A power distribution box state monitoring method and system based on dynamic baseline self-learning

Abstract

This invention belongs to the field of distribution box condition monitoring technology, and discloses a method and system for distribution box condition monitoring based on dynamic baseline self-learning. It aims to solve the technical problems of baseline drift and high false alarm rate caused by variable operating conditions and interference coupling in existing distribution box monitoring. The method includes: collecting real-time electrical parameters and fixed-point temperature parameters; clustering and dividing independent monitoring periods based on the daily cycle pattern of the distribution load; constructing a dynamic baseline model for each period; calculating multi-dimensional deviation and triggering initial early warning; and executing dual-condition coordinated alarm for electricity and heat and baseline self-updating. The system includes modules such as multi-source data acquisition and period clustering. This invention achieves dynamic baseline monitoring adapted to operating conditions, reduces false alarms and baseline drift, improves the accuracy of contact degradation fault identification, and is suitable for safety status monitoring of various distribution boxes.

Description

Technical Field

[0001] This invention belongs to the field of distribution box status monitoring technology, specifically relating to a distribution box status monitoring method and system based on dynamic baseline self-learning. Background Technology

[0002] As a fundamental physical node for power distribution and control at the end of a power system, the safety and stability of the distribution box's operation directly affects overall power safety. Currently, the industry commonly monitors the distribution box's status by collecting electrical parameters such as voltage and current inside the box, as well as temperature parameters of the box or the environment. With the development of intelligent monitoring technology, traditional fixed threshold alarm methods are gradually being replaced by dynamic baseline monitoring. For example, patent document CN121410419A proposes a distribution box insulation monitoring method based on dynamic threshold analysis. This scheme updates the warning baseline in real time using historical data within a sliding window, which can adapt to changes in the distribution box's operating environment to a certain extent and generate corresponding over-limit alarm signals.

[0003] However, the aforementioned existing technologies still have shortcomings in practical applications. Since the load of distribution boxes typically exhibits a significant daily cycle, the globally continuous sliding time window does not distinguish between different operating conditions under normal load modes such as peak and off-peak periods. When the power consumption conditions undergo periodic normal switching, false alarms are easily triggered due to parameter fluctuations crossing the global baseline. Simultaneously, existing baseline update mechanisms typically roll over all historical data, lacking the ability to identify historical anomalies, particularly transient fluctuations that only trigger initial warnings but do not evolve into actual faults. This lack of classification and isolation mechanisms makes the updated training sample set susceptible to contamination, leading to baseline model drift and a significant decrease in model reliability over long-term operation. Furthermore, the transient electromagnetic interference caused by the start-up and shutdown of large equipment and actual physical contact degradation exhibit similarities in a single electrical parameter, making it difficult for existing solutions to accurately distinguish between the two, further increasing the false alarm rate. Therefore, it is urgent to improve existing technologies to address the problems of baseline drift and high false alarm rates caused by variable operating conditions and interference coupling in distribution box monitoring. Summary of the Invention

[0004] The purpose of this application is to provide a method and system for monitoring the status of distribution boxes based on dynamic baseline self-learning, which has the advantages of improving the accuracy of distribution box status monitoring, reducing false alarm rate and enhancing model stability.

[0005] Firstly, this application provides a method for monitoring the status of distribution boxes based on dynamic baseline self-learning, the technical solution of which is as follows: Includes the following steps: S100: Collects real-time electrical parameters of each circuit in the distribution box, and collects fixed-point temperature parameters through a non-contact temperature sensor deployed on the metal connection surface of the circuit breaker outgoing terminal in the distribution box. S200: Acquire historical operating data of the distribution box, perform cluster analysis on the daily operating time in the historical operating data based on the daily cycle pattern of the distribution load of the distribution box, and divide the daily operating time into multiple independent monitoring periods; S300: Perform an anomaly removal process on the historical operating data within each independent monitoring period to construct a clean and normal sample set, and construct an independent dynamic baseline model for each independent monitoring period based on the clean and normal sample set; S400: Calculates the multidimensional deviation of real-time electrical parameters and fixed-point temperature parameters relative to the dynamic baseline model corresponding to the independent monitoring period to which the current time belongs, and triggers an initial warning when the multidimensional deviation exceeds the limit for a preset number of consecutive times; S500: After triggering the initial warning, execute the electrothermal dual-condition collaborative upgrade alarm logic to determine whether there is a real contact degradation fault, and start the baseline self-update mechanism during the preset low load period to recalibrate and reflow the real-time electrical parameters and fixed-point temperature parameters that trigger the initial warning but have not been upgraded to a real contact degradation fault.

[0006] Furthermore, this application also proposes that the step S200, which involves clustering the daily operating time in historical operating data based on the daily cycle pattern of the power distribution load of the distribution box, specifically includes: extracting multidimensional statistical features within multiple complete daily cycles in the historical operating data, inputting the multidimensional statistical features into an unsupervised clustering algorithm for feature space partitioning, identifying peak load periods, off-peak load periods, and transition periods with significant steady-state characteristics based on the clustering results, and treating the peak load periods, off-peak load periods, and transition periods as independent monitoring periods that do not overlap.

[0007] Furthermore, this application also proposes that step S300, which involves constructing an independent dynamic baseline model for each independent monitoring period based on a clean and normal sample set, specifically includes: extracting multi-scale statistical features at different time scales for the clean and normal sample set of each independent monitoring period, configuring corresponding weight coefficients for the multi-scale statistical features for weighted fusion, generating a baseline vector characterizing the normal fluctuation characteristics within the independent monitoring period, and using the baseline vector as the dynamic baseline model for the corresponding period.

[0008] Furthermore, this application also proposes that the calculation of the multidimensional deviation of the real-time electrical parameters and fixed-point temperature parameters relative to the independent monitoring period to which the current time belongs in step S400 specifically includes: obtaining the covariance matrix of the dynamic baseline model, using the Mahalanobis distance algorithm to calculate the Mahalanobis distance between the state vector composed of the real-time electrical parameters and the fixed-point temperature parameters and the center point of the dynamic baseline model, and using the Mahalanobis distance as the multidimensional deviation.

[0009] Furthermore, this application also proposes that the initial warning triggered in step S400 when the multidimensional deviation exceeds the limit for a preset number of consecutive times specifically includes: setting a stabilization counter, incrementing the stabilization counter when the multidimensional deviation is higher than the dynamic threshold corresponding to the dynamic baseline model, clearing the stabilization counter when the multidimensional deviation falls back to no higher than the dynamic threshold, and triggering the initial warning only when the cumulative value of the stabilization counter reaches the preset number of times.

[0010] Furthermore, this application also proposes that the execution of the electrothermal dual-condition collaborative upgrade alarm logic in step S500 to determine whether there is a real contact degradation fault specifically includes: obtaining the deviation of the real-time electrical parameters when the initial warning is triggered, and obtaining the current absolute temperature value and temperature rise rate according to the fixed-point temperature parameters. Only when the deviation of the real-time electrical parameters meets the preset electrical abnormality conditions and the current absolute temperature value and temperature rise rate simultaneously meet the preset thermodynamic degradation conditions, it is determined that there is a real contact degradation fault and a high-level alarm is output.

[0011] Furthermore, this application also proposes that step S500, which initiates a baseline self-update mechanism during a preset low-load period, specifically includes remarking and recirculating real-time electrical parameters and fixed-point temperature parameters that trigger initial warnings but do not escalate into actual contact degradation faults. This includes: during an independent monitoring period when the distribution box is under low load, extracting real-time electrical parameters and fixed-point temperature parameters from historical records that only trigger initial warnings but are intercepted by the dual-condition electrothermal alarm logic, remarking them as benign abnormal samples, and recirculating the benign abnormal samples back to the clean normal sample set of the corresponding independent monitoring period to retrain and update the dynamic baseline model.

[0012] Furthermore, this application also proposes that the method further includes a cold start transition step: during the initial commissioning phase of the distribution box and when the historical operating data has not reached the data volume requirement for cluster analysis, a preset fixed empirical threshold is used as a temporary monitoring baseline. After the accumulated historical operating data meets the data volume requirement for cluster analysis, the method automatically switches to the dynamic baseline model corresponding to each independent monitoring period.

[0013] Secondly, this application also provides a distribution box status monitoring system based on dynamic baseline self-learning, comprising: The multi-source data acquisition module is used to collect real-time electrical parameters of each circuit in the distribution box, and to collect fixed-point temperature parameters through a non-contact temperature sensor deployed on the metal connection surface of the circuit breaker outgoing terminal in the distribution box. The time period clustering and segmentation module communicates with the multi-source data acquisition module and is used to perform cluster analysis based on the daily cycle pattern of the power distribution load in the historical operation data of the power distribution box, dividing the daily operation time into multiple independent monitoring time periods; The segmented baseline modeling module communicates with the time period clustering module. It is used to perform anomaly removal procedures on the historical operating data within each independent monitoring period to construct a clean and normal sample set, and to construct an independent dynamic baseline model for each independent monitoring period based on the clean and normal sample set. The deviation calculation and early warning module is connected to the multi-source data acquisition module and the segmented baseline modeling module respectively. It is used to calculate the multidimensional deviation of real-time electrical parameters and fixed-point temperature parameters relative to the dynamic baseline model corresponding to the independent monitoring period to which the current time belongs, and to trigger an initial early warning when the multidimensional deviation exceeds the limit for a preset number of consecutive times. The collaborative diagnosis and self-learning module communicates with the deviation calculation and early warning module. It is used to execute the electrothermal dual-condition collaborative upgrade alarm logic after the initial warning is triggered to determine whether there is a real contact deterioration fault. It also starts the baseline self-update mechanism during the preset low load period to recalibrate and reflow the real-time electrical parameters and fixed-point temperature parameters that trigger the initial warning but have not been upgraded to a real contact deterioration fault.

[0014] The method and system for monitoring the status of distribution boxes based on dynamic baseline self-learning disclosed in this application are based on dividing independent monitoring periods according to the daily cycle of the power distribution load, constructing segmented dynamic baselines, and solving the shortcomings of existing technologies through a dual-condition alarm mechanism of electric and thermal conditions and a baseline self-update mechanism. Compared to existing technologies, this invention offers significant advantages: By using cluster analysis to divide monitoring periods into peak, trough, and transition periods, independent dynamic baselines are constructed for each period, avoiding the problem of a global baseline failing to adapt to periodic changes in operating conditions and effectively reducing false alarms caused by normal fluctuations in operating conditions. By eliminating abnormal data to construct a clean, normal sample set, combined with a baseline self-update mechanism during low-load periods, benign abnormal samples are recalibrated and re-flowed, avoiding baseline drift caused by sample contamination and improving the reliability of the model over long-term operation. Through the coordinated upgrade of alarm logic based on both electrical and thermal conditions, combined with dual judgment of electrical parameter deviation and temperature parameters (absolute temperature and temperature rise rate), it can effectively distinguish between instantaneous electromagnetic interference and actual contact degradation faults, further reducing the false alarm rate. The cold start transition step solves the problem of insufficient historical data to construct a dynamic baseline during the initial commissioning phase of the distribution box, ensuring the continuity of monitoring work. The overall solution achieves accurate monitoring of the distribution box status, improves the accuracy of contact degradation fault identification, ensures the safe and stable operation of the distribution box, and is suitable for status monitoring scenarios of various industrial and civil distribution boxes. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart illustrating a method for monitoring the status of a distribution box based on dynamic baseline self-learning provided by the present invention. Figure 2 This is a schematic diagram of the structure of a distribution box status monitoring system based on dynamic baseline self-learning provided by the present invention; Figure 3 This is a flowchart of the electrothermal dual-condition collaborative upgrade alarm logic decision-making process of the present invention.

[0017] The following are the markings in the attached diagram: 100. Multi-source data acquisition module; 200. Time-period clustering and segmentation module; 300. Segmented baseline modeling module; 400. Deviation calculation and early warning module; 500. Collaborative diagnosis and self-learning module. Detailed Implementation

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

[0019] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention.

[0020] In the field of distribution box monitoring, due to the significant daily cyclical pattern of the load, the global continuous sliding time window fails to distinguish between different operating modes such as peak, off-peak, and transition periods. This leads to parameter fluctuations crossing the baseline during periodic switching of operating conditions, triggering false alarms. Furthermore, the baseline update mechanism lacks the ability to identify historical anomalies, and transient fluctuation data is included in the training sample set, causing model contamination and leading to baseline drift. Moreover, the transient electromagnetic interference generated by the start-up and shutdown of large equipment and the degradation of actual physical contact exhibit highly similar characteristics in a single electrical parameter, making them difficult to distinguish effectively and resulting in a continuously increasing false alarm rate. These problems directly affect the reliability and early warning accuracy of the monitoring system, causing normal operating condition changes to be misjudged as fault events, while potentially masking real hidden dangers.

[0021] To address this, this application proposes a method for monitoring the status of distribution boxes based on dynamic baseline self-learning, such as... Figure 1 As shown, it includes the following steps: S100: Collects real-time electrical parameters of each circuit in the distribution box, and collects fixed-point temperature parameters through a non-contact temperature sensor deployed on the metal connection surface of the circuit breaker outgoing terminal in the distribution box. S200: Acquire historical operating data of the distribution box, perform cluster analysis on the daily operating time in the historical operating data based on the daily cycle pattern of the power distribution load, and divide the daily operating time into multiple independent monitoring periods; S300: Perform an anomaly removal process on the historical operating data within each independent monitoring period to construct a clean and normal sample set, and construct an independent dynamic baseline model for each independent monitoring period based on the clean and normal sample set; S400: Calculates the multidimensional deviation of real-time electrical parameters and fixed-point temperature parameters relative to the dynamic baseline model corresponding to the independent monitoring period to which the current time belongs, and triggers an initial warning when the multidimensional deviation exceeds the limit for a preset number of consecutive times; S500: After triggering the initial warning, execute the electrothermal dual-condition collaborative upgrade alarm logic to determine whether there is a real contact degradation fault, and start the baseline self-update mechanism during the preset low load period to recalibrate and reflow the real-time electrical parameters and fixed-point temperature parameters that trigger the initial warning but have not been upgraded to a real contact degradation fault.

[0022] This embodiment provides a method for monitoring the status of distribution boxes based on dynamic baseline self-learning. This method divides the daily operating time into multiple independent monitoring periods to adapt to the daily cyclical patterns of the power distribution load, thereby solving the baseline drift problem caused by changing operating conditions. In practical applications, independent monitoring periods refer to non-overlapping time periods divided according to the characteristics of the power distribution load. These can be achieved using a fixed-length segmentation method based on timestamps, such as dividing the day into 24 equal-length time periods, or identifying intervals with similar load patterns through the statistical distribution of historical data. The main purpose is to ensure relatively consistent load characteristics within each time period, avoiding false alarms caused by global baseline changes due to operating conditions. Furthermore, a dynamic baseline model is used to characterize the normal operating status within a specific time period. It can be constructed using the mean and standard deviation of historical data to build a Gaussian distribution model, or a probability distribution can be generated using kernel density estimation methods. The main purpose is to provide a benchmark reference for that time period to detect parameter anomalies. Specifically, multidimensional deviation refers to the overall deviation of real-time electrical parameters and fixed-point temperature parameters from the dynamic baseline model. It can be calculated using vector norms, such as L2 or L1 norm, to determine the distance between the state vector and the baseline center. This is primarily for comprehensively evaluating the combined changes of multiple parameters. As a preferred implementation, the electrothermal dual-condition collaborative upgrade alarm logic simultaneously verifies abnormal conditions of electrical and temperature parameters to confirm faults. It can set threshold ranges for electrical parameters and a threshold for the rate of temperature change; only when both exceed the limits is it considered a true contact degradation fault. This is mainly to distinguish between equipment start-up / shutdown interference and actual physical degradation. Therefore, the baseline self-update mechanism refers to the process of updating the dynamic baseline model under specific conditions. During periods of low system load, data that triggers initial warnings but is not confirmed as faults can be re-marked as benign anomaly samples and included in the training set for the corresponding period. For example, data can be returned after manual review. This is mainly to optimize the model using benign data and avoid baseline drift caused by historical anomaly data contamination. This embodiment effectively solves the problem of high false alarm rates caused by the coupling of daily load cycles and electromagnetic interference through the above mechanisms, ensuring the reliability of monitoring.

[0023] During the status monitoring of the distribution box, real-time electrical parameters are collected from each circuit within the box. Simultaneously, non-contact temperature sensors deployed within the box on the metal connection surfaces of the circuit breaker outgoing terminals acquire fixed-point temperature parameters, which directly reflect the thermal state changes at key contact points. Based on historical operating data from the distribution box, the daily cyclical patterns of the power distribution load are used for cluster analysis of daily operating time, dividing the daily operating time into multiple non-overlapping independent monitoring periods, such as peak load periods, off-peak load periods, and transition periods, with a high degree of consistency in load patterns within each period. Furthermore, an anomaly removal process is performed on the historical operating data within each independent monitoring period to construct a clean, normal sample set. This process ensures the reliability of the sample set by excluding historical anomalies. Based on the clean, normal sample set, an independent dynamic baseline model is constructed for each independent monitoring period. This model generates a baseline vector through weighted fusion of multi-scale statistical features to accurately characterize the normal fluctuation characteristics within that period. The multidimensional deviations of real-time electrical parameters and fixed-point temperature parameters relative to the dynamic baseline model of the independent monitoring period to which the current time belongs are calculated. The Mahalanobis distance algorithm is used to evaluate the overall deviation of the state vector. An anti-jitter counter mechanism requires that the multidimensional deviation exceeds the limit for a preset number of consecutive times before triggering the initial warning, thereby effectively filtering out transient fluctuations such as instantaneous electromagnetic interference. When the initial warning is triggered, the electrothermal dual-condition coordinated upgrade alarm logic is executed. Only when the deviation of the real-time electrical parameters meets the preset electrical anomaly conditions and the current absolute temperature value and temperature rise rate simultaneously meet the preset thermodynamic degradation conditions is a real contact degradation fault determined. Otherwise, during the preset low-load period, the real-time electrical parameters and fixed-point temperature parameters that triggered the initial warning but did not upgrade to a real fault are re-marked as benign anomaly samples and flowed back to the corresponding clean normal sample set to start the baseline self-update mechanism.

[0024] In practical applications, non-contact temperature sensors can be specifically deployed as infrared temperature sensors on the metal connection surfaces of circuit breaker outgoing terminals. They directly capture the radiant heat signal from the metal connection surfaces in a non-contact manner, avoiding the shortcomings of traditional ambient temperature acquisition which cannot reflect localized overheating. The clustering analysis process can be implemented using unsupervised clustering algorithms, such as partitioning the feature space based on multidimensional statistical features from historical operating data to identify time-bound periods with significant steady-state characteristics. In the construction of the dynamic baseline model, the weighted fusion of multi-scale statistical features can be configured with specific time-scale weight coefficients; for example, assigning higher weights to short-term fluctuation features enhances the model's adaptability to transient changes. The calculation of multidimensional deviation involves the covariance matrix of the dynamic baseline model. The Mahalanobis distance can be solved by normalizing the covariance of the state vector and the center point of the baseline vector, thereby improving the robustness of the deviation assessment. When the baseline self-update mechanism is activated during low-load periods, the relabeling and reflow of benign abnormal samples can be automatically completed through a data labeling system; for example, reclassifying parameter records that have not been upgraded as training samples.

[0025] Therefore, the time-segmented dynamic baseline modeling mechanism effectively eliminates misjudgments caused by the daily cyclical patterns of power distribution load during operating condition switching, enabling the dynamic baseline to accurately adapt to normal fluctuations under different load modes such as peak and off-peak periods. The electrothermal dual-condition collaborative diagnostic mechanism, by simultaneously verifying both electrical parameter deviations and thermodynamic parameters, accurately distinguishes between transient electromagnetic interference caused by equipment start-up and shutdown and actual physical contact degradation faults, significantly reducing the false alarm rate. The baseline self-updating mechanism recalibrates and reverts benign abnormal data during low-load periods, avoiding contamination of the clean normal sample set by historical abnormal data, thereby preventing long-term drift of the dynamic baseline model and ensuring the reliability and stability of the monitoring system during long-term operation.

[0026] This application further proposes that step S200, based on the daily cycle pattern of the power distribution load of the distribution box, performs cluster analysis on the daily operating time in the historical operating data. Specifically, this includes: extracting multidimensional statistical features within multiple complete daily cycles in the historical operating data, inputting the multidimensional statistical features into an unsupervised clustering algorithm for feature space partitioning, identifying peak load periods, off-peak load periods, and transition periods with significant steady-state characteristics based on the clustering results, and treating the peak load periods, off-peak load periods, and transition periods as independent monitoring periods that do not overlap.

[0027] Specifically, multidimensional statistical features refer to a set of statistics that characterize the fluctuation characteristics of power distribution loads across different time dimensions. These can be achieved using statistical features of electrical parameters such as mean current, current variance, mean voltage, voltage variance, and power factor. Unsupervised clustering algorithms can be understood as machine learning methods that automatically discover the inherent structure of data without pre-labeling sample categories. These can be implemented using K-means clustering, hierarchical clustering, or DBSCAN clustering algorithms. In practical applications, significant steady-state features specifically refer to the statistical characteristics of electrical parameters exhibiting a relatively stable state within a specific load period. This can be understood as the characteristic that the current fluctuation range is less than a preset threshold during peak load periods and the voltage fluctuation range is less than a preset threshold during off-peak load periods. Specifically, non-overlapping independent monitoring periods refer to a method of dividing monitoring intervals where there is no time overlap between periods. This can be achieved by dividing time periods in hours, ensuring that each time point belongs to only one specific monitoring period.

[0028] Specifically, the proposed solution effectively avoids interference from single-day abnormal data on clustering results by extracting multidimensional statistical features across multiple complete daily cycles. Simultaneously, the introduction of multidimensional features comprehensively captures the multidimensional fluctuation characteristics of power distribution load. Inputting these features into an unsupervised clustering algorithm for feature space partitioning automatically identifies the inherent structure of the data, objectively distinguishing different load states without pre-defined labels. Based on the clustering results, peak, trough, and transition periods with significant steady-state characteristics are identified. Focusing on steady-state features eliminates transient fluctuation interference, fully covering the entire power distribution operation cycle. Treating these periods as independent, non-overlapping monitoring periods eliminates the ambiguity of overlapping periods, ensuring each period corresponds to a unique load pattern. This lays the foundation for subsequently constructing independent dynamic baseline models for each period, thereby adapting to differences in operating conditions under daily cycle patterns and reducing false alarms caused by normal load switching.

[0029] As a specific implementation method, the solution of this application is implemented as follows: In the distribution box monitoring system, the historical operation data acquisition module first extracts the hourly mean current, current standard deviation, voltage mean, and voltage standard deviation as multidimensional statistical features from multiple complete daily cycles. These features are input into the K-means clustering algorithm, which automatically divides each 24 hours into multiple clusters. The system analyzes the feature distribution of each cluster, identifying periods with high current levels and small fluctuations as peak load periods, periods with low current levels and small fluctuations as off-peak load periods, and the remaining periods as transition periods. Finally, the system divides each day into non-overlapping peak load periods, off-peak load periods, and transition periods as independent monitoring periods for the subsequent construction of a dynamic baseline model. In this clustering process, let a certain moment... The extracted multidimensional feature vector is ,in is the feature dimension. The objective function adopted by the system is to minimize the sum of squared errors within the cluster, and its mathematical expression is: in, The preset number of clusters (in this embodiment) (corresponding to peak, trough, and transition periods respectively). Indicates the first Data clusters, Indicates the first The cluster center vectors of each cluster are obtained. By iteratively optimizing this objective function, the system can adaptively find the boundaries of time segments with similar load characteristics, thereby achieving accurate division of independent monitoring periods. For ease of understanding, Table 1 below shows the cluster center feature values ​​for each time period and their corresponding time division results obtained after iterative convergence of the clustering algorithm for a commercial building's electrical distribution box: Table 1. Cluster centers and time division examples for independent monitoring periods. As can be seen from the specific values ​​mentioned above, the clustering algorithm minimizes the error and... It automatically classifies the daytime office hours with high current and low fluctuations as peak load periods, thereby achieving accurate division of independent monitoring periods.

[0030] Through the above scheme, this application can accurately identify the peak, trough and transition periods of power distribution load, ensure that the division of independent monitoring periods is reasonable, enable the dynamic baseline model to adapt to changes in actual operating conditions, effectively distinguish between normal load mode switching and abnormal state, and significantly reduce the false alarm rate caused by the inability of the global baseline to distinguish different load modes.

[0031] This application further proposes that step S300, which involves constructing an independent dynamic baseline model for each independent monitoring period based on a clean and normal sample set, specifically includes: extracting multi-scale statistical features at different time scales for the clean and normal sample set of each independent monitoring period, configuring corresponding weight coefficients for the multi-scale statistical features for weighted fusion, generating a baseline vector characterizing the normal fluctuation characteristics within the independent monitoring period, and using the baseline vector as the dynamic baseline model for the corresponding period.

[0032] Specifically, multi-scale statistical features refer to statistics calculated at different time window lengths. These can be achieved using methods such as moving average windows, exponential smoothing, or wavelet transforms. Their purpose is to simultaneously capture the complete behavioral patterns of electrical parameters in terms of instantaneous fluctuations and long-term trends. Weight coefficient configuration refers to assigning contribution to features at different time scales. This can be achieved using fixed weights trained based on historical data or adaptive weights dynamically adjusted according to the current load. Its purpose is to optimize the feature fusion process based on the load characteristics of each time period. Baseline vector generation refers to combining the weighted fused features into a vector. This can be achieved using principal component analysis or feature concatenation. Its purpose is to accurately characterize the normal fluctuation boundaries of electrical parameters within the independent monitoring period. The dynamic baseline model refers to a reference model that stores baseline vectors. This can be achieved using database storage or memory caching. Its purpose is to provide a highly time-specific reference benchmark for subsequent deviation calculations.

[0033] Specifically, the proposed solution first extracts multi-scale statistical features from the clean, normal sample set for each independent monitoring period to capture the complete behavioral patterns of instantaneous fluctuations at the second level and trend evolution at the hour level, avoiding the feature bias caused by fixed time windows. Then, weighted coefficients are assigned to the multi-scale statistical features for weighted fusion. The weights are dynamically allocated according to the load characteristics of the time period. For example, during peak load periods, the weight of short-term fluctuation features is automatically increased to enhance the fault tolerance capability against transient interference, while during off-peak load periods, the weight of long-term stable features is emphasized to strengthen the sensitivity to identify minor anomalies. Next, a baseline vector representing the normal fluctuation characteristics within the independent monitoring period is generated. This baseline vector accurately defines the boundaries of normal fluctuations by integrating multi-dimensional fluctuation information. Finally, this baseline vector is used as a dynamic baseline model to provide a reference benchmark for deviation calculation. Through this multi-scale feature fusion mechanism, the dynamic baseline can adapt to the drastic short-term fluctuations during peak load periods and the gradual long-term changes during off-peak load periods, eliminating the risk of baseline drift caused by periodic switching of operating conditions.

[0034] In one specific embodiment, for peak load periods, the mean and standard deviation of the current over a shorter time window are extracted as short-term fluctuation features, while the mean over a longer time window is extracted as a long-term trend feature. A higher weighting coefficient is assigned to the short-term fluctuation feature, and a lower weighting coefficient is assigned to the long-term trend feature. The weighted and fused features are used to generate a baseline vector, which is then used as the dynamic baseline model for that period. Specifically, let the short-term fluctuation feature vector be... The long-term trend feature vector is The corresponding weight coefficients configured are as follows: and (and satisfy) The weighted fused baseline center vector is then... It can be represented as: For example, during the peak load period in Table 1, if a short-term current fluctuation characteristic of 5 minutes is extracted... Long-term current trend characteristics over 1 hour To enhance the system's adaptability to transient fluctuations, the following settings are configured: , The average reference current for that period is then calculated to be... .

[0035] At the same time, in order to fully depict the first The system further calculates the dispersion of data within each independent monitoring period, and then calculates the concentration of clean, normal samples. The state vector of each sample Relative to the baseline center vector covariance matrix : Finally, from the baseline center vector Covariance Matrix Together, they constitute the dynamic baseline model for this independent monitoring period, providing a complete mathematical benchmark for subsequent multidimensional anomaly detection.

[0036] Through the above technical solution, this application can comprehensively capture the composite fluctuation characteristics of electrical parameters in each independent monitoring period, so that the dynamic baseline can not only accommodate parameter fluctuations within a reasonable range, but also respond to abnormal deviations in a timely manner, effectively reducing false alarms caused by operating condition switching and missed alarms caused by slow response to gradual degradation.

[0037] This application further proposes that the calculation of the multidimensional deviation of the dynamic baseline model corresponding to the independent monitoring period to which the real-time electrical parameters and fixed-point temperature parameters belong in step S400 specifically includes: obtaining the covariance matrix of the dynamic baseline model, using the Mahalanobis distance algorithm to calculate the Mahalanobis distance between the state vector composed of the real-time electrical parameters and the fixed-point temperature parameters and the center point of the dynamic baseline model, and using the Mahalanobis distance as the multidimensional deviation.

[0038] Specifically, obtaining the covariance matrix of the dynamic baseline model refers to obtaining a matrix that characterizes the statistical correlation between real-time electrical parameters and fixed-point temperature parameters. This can be achieved by using a covariance matrix calculated based on a historical set of clean and normal samples. The aim is to accurately capture the inherent correlation characteristics between multidimensional parameters. The Mahalanobis distance algorithm can be understood as a distance calculation method that normalizes the parameter space using the covariance matrix. Specifically, it can be achieved by calculating the quadratic form of the difference between the state vector and the center point after solving the inverse of the covariance matrix. The purpose is to eliminate the interference of different parameter dimensions and correlations, ensuring that the deviation calculation truly reflects the degree of anomaly in the multidimensional space. In practical applications, using Mahalanobis distance as a multidimensional deviation means directly using the calculated Mahalanobis distance value as a deviation index. For example, it can be directly input into the early warning logic module without additional conversion. The purpose is to simplify the early warning judgment process and ensure a high degree of matching between the deviation index and the normal fluctuation characteristics within the independent monitoring period.

[0039] Specifically, the solution in this application accurately characterizes the statistical distribution characteristics of real-time electrical parameters and fixed-point temperature parameters under normal conditions by obtaining the covariance matrix of the dynamic baseline model. Based on this, when calculating the distance between the state vector and the center point of the dynamic baseline model using the Mahalanobis distance algorithm, the parameters are normalized using the covariance matrix, effectively eliminating the influence of dimensions and correlation, so that the calculated deviation can truly reflect the degree of abnormality of the current state relative to the normal fluctuation range. Applying Mahalanobis distance directly as a multidimensional deviation to the early warning logic ensures that the system can accurately distinguish between normal fluctuations and real anomalies in the scenario of periodic changes in power distribution load, thereby significantly improving the accuracy of early warning judgment.

[0040] As a preferred embodiment, the solution of this application is implemented as follows: In the distribution box status monitoring scenario, the status vector may include the effective current value, effective voltage value, and fixed-point temperature value of the metal connection surface of the circuit breaker outgoing terminal of each circuit; the covariance matrix of the dynamic baseline model is calculated through a historical clean normal sample set, reflecting the normal fluctuation relationship of these parameters during peak load periods; when calculating the Mahalanobis distance of the real-time status vector, the algorithm automatically considers the positive correlation between current and temperature, avoiding misjudgment caused by fluctuation of a single parameter.

[0041] In the specific algorithm implementation, let the current time be... The collected and constructed real-time state vector is The system retrieves the current time's corresponding number. Baseline center vector for each independent monitoring period and covariance matrix The Mahalanobis distance is calculated using the following formula. As a multidimensional deviation: This formula introduces the inverse of the covariance matrix. This not only unified the dimensions of different physical quantities (amperes, volts, degrees Celsius), but also effectively eliminated the coupling correlation between parameters, thus improving the multidimensional deviation. This objectively reflects the true Eulerian space mapping distance that indicates the current state is deviating from the normal cluster. Assuming only ordinary Eulerian distance is used, a small 8°C temperature rise might be masked by a 15A current difference; however, when calculating the aforementioned Mahalanobis distance, if historical data indicates that the temperature should not exceed 45°C under this condition, the inverse matrix operation will significantly amplify the weight of the abnormal temperature component, ultimately calculating the deviation. This makes the multidimensional deviation... It can objectively reflect the true Eulerian space mapping distance of the current state from the normal cluster.

[0042] Through the above solution, this application can effectively reduce the false alarm rate and improve the reliability of the early warning system in scenarios of periodic switching of power distribution load or transient interference, ensuring that the initial warning is triggered only when the actual deviation from the normal fluctuation range is realized.

[0043] This application further proposes that the initial warning triggered when the multidimensional deviation exceeds the limit for a preset number of consecutive times in step S400 includes: setting a stabilization counter, incrementing the stabilization counter when the multidimensional deviation is higher than the dynamic threshold corresponding to the dynamic baseline model, clearing the stabilization counter when the multidimensional deviation falls back to no higher than the dynamic threshold, and triggering the initial warning only when the cumulative value of the stabilization counter reaches the preset number of times.

[0044] Among them, the anti-shake counter refers to a counting device used to track the state of continuous over-limit events. It can be implemented by software counters or dedicated hardware counting circuits. Its purpose is to distinguish between transient interference and persistent faults by quantifying the number of continuous abnormal events. The dynamic threshold can be understood as the threshold boundary calculated in real time based on the dynamic baseline model. It can be dynamically adjusted based on the statistical distribution characteristics of historical data. Its purpose is to adapt to changes in the operating conditions of the distribution box and avoid false alarms caused by fixed thresholds. The increment operation refers to the process of increasing the counter value when the multidimensional deviation is continuously higher than the dynamic threshold. It can be implemented by incrementing by 1 each time an over-limit is detected. Its purpose is to accumulate abnormal evidence to verify the persistence of the fault. The zeroing operation refers to the operation of resetting the counter to zero when the multidimensional deviation falls back to below the dynamic threshold. It can be implemented by monitoring the deviation state in real time and resetting it immediately. Its purpose is to eliminate the impact of transient fluctuations on early warning decisions in a timely manner. The preset number of times refers to the pre-set minimum threshold for the number of consecutive over-limit events. It can be determined based on system operating experience or experimental data. Its purpose is to set strict conditions for triggering early warnings and ensure that early warnings are activated only when the abnormal state persists.

[0045] Specifically, the solution in this application uses a stabilization counter mechanism to continuously verify multidimensional deviation exceeding limits. When the multidimensional deviation exceeds a dynamic threshold, the stabilization counter increments to accumulate abnormal evidence; when the multidimensional deviation falls back to below the dynamic threshold, the stabilization counter is reset to zero; an initial warning is triggered only when the accumulated value of the stabilization counter reaches a preset number of times, thus ensuring that the warning is activated only when the abnormal state persists. This design, by introducing continuous verification in the time dimension, effectively distinguishes between instantaneous electromagnetic interference and actual contact degradation faults, avoiding the vulnerability of single-threshold judgment.

[0046] As a specific implementation method, the solution of this application is implemented as follows: The anti-shake counter can be specifically a software counter implemented in the microcontroller of the distribution box monitoring system. For example, when the system detects that the multidimensional deviation is higher than the dynamic threshold three times consecutively, the anti-shake counter increments from zero to three and triggers an initial warning; if the deviation falls back below the dynamic threshold after the second exceedance, the anti-shake counter is immediately reset to zero to avoid misjudging the instantaneous electromagnetic interference caused by the start-up and shutdown of large equipment as a fault. The underlying counting and warning judgment logic expression is as follows: Let time... Multidimensional deviation is The corresponding dynamic threshold is The current count value of the image stabilization counter is The preset number of times is The counter update logic is as follows: At the same time, the initial warning status flag bit The triggering logic is as follows: To illustrate the image stabilization effect visually, let's assume the system is set with a dynamic threshold. Preset number of times The specific evolution process is shown in Table 2: Table 2 Examples of the state evolution process of the image stabilization counter Through this coherent mathematical logic and the evolution of the above examples, the system rigorously ensures that the early warning action is protected against high-frequency transient interference (as shown in the table). to (stage) of immunity.

[0047] Through the above technical solution, this application effectively reduces the false triggering of initial warnings caused by transient electromagnetic interference or transient load changes, and significantly improves the accuracy and reliability of distribution box status monitoring.

[0048] This application further proposes the following steps in step S500: executing the electrothermal dual-condition collaborative upgrade alarm logic to determine whether a real contact degradation fault exists: The system obtains the deviation of real-time electrical parameters when the initial warning is triggered, and obtains the current absolute temperature value and temperature rise rate based on the fixed-point temperature parameter. Only when the deviation of real-time electrical parameters meets the preset electrical anomaly conditions and the current absolute temperature value and temperature rise rate simultaneously meet the preset thermodynamic degradation conditions, it determines that there is a real contact degradation fault and outputs a high-level alarm.

[0049] Specifically, the deviation of real-time electrical parameters refers to the degree of deviation of real-time collected electrical parameters such as current and voltage from the dynamic baseline model. It can be achieved using Mahalanobis distance calculation or standardized difference analysis, with the aim of quantifying the severity of electrical anomalies. The current absolute temperature value refers to the instantaneous temperature value of the metal connection surface of the circuit breaker outgoing terminal, directly measured by a non-contact temperature sensor. It can be obtained using infrared thermometry or thermal radiation sensing technology, with the aim of reflecting the thermal state of the contact point. The temperature rise rate refers to the rate of temperature change per unit time. It can be achieved by performing differential operations on continuous temperature sampling values ​​or fitting the slope of a sliding window, with the aim of capturing the dynamic trend of temperature changes. Electrical anomaly conditions refer to the preset thresholds or rule sets used to determine electrical parameter anomalies. They can be configured as dynamic thresholds or fixed threshold ranges based on historical data, with the aim of setting effective criteria for determining electrical anomalies. Thermodynamic degradation conditions refer to the preset composite conditions used to determine thermodynamic anomalies. They can include a logical combination of absolute temperature thresholds and temperature rise rate thresholds, with the aim of ensuring that temperature anomalies conform to the characteristics of true degradation.

[0050] The proposed solution provides a quantitative basis for electrical anomalies by obtaining the deviation of real-time electrical parameters when the initial warning is triggered. However, since interference such as the start-up and shutdown of large equipment often causes instantaneous fluctuations in electrical parameters, relying solely on this deviation can easily lead to misjudging interference as a fault. Simultaneously, the solution obtains the current absolute temperature value and temperature rise rate based on fixed-point temperature parameters, introducing dynamic monitoring from a thermodynamic perspective. The absolute temperature value reflects the instantaneous thermal state of the contact point, while the temperature rise rate captures the dynamic trend of temperature changes. Only when the deviation of the real-time electrical parameters meets preset electrical anomaly conditions is the electrical anomaly confirmed to have reached a valid threshold. Furthermore, the current absolute temperature value and temperature rise rate must simultaneously meet preset thermodynamic degradation conditions to ensure that the temperature anomaly not only reaches a dangerous level but also that the rate of change conforms to degradation characteristics. Finally, only when both conditions are simultaneously met is a genuine contact degradation fault determined and a high-level alarm output. This rigorous logical coordination mechanism forces electrical and thermodynamic evidence to corroborate each other, significantly improving the accuracy of fault determination.

[0051] In some preferred embodiments, the processing unit may specifically be an ARM Cortex-M7 series microcontroller, which connects to an infrared temperature sensor array via a digital communication interface to obtain fixed-point temperature parameters. When the system detects a significant increase in current or voltage deviation and triggers an initial warning, the microcontroller calculates the absolute temperature value and temperature rise trend in real time. Only when the temperature reaches a preset danger threshold and the temperature rise rate shows a continuous upward trend is it confirmed as a true contact degradation fault and a high-level alarm is output. Only when both electrical abnormality conditions and two thermodynamic conditions are simultaneously met will the module determine that a true contact degradation fault exists and issue a high-level alarm to maintenance personnel. Figure 3 As shown, after the initial warning is triggered, the system verifies the electrical deviation condition and thermodynamic condition (absolute temperature, rate of temperature rise) in parallel. Only when both main conditions are met will a high-level alarm be output; otherwise, the event will be classified as a benign anomaly and await subsequent backflow processing. The specific logical judgment algorithm is as follows: First, the real-time current deviation is calculated. ,in The baseline current mean is used; then, based on the time step... Calculate the rate of temperature rise The system defines a true contact degradation fault flag. For logic to be true (1), the following multiple inequalities must be satisfied simultaneously: in, To preset the electrical deviation rate threshold, To preset the absolute temperature alarm value, To preset the critical value of the temperature rise rate, This represents a logical AND operation. The microcontroller will only report a high-level alarm to the main control center when the above compound logical expression is true. For example, suppose the current measured current causes a deviation... (greater than 20%), and the temperature of the metal connection surface (Greater than 80°C), and the temperature in the previous minute was 80°C, resulting in a temperature rise rate of 5°C / min (greater than 3°C / min). Since all three of the above compound logic expressions are true, the microcontroller will then determine... It will report a high-level alarm to the main control center; if the temperature is only 60°C, it indicates that it is a pure electrical fluctuation and will not trigger a high-level alarm.

[0052] The above technical solutions effectively filter out electrical parameter anomalies caused solely by electromagnetic interference, significantly reducing the false alarm rate and improving the accuracy and reliability of distribution box status monitoring.

[0053] This application further proposes to extract real-time electrical parameters and fixed-point temperature parameters from historical records that only triggered initial warnings but were blocked by the dual-condition coordinated upgrade alarm logic of electricity and heat during independent monitoring periods when the distribution box is under low load. These parameters are then re-marked as benign abnormal samples, and the benign abnormal samples are fed back to the clean normal sample set of the corresponding independent monitoring period to retrain and update the dynamic baseline model.

[0054] Among them, benign abnormal samples refer to abnormal data that are judged by the electrothermal dual-condition collaborative upgrade alarm logic as non-real contact deterioration faults. They can be identified by the deviation of electrical parameters not meeting the preset electrical abnormal conditions or the thermodynamic conditions not simultaneously meeting the deterioration conditions. The purpose is to accurately distinguish between transient fluctuations and real faults. Reflow refers to the operation of re-incorporating benign abnormal samples into the clean normal sample set. This can be achieved by updating database records or adding and merging sample sets. The purpose is to expand the normal sample set to reflect benign fluctuation characteristics without introducing contamination.

[0055] Specifically, the proposed solution initiates a baseline self-update mechanism during independent monitoring periods when the distribution box is under low load, ensuring that the update process is conducted in an environment with stable system operation and minimal external interference. Based on this, parameters from historical records that only triggered initial warnings but were intercepted by the dual-condition coordinated escalation alarm logic of electricity and heat are extracted, and a dual verification mechanism is used to accurately identify benign anomalies. Subsequently, these benign anomaly samples are relabeled and fed back to the clean, normal sample set of their original independent monitoring periods, allowing the model update to target the operating characteristics of specific time periods. Finally, the dynamic baseline model is retrained based on the expanded sample set, enabling the model to retain historical normal characteristics while absorbing benign fluctuation information, effectively suppressing baseline rigidity and preventing drift.

[0056] As a preferred embodiment, the solution of this application is implemented as follows: During the off-peak load period of the distribution box at night, the system automatically retrieves data points from the historical records that only triggered the initial warning but were blocked by the dual-condition coordinated escalation alarm logic of electricity and heat, such as parameters whose current parameters deviate from the baseline but whose absolute temperature value and temperature rise rate do not meet the thermodynamic degradation conditions; these data points are re-marked as benign abnormal samples; subsequently, the marked samples are fed back to the clean normal sample set during the off-peak load period; finally, the dynamic baseline model for this period is retrained using the updated sample set. During the retraining process, in order to prevent drastic changes in the model, the system uses the exponential moving average (EMA) algorithm to moderately update the dynamic baseline. Let the extracted benign abnormal sample state vector be... The system assigns a small learning rate parameter. The updated new baseline center vector With the new covariance matrix Iterate according to the following formulas respectively: This smooth update mechanism (wherein) and (These are the parameters of the original model), enabling the model to both learn and accommodate new benign fluctuation characteristics while ensuring the robustness of baseline evolution. For example, the original mean value of the trough current baseline. If the current of a benign abnormal sample during return is Set the learning rate The updated baseline mean is then smoothly corrected to This smooth update mechanism allows the model to learn and slowly incorporate new benign fluctuations while ensuring the absolute robustness of baseline evolution, effectively preventing model contamination.

[0057] This application further proposes a cold start transition step: during the initial commissioning phase of the distribution box and when the historical operating data has not reached the data volume requirement for cluster analysis, a preset fixed empirical threshold is used as a temporary monitoring baseline. After the accumulated historical operating data meets the data volume requirement for cluster analysis, the system automatically switches to the dynamic baseline model corresponding to each independent monitoring period.

[0058] The cold start transition step refers to the temporary monitoring mechanism adopted by the system during the initial operation phase. It can be implemented using a monitoring strategy based on preset thresholds, with the aim of ensuring that the system can still maintain basic monitoring functions during periods of scarce historical data. The preset fixed experience threshold can be understood as a range of parameter thresholds set based on long-term industry operating experience. It can be implemented using default values ​​stored in the system firmware, with the aim of providing a simple and reliable monitoring benchmark to distinguish between severe anomalies and normal fluctuations. The automatic switching mechanism refers to the baseline model switching process that is automatically triggered by the system according to the progress of historical data accumulation. It can be implemented using a data volume detection module in conjunction with the switching logic, with the aim of achieving a smooth transition of monitoring strategies to ensure continuous improvement in monitoring accuracy.

[0059] Specifically, the solution proposed in this application identifies the state of insufficient historical operating data during the initial commissioning stage of the distribution box and automatically activates a preset fixed experience threshold as a temporary monitoring baseline, thereby avoiding monitoring gaps caused by missing data. When the system detects that the accumulated historical operating data meets the data volume requirements for cluster analysis, it automatically switches to a dynamic baseline model built based on each independent monitoring period, realizing a seamless connection from temporary monitoring to precise monitoring and ensuring the continuity and reliability of the system's monitoring function throughout its entire life cycle.

[0060] As a preferred embodiment, the solution of this application is implemented as follows: In the initial stage of the power distribution box operation, the system detects the amount of historical operating data during initialization. If the amount of data does not meet the minimum data requirement for cluster analysis, a pre-stored fixed empirical threshold is loaded as a temporary monitoring baseline. The system continuously monitors the data accumulation progress. When the historical operating data meets the data requirement for cluster analysis, the system automatically executes the switching process to smoothly transition the monitoring baseline from the fixed threshold to the dynamic baseline model corresponding to each independent monitoring period.

[0061] Specifically, let the minimum historical daily training period set by the system be... (For example, set to 14 days), the number of historical operating days that the current distribution box has fully collected is: The system is configured with a step-type transition control weight factor. : The deviation of the final execution decision calculated by the system during operation Incorporating fixed experience deviation Deviation between (static difference calculated based on fixed values) and dynamic Mahalanobis distance Its expression is For example, when the equipment has been running for 5 days ( ),because ,at this time , The system relies entirely on conservative, fixed empirical thresholds configured at the factory for monitoring; when the equipment has been running stably for 15 days ( The amount of data already meets the clustering requirements. Instantly becomes 1, At this point, the model completes its autonomous evolution. This is a process of... The hard switching mechanism of the control ensures that the system can provide a rough but safe fallback warning when there is no historical data in the early stage, and automatically switch to the fine monitoring of the main logic as soon as sufficient data is available.

[0062] In another embodiment, this application also discloses a distribution box status monitoring system based on dynamic baseline self-learning, such as... Figure 2 As shown, the system includes: The multi-source data acquisition module 100 is used to acquire real-time electrical parameters of each circuit in the distribution box, and to acquire fixed-point temperature parameters through a non-contact temperature sensor deployed on the metal connection surface of the circuit breaker outgoing terminal in the distribution box. The time period clustering and segmentation module 200 is communicatively connected to the multi-source data acquisition module 100. It is used to perform clustering analysis based on the daily cycle pattern of the power distribution load in the historical operation data of the power distribution box, and to divide the daily operation time into multiple independent monitoring time periods. The segmented baseline modeling module 300 is communicatively connected to the time period clustering and partitioning module 200. It is used to perform anomaly data removal procedures on the historical operating data within each independent monitoring period to construct a clean and normal sample set, and to construct an independent dynamic baseline model for each independent monitoring period based on the clean and normal sample set. The deviation calculation and early warning module 400 is communicatively connected to the multi-source data acquisition module 100 and the segmented baseline modeling module 300, respectively. It is used to calculate the multidimensional deviation of real-time electrical parameters and fixed-point temperature parameters relative to the dynamic baseline model corresponding to the independent monitoring period to which the current time belongs, and to trigger an initial early warning when the multidimensional deviation exceeds the limit for a preset number of consecutive times. The collaborative diagnosis and self-learning module 500 is connected to the deviation calculation and early warning module 400. It is used to execute the electrothermal dual-condition collaborative upgrade alarm logic after the initial early warning is triggered to determine whether there is a real contact deterioration fault. It also starts the baseline self-update mechanism during the preset low load period to recalibrate and reflow the real-time electrical parameters and fixed-point temperature parameters that trigger the initial early warning but have not been upgraded to a real contact deterioration fault.

[0063] The core innovation of this embodiment lies in combining time-period clustering with segmented baseline modeling in a time-period independent modeling manner, while introducing electrical and thermal dual-condition collaborative upgrade alarm logic and baseline self-update mechanism. This allows for accurate adaptation to the daily cycle of power distribution load, effective differentiation between instantaneous electromagnetic interference and actual contact degradation faults, and prevention of baseline model drift caused by historical abnormal data pollution, thus achieving a significant reduction in false alarm rate and improved long-term monitoring reliability. Specifically, the time-period clustering module 200 identifies steady-state load characteristics based on historical data, dividing the daily operating time into independent monitoring periods that do not overlap. This allows the dynamic baseline model to model the normal fluctuation range of specific operating conditions such as peaks and troughs, avoiding false alarms triggered by normal parameter fluctuations crossing the baseline during operating condition switching. The segmented baseline modeling module 300 constructs a clean normal sample set through anomaly data removal, ensuring that the baseline model only reflects the normal characteristics of each time period and preventing historical anomaly data from contaminating the training set. The collaborative diagnosis and self-learning module 500 upgrades alarm logic through dual-condition electrothermal upgrades, determining a real fault only when the deviation of electrical parameters and the absolute temperature value and temperature rise rate of temperature parameters simultaneously meet the abnormal conditions. This effectively distinguishes between instantaneous electromagnetic interference and physical contact degradation caused by equipment start-up and shutdown. At the same time, benign anomaly data is recalibrated and reflowed during low-load periods, preventing real fault data from contaminating the baseline model and thus maintaining the long-term stability of the model.

[0064] 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 monitoring the status of a distribution box based on dynamic baseline self-learning, characterized in that, Includes the following steps: Step S100: Collect real-time electrical parameters of each circuit in the distribution box, and collect fixed-point temperature parameters through a non-contact temperature sensor deployed on the metal connection surface of the circuit breaker outgoing terminal in the distribution box. Step S200: Obtain the historical operating data of the distribution box, and perform cluster analysis on the daily operating time in the historical operating data based on the daily cycle pattern of the power distribution load of the distribution box, dividing the daily operating time into multiple independent monitoring periods; Step S300: Perform an anomaly removal process on the historical operating data in each independent monitoring period to construct a clean and normal sample set, and construct an independent dynamic baseline model for each independent monitoring period based on the clean and normal sample set; Step S400: Calculate the multidimensional deviation of the real-time electrical parameters and the fixed-point temperature parameters relative to the dynamic baseline model corresponding to the independent monitoring period to which the current time belongs, and trigger an initial warning when the multidimensional deviation exceeds the limit for a preset number of consecutive times; Step S5 00: After triggering the initial warning, execute the electrothermal dual-condition collaborative upgrade alarm logic to determine whether there is a real contact degradation fault, and start the baseline self-update mechanism during the preset low load period to recalibrate and reflow the real-time electrical parameters and the fixed-point temperature parameters that triggered the initial warning but did not upgrade to the real contact degradation fault.

2. The method for monitoring the status of a distribution box according to claim 1, characterized in that, The step S200, which involves clustering the daily operating time in the historical operating data based on the daily cycle pattern of the power distribution load of the distribution box, specifically includes: extracting multidimensional statistical features within multiple complete daily cycles in the historical operating data; inputting the multidimensional statistical features into an unsupervised clustering algorithm for feature space partitioning; identifying peak load periods, off-peak load periods, and transition periods with significant steady-state characteristics based on the clustering results; and treating the peak load periods, off-peak load periods, and transition periods as independent monitoring periods that do not overlap.

3. The method for monitoring the status of a distribution box according to claim 1, characterized in that, The step S300, which involves constructing an independent dynamic baseline model for each independent monitoring period based on the clean and normal sample set, specifically includes: extracting multi-scale statistical features at different time scales for the clean and normal sample set for each independent monitoring period, configuring corresponding weight coefficients for the multi-scale statistical features and performing weighted fusion to generate a baseline vector characterizing the normal fluctuation characteristics within the independent monitoring period, and using the baseline vector as the dynamic baseline model for the corresponding period.

4. The method for monitoring the status of a distribution box according to claim 1, characterized in that, The step S400 of calculating the multidimensional deviation of the real-time electrical parameters and the fixed-point temperature parameters relative to the independent monitoring period to which the current time belongs specifically includes: obtaining the covariance matrix of the dynamic baseline model, using the Mahalanobis distance algorithm to calculate the Mahalanobis distance between the state vector composed of the real-time electrical parameters and the fixed-point temperature parameters and the center point of the dynamic baseline model, and using the Mahalanobis distance as the multidimensional deviation.

5. The method for monitoring the status of a distribution box according to claim 4, characterized in that, The step S400 of triggering an initial warning when the multidimensional deviation exceeds the limit for a preset number of consecutive times specifically includes: setting a stabilization counter, incrementing the stabilization counter when the multidimensional deviation is higher than the dynamic threshold corresponding to the dynamic baseline model, clearing the stabilization counter when the multidimensional deviation falls back to no higher than the dynamic threshold, and triggering the initial warning only when the cumulative value of the stabilization counter reaches the preset number of times.

6. The method for monitoring the status of a distribution box according to claim 1, characterized in that, The step S500 of executing the electrothermal dual-condition collaborative upgrade alarm logic to determine whether there is a real contact degradation fault specifically includes: obtaining the deviation of the real-time electrical parameters when the initial warning is triggered, and obtaining the current absolute temperature value and temperature rise rate according to the fixed-point temperature parameter. Only when the deviation of the real-time electrical parameters meets the preset electrical abnormality conditions and the current absolute temperature value and the temperature rise rate simultaneously meet the preset thermodynamic degradation conditions, it is determined that there is a real contact degradation fault and a high-level alarm is output.

7. The method for monitoring the status of a distribution box according to claim 1, characterized in that, In step S500, the baseline self-update mechanism is activated during a preset low-load period. The relabeling and recirculation of the real-time electrical parameters and the fixed-point temperature parameters that triggered the initial warning but did not escalate into the actual contact degradation fault specifically includes: during the independent monitoring period when the distribution box is under low load, extracting the real-time electrical parameters and the fixed-point temperature parameters that only triggered the initial warning but were intercepted by the electrothermal dual-condition collaborative escalation alarm logic from the historical records, relabeling them as benign abnormal samples, and recirculating the benign abnormal samples back to the clean normal sample set of the corresponding independent monitoring period to retrain and update the dynamic baseline model.

8. The method for monitoring the status of a distribution box according to claim 1 or 7, characterized in that, It also includes a cold start transition step: during the initial commissioning phase of the distribution box and when the historical operating data has not reached the data volume requirement for the cluster analysis, a preset fixed empirical threshold is used as a temporary monitoring baseline. After the accumulated historical operating data meets the data volume requirement for the cluster analysis, it automatically switches to the dynamic baseline model corresponding to each independent monitoring period.

9. A distribution box status monitoring system based on dynamic baseline self-learning, characterized in that, include: The multi-source data acquisition module is used to collect real-time electrical parameters of each circuit in the distribution box, and to collect fixed-point temperature parameters through a non-contact temperature sensor deployed on the metal connection surface of the circuit breaker outgoing terminal in the distribution box. The time period clustering and segmentation module is communicatively connected to the multi-source data acquisition module. It is used to perform clustering analysis based on the daily cycle pattern of the power distribution load in the historical operation data of the power distribution box, and to divide the daily operation time into multiple independent monitoring time periods. The segmented baseline modeling module is communicatively connected to the time period clustering and partitioning module. It is used to perform anomaly data removal procedures on the historical operating data in each independent monitoring time period to construct a clean and normal sample set, and to construct an independent dynamic baseline model for each independent monitoring time period based on the clean and normal sample set. The deviation calculation and early warning module is communicatively connected to the multi-source data acquisition module and the segmented baseline modeling module, respectively. It is used to calculate the multidimensional deviation of the real-time electrical parameters and the fixed-point temperature parameters relative to the dynamic baseline model corresponding to the independent monitoring period to which the current time belongs, and to trigger an initial early warning when the multidimensional deviation exceeds the limit for a preset number of consecutive times. The collaborative diagnosis and self-learning module is communicatively connected to the deviation calculation and early warning module. It is used to execute the electrothermal dual-condition collaborative upgrade alarm logic after the initial early warning is triggered to determine whether there is a real contact degradation fault. It also starts the baseline self-update mechanism during a preset low load period to recalibrate and reflow the real-time electrical parameters and the fixed-point temperature parameters that triggered the initial early warning but did not upgrade to the real contact degradation fault.

Images