Analysis system based on measurement model

Abstract

The invention discloses an analysis system based on a metering model, belongs to the field of anomaly analysis, and aims to solve the problems of monitoring equipment anomaly in real time and automatically diagnosing the equipment anomaly type. The system comprises an acquisition terminal, an electric energy meter, intelligent sensing equipment and a data management platform, the acquisition terminal is used for acquiring operation data of at least one sensing device, processing the acquired operation data and transmitting the processed operation data to the data management platform; the data management platform is used for receiving the processed operation data and carrying out anomaly detection and diagnosis on the operation data; the data management platform comprises an online detection module, an automatic diagnosis module and an abnormity recheck module; the online detection module is used for carrying out anomaly detection on the operation data received by the data management platform and outputting an abnormal equipment number; the automatic diagnosis module is used for performing automatic abnormality diagnosis on the abnormal equipment and judging the abnormality type of the equipment; and the abnormity recheck module is used for carrying out abnormity recheck diagnosis on the maintained equipment.

Description

Technical Field

[0001] The present invention relates to the field of abnormal analysis, and more specifically to an analysis system based on a metrological model. Background Art

[0002] In today's society, the power system is an indispensable infrastructure in life and industry, and its stability and reliability are crucial to daily life and economic activities. In modern society, electricity is not only the infrastructure needed by families for daily life, but also the power source of industrial production and commercial activities. From lighting to electronic devices, from enterprise production to data transmission, electricity is everywhere, and its stable supply is the basic guarantee for social operation. With the popularization of smart meters and sensor technology, the data generated by the power system has shown explosive growth. These data include consumers' electricity consumption patterns, the status information of the transmission network, and the output of distributed energy, providing unprecedented details and perspectives for the monitoring and analysis of the power system. Faced with such a large amount of data, it is a major challenge to quickly and accurately identify potential faults and anomalies. Due to the highly complex and dynamic characteristics of the power system, traditional thresholds or rules often cannot fully reflect the actual state of the system, which may lead to false alarms or missed alarms, posing risks to the safety of the power grid. In order to improve the efficiency and accuracy of anomaly detection, automated anomaly analysis systems have become a hot topic of research. Such systems can process massive amounts of data in real time, automatically identify abnormal patterns through advanced data analysis techniques and machine learning algorithms, and issue warnings in a timely manner, thereby ensuring the stable operation of the power grid.

[0003] The invention patent with publication number CN118501800A discloses a fault detection method and system based on an electric meter, which relates to the technical field of electric meter fault detection. The fault detection method based on an electric meter includes the following steps: analyzing and setting the electric meter fault warning level according to the evaluation results of the abnormality of the electric meter power parameters; analyzing and setting the electric meter fault warning level according to the evaluation results of the hidden dangers of the sudden change level of the electric meter position; analyzing, judging and setting the electric meter fault warning level according to the evaluation results of the abnormality of the magnetic parameters of the electric meter; and determining the corresponding electric meter fault warning scheme according to the electric meter fault warning level.

[0004] Although the prior art obtains the meter fault warning level through comprehensive analysis through meter power parameter detection, meter position parameter detection and meter magnetic parameter detection, and then determines the corresponding meter fault warning scheme, it has achieved the effect of improving the adaptability of the meter fault detection method to exogenous fault detection, and solved the problem of insufficient adaptability of the meter fault detection method to exogenous fault detection in the prior art. It still fails to solve the problem of real-time monitoring of equipment abnormalities and automatic diagnosis of equipment abnormality types. Therefore, in order to overcome these limitations, the present invention proposes an analysis system based on a metering model. Summary of the invention

[0005] In view of the shortcomings of the prior art, the purpose of the present invention is to provide an analysis system based on a metrology model, which solves the problem of real-time monitoring of equipment anomalies and automatic diagnosis of equipment anomaly types. The equipment operation data is collected through the acquisition terminal and transmitted to the data management platform after local processing. The online detection module performs mutation detection and comparative analysis on the operation data to find anomalies, the automated diagnosis module further determines the anomaly type, and the anomaly re-inspection module evaluates the user's on-site equipment status after the fault is eliminated.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] An analysis system based on a metering model, including a data collection terminal, an electric energy meter, an intelligent sensing device and a data management platform;

[0008] The collection terminal is used to collect operating data of at least one device, locally process the collected operating data, and transmit the processed operating data to the data management platform;

[0009] The data management platform is used to receive the processed operation data and perform abnormality detection and diagnosis on the operation data; the data management platform includes an online detection module, an automatic diagnosis module and an abnormality re-inspection module;

[0010] The online detection module is used to perform anomaly detection on the operation data received by the data management platform and output the device number with the anomaly;

[0011] The automatic diagnosis module is used to perform automatic abnormality diagnosis on the equipment with abnormality and determine the type of equipment abnormality;

[0012] The abnormality recheck module is used to perform abnormality recheck and diagnosis on the equipment according to the identified equipment abnormality type after the on-site fault elimination is completed.

[0013] Specifically, the data management platform performs anomaly detection on the received operation data, including:

[0014] Build time series data through equipment operation data, perform mutation detection on the operation data, and output the equipment number and user number with mutation in the mutation detection;

[0015] Construct a device comparison group for devices with mutations, conduct comparative analysis, and output the device numbers with differences in the comparative analysis;

[0016] The mutation point is scored abnormally based on the mutation amplitude and the degree of difference between the comparison groups, and the abnormal device number is output according to the abnormal score.

[0017] Specifically, the mutation detection includes:

[0018] For time series data A, configure a window size ω; for the jth window, calculate the angle θ between all adjacent data points in the window j,j+1 , the calculation formula is:

[0019]

[0020] in, A j+1 is the j+1th data point, A j is the jth data point;

[0021] Calculate the average value of the angles in the window to obtain a value representing the overall change trend in the window as the mutation value. The calculation formula is:

[0022]

[0023] in, is the average value of the angle of the i-th window, i.e., the mutation value;

[0024] Configure mutation threshold T θ , when the average angle in the window exceeds the mutation threshold, it is determined that a mutation has occurred at the center point of the window, that is:

[0025]

[0026] When D(i) takes the value of 1, the i-th point is determined to be a mutation point, and when D(i) takes the value of 0, the i-th point is determined to be a non-mutation point.

[0027] Specifically, the comparative analysis includes:

[0028] Configure the time interval T to extract the mutation point Q from the mutation device Q si The data series is extracted from the comparison equipment group according to the timestamp of the mutation point. in Represents the mth device in the comparison device group, with timestamp s i The data sequence is centered and T is the time interval. For each data sequence, de-radix standardization is performed, the mean of each data sequence is calculated, and the mean is subtracted from each data point. The calculation formula is as follows:

[0029]

[0030]

[0031] in, is the mutation device data sequence after base-denormalization and standardization. It is the data series of the comparison equipment group after de-base standardization;

[0032] Calculate the similarity between the data sequences of the mutant device and the comparison device group. The calculation formula is as follows:

[0033]

[0034] in, is the similarity between the data sequences of the mutant device and the comparison device group, and the average similarity between all comparison device groups and the mutant device is calculated As a comprehensive similarity;

[0035] Configure the difference threshold T ρ , when the average similarity of the mutation device Less than the difference threshold T ρ , then there is a difference in the mutation device, and the difference device number is output.

[0036] Specifically, the abnormality score includes:

[0037] For the difference device W, obtain the mutation values ​​of all mutation points in the difference device And obtain the comprehensive similarity of all mutation points in the difference device According to the number of mutation points in the difference device, the mutation value and comprehensive similarity of the mutation point, the anomaly score is calculated. The calculation formula is as follows:

[0038]

[0039] Where S is the anomaly score and the anomaly threshold T is configured. S , if the anomaly score S is greater than the anomaly threshold T S , then the difference device is determined to be an abnormal device, and the abnormal devices are sorted according to the abnormality score, so that the devices with high scores are ranked first, and the abnormal device number list is output;

[0040] Specifically, the automated abnormality diagnosis and determination of the device abnormality type specifically includes:

[0041] Obtain the operation data of the equipment mutation points in the abnormal equipment number list and the operation data of the historical abnormal equipment mutation points, and perform preprocessing and feature extraction;

[0042] Establish a preliminary prediction model, and output the probability value of the preliminary classification abnormality type according to the preliminary prediction model;

[0043] Configure the probability threshold T p , select the anomaly type whose probability value of the preliminary classification is greater than the probability threshold, perform sub-classification based on the measurement anomaly model in turn, and output the probability value of the sub-classified anomaly type;

[0044] The probability values ​​of the preliminary classification and the detailed classification are combined, and the probability value of each detailed classification in the preliminary classification to which it belongs is multiplied by the probability value of the detailed classification to obtain the comprehensive probability value of the detailed classification. The comprehensive probability values ​​of all detailed classifications in the preliminary classification are summed to obtain the comprehensive probability value of the preliminary classification. The abnormal types are sorted in turn according to the comprehensive probability values ​​of the preliminary classification and the comprehensive probability values ​​of the detailed classifications, and the abnormal type sequence of the abnormal equipment is output.

[0045] Specifically, the preliminary prediction model obtains the probability value of the preliminary classification abnormality type specifically including:

[0046] Mark the operating data of historical abnormal equipment by abnormal classification and divide them into training set and test set;

[0047] Select the gradient boosting tree as the preliminary prediction model and perform model training. In each round of training, one tree is trained for each anomaly type, and a total of K trees are trained. For each running data sample, the output calculated by all trees is represented as a score vector Z, where each element represents the output score of the corresponding category;

[0048] Convert the score vector to a probability distribution. For each category K, the probability of being predicted as that category is:

[0049]

[0050] in, is the probability value of the i-th category, z i is the original output of the i-th category, and K is the total number of categories.

[0051] Specifically, the measurement anomaly model obtains the probability value of the sub-classification specifically including:

[0052] Configure the cut-off threshold T t , if the timestamp of the abnormal device mutation point is T a , then get the abnormal device at the timestamp [T a -vT t ,T a -uT t ], where v is greater than u, and a measurement anomaly model is established for the timestamp [T a -vT t ,T a -uT t ]’s running data are fitted;

[0053] The timestamp predicted by the measurement anomaly model is [T a -uT t ,T a +uT t] and calculate the deviation between the predicted value of the running data and the actual value of the running data, and derive the probability value of the sub-classified abnormal type according to the deviation type.

[0054] Specifically, the establishment of the measurement anomaly model includes:

[0055] Define similar day rules and define similar day adjustment factor SDF, namely:

[0056]

[0057] Among them, D sim(t) is a set of similar days selected according to the environmental conditions of t, y d is the electricity consumption on similar days in history, is the average electricity consumption on similar days, m t is the number of similar days selected;

[0058] Integrate SDF into the Prophet model to establish a measurement anomaly model, namely:

[0059]

[0060] Among them, g(t) is the trend term, s(t) is the seasonal term, h(t) is the holiday term, ∈ t is the error term.

[0061] Specifically, the calculation of the probability value of the sub-classified abnormal type specifically includes:

[0062] Extract deviation features, identify deviation patterns, classify deviations according to the deviation model, and mark training data set sub-classification labels according to deviation types; use labeled data sets to train probability models, and use probability models to output probability values ​​of deviation sub-classifications of actual values ​​of running data.

[0063] Specifically, the abnormal re-examination diagnosis specifically includes:

[0064] Collect the operation data of abnormal user equipment after the fault is eliminated, and conduct re-inspection and analysis through the online detection module and automatic diagnosis module;

[0065] If the re-inspection and analysis passes, the knowledge base will be improved to record the equipment abnormality type and maintenance experience; if the re-inspection and analysis fails, the update process will be triggered, including updating the preliminary prediction model and updating the detailed classification model, and repeating the maintenance and re-inspection analysis until the equipment returns to normal, and the maintenance process is optimized regularly.

[0066] Beneficial effects of the present invention:

[0067] 1. The online detection module monitors the operation status of the equipment in real time by constructing time series data and applying mutation detection algorithms to quickly identify potential abnormal changes. This ensures that problems can be discovered in a timely manner and prevents small problems from turning into major failures. By comparing and analyzing the differences in the operation data of similar equipment and scoring anomalies based on the magnitude and degree of mutation, it can accurately identify equipment with abnormalities. This improves the accuracy of anomaly detection and reduces the possibility of false alarms.

[0068] 2. After the abnormal equipment is identified, the automated diagnosis module automatically conducts an in-depth analysis of the cause of the fault. The prediction model is established using historical and real-time data to output the probability values ​​of various types of abnormalities, and then the sub-classification diagnosis is performed based on the probability values. This not only improves the speed of fault handling, but also improves the accuracy of diagnosis. By using a combination of multiple algorithms to identify specific abnormal causes and output sub-classified probability values, the automated diagnosis module provides decision support for maintenance personnel to locate problems faster and perform effective repairs. This helps reduce equipment maintenance time and improve the overall reliability of the equipment.

[0069] 3. The abnormal re-inspection module focuses on the evaluation of the equipment status after the fault is eliminated to ensure that the equipment has returned to normal. This process is achieved by collecting the equipment operation data again and comparing it with the data before maintenance. Such a re-inspection process helps to verify the effectiveness of maintenance activities and ensure that the problem is completely solved. If the re-inspection fails, the update process will be triggered, including updating the prediction model and sub-classification model, and repeating the maintenance and re-inspection analysis until the equipment returns to normal operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 It is a schematic diagram of the structure of an analysis system based on a measurement model;

[0071] Figure 2 Flowchart for anomaly detection;

[0072] Figure 3 A schematic diagram for numbering the equipment;

[0073] Figure 4 Flowchart for abnormal diagnosis;

[0074] Figure 5 Flowchart for abnormal retest diagnosis. DETAILED DESCRIPTION

[0075] See also Figure 1 ,This embodiment introduces an analysis system based on a metrology model, including a collection terminal and a data management platform;

[0076] The collection terminal is used to collect operating data of at least one device, locally process the collected operating data, and transmit the processed operating data to the data management platform;

[0077] The data management platform is used to receive the processed operation data and perform abnormality detection and diagnosis on the operation data; the data management platform includes an online detection module, an automatic diagnosis module and an abnormality re-inspection module;

[0078] The online detection module is used to perform anomaly detection on the operation data received by the data management platform and output the device number with the anomaly;

[0079] The automatic diagnosis module is used to perform automatic abnormality diagnosis on the equipment with abnormalities and determine the type of equipment abnormality.

[0080] The abnormality recheck module is used to perform abnormality recheck diagnosis on the equipment according to the identified equipment abnormality type after the equipment maintenance is completed.

[0081] In this embodiment, the operation data of the smart meter is collected through the collection terminal. The operation data includes current, voltage, power, etc. The collection terminal performs standardized processing on the operation data and sends the processed data to the data management platform through a secure communication network. The data management platform integrates an online detection module, an automated diagnosis module, and an abnormal re-inspection module to deeply analyze the processed operation data and realize abnormal detection and diagnosis. The online detection module and the automated diagnosis module continuously receive data transmitted by the collection terminal and perform real-time monitoring. Statistical methods are used to identify abnormal patterns in the data, such as mutation detection and time series analysis. Once an abnormality is detected, the abnormal data will be marked and the automated diagnosis module will be notified. The automated diagnosis module classifies according to the type of abnormality, such as overload, short circuit, voltage abnormality, etc. The possible causes of the abnormality are analyzed using historical maintenance records, equipment parameters and other data. After the maintenance is completed, the re-inspection process is automatically or manually triggered. The operation data of the equipment after maintenance is collected and re-checked and analyzed through the online detection module and the automatic diagnosis module to evaluate the maintenance effect and decide whether further maintenance is needed; if the re-check analysis passes, the knowledge base is improved to record the equipment abnormality type and maintenance experience; if the re-check analysis fails, the update process is triggered, including updating the preliminary prediction model and updating the detailed classification model, recording the new failure mode and maintenance experience. Through efficient data collection, accurate anomaly detection and diagnosis, and meticulous anomaly re-check, the operating reliability and maintenance efficiency of power equipment have been significantly improved. Through continuous data accumulation and model optimization, it can adapt to the ever-changing power environment and equipment conditions, and provide strong technical support for the stable operation of the power system.

[0082] The locally processing the collected operation data includes converting the collected operation data into standard units and marking timestamps and device numbers;

[0083] See also Figure 2, the data management platform performs abnormality detection on the received operation data specifically including:

[0084] Build time series data through equipment operation data, perform mutation detection on the operation data, and output the equipment number with mutation in the mutation detection;

[0085] Construct a device comparison group for devices with mutations, conduct comparative analysis, and output the device numbers with differences in the comparative analysis;

[0086] The mutation point is scored abnormally based on the mutation amplitude and the degree of difference between the comparison groups, and the abnormal device number is output according to the abnormal score.

[0087] In this example, see Figure 3 , filter out the mutation device numbers with mutations from the device numbers, filter out the difference device numbers from the mutation device numbers, and finally filter out the abnormal device numbers from the difference device numbers; the specific process is as follows: according to the device number, arrange the collected operation data, such as current, voltage, etc., in chronological order to construct time series data. Select a mutation detection algorithm to identify the mutation points in the data based on the sliding window of the time series. Apply the selected algorithm to analyze the time series data, identify the device numbers with mutations, and record the time point when the mutation occurs. For each device with mutation, select similar devices, such as devices of the same region and the same model, to construct a device comparison group. And synchronize the data of the device comparison group to ensure that the data of the devices in the comparison group are synchronized in time to facilitate comparative analysis. Calculate the difference in operation data between the mutation device and other devices in its comparison group in the same time period. Determine whether the difference is significant based on the set threshold, and output the device number with significant difference. Quantify the mutation amplitude of each mutation point, such as calculating the absolute value or relative percentage of the current mutation. And quantify the degree of difference in the comparison group, such as calculating the standard deviation of the mean. Combine the mutation amplitude and the degree of difference to calculate the abnormality score. All devices marked as abnormal are sorted according to the abnormality score. The higher the score, the higher the abnormality priority of the device.

[0088] Specifically, the mutation detection includes:

[0089] For time series data A=[A1,A2,A3,...,A n ], configure a window size ω, ω is a positive odd number greater than 1, the window size is used to measure the number of data points used each time for mutation detection; the window size determines the time range of the analysis. Larger windows can provide a wider range of time series context, which helps to identify long-term trends and continuous change patterns, but may reduce sensitivity to short-term changes. On the contrary, smaller windows can capture the details of rapid changes, but may be overly sensitive to noise, resulting in false positive mutations;

[0090] For the jth window, when When , the data points in the window are when When the data point in the window is [A j ,A j+1 ], calculate the angle θ between all adjacent data points in the window j,j+1 , the calculation formula is:

[0091]

[0092] in, By considering the angles between all adjacent data points in the window, the relative changes between data points are quantified to provide richer and more stable information, which helps to identify complex change patterns;

[0093] Calculate the average value of the angles in the window to obtain a value representing the overall change trend in the window as the mutation value. The calculation formula is:

[0094]

[0095] in, is the average value of the angles in the i-th window, i.e., the mutation value; by averaging the angles in the window, a single indicator representing the trend of data changes in the entire window period is obtained, which helps to smooth short-term fluctuations, making it easier to identify real trend changes or abnormal changes;

[0096] Configure mutation threshold T θ , when the average angle in the window exceeds the mutation threshold, it is determined that a mutation has occurred at the center point of the window, that is:

[0097]

[0098] Among them, when D(i) takes 1, the i-th point is determined to be a mutation point, and when D(i) takes 0, the i-th point is determined to be a non-mutation point. The mutation threshold needs to be adjusted according to the specific application scenario and data characteristics. An appropriate threshold can avoid frequent false alarms caused by data noise while maintaining high sensitivity. When the average angle in the window exceeds this threshold, it is considered that a significant change has occurred at the center point of the window, which is a mutation point.

[0099] Specifically, the comparative analysis includes:

[0100] Configure the time interval T to extract the mutation points from the mutation device Q The data sequence is And according to the timestamp of the mutation point, extract the data points with the same timestamp from the comparison equipment group, that is, in Represents the mth device in the comparison device group, with timestamp s i The data sequence is centered and T is the time interval. The time interval T ensures that enough data points can be captured to observe and analyze the changing trend of the data.

[0101] For each data series, debase normalization is performed, the mean of each data series is calculated, and the mean is subtracted from each data point. The calculation formula is as follows:

[0102]

[0103]

[0104] in, is the mutation device data sequence after base-denormalization and standardization. It is the data series of the comparison equipment group after de-basement and standardization; it eliminates the baseline effect in the data, makes the change trend more obvious, strengthens the change information in the data, and reduces the impact of external factors;

[0105] Calculate the similarity between the data sequences of the mutant device and the comparison device group. The calculation formula is as follows:

[0106]

[0107] in, is the similarity between the data sequences of the mutant device and the comparison device group, and the average similarity between all comparison device groups and the mutant device is calculated As a comprehensive similarity; the comprehensive similarity provides an overall assessment of how similar the behavior of the mutant device is to the comparison device group, providing a quantitative similarity score that helps with subsequent data analysis and decision-making;

[0108] Configure the difference threshold T ρ , when the average similarity of the mutation device Less than the difference threshold T ρ , then there is a difference in the mutation device, and the difference device number is output. Being able to identify potential abnormal devices in a timely manner is crucial to maintaining the stability and reliability of the system.

[0109] Specifically, the abnormality score includes:

[0110] For the difference device W, obtain the mutation values ​​of all mutation points in the difference device And obtain the comprehensive similarity of all mutation points in the difference device According to the number of mutation points in the difference device, the mutation value and comprehensive similarity of the mutation point, the anomaly score is calculated. The calculation formula is as follows:

[0111]

[0112] Where S is the anomaly score and the anomaly threshold T is configured. S , if the anomaly score S is greater than the anomaly threshold T S , the difference device is determined to be an abnormal device, and the abnormal devices are sorted according to the abnormality score, so that the devices with high scores are ranked first, and a list of abnormal device numbers is output; in order to more effectively identify and evaluate the abnormal status of the device and take corresponding measures.

[0113] See also Figure 4 The automatic abnormal diagnosis and judgment of the abnormal type of the equipment specifically includes:

[0114] Obtain the operation data of the equipment mutation points in the abnormal equipment number list and the operation data of the historical abnormal equipment mutation points, and perform preprocessing and feature extraction;

[0115] Establish a preliminary prediction model, and output the probability value of the preliminary classification abnormality type according to the preliminary prediction model;

[0116] Configure the probability threshold T p , select the anomaly type whose probability value of the preliminary classification is greater than the probability threshold, perform sub-classification based on the measurement anomaly model in turn, and output the probability value of the sub-classified anomaly type;

[0117] The probability values ​​of the preliminary classification and the detailed classification are combined, and the probability value of each detailed classification in the preliminary classification to which it belongs is multiplied by the probability value of the detailed classification to obtain the comprehensive probability value of the detailed classification. The comprehensive probability values ​​of all detailed classifications in the preliminary classification are summed to obtain the comprehensive probability value of the preliminary classification. The abnormal types are sorted in turn according to the comprehensive probability values ​​of the preliminary classification and the comprehensive probability values ​​of the detailed classifications, and the abnormal type sequence of the abnormal equipment is output.

[0118] In this embodiment, the real-time operation data of the equipment in the abnormal equipment number list is obtained, including indicators such as current, voltage, and power factor. The operation data of the abnormal equipment in history is extracted as a reference. The collected data is cleaned, the noise is removed, and normalized for subsequent feature extraction. Features that are helpful in judging the type of abnormality are extracted from the data, such as the current change rate at the mutation point, voltage stability, power factor deviation, etc. The most influential features are selected using a machine learning algorithm to establish a preliminary prediction model to identify major abnormalities such as overload, voltage abnormality, and current abnormality. The model outputs the probability value of each major abnormality, for example, 60% overload, 30% voltage abnormality, and 10% current abnormality. A threshold is set, such as 0.5, and the major abnormalities exceeding this threshold are selected for subclassification. For the selected major abnormalities, the metering abnormality model is used for subclassification to identify the specific abnormal causes, such as the overload cause is subdivided into equipment aging, sudden increase in power equipment, etc. The probability value of each subclassification in the major category to which it belongs is multiplied by the probability value of the subclassification to obtain the comprehensive probability value of each subclassification. Sum the comprehensive probability values ​​of all subcategories within each major category to obtain the comprehensive probability value of the major category. Sort the possible abnormal types according to the calculated preliminary classification and the comprehensive probability values ​​of the subcategories. Output the abnormal type sequence of the abnormal equipment and its corresponding comprehensive probability value for reference by maintenance personnel. This can automatically and accurately diagnose the abnormal equipment and arrange the possible abnormal causes in order of priority. This method not only improves the speed and accuracy of fault handling, but also provides strong support for the stable operation of the equipment.

[0119] The preliminary prediction model obtains the probability value of the preliminary classification abnormality type specifically including:

[0120] The operation data of historical abnormal equipment is marked for abnormal classification and divided into training set and test set to ensure that each abnormal type has a corresponding label, allowing the model to learn to distinguish different abnormal states to ensure that each abnormal type has a corresponding label;

[0121] The gradient boosting tree is selected as the preliminary prediction model for model training. In each round of training, one tree is trained for each anomaly type, and a total of K trees are trained. For each running data sample, the output calculated by all trees is expressed as a score vector, that is, Z = (z1, z2, z3, ..., z K ), where each element represents the output score of the corresponding category; this ensures that the model can perform specialized learning for each anomaly type, improving the performance of the model in multi-category classification problems. The score vector contains the model's prediction information for each category and is the basis for subsequent probability conversion;

[0122] Convert the score vector to a probability distribution. For each category K, the probability of being predicted as that category is:

[0123]

[0124] in, is the probability value of the i-th category, z i is the original output of the i-th category, and K is the total number of categories; the score vector is converted into a probability distribution, which ensures that the sum of the probabilities of all categories is 1, and the probability value of each category is between 0 and 1, providing an intuitive and reasonable probability basis for the final classification decision;

[0125] The cross entropy loss function is used during training to measure the gap between the predicted probability and the true label. The cross entropy loss function is:

[0126]

[0127] Among them, y k It is to run the data sample for the true label of the kth category; to effectively measure the closeness of the model output probability to the true label, and guide the training optimization of the model.

[0128] The probability value of obtaining the sub-classification by the measurement anomaly model specifically includes:

[0129] Configure the cut-off threshold T t , if the timestamp of the abnormal device mutation point is T a , then get the abnormal device at the timestamp [T a -vT t ,T a -uT t ], where v is greater than u and both are normal numbers, and a measurement anomaly model is established for the timestamp [T a -vT t ,T a -uT t ]’s running data are fitted;

[0130] The timestamp predicted by the measurement anomaly model is [T a -uT t ,T a +uT t ] and calculate the deviation between the predicted value of the running data and the actual value of the running data, and derive the probability value of the sub-classified abnormal type according to the deviation type;

[0131] The establishment of the measurement anomaly model includes:

[0132] Define similar day rules, including weather conditions, working days, and rest days; define similar day adjustment factor SDF, namely:

[0133]

[0134] Among them, D sim(t) is a set of similar days selected according to the environmental conditions of t, y d is the electricity consumption on similar days in history, is the average electricity consumption on similar days, m t is the number of similar days selected; by introducing the similar day adjustment factor SDF, the electricity consumption data under similar conditions in history can be used to adjust the current forecast. It is particularly effective in dealing with changes in electricity consumption patterns caused by special weather or events, thereby significantly improving the prediction accuracy of the model when facing abnormal or unstable factors. For example, if the temperature and humidity on a certain day are very close to those on certain days in historical data, and the electricity consumption on these days has increased significantly, then the model will adjust the prediction value accordingly to more accurately reflect the actual electricity demand;

[0135] Integrate SDF into the Prophet model to establish a measurement anomaly model, namely:

[0136]

[0137] Among them, g(t) is the trend term, s(t) is the seasonal term, h(t) is the holiday term, ∈ t is an error term, the trend term can be described by a logistic regression growth model, the seasonal term can be represented by a trigonometric function, and the holiday term can be represented by an indicator function; the introduction of the similar day method increases the model's adaptability to different environmental conditions. It can not only learn the periodicity and trend changes of time series, but also make real-time adjustments based on specific environmental conditions, such as weather and holidays, so that the model can maintain good prediction performance in various complex situations. The improvement in adaptability greatly enhances the model's flexibility in dealing with seasonal changes or emergencies, and reduces the prediction errors caused by the model being too rigid.

[0138] The calculation of the probability value of the sub-classified abnormal type specifically includes:

[0139] Extract deviation features and identify deviation patterns, including deviation size, deviation direction, and deviation periodicity. Classify deviations according to deviation models, including instantaneous spikes, persistent offsets, and periodic deviations. Label training data sets with sub-classification labels according to deviation types. Use labeled data sets to train probability models, including Gaussian mixture models, naive Bayes, random forests, etc. Output probability values ​​of deviation sub-classifications of actual values ​​of running data through probability models.

[0140] See also Figure 5 The abnormal re-examination diagnosis specifically includes:

[0141] Collect the operating data of abnormal equipment after maintenance, and conduct re-inspection and analysis through the online detection module and automatic diagnosis module;

[0142] If the re-inspection and analysis passes, the knowledge base will be improved to record the equipment abnormality type and maintenance experience; if the re-inspection and analysis fails, the update process will be triggered, including updating the preliminary prediction model and updating the detailed classification model, and repeating the maintenance and re-inspection analysis until the equipment returns to normal, and the maintenance process is optimized regularly.

[0143] The re-examination and analysis process includes:

[0144] After equipment maintenance, collect and preprocess the operating data to ensure data quality and consistency. Extract key data features from the preprocessed operating data for subsequent re-inspection and analysis;

[0145] Use the same preliminary prediction model as before maintenance to perform anomaly detection on the processed data and generate a preliminary classification probability value;

[0146] Set a re-inspection threshold. If all preliminary classification probability values ​​are lower than this threshold, the re-inspection is considered to have passed. If any preliminary classification probability value is higher than this threshold, the re-inspection has failed and further analysis is required. Setting the re-inspection threshold reduces unnecessary maintenance costs while ensuring the safety and stability of equipment operation.

[0147] The update process includes:

[0148] If the equipment fails the re-inspection, it means that the anomaly has not been completely repaired or a new anomaly type has appeared, and it is necessary to re-enter the anomaly analysis process. For data that fails the re-inspection, continue to use the sub-classification model to classify the anomaly, output the probability value of the sub-classification, and obtain the anomaly classification list;

[0149] Perform maintenance based on the abnormal classification list, record new failure modes and maintenance experience, update the preliminary prediction model and update the detailed classification model; continuous updating of the model ensures that the system continues to improve over time and improves long-term operation and maintenance efficiency;

[0150] Continue to re-inspect and diagnose the equipment after maintenance until the equipment resumes normal operation; the cyclic inspection and diagnosis process ensures the long-term stable operation of the equipment and reduces the risk of sudden downtime.

[0151] Working principle and its effect:

[0152] An analysis system based on a metrology model includes a collection terminal and a data management platform, wherein the collection terminal is responsible for collecting equipment operation data and transmitting it to the data management platform after local processing. The data management platform includes an online detection module, an automatic diagnosis module, and an abnormal re-inspection module, which are used to receive the processed operation data for abnormality detection and diagnosis. The abnormal equipment is identified by constructing time series data, mutation detection, comparative analysis, and abnormality scoring. The automatic diagnosis module further classifies the abnormal equipment and determines the abnormal type of the equipment. The equipment after maintenance will go through a re-inspection process to evaluate the maintenance effect and determine whether further maintenance is required. It also supports updating the prediction model based on the data after maintenance and the re-inspection results, and improves the performance and reliability of the entire system through continuous optimization. Overall, the system can effectively monitor the operation status of the equipment, diagnose abnormalities in a timely manner, guide maintenance work, and continuously optimize the maintenance process and improve system performance through data analysis.

[0153] The acquisition terminal is responsible for collecting and preprocessing the equipment operation data, such as current and voltage. The data is then transmitted to the data management platform, which is equipped with an online detection module, an automated diagnosis module, and an abnormal recheck module. The online detection module first constructs time series data by analyzing the equipment operation data. A mutation detection algorithm is used to identify mutation points in the data, indicating that the equipment behavior has unexpectedly changed. In order to identify the equipment that does have anomalies, a comparative analysis is performed, that is, similar equipment is selected and the difference in operation data between them is calculated. Based on the mutation amplitude and the degree of difference in the comparison group, the online detection module will score the mutation points for anomalies and output the device number that may have anomalies. Once the equipment is detected to be abnormal, the automated diagnosis module will intervene to determine the specific type of anomaly. The historical and real-time operation data of the abnormal equipment are preprocessed and feature extracted, and then a preliminary prediction model is established. The probability values ​​of various types of anomalies are output, and based on these probability values, the anomaly types for further refined diagnosis are selected. Refined diagnosis is performed through the measurement anomaly model, which uses a combination of multiple algorithms to identify the specific cause of the anomaly and output the probability values ​​of the sub-classification. Finally, the automated diagnosis module will sort the possible anomaly types based on these probability values ​​to provide decision support for maintenance personnel. The anomaly recheck module focuses on evaluating the equipment status after maintenance. It will collect new operating data for recheck analysis after the equipment maintenance is completed and put back into use. This process is designed to evaluate the maintenance effect and ensure that the equipment has returned to normal. If the recheck fails, that is, the equipment still exhibits anomalies or new anomaly types appear, the system will trigger an update process. The update process includes updating the preliminary prediction model and the detailed classification model, and repeating the maintenance and recheck analysis until the equipment returns to normal operation.

[0154] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. An analysis system based on a metrological model, characterized in that: Including collection terminal and data management platform; The collection terminal is used to collect operating data of at least one device, locally process the collected operating data, and transmit the processed operating data to the data management platform; The data management platform is used to receive the processed operation data and perform abnormality detection and diagnosis on the operation data; The data management platform includes an online detection module, an automatic diagnosis module and an abnormality recheck module; The online detection module is used to perform anomaly detection on the operation data received by the data management platform and output the device number with the anomaly; The automatic diagnosis module is used to perform automatic abnormality diagnosis on the equipment with abnormality and determine the type of equipment abnormality; The abnormality recheck module is used to perform abnormality recheck diagnosis on the equipment according to the identified equipment abnormality type after the equipment maintenance is completed.

2. The analysis system based on the measurement anomaly model according to claim 1, characterized in that: The data management platform performs abnormality detection on the received operation data specifically including: Build time series data through equipment operation data, perform mutation detection on the operation data, and output the equipment number with mutation in the mutation detection; Construct a device comparison group for devices with mutations, conduct comparative analysis, and output the device numbers with differences in the comparative analysis; The mutation point is scored abnormally based on the mutation amplitude and the degree of difference between the comparison groups, and the abnormal device number is output according to the abnormal score.

3. The analysis system based on the measurement model as claimed in claim 2, characterized in that: The mutation detection comprises: For time series data A, configure a window size ω; for the jth window, calculate the angle θ between all adjacent data points in the window j,j+1 , the calculation formula is: in, A j+1 is the j+1th data point, A j is the jth data point; Calculate the average value of the angles in the window to obtain a value representing the overall change trend in the window as the mutation value. The calculation formula is: in, is the average value of the angle of the i-th window, i.e., the mutation value; Configure mutation threshold T θ , when the average angle in the window exceeds the mutation threshold, it is determined that a mutation has occurred at the center point of the window, that is: When D(i) takes the value of 1, the i-th point is determined to be a mutation point, and when D(i) takes the value of 0, the i-th point is determined to be a non-mutation point.

4. The analysis system based on the measurement model according to claim 2, characterized in that: The comparative analysis includes: Configure the time interval T to extract the mutation point Q from the mutation device Q si The data series is extracted from the comparison equipment group according to the timestamp of the mutation point. in Represents the mth device in the comparison device group, with timestamp s i The data sequence is centered and T is the time interval. For each data sequence, de-radix standardization is performed, the mean of each data sequence is calculated, and the mean is subtracted from each data point. The calculation formula is as follows: in, is the mutation device data sequence after base-denormalization and standardization. It is the data series of the comparison equipment group after base-removal and standardization; Calculate the similarity between the data sequences of the mutant device and the comparison device group. The calculation formula is as follows: in, is the similarity between the data sequences of the mutant device and the comparison device group, and the average similarity between all comparison device groups and the mutant device is calculated As a comprehensive similarity; Configure the difference threshold T ρ , when the average similarity of the mutated device Less than the difference threshold T ρ , then there is a difference in the mutation device, and the difference device number is output.

5. The analysis system based on the measurement model according to claim 2, characterized in that: The abnormality score includes: For the difference device W, obtain the mutation values ​​of all mutation points in the difference device And obtain the comprehensive similarity of all mutation points in the difference device According to the number of mutation points in the difference device, the mutation value and comprehensive similarity of the mutation point, the anomaly score is calculated. The calculation formula is as follows: Where S is the anomaly score and the anomaly threshold T is configured. S , if the anomaly score S is greater than the anomaly threshold T S , then the difference device is determined to be an abnormal device, and the abnormal devices are sorted according to the abnormality score, so that the devices with high scores are ranked in front, and the abnormal device number list is output.

6. The analysis system based on the measurement model according to claim 1, characterized in that: The automatic abnormal diagnosis and judgment of the abnormal type of the equipment specifically includes: Obtain the operation data of the equipment mutation points in the abnormal equipment number list and the operation data of the historical abnormal equipment mutation points, and perform preprocessing and feature extraction; Establish a preliminary prediction model, and output the probability value of the preliminary classification abnormality type according to the preliminary prediction model; Configure the probability threshold T p , select the anomaly type whose probability value of the preliminary classification is greater than the probability threshold, perform sub-classification based on the measurement anomaly model in turn, and output the probability value of the sub-classified anomaly type; The probability values ​​of the preliminary classification and the detailed classification are combined, and the probability value of each detailed classification in the preliminary classification to which it belongs is multiplied by the probability value of the detailed classification to obtain the comprehensive probability value of the detailed classification. The comprehensive probability values ​​of all detailed classifications in the preliminary classification are summed to obtain the comprehensive probability value of the preliminary classification. The abnormal types are sorted in turn according to the comprehensive probability values ​​of the preliminary classification and the comprehensive probability values ​​of the detailed classifications, and the abnormal type sequence of the abnormal equipment is output.

7. The analysis system based on the metrological model according to claim 6, characterized in that: The preliminary prediction model obtains the probability value of the preliminary classification abnormality type specifically including: Mark the operating data of historical abnormal equipment by abnormal classification and divide them into training set and test set; Select the gradient boosting tree as the preliminary prediction model and perform model training. In each round of training, one tree is trained for each anomaly type, and a total of K trees are trained. For each running data sample, the output calculated by all trees is represented as a score vector Z, where each element represents the output score of the corresponding category; Convert the score vector to a probability distribution. For each category K, the probability of being predicted as that category is: in, is the probability value of the i-th category, z i is the original output of the ith category, and K is the total number of categories.

8. The analysis system based on the measurement model according to claim 6, characterized in that: The probability value of obtaining the sub-classification by the measurement anomaly model specifically includes: Configure the cut-off threshold T t , if the timestamp of the abnormal device mutation point is T a , then get the abnormal device at the timestamp [T a -vT t ,T a -uT t ], where v is greater than u, and a measurement anomaly model is established for the timestamp [T a -vT t ,T a -uT t ]’s running data are fitted; The timestamp predicted by the measurement anomaly model is [T a -uT t ,T a +uT t ] and calculate the deviation between the predicted value of the running data and the actual value of the running data, and derive the probability value of the sub-classified abnormal type according to the deviation type.

9. The analysis system based on the metrology model according to claim 8, characterized in that: The establishment of the measurement anomaly model includes: Define similar day rules and define similar day adjustment factor SDF, namely: Among them, D sim(t) is a set of similar days selected according to the environmental conditions of t, y d is the electricity consumption on similar days in history, is the average electricity consumption on similar days, m t is the number of similar days selected; Integrate SDF into the Prophet model to establish a measurement anomaly model, namely: Among them, g(t) is the trend term, s(t) is the seasonal term, h(t) is the holiday term, ∈ t is the error term.

10. The analysis system based on the measurement model according to claim 8, characterized in that: The calculation of the probability value of the sub-classified abnormal type specifically includes: Extract deviation features, identify deviation patterns, classify deviations according to the deviation model, and mark training data set sub-classification labels according to deviation types; use labeled data sets to train probability models, and use probability models to output probability values ​​of deviation sub-classifications of actual values ​​of running data.

Images