Intelligent comprehensive distribution box fault intelligent detection system

Abstract

The invention relates to the field of electric meter box detection, in particular to an intelligent fault detection system for an intelligent comprehensive distribution box. The system mainly comprises a data acquisition module, a data processing module, a risk assessment module and a grading processing module, according to the invention, through a dual-acquisition mode of basic acquisition and active acquisition, and in combination with dual judgment of a static threshold and a dynamic threshold, the problems of inflexible acquisition and early hidden danger report omission caused by mechanical fixation of the threshold in the prior art are effectively solved. According to the invention, a distribution box fault chain time sequence rule base of a preset rule and self-learning unit is constructed through a risk assessment module, accurate identification of a fault type and a current stage is realized in combination with a DTW algorithm, and a subsequent fault chain evolution link and residual transition time are pre-judged at the same time; a linear weighting algorithm is adopted by a risk assessment module, three dimensions of a fault chain stage, residual transition time and equipment importance are quantized and assigned, risk values are calculated according to weights, and the risk values are divided into low, medium, high and extremely high risk levels according to the risk values.

Description

Technical Field

[0001] This invention relates to the field of meter box testing, and in particular to an intelligent fault detection system for intelligent integrated distribution boxes. Background Technology

[0002] As the hub of the power distribution system, the operational stability of intelligent integrated distribution boxes is crucial to the security and continuity of power supply, and fault detection is an important means to ensure their reliable operation. However, existing fault detection methods for distribution boxes have many shortcomings and are difficult to meet the needs of intelligent and precise operation and maintenance.

[0003] To address the aforementioned issues, existing technologies have proposed solutions. For example, invention patent CN121069259A discloses a leakage current detection method, device, and storage medium for distribution boxes. The method includes: establishing an identification influence relationship between the detection environment and leakage current detection, including incremental and attenuation relationships and corresponding relationship coefficients; collecting current detection environment parameters; using the identification influence relationship to perform gain analysis on the detection environment parameters to obtain target gain parameters and gain adjustment amounts; adjusting the gain of the leakage current detection sensor based on the target gain parameters and gain adjustment amounts to acquire monitoring signals; performing leakage current detection based on the leakage current monitoring signals; and identifying the leakage current detection results and leakage current location information of the distribution box. This technical solution solves the technical problem that leakage current detection is easily affected by environmental factors in the existing technology, resulting in insufficient detection sensitivity and accuracy, and achieves the technical effect of improving the sensitivity and accuracy of leakage current detection. Although this technical solution can solve the problem of leakage current in the distribution box, there are still the following problems: (1) Existing technologies mostly use fixed frequency to collect operating parameters, which cannot adapt to the sudden data fluctuations of fault precursors, which can easily lead to the failure to report early hidden dangers or generate a large amount of invalid data. Moreover, it mostly relies on fixed thresholds to make abnormal judgments, without considering dynamic scenarios such as equipment aging and load fluctuations, resulting in a high false alarm rate and inability to accurately lock fault precursors.

[0004] (2) Existing technologies mostly make isolated judgments on a single fault point, ignoring the temporal progression characteristics of the distribution box fault, and cannot predict the fault chain evolution path and subsequent high-risk links, resulting in delayed operation and maintenance response and difficulty in preventing the fault from expanding; at the same time, risk assessments mostly consider only a single fault parameter, without combining multiple factors such as the stage of the fault, the remaining evolution time and the importance of the equipment, resulting in poor accuracy of the assessment results.

[0005] (3) Existing technologies often adopt a uniform approach to handle faults of different risk levels, which can easily lead to insufficient or excessive handling, affecting power supply continuity and operation and maintenance efficiency. Summary of the Invention

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: an intelligent integrated distribution box fault intelligent detection system, comprising: a data acquisition module, used to acquire the operating parameters of the distribution box at a basic acquisition frequency, and to initiate active acquisition when the operating parameters trigger a corresponding threshold to obtain the operating parameters within a set time period before and after the trigger.

[0007] The data processing module preprocesses the operating parameters and normalizes each sub-parameter in the operating parameters to form a fault feature dataset; it then marks the fault development points in the fault feature dataset.

[0008] The risk assessment module constructs a rule base for the fault chain sequence of the distribution box, matches the fault development point with the rule base, identifies the current fault and its current time sequence stage, analyzes the subsequent triggered fault chain stages and the transition time of the remaining stages, and assesses the fault chain risk level.

[0009] The graded processing module formulates different handling strategies based on the fault cascading risk level.

[0010] Furthermore, the operating parameters include electrical parameters and environmental parameters, wherein the electrical parameters include three-phase voltage, neutral line voltage, line voltage, three-phase current, neutral line current, and leakage current; and the environmental parameters include temperature and humidity.

[0011] The active acquisition is initiated as follows: when any operating parameter triggers a static or dynamic threshold, the acquisition frequency is immediately increased from the basic acquisition frequency to the active acquisition frequency.

[0012] Furthermore, the static threshold is the rated parameter of the device; the dynamic threshold is determined based on historical operating parameters within a sliding window.

[0013] Furthermore, the method for marking the fault development points in the fault feature dataset is as follows: the data in the fault feature dataset are arranged in order of collection time.

[0014] When the change of any sub-parameter exceeds the mutation threshold within a specified time, the sub-parameter is marked as a fault development point.

[0015] Record the sub-parameter types, numerical change ranges, and occurrence time nodes corresponding to each fault development point, and generate a fault development trajectory with time series markers.

[0016] Furthermore, the rule base for constructing the fault chain sequence of the distribution box is as follows: fault chain rules are pre-set based on the fault mechanism of the distribution box and industry experience. Each rule includes: fault type, stage division, fault chain sequence stage and stage transition condition.

[0017] The rule base for the fault chain sequence of the distribution box also includes a self-learning unit. Based on historical fault data, manually labeled data and emergency repair records during system operation, the self-learning unit uses the DTW algorithm to perform cluster analysis on the fault development point label sequence in the historical fault feature dataset to identify the common time sequence features of similar faults; calculates the support and confidence between fault label points, determines new fault chain sequence rules and stores them in the rule base.

[0018] Furthermore, the specific method for identifying the current fault and its stage is as follows: matching the fault development point's sub-parameter type with the fault chains in the rule base to filter out candidate fault chains.

[0019] The DTW algorithm is used to calculate the temporal similarity between the time series of each fault development point in the current fault feature dataset and the candidate fault chain.

[0020] Based on temporal similarity, it is determined whether the current fault development point belongs to a certain candidate fault chain. If it does, the current fault development point is located to the specific temporal stage of the fault chain. If it does not, the fault development trajectory is submitted as a new fault sample to the self-learning unit for processing.

[0021] Furthermore, the specific method for triggering subsequent fault chain stages and remaining stage transition times is as follows: based on the fault type associated with the current fault development point and the located time sequence stage, retrieve the complete fault chain time sequence stage sequence corresponding to the fault type from the rule base, and locate all subsequent stages after the current stage.

[0022] For each subsequent stage, based on the magnitude of the numerical change of the fault development point and the time node of its occurrence, the earliest and latest time points that trigger the subsequent stage are calculated as the prediction range of the transition time of the subsequent stage.

[0023] Furthermore, the assessment method for the fault cascading risk level is as follows: based on the current fault chain stage, the remaining stage transition time, and the importance of the equipment, a linear weighted algorithm is used to calculate the risk value, and the cascading risk level is determined based on the risk value.

[0024] Chain risk levels are classified into four categories: low risk, medium risk, high risk, and extremely high risk.

[0025] Furthermore, the importance of the equipment is determined by the system based on the importance assessment value of the power supply object of the distribution box and the importance assessment value of the regional power supply.

[0026] Furthermore, the specific working method of the graded processing module is as follows: when the fault cascading risk level is low, a fault warning message is generated and pushed to the operation and maintenance management platform to prompt operation and maintenance personnel to pay attention to changes in equipment status.

[0027] When the risk level is medium, an early warning message is pushed, which includes the equipment location, fault type, and suggested handling time limit; and a preliminary fault handling suggestion plan is generated, along with historical handling cases of similar faults.

[0028] If the risk level reaches high risk, the audible and visual alarm device will be triggered, an emergency alarm message will be sent to the operation and maintenance manager, non-core load circuits will be cut off according to the preset handling strategy, and the nearest operation and maintenance personnel will be dispatched to the site.

[0029] When the risk is assessed as extremely high, in addition to implementing high-risk response measures, the main power supply to the distribution box is cut off via the remote control module, the fire protection system is activated, and the situation is reported to the power dispatch center and safety management department to activate the emergency plan.

[0030] The beneficial effects of this system are as follows: This invention effectively solves the problems of inflexible acquisition and mechanically fixed thresholds in existing technologies, which lead to missed early hidden dangers, by using a dual acquisition mode of basic acquisition and active acquisition, combined with dual judgment of static thresholds and dynamic thresholds. When the parameters trigger the threshold, the active acquisition automatically increases the frequency and locks the operating parameters within a set time period before and after the trigger. The dual threshold design adapts to dynamic scenarios such as equipment aging and load fluctuations, reducing the false alarm rate and the missed alarm rate.

[0031] This invention processes multi-dimensional operating parameters through a data processing module and generates a fault development trajectory with time-series marking based on time-series logic marking fault development points. This effectively solves the problems of cross-dimensional analysis of parameters with different dimensions and the inability to correlate fault feature points in time sequence.

[0032] This invention constructs a fault chain temporal rule library for distribution boxes using a risk assessment module and a self-learning unit. Combined with the DTW algorithm, it achieves accurate identification of fault types and their current stages, while simultaneously predicting subsequent fault chain evolution stages and remaining transition times. The self-learning unit dynamically optimizes the rule library based on historical fault data, manually labeled data, and repair records, using cluster analysis and support / confidence filtering to ensure the inclusion of new fault types. The temporal analysis reflects the entire evolution path of a fault from its current stage to the accident stage.

[0033] This invention employs a linear weighted algorithm in its risk assessment module to quantify and score three dimensions: fault chain stage, remaining transition time, and equipment importance. Risk values ​​are calculated based on these weights, and these values ​​are used to classify risks into four levels: low, medium, high, and extremely high. This addresses the problem of existing risk assessment methods being limited to a single dimension and unable to be quantified. Furthermore, this invention establishes different handling strategies matching different risk levels, developing corresponding handling plans for different risk levels, effectively solving the problem of unreasonable existing handling strategies. Attached Figure Description

[0034] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0035] Figure 1 This is a schematic diagram showing the connection of each module in the intelligent integrated power distribution box fault detection system of the present invention.

[0036] Figure 2 This is a schematic diagram of the fault chain rule of the present invention. Detailed Implementation

[0037] The embodiments of the present invention are described in detail below. The embodiments described below are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. Where specific techniques or conditions are not specified in the embodiments, they shall be performed in accordance with the techniques or conditions described in the literature in the art or in accordance with the product manual.

[0038] Please see Figure 1 A smart integrated distribution box fault detection system mainly includes: a data acquisition module, a data processing module, a risk assessment module, and a hierarchical processing module; the modules are connected and communicate with each other through a data bus or an industrial network.

[0039] The data acquisition module is used to collect the operating parameters of the distribution box at a basic acquisition frequency. When the operating parameters trigger the corresponding threshold, active acquisition is initiated to obtain the operating parameters within a set time period before and after the trigger.

[0040] Considering that conventional fault detection of distribution boxes mostly adopts fixed frequency data acquisition or manual inspection mode, the above methods have the following defects: First, the data acquisition method adopts the relatively conventional fixed frequency data acquisition. The fixed frequency cannot adapt to the sudden data fluctuations of the precursor to the fault, and it is easy to miss early hidden dangers or generate a large amount of invalid data; Second, the existing mechanical judgment based on fixed thresholds is difficult to cope with dynamic scenarios such as equipment aging and load fluctuations, resulting in a high false alarm rate.

[0041] As a data acquisition channel, the data acquisition module needs to cover all parameters related to the distribution box and adapt to non-fixed thresholds such as equipment status and load fluctuations. It should be able to continuously capture normal operating data of the equipment and accurately lock in the fluctuations that are precursors to faults, providing support for subsequent data processing with time-series raw data.

[0042] In embodiments of the present invention, the operating parameters include electrical parameters and environmental parameters, wherein the electrical parameters include three-phase voltage, neutral line voltage, line voltage, three-phase current, neutral line current, and leakage current; and the environmental parameters include temperature and humidity.

[0043] This invention synchronously collects real-time values ​​of all operating parameters through voltage sensors, current sensors, temperature and humidity sensors built into the distribution box. The data acquisition of the above sensors is existing technology and will not be described in detail here. The data collected by the above sensors is stored in the local cache unit of the data acquisition module.

[0044] Meanwhile, the basic acquisition frequency of this invention is set to 500Hz for rapidly changing electrical parameters such as current and voltage, and 5Hz for environmental parameters such as temperature and humidity that change more slowly. The active acquisition frequency is set differently according to different types of operating parameters. For rapidly changing electrical parameters such as current and voltage, the active acquisition frequency is set to 1000Hz, and for environmental parameters such as temperature and humidity, the active acquisition frequency is set to 10Hz.

[0045] It should be noted that in this embodiment, the time period before and after the trigger is set to 30 seconds before the trigger and 60 seconds after the trigger. Staff can make minor adjustments to the specific time period according to the actual situation.

[0046] In an embodiment of the present invention, the active acquisition is initiated as follows: when any operating parameter triggers a static threshold or a dynamic threshold, the acquisition frequency is immediately increased from the basic acquisition frequency to the active acquisition frequency.

[0047] For example, the static threshold is the rated parameter of the device, such as a leakage current static threshold of 30mA, a temperature static threshold of 60℃, and a voltage deviation static threshold of ±10% of the rated voltage.

[0048] The dynamic threshold is determined based on historical operating parameters within a sliding window. For example, based on operating data from the past 7 days, the dynamic threshold for the operating parameter is the sum of the average value within the sliding window and three times the standard deviation of the corresponding sub-parameter. Determining the dynamic threshold based on the three-standard-deviation principle is a statistical method widely used in monitoring and anomaly detection. It uses the mean and standard deviation of data within the sliding time window to dynamically adjust the warning boundary, rather than using a fixed static threshold; this allows the invention to handle dynamic scenarios such as equipment aging and load fluctuations.

[0049] The data processing module preprocesses the operating parameters and performs min-max normalization on each sub-parameter of the operating parameters to form a fault feature dataset; it also marks the fault development points in the fault feature dataset.

[0050] In embodiments of the present invention, the preprocessing of operating parameters includes noise removal, missing values ​​removal, and outlier removal. The noise removal process can employ a combination of wavelet transform and Kalman filtering algorithms to remove power frequency interference and random noise from transient parameters such as voltage and current, and to employ first-order hysteresis filtering for slowly varying parameters such as temperature and humidity. Missing value filling can be achieved by using linear interpolation to fill in missing data.

[0051] Normalization, on the other hand, maps sub-parameters of different dimensions to the [0, 1] interval using a normalization and standardization method. Through normalization, parameters such as voltage, current, temperature, humidity, and leakage current, which were originally of different orders of magnitude and dimensions, can be compared and analyzed under the same dimension, providing a unified data foundation for subsequent fault development point marking and fault chain matching.

[0052] Considering the temporal progression of fault development, the marking of fault development points should be based on time to ensure the temporal correlation of each feature point. At the same time, it is necessary to set different mutation thresholds and time windows for different types of sub-parameters to avoid omissions or mislabeling due to differences in parameter change rates.

[0053] It should be noted that the mutation threshold and time window can be determined based on the equipment's rated parameters and historical fault data statistics. Implementers can make reasonable adjustments according to the on-site scenario. For example, the voltage parameter mutation threshold is ±5% of the rated voltage, the current parameter mutation threshold is ±8% of the rated current, and the temperature mutation threshold is 5℃ / 5min. The specified time window is set to 1 second for electrical parameters and 30 seconds for temperature and humidity parameters to ensure adaptation to the change rate of different parameters.

[0054] The method for marking fault development points in the fault feature dataset is as follows: the data in the fault feature dataset are arranged in ascending order according to the collection time to form a time-series data matrix with time as the horizontal axis and each sub-parameter as the vertical axis; the data matrix must include four pieces of information: sub-parameter name, collection timestamp, normalized value, and original value, to ensure that the time sequence is without deviation.

[0055] For each sub-parameter, a corresponding mutation threshold and a specified time window are preset, and the numerical change of the sub-parameter within each consecutive time window is calculated; when the change of each sub-parameter within the specified time exceeds the mutation threshold, the sub-parameter is marked as a fault development point, and the end time of the time window is the fault development point trigger time.

[0056] It should be noted that the implementer of the mutation threshold can dynamically adjust it according to the aging of the equipment and load fluctuations, but the setting standard of the same type of sub-parameter must be consistent.

[0057] Record the sub-parameter types, numerical change ranges, and occurrence time nodes corresponding to each fault development point, and generate a fault development trajectory with time series markers.

[0058] This system focuses on fault chain reaction analysis, and the fault development point is the basic unit that constitutes the fault chain. Without accurate marking, it is impossible to achieve subsequent functions such as precise stage location. At the same time, the dimensions and changes of electrical and environmental parameters are very different. The marking process transforms the data into unified features through a unified time frame and storage format, so as to achieve cross-parameter fault chain matching.

[0059] The risk assessment module constructs a rule base for the fault chain sequence of the distribution box, matches the fault development point with the rule base, identifies the current fault and its current time sequence stage, analyzes the subsequent triggered fault chain stages and the transition time of the remaining stages, and assesses the fault chain risk level.

[0060] Considering the diversity and complexity of power distribution box faults, preset rules alone cannot cover all scenarios. Therefore, the rule base must include self-learning units. At the same time, the self-learning process must be based on real operating data and go through a self-learning process of feature extraction, cluster analysis, threshold screening, and standardized data entry to ensure the effectiveness and reliability of new rules.

[0061] It should be noted that fault mechanism analysis, DTW algorithm clustering, and support and confidence calculation are all common analysis algorithms that can be implemented by staff using conventional programming tools without the need to develop proprietary algorithms.

[0062] The rule update strategy of the self-learning unit is as follows: After a newly generated candidate rule passes the support and confidence thresholds, the system first compares it with existing rules in the rule base. If the new rule is basically consistent with the fault type, stage characteristics, and transition conditions described by the existing rules, the parameters of the existing rules (such as the mean and standard deviation of the transition conditions) are updated with the new rule. If the new rule is inconsistent with any existing rule, it is stored as a new rule in the rule base. The self-learning process is usually executed periodically in offline and batch processing modes, or triggered when a certain number of unidentified fault samples accumulate, to avoid affecting the performance of real-time diagnosis.

[0063] In an embodiment of the present invention, the rule base construction rules for the fault chain sequence of the distribution box are as follows: fault chain rules are pre-set based on the fault mechanism of the distribution box and industry experience, see [reference]. Figure 2 Each rule includes: fault type, phase division, fault chain timing phase, and phase transition conditions.

[0064] For example, in each rule, the fault type is named using a composite naming method of fault phenomenon-possible cause, such as leakage fault-insulation aging type, over-temperature fault-loose connection type, which makes it easy to distinguish fault modes and avoid the same fault with the same name or the same fault with different names.

[0065] Phase division: All rules are uniformly divided into 4 phases; for example, Level 1: early hidden danger, Level 2: development stage, Level 3: critical stage, and Level 4: accident stage, consistent with the system failure chain phase definition.

[0066] Timing stages: Describe the main characteristics of each stage in chronological order. For example, the first-level timing stage is characterized by a slight increase in leakage current and stable temperature.

[0067] Transition conditions: Clarify the quantitative judgment criteria for stage transition, such as the transition condition from level one to level two: leakage current is not less than 5mA and lasts for 10 seconds.

[0068] It should be noted that the basic rule data is mainly collected through the following methods to obtain typical fault information: fault mechanism analysis: based on electrical theories such as Ohm's law and the principle of electromagnetic induction, the fault evolution law is derived; for example, the fault evolution law is the logic of leakage fault: insulation resistance decreases → leakage current increases → arc generation.

[0069] Industry experience summary: Refer to industry manuals and historical fault cases from power operation and maintenance companies.

[0070] Equipment manufacturer data: Collect and determine the fault characteristic parameters of the main components of the distribution box.

[0071] The rule base for the fault chain sequence of the distribution box also includes a self-learning unit, which is based on historical fault data, manually labeled data and emergency repair records during system operation.

[0072] In an embodiment of the present invention, the self-learning unit collects the following data during system operation to form a self-learning sample set.

[0073] Historical fault data: a dataset of fault features that the system has identified and processed; manually labeled data: suspected fault data manually marked by maintenance personnel; emergency repair record data: debriefing data after emergency repairs.

[0074] The above data was cleaned to remove invalid data, and data from different sources were converted into the format of fault development point sequence, timestamp, and processing result. At the same time, for unlabeled data, the operators added fault type labels to ensure that the sample set labels were complete.

[0075] After processing, the data is stored in a local cache unit for quick access by the self-learning unit.

[0076] For each fault sample in the sample set, extract the sub-parameter type features, numerical change amplitude features, and time interval features of the fault development point sequence to form a feature vector; after feature extraction, each fault sample corresponds to a three-dimensional feature vector, namely sub-parameter type, change amplitude, and time interval, which is used for subsequent cluster analysis.

[0077] The DTW algorithm is used to perform cluster analysis on the fault development point marker sequences in the historical fault feature dataset to identify common temporal features of similar faults.

[0078] The feature vectors of multiple unclassified samples are randomly selected from the sample set as initial cluster centers; for example, when the number of historical fault samples is 100-500 groups, 5 initial cluster centers are selected, and when it is greater than 500 groups, 8 initial cluster centers are selected.

[0079] The DTW algorithm is used to calculate the similarity distance between each unclassified sample and multiple cluster centers. The formula is as follows: Where X is the feature vector of the unclassified sample at the current fault development point, and Y is the feature vector of the cluster center. As weights, the weight for parameter type is set to 0.4, the weight for change magnitude is set to 0.3, and the weight for time interval is set to 0.3. denoted as Euclidean distance between two points; i represents the fault development point number, n represents the number of fault development points; j represents the time sequence number of the candidate fault chain in the rule base, and m represents the time sequence number of the candidate fault chain in the rule base.

[0080] It should be noted that the above weight settings are intended to highlight the primary influence of the sub-parameter type, since the sub-parameter type determines the fault type and has the highest weight, while the remaining two feature vectors are set on an average basis.

[0081] Each unclassified sample is assigned to the cluster with the smallest similarity distance, and then the mean of the feature vector of each cluster is recalculated as the new cluster center.

[0082] Repeat the above steps until the change in cluster centers is no greater than 0.01, indicating that the cluster centers have stabilized. Stop iterating at this point. Each cluster corresponds to a type of fault sample with common temporal characteristics. In this embodiment, the number of iterations is preset to 10. If the clustering is not stable within 10 iterations, the clustering result of the 10th iteration is taken to ensure a balance between clustering efficiency and effect.

[0083] Calculate the support and confidence between fault markers, determine new fault chain timing rules, and store them in the rule base.

[0084] For example, candidate rule generation: Based on the common temporal features of clustered samples, candidate rules are formulated according to the four classification formats of rules (i.e., fault type, stage division, fault chain temporal stage, stage transition condition); the transition condition is: the sample mean of each stage feature is plus or minus 1 standard deviation.

[0085] Support reflects the probability that events at each stage of a candidate rule will occur simultaneously. The formula for calculating support is: Among them, A, B, and C are the consecutive temporal stages of the candidate rules, namely the first, second, and third levels respectively.

[0086] Confidence level reflects the conditional probability that a subsequent event will occur after a previous event has occurred. The formula is: .

[0087] Set a support threshold of no less than 0.6 and a confidence threshold of no less than 0.8. If the confidence of all consecutive stages of a candidate rule is no less than 0.8 and the overall support is no less than 0.6, it is determined to be a valid new rule; otherwise, the candidate rule is removed.

[0088] It should be noted that the rule for the distribution box fault chain is that once the previous stage is triggered, the next stage is highly likely to follow. This evolution is not a random event, but a strong causal relationship determined by the electrical fault mechanism. The confidence level reflects the conditional probability of the next stage occurring after the previous stage occurs. If the confidence level is below 0.8, it means that after the previous stage occurs, the next stage has only a 70% probability of occurring, and the correlation will be weakened. Moreover, this value setting is based on long-term applications in fields such as power equipment fault diagnosis and industrial time-series fault chain identification, which conforms to the strong causal law of fault chain stage transitions and can ensure the reliability of the rule's prediction of subsequent stages. At the same time, the support level reflects the probability of all stages of a fault chain occurring simultaneously, that is, the commonness of the fault chain in real-world scenarios. Therefore, there is no need to set an excessively high threshold. As long as the rule covers most common scenarios, it can meet the system requirements.

[0089] This invention ensures the accuracy of fault identification during initial operation through a basic rule base and achieves dynamic optimization throughout the entire cycle through a self-learning unit, making the fault chain time sequence rule base more comprehensive and the judgment more accurate, thus providing support for subsequent fault chain matching and stage positioning.

[0090] Considering that the current fault may be a typical fault or a novel fault, the candidate fault chain can be screened by parameter type, which can significantly narrow the matching range and improve the identification efficiency. The DTW algorithm is used to calculate the temporal similarity, which can solve the problem of temporal alignment between the incomplete development of actual faults and the full stages of standard fault chains. Unmatched samples are submitted to the self-learning unit, which can realize the dynamic expansion of the rule base and adapt to novel faults.

[0091] In an embodiment of the present invention, the specific method for identifying the current fault and its stage is as follows: matching the fault development point's sub-parameter type with the fault chain in the rule base to filter out candidate fault chains.

[0092] Specifically, the 2-3 most prevalent sub-parameter types are extracted from the fault development trajectory as the main matching features. The filtered rule subset is traversed, and the main parameters of the time-series stage in the rule are compared with the main sub-parameters of the current fault. If the overlap is not less than 60%, the fault chain corresponding to the rule is included in the candidate fault chain list. It should be noted that since this is a coarse matching, the overlap standard is set to 60%.

[0093] Candidate fault chains are sorted in descending order based on the sum of their support and confidence scores according to the rules, with priority given to matching candidate chains with higher priority.

[0094] The DTW algorithm is used to calculate the temporal similarity between the time series of each fault development point in the current fault feature dataset and the candidate fault chain.

[0095] Specifically, three-dimensional feature vectors are constructed for the current fault development trajectory and the time-series template of each candidate fault chain.

[0096] Using the same method as the DTW algorithm described above, the temporal similarity distance between the current fault development trajectory and the time series of each candidate fault chain is calculated.

[0097] The similarity distance is converted into a temporal similarity in the interval [0, 1]. The similarity distance is then compared with the maximum similarity distance, and the difference between 1 and this ratio is taken as the temporal similarity.

[0098] Based on temporal similarity, it is determined whether the current fault development point belongs to a certain candidate fault chain. If it does, the current fault development point is located to the specific temporal stage of the fault chain. If it does not, the fault development trajectory is submitted as a new fault sample to the self-learning unit for processing.

[0099] Specifically, by reading the temporal similarity results of candidate fault chains, if a candidate chain has a temporal similarity of not less than 85%, it is considered a successful match, and the fault type corresponding to that candidate chain is the current fault type; if the temporal similarity of all candidate chains is less than 85%, it is considered a failed match. It should be noted that the 85% similarity threshold is the optimal value verified by engineering testing and can be applied to various working conditions.

[0100] For a successfully matched candidate fault chain, extract the main parameters of each fault development point in the current fault development trajectory (such as leakage current value, temperature value, etc., which are equal to the parameters associated with the match) and compare them with the transition conditions of each time stage of the candidate chain.

[0101] The stage in which the current fault occurs is the one where all development point parameters meet the characteristics of the current stage and the transition conditions for the next stage are not triggered.

[0102] If a development point covers multiple stage characteristics at the same time, the stage corresponding to the last failure development point shall be the current stage (e.g., if the last development point triggers the transition condition from level 2 to level 3 but does not meet the duration, it shall be positioned as level 2).

[0103] For failure trajectories that fail to match, they are marked as unclassified failure samples. The unclassified failure samples are compared with the feature vectors of all known failure chains in the rule base, and the difference is calculated. The difference is the difference between 1 and the mean of temporal similarity. If the difference is not less than 0.7, it is judged as a potential new failure chain sample.

[0104] Potential novel fault chain samples are pushed to the self-learning unit and processed according to the process of feature extraction, cluster analysis, support and confidence calculation, and rule generation. If the difference is less than 0.7, it is judged as a suspected abnormal data sample and pushed to the operation and maintenance platform for manual review.

[0105] In an embodiment of the present invention, the specific method for the subsequent triggered fault chain stage and the transition time of the remaining stage is as follows: based on the fault type associated with the current fault development point and the located timing stage, retrieve the complete fault chain timing stage sequence that completely matches the fault type from the rule base of the power distribution box fault chain timing, and locate all subsequent evolution stages after the current stage based on the timing sequence logic.

[0106] The core basis for retrieving the corresponding sequence is the identified fault type and the current stage, ensuring that subsequent steps are not missed or deviated from. Only by clearly defining the complete evolution path can the risk of fault propagation be fully predicted.

[0107] For each subsequent stage, based on the magnitude of the numerical change of the fault development point and the time node of its occurrence, the earliest and latest time points that trigger the subsequent stage are calculated as the prediction range of the transition time of the subsequent stage.

[0108] Specifically, for each transition parameter in a single subsequent stage, the time required for the parameter to reach the threshold is calculated as follows: calculate the difference between the parameter threshold and the current value of the parameter, and use the ratio of the difference to the parameter change amplitude as the time required for the parameter to reach the threshold; where the change amplitude is the real-time change rate of the current fault development point, and the ratio of the difference between the current fault development point value and the value of the previous acquisition cycle to the acquisition cycle value is used as the real-time change rate of the parameter.

[0109] Determining the earliest time point of this subsequent stage requires multiple parameters to be satisfied simultaneously: the sum of the time point when the fault development point occurs and the minimum time required for all parameters to meet the criteria.

[0110] Determining the latest time point of this subsequent stage requires multiple parameters to be satisfied simultaneously: the sum of the time point when the fault development point occurs and the maximum time required for all parameters to meet the criteria.

[0111] The advantages of the aforementioned subsequent triggered fault chain stages and remaining stage transition times are as follows: First, by retrieving the complete fault chain time sequence, the entire chain evolution path of the fault from the current stage to the final accident stage can be clearly presented, avoiding the risk of ignoring subsequent high-risk links due to focusing only on the current stage; Second, the transition time prediction interval calculated based on the parameter change amplitude comprehensively considers the time differences of different parameters reaching the threshold, providing both the earliest warning node where the fault may accelerate its evolution and the latest boundary of fault development, enabling maintenance personnel to reasonably allocate emergency repair resources according to actual working conditions and optimize emergency repair scheduling efficiency while ensuring safety.

[0112] Considering the immediacy and fundamental nature of the factors influencing the risk of cascading failures, it is necessary to complete an equipment importance analysis and then conduct a risk assessment in conjunction with dynamic failure indicators. Among these factors, equipment importance determines the minimum scope and severity of the failure's impact, while the current stage and remaining time determine the immediate urgency of the failure. Combining these three factors can comprehensively cover the scope of the risk.

[0113] In an embodiment of the present invention, the assessment method for the fault cascading risk level is as follows: based on the current fault chain stage, the remaining stage transition time, and the importance of the equipment, a linear weighted algorithm is used to calculate the risk value, and the cascading risk level is determined based on the risk value.

[0114] In the above analysis, it is necessary to quantify the current fault chain stage, the transition time of the remaining stages, and the importance of the equipment.

[0115] Since there is no unified standard for measuring the current fault chain stage, the transition time of the remaining stage, and the importance of the equipment, direct superposition analysis would lead to logical contradictions. This invention adopts a scoring method to unify the three types of heterogeneous factors into standardized scores of [0, 100] through pre-defined quantitative rules, providing an operable unified data foundation for subsequent linear weighted calculation of risk values.

[0116] Specifically, the quantification of the current failure chain stage is based on the stages identified in the failure chain above, assigning corresponding scores. The later the stage, the higher the risk. Level 1, the early hidden danger, is assigned 10 points; Level 2, the development stage, is assigned 30 points; Level 3, the critical stage, is assigned 60 points; and Level 4, the accident stage, is assigned 100 points. The gradient design of the stage quantification scores is to reflect the increasing risk effect of stage transitions. The critical stage and the accident stage have higher score weights, which is consistent with the law of failure evolution.

[0117] The remaining stage transition time is quantified based on the median of the predicted time interval, using negative correlation quantification, that is, the shorter the time, the higher the risk. The specific criteria are as follows: 10 points are assigned for a remaining stage transition time greater than or equal to 30 minutes; 30 points are assigned for a remaining stage transition time greater than or equal to 10 minutes and less than 30 minutes; 60 points are assigned for a remaining stage transition time greater than or equal to 1 minute and less than 10 minutes; and 100 points are assigned for a remaining stage transition time less than 1 minute.

[0118] The importance of the equipment is determined by the system based on the importance assessment values ​​of the power supply objects supplied by the distribution box and the regional power supply importance assessment values. The total score of the equipment importance is the same as the transition time of the current fault chain stage and the remaining stage, and is assigned a total score of 100 points. Among them, the corresponding level is determined according to the load type connected to the distribution box and the corresponding score is directly assigned, with a total score of 60 points; the level is determined according to the location of the distribution box in the power distribution system and the score is assigned according to the grading standard, with a total score of 40 points.

[0119] For example, the importance assessment value of the power supply object of the distribution box is assigned a score according to its importance. The score is assigned according to the preset load importance classification standard. For example, the distribution box that supplies power to key production equipment such as factory production line is assigned the highest score (e.g., 60 points); the distribution box that supplies power to major loads such as commercial lighting is assigned a medium score (e.g., 40 points); and the distribution box that supplies power to ordinary loads such as residential buildings is assigned a basic score (e.g., 20 points).

[0120] The regional power supply importance assessment value is based on the power supply level of the distribution box in the power distribution network topology and its power supply coverage. The assessment criteria are whether the distribution box is connected to a secondary distribution box downstream and the power supply area affected. Specifically, the score is determined according to the number of secondary distribution boxes covered. For example, covering no less than 3 secondary distribution boxes is given 40 points, covering 1-2 secondary distribution boxes is given 25 points, and no secondary distribution boxes are given 10 points.

[0121] The sum of the importance assessment value of the power supply object of the distribution box and the importance assessment value of the regional power supply is used as the equipment importance. This equipment importance serves as a basic indicator for measuring the potential impact range and severity of consequences of a fault in risk assessment.

[0122] The current fault chain stage, remaining stage transition times, and equipment importance are weighted and summed. The current fault chain stage and remaining stage transition times have a weight of 0.4, while equipment importance has a weight of 0.2. These weights are based on the following: the current fault chain stage and remaining stage transition times directly reflect the urgency and rate of change of the fault, and have a more significant impact on the immediacy of repair decisions, hence they are given higher weights. Equipment importance, as a fundamental influencing factor, determines the potential scale of the fault's consequences; its weight is slightly lower but indispensable. The weight allocation prioritizes urgency over the scope of impact and can be fine-tuned according to the actual scenario.

[0123] The flexibility of the weighted summation analysis method described above lies in its adaptability to different scenarios. For example, in industrial scenarios, if the power supply priority of the production line is extremely high, the weight of equipment importance can be adjusted to 0.3, and the weight of the current stage can be adjusted to 0.35. In residential power consumption scenarios, the urgency of the remaining time is more critical, so the original weight can be maintained. In commercial scenarios, if the regional power supply coverage has a greater impact, the weighting of the regional power supply importance can be appropriately increased. This flexible adjustment allows the system to adapt to the needs of different power distribution scenarios such as industrial, commercial, and residential, avoiding the problem that a single assessment cannot cover different scenarios.

[0124] In embodiments of the present invention, the risk levels of the chain are classified into four categories: low risk, medium risk, high risk, and extremely high risk.

[0125] The risk scores are as follows: 85 or higher is considered extremely high risk; 60 or higher and less than 85 is considered high risk; 30 or higher and less than 60 is considered medium risk; and less than 30 is considered low risk.

[0126] To address the shortcomings of existing technologies in risk assessment due to their limited dimensions, this invention quantifies and scores three types of factors: the fault chain stage reflecting the severity of the fault, the remaining transition time reflecting the urgency, and the equipment importance reflecting the scope of impact. Through linear weighting, it achieves multi-dimensional risk analysis, comprehensively covering the scope of risk assessment.

[0127] The graded processing module formulates different handling strategies based on the fault cascading risk level.

[0128] The specific working method of the graded processing module is as follows: when the fault cascading risk level is low, only early warning information containing fault characteristics is generated and pushed to the operation and maintenance management platform, without triggering any active control operations.

[0129] When the risk level is medium risk, an early warning work order with a handling time limit is generated. The early warning work order shall include at least the equipment location information, the fault type identifier, and the suggested handling time window.

[0130] If the risk level reaches high risk, the local alarm device is triggered and a remote emergency alarm is generated. At the same time, non-critical load loops are selectively cut off based on the preset load priority strategy, and nearby operation and maintenance resources are automatically dispatched.

[0131] When the risk is assessed as extremely high, systemic protection operations are performed, including cutting off the main power supply to the distribution box, activating the fire protection and security system, reporting the fault status to the superior monitoring center, and activating the corresponding emergency plan.

[0132] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention, which are still covered within the protection scope of the present invention.

Claims

1. An intelligent fault detection system for an intelligent integrated distribution box, characterized in that, include: The data acquisition module is used to collect the operating parameters of the distribution box at a basic acquisition frequency. When the operating parameters trigger the corresponding threshold, active acquisition is started to obtain the operating parameters within a set time period before and after the trigger. The data processing module preprocesses the operating parameters and normalizes each sub-parameter in the operating parameters to form a fault feature dataset. Mark the fault development points in the fault feature dataset; The risk assessment module constructs a rule base for the fault chain sequence of the distribution box, matches the fault development point with the rule base, identifies the current fault and its current time sequence stage, analyzes the subsequent triggered fault chain stages and the transition time of the remaining stages, and assesses the fault chain risk level. The graded processing module formulates different handling strategies based on the fault cascading risk level.

2. The intelligent integrated distribution box fault detection system according to claim 1, characterized in that, The operating parameters include electrical parameters and environmental parameters. The electrical parameters include three-phase voltage, neutral line voltage, line voltage, three-phase current, neutral line current, and leakage current. The environmental parameters include temperature and humidity. The active acquisition is initiated as follows: when any operating parameter triggers a static or dynamic threshold, the acquisition frequency is immediately increased from the basic acquisition frequency to the active acquisition frequency.

3. The intelligent integrated distribution box fault detection system according to claim 2, characterized in that, The static threshold is the rated parameter of the device; the dynamic threshold is determined based on the historical operating parameters within the sliding window.

4. The intelligent integrated distribution box fault detection system according to claim 1, characterized in that, The method for marking fault development points in the fault feature dataset is as follows: Arrange the data in the fault feature dataset in chronological order of collection time; When the change of any sub-parameter exceeds the mutation threshold within a specified time, the sub-parameter is marked as a fault development point. Record the sub-parameter types, numerical change ranges, and occurrence time nodes corresponding to each fault development point, and generate a fault development trajectory with time series markers.

5. The intelligent fault detection system for an intelligent integrated distribution box according to claim 4, characterized in that, The rule base for constructing the fault chain sequence of the distribution box is as follows: Based on the fault mechanism of the distribution box and industry experience, fault chain rules are pre-set. Each rule includes: fault type, stage division, fault chain timing stage and stage transition conditions. The rule base for the fault chain sequence of the distribution box also includes a self-learning unit. Based on historical fault data, manually labeled data and emergency repair records during system operation, the self-learning unit uses the DTW algorithm to perform cluster analysis on the fault development point label sequence in the historical fault feature dataset to identify the common time sequence features of similar faults; calculates the support and confidence between fault label points, determines new fault chain sequence rules and stores them in the rule base.

6. The intelligent fault detection system for an intelligent integrated distribution box according to claim 5, characterized in that, The specific methods for identifying the current fault and its stage are as follows: Candidate fault chains are selected by matching the sub-parameter types of the fault development point with the fault chains in the rule base. The DTW algorithm is used to calculate the temporal similarity between the time series of each fault development point in the current fault feature dataset and the candidate fault chain. Based on temporal similarity, it is determined whether the current fault development point belongs to a certain candidate fault chain. If it does, the current fault development point is located to the specific temporal stage of the fault chain. If it does not, the fault development trajectory is submitted as a new fault sample to the self-learning unit for processing.

7. The intelligent fault detection system for an intelligent integrated distribution box according to claim 6, characterized in that, The specific methods for the subsequent triggered fault chain phases and the transition times of the remaining phases are as follows: Based on the fault type and the time sequence stage associated with the current fault development point, retrieve the complete fault chain time sequence stage sequence corresponding to the fault type from the rule base, and locate all subsequent stages after the current stage. For each subsequent stage, based on the magnitude of the numerical change of the fault development point and the time node of its occurrence, the earliest and latest time points that trigger the subsequent stage are calculated as the prediction range of the transition time of the subsequent stage.

8. The intelligent fault detection system for an intelligent integrated distribution box according to claim 1, characterized in that, The assessment method for the cascading failure risk level is as follows: Based on the current stage of the failure chain, the transition time of the remaining stages, and the importance of the equipment, a linear weighted algorithm is used to calculate the risk value, and the cascading risk level is determined based on the risk value. Chain risk levels are classified into four categories: low risk, medium risk, high risk, and extremely high risk.

9. The intelligent fault detection system for an intelligent integrated distribution box according to claim 8, characterized in that, The importance of the equipment is determined using a quantitative scoring method, including: The importance assessment value of the power supply object of the distribution box is determined based on the type of load connected to the distribution box; The importance assessment value of regional power supply is determined based on the power supply level and coverage of the distribution box in the power distribution network topology; The quantitative score of the equipment importance is calculated based on the importance assessment value of the power supply object and the importance assessment value of the regional power supply.

10. The intelligent integrated distribution box fault detection system according to claim 8, characterized in that, The specific working method of the hierarchical processing module is as follows: When the risk level of the fault cascading is low, only early warning information containing fault characteristics is generated and pushed to the operation and maintenance management platform, without triggering any active control operations. When the risk level is medium risk, an early warning work order with a handling time limit is generated. The early warning work order shall at least include equipment location information, fault type identifier and suggested handling time window. If the risk level reaches high risk, the local alarm device is triggered and a remote emergency alarm is generated. At the same time, non-critical load loops are selectively cut off based on the preset load priority strategy, and nearby operation and maintenance resources are automatically dispatched. When the risk is assessed as extremely high, systemic protection operations are performed, including cutting off the main power supply to the distribution box, activating the fire protection and security system, reporting the fault status to the superior monitoring center, and activating the corresponding emergency plan.

Images