Power distribution network user voltage abnormity diagnosis method and system

Abstract

The invention provides a power distribution network user voltage abnormity diagnosis method and system, and the method comprises the steps: extracting basic features and electrical features when a low-voltage event occurs; acquiring an event type corresponding to the low-voltage event according to the at least one key factor under a first preset rule; performing cause diagnosis on the low-voltage event under a second preset rule, and performing primary marking on a cause diagnosis result based on a third preset rule; constructing a final evaluation model; inputting the basic features and the electrical features into the final evaluation model according to a primary marking result to obtain a first diagnosis result corresponding to the primary marking result; and verifying and fusing the first diagnosis result and the cause diagnosis result under a fourth preset rule so as to output a diagnosis report about the low-voltage event according to a verification and fusion result. According to the invention, the efficiency and accuracy of low-voltage diagnosis can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of low - voltage diagnosis of distribution networks, and particularly to a method and system for diagnosing abnormal user voltage in a distribution network. Background Art

[0002] In traditional power systems, the monitoring and diagnosis of user low - voltage problems are relatively simple. It mainly relies on the voltage monitoring of fixed substations and a small number of key nodes, and generates alarms by setting fixed thresholds. The cause analysis depends more on the experience judgment of operation and maintenance personnel and manual inspections, with low diagnosis efficiency and accuracy, and it is difficult to deal with complex and concealed faults. In addition, the data source in the diagnosis process is single, mainly the steady - state operation data collected by the SCADA system, lacking refined measurement information on the user side and the distribution transformer side, further reducing the diagnosis accuracy. Summary of the Invention

[0003] The purpose of the present invention is to provide a method and system for diagnosing abnormal user voltage in a distribution network, aiming to solve the problems of accuracy, real - time performance, and interpretability of user low - voltage cause analysis under the background of a new power system, so as to improve the intelligent level of power grid operation and power supply reliability.

[0004] In the first aspect, the present invention provides a method for diagnosing abnormal user voltage in a distribution network, and the method includes: When a low - voltage event occurs, collect user - side electrical data and sub - station area electrical data, and extract basic features and electrical features from the user - side electrical data and the sub - station area electrical data; Obtain at least one key factor regarding the low - voltage event, and obtain an event type corresponding to the low - voltage event according to the at least one key factor under a first preset rule; Under a second preset rule, perform cause diagnosis on the low - voltage event according to the basic features, the electrical features, and the event type to obtain a cause diagnosis result, and perform a first marking on the cause diagnosis result based on a third preset rule; Construct an initial evaluation model, obtain multiple historical low - voltage events, perform feature extraction and cause marking on the historical low - voltage events to obtain a data set, and train the initial evaluation model according to the data set to obtain a final evaluation model; Input the basic features and electrical features into the final evaluation model according to the first marking result to obtain a first diagnosis result corresponding to the first marking result; Under a fourth preset rule, perform verification and fusion on the first diagnosis result and the cause diagnosis result, and output a diagnosis report regarding the low - voltage event according to the verification and fusion result.

[0005] In a second aspect, the present invention provides a distribution network user voltage anomaly diagnosis system, which includes: A feature extraction module, configured to collect user-side electrical data and substation electrical data when a low voltage event occurs, and extract basic features and electrical features from the user-side electrical data and the substation electrical data; An event type acquisition module, configured to obtain at least one key factor regarding the low voltage event, and obtain an event type corresponding to the low voltage event according to the at least one key factor under a first preset rule; A marking module, configured to perform cause diagnosis on the low voltage event according to the basic features, the electrical features, and the event type under a second preset rule to obtain a cause diagnosis result, and perform a primary marking on the cause diagnosis result based on a third preset rule; A model training module, configured to construct an initial evaluation model, obtain a plurality of historical low voltage events, perform feature extraction and cause marking on the historical low voltage events to obtain a data set, and train the initial evaluation model according to the data set to obtain a final evaluation model; A machine recognition module, configured to input the basic features and the electrical features into the final evaluation model according to the primary marking result to obtain a first diagnosis result corresponding to the primary marking result; A diagnosis result output module, configured to verify and fuse the first diagnosis result and the cause diagnosis result under a fourth preset rule, and output a diagnosis report regarding the low voltage event according to the verification and fusion result.

[0006] In a third aspect, the present invention provides a storage medium, which stores one or more programs, and when the program is executed by a processor, the above-mentioned distribution network user voltage anomaly diagnosis method is implemented.

[0007] In a fourth aspect, the present invention provides an electronic device, which includes a memory and a processor, wherein: The memory is used to store a computer program; The processor is configured to implement the above-mentioned distribution network user voltage anomaly diagnosis method when executing the computer program stored on the memory.

[0008] Compared with the prior art, the present invention has the following advantages: 1. The present invention constructs a feature engineering system driven by electrical mechanism; designs a hierarchical positioning rule for low voltage event types based on distributed state deduction; and adopts a hybrid model architecture combining rule pre-diagnosis and machine learning supplementary diagnosis; finally, a diagnosis result verification and conflict fusion decision mechanism based on key feature quantization review is introduced, thereby greatly improving the accuracy, real-time performance, and interpretability of user low voltage cause analysis.

[0009] 2. By deeply integrating electrical mechanism, rule reasoning, and machine learning, the ability to identify the causes of various low-voltage situations (especially complex and ambiguous scenarios) is enhanced; through topological relationship and distributed state deduction, the level where a low-voltage event occurs (the end of a low-voltage line, the outlet of a distribution transformer, the end of a medium-voltage line, the outlet of a substation) can be accurately located; by combining the clear logic of rules and the generalization ability of machine learning, and introducing features driven by electrical mechanism, the diagnostic process becomes more transparent and the results are more reliable; for cases where the rule diagnosis is inconsistent with the prediction of the machine learning model, a fusion decision-making process based on key feature quantization review is designed to improve the reliability of the final diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 It is a flowchart of a method for diagnosing abnormal voltage of distribution network users proposed in an embodiment of the present invention; Figure 2 It is a schematic structural diagram of a system for diagnosing abnormal voltage of distribution network users proposed in an embodiment of the present invention.

[0011] The following specific embodiments will further illustrate the present invention in conjunction with the above-mentioned drawings. SPECIFIC EMBODIMENTS

[0012] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention. Unless otherwise defined, the technical terms or scientific terms used herein shall have the ordinary meaning as understood by those of ordinary skill in the art in the field to which the present invention belongs. The words such as "including" used herein mean that the elements or items appearing before the word cover the elements or items listed after the word and their equivalents, without excluding other elements or items.

[0013] Traditional analysis methods for low-voltage problems of distribution network users mainly include: (1) Alarm method based on threshold: This is the most basic method. By monitoring user voltage data, when the voltage value is lower than a preset threshold (such as -10% or -7% of the nominal voltage) and lasts for a certain period of time, a low-voltage alarm is generated. This method can only detect the low-voltage phenomenon and cannot determine its cause.

[0014] (2) Diagnostic methods based on simple rules: On the basis of threshold alarms, some simple logical rules are added. For example, if low voltage occurs simultaneously among multiple users in a certain district, it may be judged as a district-level problem; if the outlet voltage of the district is normal while the user voltage is low, it may be judged as a low-voltage line problem. These rules are usually set based on experience, lacking in-depth utilization of electrical principles, and with fixed thresholds, making it difficult to adapt to complex operating conditions and changing fault modes. For example, it may simply judge according to the voltage drop amplitude, but does not distinguish the voltage performance differences of transformers with different connection groups during single-phase loss, nor does it consider the voltage characteristics of specific faults such as neutral line disconnection.

[0015] (3) Single machine learning applications: Some studies have tried to directly use machine learning (such as SVM, neural networks, etc.) to classify low-voltage events. Usually, the statistical features of basic measurement data such as voltage and current are used as inputs. Although machine learning can handle non-linear relationships, there are the following problems: ① A large number of high-quality labeled samples are required, and the acquisition cost is high; ② The model may become a "black box", lacking physical interpretability and making it difficult to gain the trust of operators; ③ It is highly dependent on feature engineering. If key electrical mechanism features are not extracted, the model performance is limited; ④ It fails to effectively combine the topological structure of the power system and expert experience rules.

[0016] (4) Traditional fault recording and manual analysis: In the event of severe low voltage or a fault, rely on a fault recording device to record the waveform and have professional engineers conduct manual analysis. This method has high accuracy, but is costly, cannot cover all low-voltage events, and has a slow response speed, making it unsuitable for the management of large-scale and regular low-voltage problems.

[0017] In summary, for traditional voltage anomaly detection methods, the applicant has found the following problems: (1) Insufficient diagnostic accuracy: Simple rules have poor judgment ability for complex operating conditions and fuzzy boundary scenarios, and are prone to misjudgment or missed judgment. For example, it is difficult to distinguish whether the voltage at the end of the low-voltage line is low or the overall outlet voltage of the district is low, or it is difficult to distinguish different types of faults.

[0018] (2) Inadequate feature extraction: It fails to fully utilize electrical mechanisms to extract key electrical quantitative indicators that can directly reflect specific low-voltage causes (such as single-phase loss of different wiring groups, neutral line disconnection) as features, resulting in a lack of pertinence in the diagnostic model.

[0019] (3) Poor model adaptability: Rules based on fixed thresholds are difficult to adapt to dynamic scenarios such as changes in power grid operating modes and load fluctuations. A single machine learning model may be sensitive to data quality and have limited generalization ability.

[0020] (4)Lack of hierarchical positioning: Most methods fail to effectively combine topological information for hierarchical and graded positioning of low-voltage problems (such as distinguishing whether the problem is at the low-voltage line, substation area, medium-voltage line, or substation level), which is not conducive to accurate traceability.

[0021] (5)Weak interpretability of diagnostic results: Especially for pure machine learning methods, their decision-making processes are difficult to explain. Simple rules are easy to understand but cannot handle situations outside their rules.

[0022] (6)Lack of confidence evaluation and conflict handling: Existing methods usually only give a diagnostic conclusion, lacking a quantitative evaluation of the reliability of this conclusion. When conflicts occur between different methods or rules, there is no clear handling mechanism.

[0023] (7)Insufficient data fusion: It fails to effectively integrate multi-source heterogeneous data such as measurement data, equipment ledgers, user ledgers, and event records, resulting in low information utilization.

[0024] Based on this, the present invention proposes a method for diagnosing abnormal user voltage in a distribution network. By constructing a feature engineering system driven by electrical mechanisms; designing hierarchical positioning rules for low-voltage event types based on distributed state deduction; and adopting a hybrid model architecture that combines rule pre-diagnosis and machine learning supplementary diagnosis; finally, a diagnostic result verification and conflict fusion decision-making mechanism based on quantitative review of key features is introduced to achieve the diagnosis of low-voltage events in various complex situations, thereby greatly improving the accuracy, real-time performance, and interpretability of the analysis of the causes of user low voltage.

[0025] As Figure 1 shown, an embodiment of the present invention proposes a method for diagnosing abnormal user voltage in a distribution network. The method includes steps S101 to S106, where: Step S101: When a low-voltage event occurs, collect user-side electrical data and substation area electrical data, and extract basic features and electrical features from the user-side electrical data and the substation area electrical data; It should be noted that the user-side electrical data includes user voltage and user current, and the substation area electrical data includes substation area outlet voltage and substation area outlet current. In addition, in some embodiments, the collected data also includes distribution equipment ledger information: distribution transformer capacity , the affiliated medium-voltage line Line_MV, the medium-voltage line topology structure (superior-subordinate relationship); user ledger information: user electricity consumption nature , the affiliated substation area TP_user, the metering method (single-phase / three-phase); other data: user low-voltage event alarm records, line power outage and restoration records, operation records (such as load transfer) and other event information.

[0026] In some embodiments, after the above data is collected, preprocessing operations such as quality verification, outlier processing, time alignment, and data interpolation are performed on the acquired data; and preprocessing operations such as quality verification, outlier processing, time alignment, and data interpolation are performed on the acquired data.

[0027] In addition, in some embodiments, the basic features include the mean, variance, peak value, valley value, voltage change rate, current change rate, and autocorrelation coefficient of the user voltage, user current, substation outlet voltage, and substation outlet current; In addition, in some embodiments, key electrical quantification indicators (electrical features) reflecting specific low-voltage causes need to be calculated as input features for subsequent machine learning models, specifically including: Extract the phase loss feature of the substation according to the following formula: ; where is the first phase loss feature of the substation, returns the absolute value of a number, 、 are the two-phase voltages with small values in the substation outlet voltage, is the voltage of the phase with the largest value in the substation outlet voltage, is the second phase loss feature of the substation, is the nominal voltage, is the third phase loss feature of the substation; Extract the low-voltage line network problem feature according to the following formula: ; where is the first low-voltage line network problem feature, is the second low-voltage line network problem feature, is the peak value of the substation outlet voltage during the peak load period, is the peak value of the user voltage, is the valley value of the substation outlet voltage during the low load period, is the valley value of the user voltage; Extract the neutral line disconnection feature according to the following formula: ; where is the highest phase voltage, 、 、 are the phase A, phase B, and phase C voltages in the substation outlet voltage respectively, is the lowest phase voltage, 、 are the first and second neutral line disconnection features respectively; The abnormal acquisition features are calculated according to the following formula, that is, the average value and the minimum value of the correlation coefficients between the outlet voltage of the transformer substation area and the voltages of all subordinate users are calculated: ; Among them, is the first abnormal acquisition feature, is the second abnormal acquisition feature, is the function for calculating the average value, is the function for calculating the correlation coefficient between variables, is the outlet voltage of the transformer substation area, is the voltage of the i-th user; The connection group features of the transformer substation area are extracted according to the following formula, that is, the correlation coefficient between the change in the outlet voltage of the transformer substation area and the change in current is calculated to assist in judging the connection group: ; Among them, is the connection group feature of the transformer substation area, is the change in the outlet voltage of the transformer substation area, is the change in the outlet current of the transformer substation area; In addition, in some embodiments, the electrical feature further includes the fourth phase loss feature of the transformer substation area: , are the i-th phase and the j-th phase of the outlet voltage of the transformer substation area respectively, that is, the correlation coefficients between every two of the three-phase voltages are calculated.

[0028] Step S102: Obtain at least one key factor regarding the low voltage event, and obtain the event type corresponding to the low voltage event according to the at least one key factor under the first preset rule; It should be noted that this step involves the hierarchical classification of the user low voltage problem based on distributed state deduction. The key factors include the number of users with low voltage events occurring in the same phase within the same transformer substation area , the duration of low voltage , the starting time difference of the low voltage event , and the minimum outlet voltage of the transformer substation area during the occurrence of the low voltage event . The specific hierarchical classification process is as follows: (1) If , and , and , and , then it is determined that the event type is the low voltage event at the end of the low voltage line (Event_LV_End).

[0029] (2) If , and , and , and , it is determined that the event type is the low voltage event at the distribution transformer outlet (Event_TP_Outlet).

[0030] In addition, in some embodiments, the key factors further include the number of low voltage events at the distribution transformer outlets existing under the same medium voltage line , the starting time difference of each low voltage event at the distribution transformer outlet , and the hierarchical classification process further includes: (3) If and , it is determined that the event type is the low voltage event at the end of the medium voltage line (Event_MV_End).

[0031] In addition, in some embodiments, the key factor further includes that if and , it is determined that the event type is the low voltage event at the end of the medium voltage line, and the hierarchical classification process further includes: (4) If and , it is determined that the event type is the low voltage event at the substation outlet (Event_SS_Outlet).

[0032] Finally, the distribution level is carried out: each low voltage event of the user is assigned to the highest level event that it can be associated with. The priority order is: Event_SS_Outlet > Event_MV_End > Event_TP_Outlet > Event_LV_End.

[0033] Step S103: Under the second preset rule, perform cause diagnosis on the low voltage event according to the basic features, the electrical features, and the event type to obtain a cause diagnosis result, and perform a primary mark on the cause diagnosis result based on the third preset rule; It should be noted that the second preset rule includes multiple situations, specifically: The first situation: when the event type is the low voltage event at the end of the low voltage line, if and and , the cause diagnosis result is: the low voltage line grid problem; The second situation: when the event type is the low voltage event at the end of the medium voltage line, if load transfer and medium voltage single-phase disconnection are excluded, the cause diagnosis result is: the medium voltage line grid problem; The third situation: when the event type is the low voltage event at the distribution transformer outlet, if and and and , it is the Yyn0 connection group. At this time, if is normal and and and then the cause diagnosis result is: there is a problem of open phase on the high-voltage side; If and / or and / or and / or then it is the Dyn11 connection group. At this time, if and there are two-phase voltages within the first preset range and the other phase voltage is less than or equal to then the cause diagnosis result is: there is a problem of open phase on the high-voltage side; The fourth case: When the event type is the low-voltage event at the distribution transformer outlet, set the first condition as: and or there are two-phase voltages both less than ; The second condition is: or ; If both the first condition and the second condition are satisfied at the same time, the cause diagnosis result is: the neutral line is broken; The fifth case: When the event type is the low-voltage event at the end of the medium-voltage line, if there is a line power outage or operation record before the low-voltage event occurs, the cause diagnosis result is: load transfer; The sixth case: When the event type is the low-voltage event at the distribution outlet, if and then the cause diagnosis result is: acquisition anomaly; The seventh case: When the event type is the low-voltage event at the end of the medium-voltage line, if there is a problem of open phase on the high-voltage side of at least two substations on the same line at the same time, the cause diagnosis result is: single-phase break of the medium-voltage line; In addition, in some embodiments, the specific marking process is: If the cause diagnosis result related to the low-voltage event is only one kind, perform a first-level marking; If the cause diagnosis result related to the low-voltage event is zero, perform a second-level marking; If the cause diagnosis result related to the low-voltage event includes at least two kinds, perform a third-level marking.

[0034] Exemplarily, if the user's low-voltage event only meets the diagnostic rules of one type of low-voltage cause, it is initially determined as this cause (Cause_Rule) and marked as "High-confidence Cause Diagnosis (Rule_High_Confidence)"; If the user's low-voltage event cannot meet the diagnostic rules of any type of low-voltage cause, it is marked as "Cause Unknown (Rule_Unknown)"; If the diagnostic rules for two or more types of low-voltage causes are satisfied simultaneously, it is marked as "Cause Ambiguous (Rule_Ambiguous)".

[0035] In summary, when the traditional low-voltage diagnostic method processes a large number of alarms, if a complex analysis model is directly adopted, it may face the problem of low computational efficiency. Moreover, for some typical faults with clear electrical characteristics, rule-based judgment is often more direct and efficient. However, the existing rule system may have problems such as incomplete rule coverage, threshold setting relying on expert experience and being inflexible to adjust, and inaccurate hierarchical positioning of low-voltage events (especially in the case of incomplete coverage of metering devices), which limits the diagnostic accuracy and practicality of the rule system. Based on this, the present invention designs a hierarchical and graded method for low-voltage problems based on distributed state deduction and a rule-based rapid pre-diagnosis process based on electrical mechanism. This step first generates virtual low-voltage events at the end of the low-voltage line, the outlet of the distribution transformer, the end of the medium-voltage line, and even the outlet of the substation by gradually deducing upward through the distributed state deduction logic, based on the measurement data at the outlet of the substation area and the user side, combined with the spatio-temporal correlation characteristics of low-voltage events of the user group (such as multiple users of the same phase, the same substation area, duration, start time difference, etc.), and assigns a unique and highest-priority hierarchical attribution to each user low-voltage event, solving the problem of hierarchical positioning caused by metering missing. Secondly, for different levels and known typical low-voltage causes (such as low-voltage / medium-voltage grid problems, phase loss on the high-voltage side of the substation area, neutral line disconnection, load transfer, acquisition anomaly, single-phase disconnection of the medium-voltage line, etc.), a series of clear diagnostic rules are constructed by combining the basic features and electrical characteristics extracted in the above steps. These rules directly utilize the typical electrical appearances of faults (for example, the feature of "one phase normal and two phases low and the sum of voltages approximate" when the Dyn11 distribution transformer is phase-lost, and the feature of "voltage imbalance and specific correlation" when the neutral line is disconnected). Finally, if the user low-voltage event satisfies the diagnostic rules for a certain type of cause, it is initially marked as "High-confidence Cause Diagnosis"; if it cannot be satisfied or satisfies multiple rules simultaneously, it is marked as "Cause Unknown" or "Cause Ambiguous", providing a preliminary screening result for the subsequent steps. In addition, through the hierarchical and graded method, the information loss caused by insufficient actual metering points is compensated, and the positioning accuracy of the low-voltage problem source is improved; at the same time, by using the rules containing electrical mechanism, typical low-voltage events with obvious features can be quickly and accurately identified, greatly improving the diagnostic efficiency, and providing high-value prior information and preliminary confidence evaluation for the diagnosis of subsequent complex causes, effectively diverting the diagnostic tasks.

[0036] Step S104: Construct an initial evaluation model, obtain multiple historical low-voltage events, perform feature extraction and cause marking on the historical low-voltage events to obtain a data set, and train the initial evaluation model according to the data set to obtain a final evaluation model; It should be noted that regarding this evaluation model, it adopts the XGBoost model. The input of its input layer is multiple feature vectors, which include all basic features and electrical features, and the output of its output layer is the diagnostic results of each cause. The probability distribution of the probability distribution , which are the diagnostic results of the 1st, 2nd, and Nth causes respectively.

[0037] After selecting the initial evaluation model, it is necessary to train the initial evaluation model. Specifically: First, construct a large-scale and high-quality labeled data set. Each piece of data represents a historical low-voltage event, including its complete feature vector and the true cause label confirmed by experts or verified on-site. Then, normalize the input numerical features, and then use the labeled data set to train the XGBoost model, minimizing the difference between the predicted probability and the true label using the cross-entropy loss function. Finally, evaluate the model performance through the test set. Perform hyperparameter tuning to obtain the final evaluation model.

[0038] Step S105: Input the basic features and electrical features into the final evaluation model according to the first marking result to obtain a first diagnostic result corresponding to the first marking result; It should be noted that in some embodiments, the first diagnostic result is specifically: For the low-voltage events with the first-level marking, input the basic features and electrical features related to the low-voltage event into the final evaluation model to output the cause diagnostic result with the highest probability and its first probability value, and output the second probability value corresponding to the cause diagnostic result output by the low-voltage event under the second preset rule; For the low-voltage events with the second-level marking or the third-level marking, input the basic features and electrical features related to the low-voltage event into the final evaluation model to output the cause diagnostic result with the highest probability and its third probability value.

[0039] In summary, it should be noted that although rule-based diagnostic systems are efficient, their diagnostic capabilities are often limited when dealing with new types of faults not covered by the rules, complex fault scenarios caused by multi-factor coupling, or marginal cases with atypical feature manifestations, and it is easy to produce results of "unknown cause" or "ambiguous cause". In addition, rule systems usually rely on fixed thresholds and idealized conditions and are difficult to adapt to the diversity of actual working conditions and data noise. Based on this, the present invention introduces a supplementary diagnostic process based on machine learning. Specifically, first, a suitable machine learning model is selected and designed, such as a gradient boosting decision tree (specifically using XGBoost), which can effectively handle complex interaction relationships between features, is robust to missing data, and can output feature importance to assist understanding. Secondly, the input layer of the model is designed to extract basic statistical features and all key quantitative indicators driven by electrical mechanisms, and these features provide the model with rich physical information directly related to the causes of low voltage. The output layer of the model uses the Softmax function to output the probability distribution of the current low voltage event belonging to each predefined cause (such as low voltage grid, medium voltage grid, various types of phase loss, neutral line disconnection, etc.). Then, the model is trained with a large-scale, high-quality historical low voltage event dataset containing true cause labels, and strict evaluation and tuning are carried out. In the application stage, the cause with the highest probability output by the model will be used as the supplementary diagnostic result. The effects of this are as follows: It utilizes the powerful pattern recognition and non-linear mapping capabilities of machine learning to effectively make up for the deficiencies of the rule system in dealing with complex, atypical, and unknown fault patterns; by inputting features containing rich electrical mechanisms, the model decision-making is more physically meaningful and has the potential for interpretability; it can learn deep associations from historical data that are difficult to explicitly express by rules, thereby improving the diagnostic accuracy and coverage rate of "unknown / ambiguous cause" events, providing more comprehensive information input for subsequent diagnostic result fusion, and significantly improving the intelligence level and diagnostic performance of the entire diagnostic system.

[0040] Step S106: Check and fuse the first diagnostic result and the cause diagnostic result under the fourth preset rule, and output a diagnostic report on the low voltage event according to the check and fusion result.

[0041] In the process of checking and fusing the cause diagnostic result obtained by rule recognition and the first diagnostic result obtained by machine recognition, it is first necessary to make a preliminary comparison and confirmation of the two recognition results: (1) For low voltage events marked at the first level, if the first probability is equal to the second probability, it is determined that the machine recognition result is the same as the rule recognition result. At this time, the cause diagnostic result obtained by machine recognition or rule recognition is used as the final diagnostic result, and the confidence level assessment grade is set as high; (2) If the first probability is not equal to the second probability, but the second probability value is greater than or equal to the twenty-seventh threshold , the cause diagnosis result identified by the rule is used as the final diagnosis result, and the confidence rating is set to medium; (3) If the second probability value is less than , then the first probability value is greater than or equal to the twenty-eighth threshold , then it is determined that the machine recognition result conflicts with the rule recognition result, so as to proceed to the subsequent analysis; (4) For a low voltage event with a secondary or tertiary mark, if the third probability value is greater than or equal to the 29th threshold , the cause diagnosis result corresponding to the third probability value identified by the machine is used as the final diagnosis result, and the confidence rating is set to high; (5) If the third probability value is less than , the cause is determined to be pending, and the confidence level is set to low, and manual analysis is recommended.

[0042] Then, if the machine recognition results conflict with the rule recognition results, key feature review is performed: First, obtain the first key feature related to the rule recognition result and the second key feature related to the machine recognition result, and calculate the feature deviation based on the first key feature and the second key feature. and characteristic abnormality In some embodiments, the first key feature is the core condition for triggering the rule recognition result, and the second key feature is a highly important feature relied upon for machine recognition, such as the top three features in the XGBoost model.

[0043] Specifically, if ,and , the cause diagnosis result corresponding to the third probability value identified by the machine is used as the final diagnosis result, and the confidence rating is set to high; like ,and , the cause diagnosis result identified by the rule is used as the final diagnosis result, and the confidence rating is set to high; like ,and , the cause is determined to be pending, and the confidence level is set to low. In this case, it is recommended to switch to manual analysis.

[0044] In some embodiments, the characteristic deviation is calculated according to the following formula: ; In some embodiments, the characteristic deviation is calculated according to the following formula: ; in, is the actual calculated value or measured value of the corresponding feature, is the pre-set judgment threshold for the corresponding feature, is the first key feature, is the second key feature, is the mean value of the corresponding feature, is the standard deviation of the corresponding feature.

[0045] Exemplarily, assume that in a low-voltage event, the cause diagnosis result identified by the rule is "low-voltage line grid problem" (high confidence), while the cause diagnosis result identified by the machine is confirmed as "neutral line broken" with a high probability of 0.85. At this time, after judgment, it is confirmed that there is a diagnosis conflict. The first key features include the core electrical quantities it depends on, such as "user low-voltage level" (assuming it conforms to "the end of the low-voltage line"), "voltage difference between the substation outlet voltage and the user voltage during the peak load period " (assuming the calculated value is 23.5V, and its rule threshold is >22V, then the deviation is about 6.8%), "voltage difference between the substation outlet voltage and the user voltage during the off-peak load period " (assuming the value is 10V, meeting the rule of <22V) and "user voltage during the off-peak period " (assuming the value is 205V, meeting the rule of >198V). On the other hand, through SHAP analysis, the second key features that contribute the most to the machine-identified output of neutral line broken may include: "highest phase voltage " (assuming the measured value is 265V, if the normal mean value is 221V and the standard deviation is 5V, then its is as high as 8.8), "lowest phase voltage " (assuming the measured value is 160V, if the normal mean value is 219V and the standard deviation is 4V, then its is 14.75) and " and relative deviation relationship with the nominal voltage" (showing that one phase is significantly higher by 45V and the other phase is significantly lower by 60V, conforming to the characteristics of neutral line broken). When making a quantitative comparison, although it meets the conditions, its deviation of 6.8% is relatively small; while the and of the values (8.8 and 14.75) far exceed the statistical significance threshold (such as >3), indicating that its electrical performance is extremely abnormal. Therefore, the system will judge that the feature evidence supporting the machine learning diagnosis B is stronger and more significant, thus overthrowing the rule diagnosis, adopting "neutral line broken" as the final diagnosis result, and giving a high confidence.

[0046] In addition, in some embodiments, a structured diagnostic report is output. The report clearly gives the cause of the low voltage after calibration and fusion decision, and is accompanied by a confidence level determined according to the above decision logic. The report can also include the key quantitative index values (such as feature deviation and feature abnormality, etc.) that trigger the decision and a brief description of the decision path.

[0047] In addition, to are the first threshold to the thirty-third threshold respectively. Exemplarily, to can be successively , , , , , ; The first preset range is to ; Since these thresholds and the first preset range are related to specific usage requirements, in some other embodiments of the present invention, the above thresholds and the first preset range can also be other values.

[0048] In summary, it should be noted that when combining rule-based diagnosis and machine learning diagnosis, existing low-voltage cause diagnosis methods often face the following problems: Some methods may simply adopt a voting mechanism or preset priorities to handle conflicts, lacking careful consideration of the data evidence strength in specific events. When the applicable boundaries of rules are blurred or machine learning models encounter rare operating conditions not fully covered in the training data, this simple fusion strategy is prone to misjudgment. Although some other methods attempt to introduce more complex fusion models, their decision-making processes may not be transparent enough to trace and understand. Especially when the conclusions of rules and models conflict highly, they cannot provide clear explanations and reliable confidence assessments. Based on this, the present invention proposes a verification and fusion decision-making mechanism based on key feature quantification and review. Specifically, when there is a conflict between rule-based diagnosis (e.g., diagnosing cause A) and machine learning model diagnosis (e.g., diagnosing cause B with probability P(B) ≥ 0.75), the present invention does not simply choose one of them. Instead, it first identifies the first key features supporting cause A (derived from rule definitions) and the second key features supporting cause B (derived from model explanations, such as strongly correlated features obtained through SHAP value analysis). Secondly, it quantitatively evaluates the actual measured values or calculated values of these key features in the current event. Then, by comparing these quantitative indicators, it determines which diagnostic conclusion better conforms to the actual performance of the current data. The technical effect of this mechanism is that by focusing on the quantitative performance of the core electrical features directly related to the conflicting diagnosis, it provides a transparent, interpretable, and data-evidence-based decision-making path for resolving the diagnostic conflicts between rules and machine learning models, avoiding the drawbacks of blindly trusting a certain diagnostic source or simple voting. At the same time, it can dynamically evaluate the evidence strength according to the significance degree of feature deviation, thus making a more robust judgment in complex and marginal situations, improving the accuracy and credibility of the final diagnostic result, and being able to give clear "manual intervention" suggestions for difficult events, ensuring the reliability of the diagnosis.

[0049] Compared with the prior art, the present invention has the following advantages: 1. The present invention constructs a feature engineering system driven by electrical mechanisms; designs a hierarchical location rule for low-voltage event types based on distributed state deduction; and adopts a hybrid model architecture combining rule-based pre-diagnosis and machine learning supplementary diagnosis; finally, it introduces a diagnostic result verification and conflict fusion decision-making mechanism based on key feature quantification and review, thus greatly improving the accuracy, real-time performance, and interpretability of users' low-voltage cause analysis.

[0050] 2. By deeply integrating electrical mechanism, rule reasoning, and machine learning, improve the recognition ability of various low-voltage causes (especially in complex and fuzzy scenarios); through topology relationship and distributed state deduction, accurately locate the level where the low-voltage event occurs (the end of the low-voltage line, the outlet of the distribution transformer, the end of the medium-voltage line, the outlet of the substation); by combining the explicit logic of rules and the generalization ability of machine learning, and introducing electrical mechanism-driven features, make the diagnosis process more transparent and the results more reliable; for the situation where the rule diagnosis is inconsistent with the prediction of the machine learning model, design a fusion decision-making process based on key feature quantization review to improve the reliability of the final diagnosis.

[0051] As Figure 2 shown, an embodiment of the present invention proposes a distribution network user voltage anomaly diagnosis system, and the system includes: A feature extraction module 10, configured to collect user-side electrical data and substation area electrical data when a low-voltage event occurs, and extract basic features and electrical features from the user-side electrical data and the substation area electrical data; An event type acquisition module 20, configured to acquire at least one key factor regarding the low-voltage event, and acquire an event type corresponding to the low-voltage event according to the at least one key factor under a first preset rule; A marking module 30, configured to perform cause diagnosis on the low-voltage event according to the basic features, the electrical features, and the event type under a second preset rule to obtain a cause diagnosis result, and perform a first marking on the cause diagnosis result based on a third preset rule; A model training module 40, configured to build an initial evaluation model, acquire a plurality of historical low-voltage events, perform feature extraction and cause marking on the historical low-voltage events to obtain a data set, and train the initial evaluation model according to the data set to obtain a final evaluation model; A machine recognition module 50, configured to input the basic features and the electrical features into the final evaluation model according to the first marking result to obtain a first diagnosis result corresponding to the first marking result; A diagnosis result output module 60, configured to check and fuse the first diagnosis result and the cause diagnosis result under a fourth preset rule, and output a diagnosis report regarding the low-voltage event according to the check and fusion result.

[0052] On the other hand, the present invention also proposes a storage medium, on which one or more programs are stored, and when the program is executed by a processor, the above-mentioned distribution network user voltage anomaly diagnosis method is implemented.

[0053] On the other hand, the present invention also provides an electronic device, including a memory and a processor, where the memory is used to store a computer program, and the processor is used to execute the computer program stored on the memory to implement the above-mentioned distribution network user voltage anomaly diagnosis method.

[0054] Those skilled in the art can understand that the logic and / or steps represented in the flowchart or described in other ways herein, for example, can be considered as a definite sequence list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch instructions from the instruction execution system, apparatus, or device and execute the instructions), or in combination with these instruction execution systems, apparatuses, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.

[0055] More specific examples (non-exhaustive list) of computer-readable media include the following: an electrical connection part (electronic device) having one or more wirings, a portable computer disk cartridge (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpretation, or other suitable processing as necessary, and then stored in a computer memory.

[0056] It should be understood that each part of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in the memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application-specific integrated circuits having appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

[0057] Although the embodiments of the present invention have been described in detail above, it will be apparent to those skilled in the art that various modifications and variations can be made to these embodiments. However, it should be understood that such modifications and variations are all within the scope and spirit of the present invention as described in the claims. Moreover, the present invention as described herein may have other embodiments and can be implemented or realized in various ways.

Claims

1. A method for diagnosing abnormal voltage of distribution network users, characterized in that, The method includes: When a low-voltage event occurs, collect electrical data on the user side and electrical data in the distribution area, and extract basic features and electrical features from the electrical data on the user side and the electrical data in the distribution area; Obtain at least one key factor regarding the low-voltage event, and obtain the event type corresponding to the low-voltage event according to the at least one key factor under a first preset rule; Under a second preset rule, conduct cause diagnosis on the low-voltage event according to the basic features, the electrical features, and the event type to obtain a cause diagnosis result, and perform a first marking on the cause diagnosis result based on a third preset rule; Construct an initial evaluation model, obtain multiple historical low-voltage events, perform feature extraction and cause marking on the historical low-voltage events to obtain a data set, and train the initial evaluation model according to the data set to obtain a final evaluation model; Input the basic features and electrical features into the final evaluation model according to the first marking result to obtain a first diagnosis result corresponding to the first marking result; Under a fourth preset rule, verify and fuse the first diagnosis result and the cause diagnosis result to output a diagnosis report regarding the low-voltage event according to the verification and fusion result.

2. The method for diagnosing abnormal voltage of distribution network users according to claim 1, characterized in that, The step of when a low-voltage event occurs, collecting electrical data on the user side and electrical data in the distribution area, and extracting basic features and electrical features from the electrical data on the user side and the electrical data in the distribution area includes: The electrical data on the user side includes user voltage and user current, the electrical data in the distribution area includes the outlet voltage and outlet current of the distribution area, and the basic features include the mean, variance, peak value, valley value, voltage change rate, current change rate, and autocorrelation coefficient of user voltage, user current, outlet voltage of the distribution area, and outlet current of the distribution area; Extract the phase loss feature of the distribution area according to the following formula: ； Among them, is the single-phase loss feature of the first transformer area, returns the absolute value of a number, , are the two-phase voltages with smaller values in the outlet voltage of the transformer area, is the single-phase voltage with the largest value in the outlet voltage of the transformer area, is the single-phase loss feature of the second transformer area, is the nominal voltage; Extract the low-voltage line network problem feature according to the following formula: ； Among them, is the problem feature of the first low-voltage line network, is the problem feature of the second low-voltage line network, is the peak voltage at the substation outlet during the peak load period, is the peak voltage of the user, is the valley voltage at the substation outlet during the low load period, is the valley voltage of the user; Extract the neutral line disconnection feature according to the following formula: ； Among them, is the highest phase voltage, , , are the phase A, phase B, and phase C voltages in the substation outlet voltage respectively, is the lowest phase voltage, , are the broken wire characteristics of the first and second neutral lines respectively; Calculate the acquisition anomaly feature according to the following formula: ； Among them, is the first acquisition anomaly feature, is the second acquisition anomaly feature, is the function for calculating the average value, is the function for calculating the correlation coefficient between variables, is the outlet voltage of the substation area, is the voltage of the i-th user; Extract the connection group feature of the distribution area according to the following formula: ； Among them, is the characteristic of the transformer substation connection group, is the change in the outlet voltage of the transformer substation, is the change in the outlet current of the transformer substation.

3. The method for diagnosing abnormal voltage of distribution network users according to claim 2, wherein The step of obtaining at least one key factor regarding the low-voltage event, and obtaining the event type corresponding to the low-voltage event according to the at least one key factor under a first preset rule includes: The key factors include the number of users experiencing low - voltage events in the same phase within the same distribution area , the duration of low voltage , the starting - time difference of low - voltage events , the minimum outlet voltage of the distribution area during the low - voltage event ; The first preset rule is: If and and and , it is determined that the event type is a low-voltage event at the end of the low-voltage line; If and and and , it is determined that the event type is a low-voltage event at the distribution transformer outlet; The key factors also include the number of low-voltage events at the transformer outlet under the same medium-voltage line , the starting time difference of each low-voltage event at the transformer outlet ; If and , it is determined that the event type is a low-voltage event at the end of the medium-voltage line; The key factors also include the number of low-voltage events at the end of medium-voltage lines under the outlet of the same substation , the starting time difference of low-voltage events at the end of medium-voltage lines ; If and , it is determined that the event type is a low-voltage event at the substation outlet; Among them, to are the first to eighth thresholds respectively.

4. The method for diagnosing abnormal voltage of distribution network users according to claim 3, characterized in that The step of under a second preset rule, conducting cause diagnosis on the low-voltage event according to the basic features, the electrical features, and the event type to obtain a cause diagnosis result, and performing a first marking on the cause diagnosis result based on a third preset rule includes: The second preset rule includes multiple situations, specifically: The first case: When the event type is a low-voltage event at the end of a low-voltage line, if and and , then the cause diagnosis result is: low-voltage line grid problem; The second situation: When the event type is low voltage at the end of the medium-voltage line, if load transfer and medium-voltage single-phase disconnection are excluded, the cause diagnosis result is: medium-voltage line network problem; The third case: When the event type is the low-voltage event at the distribution transformer outlet, if and and and , then it is the Yyn0 connection group. At this time, if is normal and and and , then the cause diagnosis result is: There is a problem of one-phase loss on the high-voltage side; If and / or and / or and / or , it is Dyn11 connection group. At this time, if , and there are two-phase voltages both within the first preset range, and the voltage of the other phase is less than or equal to , the cause diagnosis result is: there is a problem of open phase on the high-voltage side; The fourth case: When the event type is the low-voltage event at the distribution transformer outlet, the first condition is set as: and , or there are two-phase voltages both less than ; The second condition is: , or ; If the first condition and the second condition are both satisfied, the cause diagnosis result is: neutral line disconnection problem; The fifth situation: When the event type is low voltage at the end of the medium-voltage line, if there is a line power outage or operation record before the low-voltage event occurs, the cause diagnosis result is: load transfer; The sixth case: When the event type is a low-voltage event at the distribution outlet, if and , the cause diagnosis result is: abnormal acquisition; The seventh case: When the event type is a low-voltage event at the end of a medium-voltage line, if the phase loss problem occurs on the high-voltage side of at least two substations under the same line simultaneously, the cause diagnosis result is: single-phase disconnection of the medium-voltage line; Among them, to are the ninth to twenty-sixth thresholds respectively.

5. The method for diagnosing abnormal voltage of distribution network users according to claim 4, wherein, The step of performing a primary marking on the cause diagnosis result based on the third preset rule includes: If there is only one cause diagnosis result related to the low-voltage event, perform a first-level marking; If there is no cause diagnosis result related to the low-voltage event, perform a second-level marking; If there are at least two cause diagnosis results related to the low-voltage event, perform a third-level marking.

6. The method for diagnosing abnormal voltage of distribution network users according to claim 5, characterized in that, The initial evaluation model uses the XGBoost model. The feature vectors in its input layer include the basic features and electrical features, and the output of its output layer is the probability distribution of each cause diagnosis result.

7. The method for diagnosing abnormal voltage of distribution network users according to claim 6, characterized in that The step of inputting the basic features and electrical features into the final evaluation model according to the primary marking result to obtain the first diagnosis result corresponding to the primary marking result includes: For the low-voltage event with a first-level marking, input the basic features and electrical features related to the low-voltage event into the final evaluation model to output the cause diagnosis result with the highest probability and its first probability value, and output the second probability value corresponding to the cause diagnosis result output by the low-voltage event under the second preset rule; For the low-voltage event with a second-level or third-level marking, input the basic features and electrical features related to the low-voltage event into the final evaluation model to output the cause diagnosis result with the highest probability and its third probability value.

8. The method for diagnosing abnormal voltage of distribution network users according to claim 7, wherein The step of verifying and fusing the first diagnosis result and the cause diagnosis result under the fourth preset rule to output a diagnosis report on the low-voltage event according to the verification and fusion result includes: For the low-voltage event with a first-level marking, if the first probability is equal to the second probability, it is determined that the machine recognition result is the same as the rule recognition result. At this time, the cause diagnosis result identified by the machine or the rule is used as the final diagnosis result; If the first probability is not equal to the second probability, but the second probability value is greater than or equal to the twenty-seventh threshold , then the cause diagnosis result identified by the rule is used as the final diagnosis result; If the second probability value is less than , then the first probability value is greater than or equal to the twenty-eighth threshold , then it is determined that the machine recognition result conflicts with the rule recognition result; For a low-voltage event of a secondary tag or a tertiary tag, if the third probability value is greater than or equal to the twenty-ninth threshold , then use the cause diagnosis result corresponding to the third probability value recognized by the machine as the final diagnosis result; If the third probability value is less than , it is determined that the cause result is pending.

9. The method for diagnosing abnormal voltage of distribution network users according to claim 8, characterized in that The method further includes: For the case where the machine recognition result conflicts with the rule recognition result, obtain the first key feature related to the rule recognition result, the second key feature related to the machine recognition result, and calculate the feature deviation degree based on the first key feature and the second key feature and the feature abnormality degree ; If and , then use the cause diagnosis result corresponding to the third probability value recognized by the machine as the final diagnosis result; If and , then the cause diagnosis result identified by the rule is used as the final diagnosis result; If and , it is determined that the cause result is pending, to are the 30th threshold to the 33rd threshold respectively.

10. A distribution network user voltage anomaly diagnosis system, characterized in that, The system includes: A feature extraction module, configured to collect user-side electrical data and substation electrical data when a low-voltage event occurs, and extract basic features and electrical features from the user-side electrical data and the substation electrical data; An event type acquisition module, configured to acquire at least one key factor regarding the low-voltage event, and obtain the event type corresponding to the low-voltage event according to the at least one key factor under the first preset rule; A marking module, configured to perform cause diagnosis on the low-voltage event according to the basic features, the electrical features, and the event type under the second preset rule to obtain a cause diagnosis result, and perform a primary marking on the cause diagnosis result based on the third preset rule; A model training module, configured to construct an initial evaluation model, acquire multiple historical low-voltage events, perform feature extraction and cause marking on the historical low-voltage events to obtain a data set, and train the initial evaluation model according to the data set to obtain a final evaluation model; The machine recognition module is used to input the basic features and electrical features into the final evaluation model according to the primary marking result, and obtain a first diagnosis result corresponding to the primary marking result; The diagnosis result output module is used to check and fuse the first diagnosis result and the cause diagnosis result under the fourth preset rule, so as to output a diagnosis report on the low-voltage event according to the result of the check and fusion.

Images