further provided.

[0066] Specifically, a health status threshold and an early warning status threshold are set, where the health status threshold is greater than the early warning status threshold. These thresholds can be flexibly adjusted based on the appliance cabinet's model, service life, and importance. If the health score is greater than or equal to the health status threshold, the appliance cabinet is determined to be in a healthy state; if the health score is less than the health status threshold but greater than or equal to the early warning status threshold, the appliance cabinet is determined to be in an early warning state; if the health score is less than the early warning status threshold, the appliance cabinet is determined to be in an abnormal state.

[0067] This completes the online detection and assessment of the mechanical fault in the electrical cabinet during this operation.

[0068] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for assessing the status of intelligent electrical cabinets based on multi-dimensional time-series data, characterized in that, include: The system acquires operational and environmental data during a single operation of the electrical cabinet; the operational data includes current timing data and instantaneous operating voltage values, and the environmental data includes temperature data. Based on the temperature difference between the temperature data of this action and the temperature data of the historical health samples of the electrical cabinet, the correction amount is calculated in combination with the waveform change rate of the historical health samples. The correction amount is used to compensate for the amplitude of the historical health samples to obtain the adaptive reference waveform of this action. By using the current timing data and the adaptive reference waveform, the dynamic time warping distance and waveform slope difference are calculated. The dynamic time warping distance and waveform slope difference are then weighted and fused to obtain the topology distortion index of this action. Calculate the current integral difference and action duration difference between the current timing data and the adaptive reference waveform. Based on the current integral difference and action duration difference, calculate the energy efficiency deviation index of this action. The topological distortion index and energy efficiency deviation index are combined into a feature vector. The Mahalanobis distance between the feature vector and the center of the historical healthy feature vector is calculated to obtain a comprehensive anomaly score. The status of the appliance cabinet is evaluated based on the comprehensive anomaly score.

2. The method for assessing the status of intelligent electrical cabinets based on multi-dimensional time-series data according to claim 1, characterized in that, The weighted fusion of the dynamic time warp distance and the waveform slope difference to obtain the topology distortion index for this action includes: Divide the dynamic time-normalized distance by the total number of sampling points to obtain the first component; Calculate the product of the square of the second derivative of the adaptive reference waveform at each time step and the preset balance coefficient, and add 1 to the product to obtain the weight corresponding to each time step; The absolute value of the waveform slope difference is calculated and multiplied by the corresponding weight at each time step to obtain the weighted slope difference at each time step. Integrate the weighted slope difference over a preset sampling time to obtain the second component; The first and second components are weighted and summed to obtain the topological distortion index for this action.

3. The method for assessing the status of intelligent electrical cabinets based on multi-dimensional time-series data according to claim 2, characterized in that, The second component satisfies the following relation: ； in, This is the second component of this action; It is the sampling duration; The adaptive reference waveform is in the 1st... The second derivative at time t; , These are the current timing data and the adaptive reference waveform at the [number]th [time series]. The first derivative at time t; These are weighting coefficients. It is the absolute value symbol; It is the balance coefficient.

4. The method for assessing the status of intelligent electrical cabinets based on multi-dimensional time-series data according to claim 1, characterized in that, The adaptive reference waveform for this action is in the [number]th [year]. Amplitude at time Satisfying the relation: ； in, It is the first Time decay weights for each historical healthy sample; It is the weight normalization factor, whose value is all weights. sum; It is the total number of historical healthy samples; It is the sensitivity coefficient; The average waveform of all historical healthy samples at the 1st Amplitude at any given moment; The average waveform is at the 1st The first derivative of a time point with respect to time; This is the temperature data for this action; It is the first Temperature data from a historical healthy sample; It is the first The historical health sample in the first Amplitude at any given moment; It is the absolute value symbol.

5. The method for assessing the status of intelligent electrical cabinets based on multi-dimensional time-series data according to claim 1, characterized in that, The calculation of the energy efficiency deviation index for this action based on the current integral difference and the action duration difference includes: The time difference of the action duration is exponentialized to obtain a time suppression factor; The energy efficiency deviation index is obtained by multiplying the current integral difference by the time suppression factor.

6. The method for assessing the status of intelligent electrical cabinets based on multi-dimensional time-series data according to claim 5, characterized in that, The energy efficiency deviation index satisfies the following relationship: ； in, This is the energy efficiency deviation index for this action; It is the first The amplitude of the current data at any given time; The adaptive reference waveform for this action is at the [number]th [number]th [number]. Amplitude at any given moment; This is the actual travel time of this action; This is the base travel time; It is the time sensitivity coefficient; It is a natural exponential function; It is the absolute value symbol; It refers to the sampling duration.

7. The method for assessing the status of intelligent electrical cabinets based on multi-dimensional time-series data according to claim 1, characterized in that, The calculation of the Mahalanobis distance between the feature vector and the center of the historical health feature vector includes: Obtain the mean vector of the two-dimensional feature vectors of the historical health samples, and the inverse matrix of the covariance matrix of the two-dimensional feature vectors of the historical health samples. Using the inverse of the mean vector and covariance matrix, the Mahalanobis distance of the two-dimensional feature vector of this action deviates from the health center, and the Mahalanobis distance is used as the comprehensive anomaly score of this action.

8. The method for assessing the status of intelligent electrical cabinets based on multi-dimensional time-series data according to claim 1, characterized in that, The acquisition of operational and environmental data during a single operation of the electrical cabinet includes: Collect operational and environmental data during a single operation of the electrical cabinet; Based on Ohm's law, the correction factor is calculated using the ratio of the rated voltage of the electrical cabinet to the instantaneous value of the operating voltage. The original current data is linearly scaled using the correction coefficient to obtain current time series data.

9. The method for assessing the status of intelligent electrical cabinets based on multi-dimensional time-series data according to claim 1, characterized in that, The assessment of the electrical cabinet's status based on the comprehensive anomaly score includes: Calculate the natural logarithm of the comprehensive anomaly score; The health score is obtained by dividing the preset health score benchmark value by the sum of the natural logarithm of the comprehensive abnormality score and 1. The condition of the appliance cabinet is assessed based on the health score.

10. The method for assessing the status of intelligent electrical cabinets based on multi-dimensional time-series data according to claim 9, characterized in that, The assessment of the appliance cabinet's status based on the health score includes: A preset warning state threshold and a health state threshold are defined, wherein the health state threshold is greater than the warning state threshold. If the health score is greater than or equal to the health status threshold, the appliance cabinet is determined to be in a healthy state. If the health score is less than the health status threshold but greater than or equal to the warning status threshold, the electrical cabinet is determined to be in a warning state. If the health score is less than the warning status threshold, the electrical cabinet is determined to be in an abnormal state.

Images