Time sequence holographic database design method, system, equipment and medium

Abstract

The invention discloses a time sequence holographic database design method, system and device and a medium, and belongs to the technical field of transformer monitoring, and the method comprises the following steps: constructing a data layer of a time sequence holographic database, and collecting multi-source time sequence data of a transformer; constructing an intermediate layer of the time sequence holographic database, performing classified transmission and compressed storage on the collected multi-source time sequence data, and establishing a data integrity verification mechanism; constructing an event layer of the time sequence holographic database, analyzing and processing the stored multi-source time sequence data, establishing an event label system, and establishing an event prediction model based on historical data; and generating an event information report of transformer equipment state monitoring according to the event label system and a prediction result of the event prediction model. According to the method, a multi-source time series data unified collection framework is constructed, so that the problem that data table modeling is difficult due to diverse data sources and non-uniform data field names and contents is solved, and a foundation is laid for optimization design of a time series database.

Description

Technical Field

[0001] This invention relates to the field of transformer monitoring technology, specifically to a time-series holographic database design method, system, device, and medium. Background Technology

[0002] Traditional transformer equipment monitoring methods that rely on manual inspections and periodic maintenance suffer from low monitoring frequency and inability to detect potential equipment faults in a timely manner. Time-series databases, as an efficient way to store and manage time-series data, provide data support for real-time status monitoring of transformer equipment.

[0003] Currently, most time-series databases can efficiently store and manage time-series data, and quickly retrieve specific data. To better apply time-series databases in the field of equipment fault diagnosis, the defect data table structure is obtained by analyzing the data sources. Then, the defect data is cleaned and deduplicated. Based on historical reports, fault and defect information is manually annotated to obtain a time-series database containing equipment defect data. Addressing the challenge of storing large amounts of data in the fault diagnosis field, the target data is partitioned according to partitioning rules, and the partitioning results are obtained. Finally, the target data is stored in the corresponding partitioned time-series database, achieving large-scale data storage capabilities.

[0004] Currently, time-series databases perform data cleaning and deduplication to meet stable data storage requirements when processing massive amounts of data. However, this data processing method may lead to the loss of key data. Therefore, it is necessary to study a time-series database for transformer condition monitoring that, while ensuring complete data storage, adds data analysis functions and ultimately realizes the function of corresponding characteristic data with equipment events. Summary of the Invention

[0005] In view of the above-mentioned problems, the present invention is proposed.

[0006] Therefore, the technical problem solved by this invention is that in the process of data collection, there are problems such as diverse data sources, different data application scenarios, and inconsistent data field names and contents, which increases the difficulty of data table modeling and makes it impossible to reasonably predict the storage capacity of time series databases.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a time-series holographic database design method, which includes the following steps:

[0008] A data layer for constructing a time-series holographic database is used to collect multi-source time-series data of transformers;

[0009] An intermediate layer for constructing a time-series holographic database is established to classify, transmit, compress, and store the collected multi-source time-series data, and to establish a data integrity verification mechanism.

[0010] Construct the event layer of the time-series holographic database, analyze and process the stored multi-source time-series data, establish an event tagging system, and build an event prediction model based on historical data;

[0011] Based on the prediction results of the event tagging system and the event prediction model, an event information report for transformer equipment status monitoring is generated.

[0012] As a preferred embodiment of the time-series holographic database design method described in this invention, the multi-source time-series data includes equipment operation time-series data, fault time-series data, energy consumption time-series data, third-party monitoring data, and transformer test data.

[0013] As a preferred embodiment of the time-series holographic database design method described in this invention, the operation steps of the data integrity verification mechanism include:

[0014] Acquire the multi-source time series data;

[0015] Calculate the rate of change of multi-source time series data between adjacent time points. When the rate of change exceeds a preset limit, it is determined to be a data anomaly.

[0016] For multi-source time series data that is identified as abnormal, the corresponding time point is obtained. Multi-source time series data of other phase lines at the same time point are checked and cross-validation is performed.

[0017] The steps of the cross-validation include:

[0018] When the multi-source time series data of other phase lines are all normal, the multi-source time series data that is determined to be abnormal is reconstructed.

[0019] An early warning will be activated if data anomalies are found in the multi-source time-series data of other phase lines.

[0020] When reconstructing multi-source time-series data, let the corrupted data point be time t. Take the value of the previous time t-1 and the value of the next time t+1, and calculate the reconstructed value according to the formula, which is expressed as:

[0021] V t =V1*α+V2*β,α+β=1;

[0022] In the formula, V t The values ​​represent the damaged values, V1 and V2 represent the values ​​at the previous time t-1 and the next time t+1, respectively, and α and β represent the corresponding weights.

[0023] As a preferred embodiment of the time-series holographic database design method described in this invention, the analysis and processing of the stored multi-source time-series data includes the following steps:

[0024] Read the compressed time-series data file from the intermediate layer and decompress it;

[0025] The decompressed time series data files are classified according to data type, and different types of time series signals are extracted respectively;

[0026] Numerical calculations are performed on the extracted time-series signals, including: calculating the effective values ​​of voltage and current in each phase, calculating the harmonic content of each phase through fast Fourier transform, calculating the phase difference between voltage and current, and calculating the instantaneous values ​​of active power and reactive power.

[0027] The calculated feature parameters are compared with a preset threshold, and the determination method includes:

[0028] Add corresponding event tags to the corresponding time series data files based on the judgment results;

[0029] The methods for determining the feature parameters against the preset threshold include:

[0030] When the total harmonic distortion rate exceeds 5%, it is considered that the harmonics exceed the standard.

[0031] When the phase difference between voltage and current exceeds ±15°, it is determined to be a phase imbalance;

[0032] When the proportion of negative sequence current to positive sequence current exceeds 10%, it is determined to be a load imbalance.

[0033] When the frequency deviates from 50Hz by more than ±0.5Hz and the duration exceeds 10 seconds, it is determined to be a frequency offset.

[0034] As a preferred embodiment of the time-series holographic database design method described in this invention, the step of establishing the event prediction model includes:

[0035] A multidimensional feature vector is constructed based on historical fault data. The multidimensional feature vector includes electrical features, thermal features, mechanical features, and chemical features.

[0036] An improved long short-term memory neural network model is used for fault prediction. The long short-term memory neural network model includes an attention mechanism layer, a neural network layer, and a fully connected layer.

[0037] The calculation formula for the attention mechanism layer is:

[0038]

[0039] e i =v T tanh(W h h i +W x xi +b);

[0040]

[0041] In the formula, A i For attention weights, e i Let h be the attention score at time i, T be the time series length, and h be the time series length. i Let x be the hidden state of the neural network. i Let W be the input feature vector at time i, c be the weighted context vector, and W be the input feature vector at time i. h and W x Let v be the hidden state weight matrix and the input feature weight matrix, and b be the attention parameter vector and the bias vector, respectively.

[0042] A fault evolution trend prediction model is established. By analyzing the characteristic change trends within multiple time windows, the probability of fault occurrence within the next 1-7 days is predicted. The formula for calculating the probability of fault occurrence is as follows:

[0043] P(fault) = σ(W) o c+b o );

[0044] In the formula, P(fault) is the probability of a fault occurring, σ is the sigmoid activation function, and W... o To output the weight matrix, b o This is the output layer bias term.

[0045] In a preferred embodiment of the time-series holographic database design method described in this invention, the intermediate layer further includes an intelligent data fusion processing module, the processing steps of which include:

[0046] Time synchronization and alignment of similar time-series signals from different monitoring devices are performed, and the least squares method is used to eliminate time deviations.

[0047] Establish a data credibility assessment mechanism and assign credibility weights to each data source based on the historical accuracy of the monitoring device, data continuity, and environmental factors;

[0048] When discrepancies are detected in the measurements of the same parameter from multiple data sources, a weighted fusion algorithm is used to integrate the data.

[0049] In a preferred embodiment of the time-series holographic database design method described in this invention, the event information report further includes an intelligent diagnostic suggestion module, which includes the following functions:

[0050] Establish a risk assessment system based on event tags, and calculate equipment health status scores according to the type and number of event tags;

[0051] The formula for calculating the equipment health status score is as follows:

[0052]

[0053] In the formula, H s Assess the health status of the device, where n is the number of event label types, and W is the number of device health status scores. i C represents the weight coefficient of the i-th type of event label. i Let be the number of times the i-th type of event label appears within the statistical period;

[0054] Establish a fault risk level classification standard, when H s A value ≥ 80 is considered normal; when 60 ≤ H s A warning state is activated when H < 80; a warning state is activated when 40 ≤ H s When H < 60, it is in alarm status; when H s A temperature below 40 indicates a dangerous situation.

[0055] Personalized maintenance recommendations are generated based on the equipment's health status score.

[0056] Another objective of this invention is to provide a time-series holographic database design system.

[0057] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a time-series holographic database design system, including a data collection module, a data storage module, a data analysis module, and a report generation module;

[0058] The data collection module is responsible for collecting multi-source timing data of the transformer;

[0059] The data storage module includes a data layer and an intermediate layer of a time-series holographic database;

[0060] The data analysis module analyzes and processes the stored multi-source time-series data;

[0061] The report generation module generates event information reports for transformer equipment status monitoring based on the prediction results of the event tagging system and the event prediction model.

[0062] The present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the time-series holographic database design method.

[0063] The present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the steps of the time-series holographic database design method.

[0064] The beneficial effects of this invention are as follows: By constructing a unified collection framework for multi-source time-series data, the field structure and content of equipment operation time-series data, fault time-series data, energy consumption time-series data, third-party monitoring data, and transformer test data are clarified. This solves the problem of difficulty in data table modeling caused by diverse data sources and inconsistent data field names and contents, and lays the foundation for the optimized design of time-series databases.

[0065] By establishing a data integrity verification mechanism and calculating the rate of change of data at adjacent time points and using cross-validation methods, abnormal data can be detected and repaired in a timely manner. This avoids the risk of missing key data during the data cleaning process in traditional time-series databases, and ensures the integrity and reliability of massive data storage.

[0066] An improved long short-term memory neural network model is used to establish an event prediction model. Combined with an attention mechanism layer for deep learning of historical fault data, this model can predict the probability of fault occurrence within the next 1-7 days. Compared to traditional single-classification methods, the prediction accuracy is improved by more than 30%, realizing a shift from passive monitoring to proactive early warning. Through an intelligent data fusion processing module, similar data from different monitoring devices are synchronized in time and their reliability is assessed, resolving the issues of time deviation and measurement error in multi-source data and providing more accurate and reliable data support for condition monitoring. Attached Figure Description

[0067] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0068] Figure 1 The above is a flowchart of a time-series holographic database design method provided in one embodiment of the present invention.

[0069] Figure 2 This is a framework diagram of a time-series holographic database design method provided in one embodiment of the present invention.

[0070] Figure 3 This invention provides an event information report for a database based on a time-series holographic database design method, as an embodiment of the present invention. Detailed Implementation

[0071] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0072] Example 1, referring to Figures 1-3 This is one embodiment of the present invention, which provides a time-series holographic database design method, including the following steps S1 to S4.

[0073] S1. Construct the data layer of the time-series holographic database to collect multi-source time-series data of the transformer;

[0074] S2. Construct the intermediate layer of the time-series holographic database, classify, transmit, compress, and store the collected multi-source time-series data, and establish a data integrity verification mechanism.

[0075] S3. Construct the event layer of the time-series holographic database, analyze and process the stored multi-source time-series data, establish an event tagging system, and build an event prediction model based on historical data;

[0076] S4. Generate an event information report for transformer equipment status monitoring based on the prediction results of the event tagging system and the event prediction model.

[0077] It should be noted that during the operation of transformer equipment, sensors are equipped at various key points to collect time-series data. The volume of time-series data is growing exponentially. At the same time, it faces the problems of diverse data sources and inconsistent data field names and contents. Traditional time-series databases cannot meet the management and storage needs of massive amounts of time-stamped data every day without special data cleaning processing. Moreover, most current time-series databases lack event tags when processing data, which weakens the application functions of time-series databases.

[0078] Therefore, to address the aforementioned issues of monitoring data management and event identification, a three-tiered time-series holographic database is constructed through steps S1-S4. The data layer completes multi-source data collection and clarifies the data structure; the intermediate layer achieves efficient data compression storage and integrity verification; and the event layer realizes intelligent status monitoring through data analysis and an event tagging system, providing comprehensive health status assessment and fault early warning functions for transformer equipment.

[0079] In this embodiment of the application, the data layer of the time-series holographic database is constructed in step S1. Multi-source time-series data of the transformer are collected by a smart monitoring device and a fault recording device to collect time-series data and clarify the data source and field structure.

[0080] In an optional implementation, the multi-source time-series data of the transformer collected in step S1 can also be pre-processed at the data source end by deploying distributed data acquisition nodes and using edge computing technology, thereby reducing the amount of data transmission and improving acquisition efficiency.

[0081] In another alternative implementation, for example, if an anomaly is found in the timing data of a transformer in a hydropower plant during operation, the fault recording device inputs the analysis results into a time-series holographic database. In the event layer of the database, event tag definition logic is used to assign corresponding tags to the data to which the analysis results belong. Finally, combining the event tags with the transformer's operating mechanism, it is determined that a two-phase short-circuit fault (phases A and B) has occurred at line 1, and an event information report is generated. Figure 3 The image shown is the event information report for this fault event.

[0082] from Figure 3 As can be seen, the data layer of the time-series holographic database saves character-type data related to this fault, as well as numerical-type data from the fault time-series data, and displays them on the report. After the data is analyzed by the data analysis algorithm package in the fault recording device, the analysis results in the event layer of the time-series holographic database, combined with the event tag logic rules defined in the database, assign event tags to the data, that is, in... Figure 3 The faulty phase was indicated as a two-phase short circuit (A and B). Finally, the interaction between the data layer, intermediate layer, and event layer generated an event information report indicating the transformer's abnormal operation. After investigation, it was confirmed that the transformer had experienced a phase-to-phase short circuit fault. This embodiment demonstrates that the time-series holographic database proposed in this invention has high efficiency and high accuracy in the field of transformer equipment condition monitoring.

[0083] Example 2, refer to Figures 1-3 This is one embodiment of the present invention, which provides a time-series holographic database design method based on the previous embodiment.

[0084] In this application embodiment, in step S1, the multi-source time-series data includes equipment operation time-series data, fault time-series data, energy consumption time-series data, third-party monitoring data, and transformer test data.

[0085] In this embodiment of the application, step S2, the operation steps of the data integrity verification mechanism include steps A1-A3:

[0086] A1: Obtain the multi-source time-series data;

[0087] A2: Calculate the rate of change of multi-source time series data between adjacent time points. When the rate of change exceeds the preset limit, it is judged as data anomaly.

[0088] A3: For multi-source time series data that is determined to be abnormal, obtain the corresponding time point, check the multi-source time series data of other phase lines at the same time point, and perform cross-validation.

[0089] Specifically, in step A3, the cross-validation steps include: when the multi-source time series data of other phase lines are all normal, the multi-source time series data that is determined to be abnormal is reconstructed; when the multi-source time series data of other phase lines has abnormal data, an early warning is activated.

[0090] When reconstructing multi-source time-series data, let the corrupted data point be time t. Take the value of the previous time t-1 and the value of the next time t+1, and calculate the reconstructed value according to the formula, which is expressed as:

[0091] V t =V1*α+V2*β,α+β=1;

[0092] In the formula, V t The values ​​represent the damaged values, V1 and V2 represent the values ​​at the previous time t-1 and the next time t+1, respectively, and α and β represent the corresponding weights.

[0093] It should be noted that the cross-validation mechanism, through horizontal comparison of multi-phase line data and vertical analysis of time series, can accurately identify the cause of data anomalies, distinguish between sensor failures and actual equipment anomalies, and avoid misjudgments and missed judgments. When a single-phase line data anomaly is detected while other phase lines are normal, weighted interpolation reconstruction is used; when multiple phase lines are detected to be abnormal simultaneously, an early warning mechanism is immediately activated to notify maintenance personnel to conduct on-site inspections.

[0094] In an optional implementation, the cross-validation in step A3 can also be combined with historical data pattern recognition. By analyzing the data change patterns under similar working conditions, a dynamic anomaly detection threshold can be established to improve the accuracy and adaptability of anomaly detection.

[0095] In another alternative implementation, data reconstruction can also employ polynomial interpolation or spline interpolation methods. The optimal interpolation algorithm is selected based on the continuity characteristics of the data. Linear interpolation is used for slowly varying signals, and higher-order interpolation is used for rapidly varying signals to ensure the accuracy of the reconstructed data.

[0096] It should be noted that this data integrity verification mechanism has a higher level of intelligence and accuracy compared to the traditional simple threshold judgment method. It can ensure data integrity while avoiding the loss of key information due to the data cleaning process, thus providing a reliable data foundation for subsequent event analysis and fault prediction.

[0097] In this embodiment of the application, step S3, which involves analyzing and processing the stored multi-source time-series data, includes the following steps:

[0098] B1: Reads the compressed time-series data file from the intermediate layer and decompresses it;

[0099] Specifically, in step B1, the decompression process employs multi-threaded parallel decompression technology, performing batch processing according to the timestamp order of the data files to ensure the temporal integrity of the data. During decompression, data format verification is performed simultaneously, checking the header information and data segment integrity of the comtrade files. For corrupted files, a backup file reading mechanism is automatically triggered.

[0100] B2: Classify the decompressed timing data files according to data type and extract different types of timing signals respectively;

[0101] In step B2, data types are categorized according to the following rules: Equipment operation sequence data includes three-phase port phase voltage, three-phase port phase current, and voltage level information; fault sequence data includes electrical sequence data for the two cycles before the fault, the fault process, and the five cycles after the fault ends; energy consumption sequence data includes parameters such as active power P, reactive power Q, active power at the generator terminal, reactive power at the generator terminal, power factor, and load rate; third-party data includes dissolved gas monitoring data in the oil, including carbon monoxide content, carbon dioxide content, hydrogen content, methane content, ethane content, ethylene content, and acetylene content, as well as temperature monitoring data, including oil temperature and chassis temperature; transformer test data includes equipment technical parameters and test results data.

[0102] B3: Perform numerical calculations on the extracted time-series signals, including: calculating the effective values ​​of voltage and current in each phase, calculating the harmonic content of each phase through fast Fourier transform, calculating the phase difference between voltage and current, and calculating the instantaneous values ​​of active power and reactive power.

[0103] In step B3, the specific implementation of the numerical calculation is as follows:

[0104] The effective value (RMS) is calculated using the sliding window RMS algorithm, with the window length set to one power frequency cycle (20ms). The formula for calculating the effective value (RMS) is as follows:

[0105]

[0106] In this calculation formula, N is the number of sampling points in one period, and x i Let be the instantaneous value of the i-th sampling point.

[0107] The Fast Fourier Transform (FFT) calculation uses a 1024-point FFT algorithm, analyzing frequencies from DC to 25.6 kHz, and can detect the 50th harmonic component. The Total Harmonic Distortion (THD) is calculated using the following formula:

[0108]

[0109] Among them, I h Ih is the effective value of the h-th harmonic current, and I1 is the effective value of the fundamental current.

[0110] The phase difference calculation adopts a combination of zero-crossing detection and correlation analysis. First, the fundamental phase is determined by zero-crossing detection, and then the phase difference is accurately measured by correlation analysis.

[0111] Power calculation uses instantaneous power theory, and the formula for calculating active power P is:

[0112]

[0113] Reactive power Q is calculated using the Hilbert transform method:

[0114]

[0115] In the formulas for calculating active power P and reactive power Q, i′(t) is the orthogonal component of the current signal after Hilbert transformation, T is the period of the AC signal, u(t) is the instantaneous value of voltage, i(t) is the instantaneous value of current, and i′(t) is the orthogonal component of the current signal after Hilbert transformation.

[0116] B4: The calculated feature parameters are compared with the preset thresholds, and the determination methods include;

[0117] Add corresponding event tags to the corresponding time series data files based on the judgment results;

[0118] In step B4, the method for determining the feature parameters against the preset threshold includes:

[0119] When the total harmonic distortion rate exceeds 5%, it is considered that the harmonics exceed the standard.

[0120] When the phase difference between voltage and current exceeds ±15°, it is determined to be a phase imbalance;

[0121] When the proportion of negative sequence current to positive sequence current exceeds 10%, it is determined to be a load imbalance.

[0122] When the frequency deviates from 50Hz by more than ±0.5Hz and the duration exceeds 10 seconds, it is determined to be a frequency offset.

[0123] In this embodiment, step S3, the establishment of the event prediction model, includes steps C1 to C3.

[0124] C1: Construct a multi-dimensional feature vector based on historical fault data. The multi-dimensional feature vector includes electrical features, thermal features, mechanical features, and chemical features.

[0125] Specifically, in step C1, the multidimensional feature vector is constructed as follows:

[0126] Electrical characteristics include, but are not limited to, the effective values ​​of three-phase voltage, the effective values ​​of three-phase current, the content of each harmonic (2nd-50th), the total harmonic distortion rate, the power factor, the active power, the reactive power, the positive sequence, the negative sequence, and the zero sequence components, the phase difference between voltage and current, and the frequency offset.

[0127] Thermal characteristics include, but are not limited to, transformer oil temperature, winding temperature, core temperature, ambient temperature, temperature rise gradient, and cooling system status.

[0128] Mechanical characteristics include, but are not limited to, vibration amplitude, vibration spectrum, acoustic signal characteristics, tap changer operation frequency, and cooling fan operating status.

[0129] Chemical characteristics include, but are not limited to, the concentration of dissolved gases in the oil (H2, CH4, C2H6, C2H4, C2H2, CO, CO2), the oil's dielectric loss factor, insulation resistance, and acid value.

[0130] C2: An improved long short-term memory neural network model is used for fault prediction. The long short-term memory neural network model includes an attention mechanism layer, a neural network layer, and a fully connected layer.

[0131] The calculation formula for the attention mechanism layer is:

[0132]

[0133] e i =v T tanh(W h h i +W x x i +b);

[0134]

[0135] In the formula, A i For attention weights, e i Let h be the attention score at time i, T be the time series length, and h be the time series length. i Let x be the hidden state of the neural network. i Let W be the input feature vector at time i, c be the weighted context vector, and W be the input feature vector at time i. h and W x Let v be the hidden state weight matrix and the input feature weight matrix, and b be the attention parameter vector and the bias vector, respectively.

[0136] C3: Establish a fault evolution trend prediction model. By analyzing the characteristic change trends within multiple time windows, predict the probability of fault occurrence within the next 1-7 days. The formula for calculating the probability of fault occurrence is:

[0137] P(fault) = σ(W) o c+b o );

[0138] In the formula, P(fault) is the probability of a fault occurring, σ is the sigmoid activation function, and W... o To output the weight matrix, b o This is the output layer bias term.

[0139] The intermediate layer also includes an intelligent data fusion processing module, whose processing steps include D1 to D3.

[0140] D1: Time synchronization and alignment of the same type of time-series signals from different monitoring devices are performed, and the least squares method is used to eliminate time deviation;

[0141] Specifically, in step D1, time synchronization alignment uses a GPS clock source as the reference time standard. Due to differences in the precision of the clock crystal oscillators used in the manufacturing of different monitoring devices, coupled with factors such as network transmission latency, time discrepancies are common between devices. Therefore, all monitoring devices are required to synchronize their time via a network time protocol to ensure synchronization accuracy at the millisecond level.

[0142] For existing time discrepancies in historical data, a least squares method is used for correction. Specifically, obvious physical events such as transformer startup and shutdown are selected as time markers. The differences in recording times of the same event by different devices are analyzed to establish a time discrepancy correction model. For example, when a transformer undergoes load switching at a certain moment, device A records the time as 10:15:30.120, while device B records the time as 10:15:30.145. By collecting multiple such synchronization markers, it is calculated that device B has a fixed delay of 25 milliseconds relative to device A, and there may also be clock drift accumulating over time. All historical data will be automatically time-corrected to ensure complete alignment of timestamps from different data sources.

[0143] The time synchronization effect is verified through correlation analysis: when two devices measure the same electrical signal, the correlation coefficient of the time-synchronized data should reach 0.98 or higher. If this standard is not met, time correction will be performed again.

[0144] D2: Establish a data credibility assessment mechanism and assign credibility weights to each data source based on the historical accuracy of the monitoring device, data continuity, and environmental factors.

[0145] In step D2, the data credibility assessment mechanism employs a multi-dimensional comprehensive evaluation method. A credibility profile is established for each data source, encompassing four dimensions: historical accuracy, data continuity, environmental adaptability, and equipment aging.

[0146] Historical accuracy assessment is achieved by comparing the device with a standard reference value. For example, for voltage measuring devices, they are periodically calibrated using a high-precision standard voltmeter, and measurement errors are recorded to establish an accuracy profile. If one device has a historical average measurement error of 0.2% and another device has 0.5%, the former receives a higher accuracy score.

[0147] Data continuity assessment measures the data missingness of statistical devices. The data completeness rate is calculated by recording the number of data missing events and their duration for each device over the past year. For example, if device A experienced 5 data interruptions with a total missing time of 2 hours in the past year, while device B experienced 15 interruptions with a total missing time of 8 hours, then device A receives a higher continuity score.

[0148] Environmental adaptability assessment considers the impact of environmental factors such as temperature, humidity, and electromagnetic interference on measurement accuracy. An environmental adaptability profile is established by recording the measurement stability of each device under different environmental conditions. For example, in high-temperature environments, some devices may experience measurement drift, and the reliability weight of these devices under high-temperature conditions will decrease accordingly.

[0149] Equipment aging assessments are based on the equipment's age and maintenance history. Newly installed equipment generally has higher reliability, while equipment that has been used for many years and has a poor maintenance record has lower reliability.

[0150] D3: When differences are detected in the measurement values ​​of the same parameter from multiple data sources, a weighted fusion algorithm is used to integrate the data.

[0151] In step D3, when significant differences are detected in the measured values ​​of the same parameter from multiple data sources, weighted fusion processing is initiated. This will be illustrated using three voltage measuring devices as an example:

[0152] Suppose that at a certain moment, three devices measure the same phase voltage as follows: Device A measures 220.5V, Device B measures 219.8V, and Device C measures 221.2V. First, check whether these values ​​are within a reasonable range, and then calculate the overall reliability weight of each device. Device A receives a reliability weight of 0.9 due to its high historical accuracy and good data continuity; Device B receives a weight of 0.8 due to a past data interruption; and Device C receives a weight of 0.85 due to its long service life.

[0153] It also dynamically adjusts based on the deviation of the current measurement from the group average. In this example, the simple average of the three measurements is 220.5V. The measurement of device A is exactly equal to the average, obtaining a full-score dynamic adjustment factor of 1.0; device B has a smaller deviation, obtaining an adjustment factor of 0.95; and device C has a relatively larger deviation, obtaining an adjustment factor of 0.88.

[0154] The final fusion result, taking into account both the reliability weight and the dynamic adjustment factor, yielded a fusion voltage value of 220.4V. This fusion value considers both the historical reliability of each device and the rationality of the current measurement.

[0155] For outlier detection, a strategy combining the three-standard-deviation criterion and box plot method is adopted. If the measurement value of a device deviates from the population mean by more than three standard deviations, or falls within the outlier range of the box plot, it will be marked as an outlier and its weight in the fusion process will be reduced. In severe cases, the data source may even be completely excluded.

[0156] The event information report also includes an intelligent diagnostic suggestion module, which includes the following functions:

[0157] Establish a risk assessment system based on event tags, and calculate equipment health status scores according to the type and number of event tags;

[0158] The formula for calculating the equipment health status score is:

[0159]

[0160] In the formula, H s Assess the health status of the device, where n is the number of event label types, and W is the number of device health status scores. i C represents the weight coefficient of the i-th type of event label. i Let be the number of times the i-th type of event label appears within the statistical period;

[0161] Establish a fault risk level classification standard, when H s A value ≥ 80 is considered normal; when 60 ≤ H s A warning state is activated when H < 80; a warning state is activated when 40 ≤ H s When H < 60, it is in alarm status; when H s A temperature below 40°C indicates a dangerous situation.

[0162] Personalized maintenance recommendations are generated based on the equipment's health status score.

[0163] In this embodiment, a certain type of power transformer is taken as an example. This transformer undertakes the key tasks of voltage transformation and power transmission in the power system. Its stable operation is crucial to the safety and reliability of the power grid. Therefore, it is necessary to monitor and evaluate its health status in real time.

[0164] The following event tag data are collected using various sensors installed on the transformer: Overheating oil alarm (weight coefficient 0.5, occurring twice within the statistical period); Abnormal oil level (weight coefficient 0.3, occurring once within the statistical period); Partial discharge exceeding the limit (weight coefficient 0.8, occurring three times within the statistical period); Cooling system failure (weight coefficient 0.4, occurring once within the statistical period). These event tag types are predetermined based on the transformer's operating characteristics and common fault modes. Their weight coefficients reflect the degree of impact of different events on the transformer's health status; the larger the weight coefficient, the more severe the impact of the event on the equipment's health status.

[0165] Substituting these parameters into the calculation formula, the output result is 95.9. According to the fault risk level classification standard, H... s A value ≥80 indicates a normal operating condition; therefore, the transformer is currently in a normal operating condition.

[0166] Personalized maintenance recommendations: Even if the equipment is in normal condition, attention should still be paid to overheating alarms and excessive partial discharge detection. It is recommended to further check whether the transformer's cooling system is working properly, investigate potential causes of overheating such as excessive load or cooler blockage, and perform appropriate maintenance or repairs. For excessive partial discharge detection, it is recommended to use more precise positioning technology to determine the location of the discharge source, analyze the possibility of insulation aging or damage, and, if necessary, conduct power outage maintenance or replace relevant components. Simultaneously, continue monitoring of other event tags and regularly update the health status score to promptly grasp changes in the equipment's operating status.

[0167] Example 3 is an embodiment of the present invention, which provides a time-series holographic database design system, including a data collection module, a data storage module, a data analysis module, and a report generation module;

[0168] The data collection module is responsible for collecting multi-source time-series data of the transformer;

[0169] The data storage module includes the data layer and the middle layer of the time-series holographic database;

[0170] The data analysis module analyzes and processes the stored multi-source time-series data;

[0171] The report generation module generates event information reports for transformer equipment status monitoring based on the prediction results of the event tagging system and the event prediction model.

[0172] This embodiment also provides an electronic device applicable to a time-series holographic database design method, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the time-series holographic database design method proposed in the above embodiment.

[0173] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, it implements a time-series holographic database design method as proposed in the above embodiments.

[0174] The storage medium proposed in this embodiment belongs to the same inventive concept as the time-series holographic database design method proposed in the above embodiments. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0175] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0176] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for designing a time-series holographic database, characterized in that: Includes the following steps, A data layer for constructing a time-series holographic database is used to collect multi-source time-series data of transformers; An intermediate layer for constructing a time-series holographic database is established to classify, transmit, compress, and store the collected multi-source time-series data, and to establish a data integrity verification mechanism. Construct the event layer of the time-series holographic database, analyze and process the stored multi-source time-series data, establish an event tagging system, and build an event prediction model based on historical data; Based on the prediction results of the event tagging system and the event prediction model, an event information report for transformer equipment status monitoring is generated.

2. The time-series holographic database design method as described in claim 1, characterized in that: The multi-source time-series data includes equipment operation time-series data, fault time-series data, energy consumption time-series data, third-party monitoring data, and transformer test data.

3. The time-series holographic database design method as described in claim 2, characterized in that: The operation steps of the data integrity verification mechanism include: Acquire the multi-source time series data; Calculate the rate of change of multi-source time series data between adjacent time points. When the rate of change exceeds a preset limit, it is determined to be a data anomaly. For multi-source time series data that is identified as abnormal, the corresponding time point is obtained. Multi-source time series data of other phase lines at the same time point are checked and cross-validation is performed. The steps of the cross-validation include: When the multi-source time series data of other phase lines are all normal, the multi-source time series data that is determined to be abnormal is reconstructed. An early warning will be activated if data anomalies are found in the multi-source time-series data of other phase lines. When reconstructing multi-source time-series data, let the corrupted data point be time t. Take the value of the previous time t-1 and the value of the next time t+1, and calculate the reconstructed value according to the formula, which is expressed as: V t =V1*α+V2*β,α+β=1; In the formula, V t The values ​​represent the damaged values, V1 and V2 represent the values ​​at the previous time t-1 and the next time t+1, respectively, and α and β represent the corresponding weights.

4. The time-series holographic database design method as described in claim 3, characterized in that: The analysis and processing of the stored multi-source time-series data includes the following steps. Read the compressed time-series data file from the intermediate layer and decompress it; The decompressed time series data files are classified according to data type, and different types of time series signals are extracted respectively; Numerical calculations are performed on the extracted time-series signals, including: calculating the effective values ​​of voltage and current in each phase, calculating the harmonic content of each phase through fast Fourier transform, calculating the phase difference between voltage and current, and calculating the instantaneous values ​​of active power and reactive power. The calculated feature parameters are compared with a preset threshold, and the determination method includes: Add corresponding event tags to the corresponding time series data files based on the judgment results; The methods for determining the feature parameters against the preset threshold include: When the total harmonic distortion rate exceeds 5%, it is considered that the harmonics exceed the standard. When the phase difference between voltage and current exceeds ±15°, it is determined to be a phase imbalance; When the proportion of negative sequence current to positive sequence current exceeds 10%, it is determined to be a load imbalance. When the frequency deviates from 50Hz by more than ±0.5Hz and the duration exceeds 10 seconds, it is determined to be a frequency offset.

5. The time-series holographic database design method as described in claim 4, characterized in that: The steps for establishing the event prediction model include: A multidimensional feature vector is constructed based on historical fault data. The multidimensional feature vector includes electrical features, thermal features, mechanical features, and chemical features. An improved long short-term memory neural network model is used for fault prediction. The long short-term memory neural network model includes an attention mechanism layer, a neural network layer, and a fully connected layer. The calculation formula for the attention mechanism layer is: e i =v T fishy(W h h i +W x x i +b); In the formula, A i For attention weights, e i Let h be the attention score at time i, T be the time series length, and h be the time series length. i Let x be the hidden state of the neural network. i Let W be the input feature vector at time i, c be the weighted context vector, and W be the input feature vector at time i. h and W x Let v be the hidden state weight matrix and the input feature weight matrix, and b be the attention parameter vector and the bias vector, respectively. A fault evolution trend prediction model is established. By analyzing the characteristic change trends within multiple time windows, the probability of fault occurrence within the next 1-7 days is predicted. The formula for calculating the probability of fault occurrence is as follows: P(fault)=σ(W o c+b o ); In the formula, P(fault) is the probability of a fault occurring, σ is the sigmoid activation function, and W... o To output the weight matrix, b o This is the output layer bias term.

6. The time-series holographic database design method as described in claim 5, characterized in that: The intermediate layer also includes an intelligent data fusion processing module, whose processing steps include: Time synchronization and alignment of similar time-series signals from different monitoring devices are performed, and the least squares method is used to eliminate time deviations. Establish a data credibility assessment mechanism and assign credibility weights to each data source based on the historical accuracy of the monitoring device, data continuity, and environmental factors; When discrepancies are detected in the measurements of the same parameter from multiple data sources, a weighted fusion algorithm is used to integrate the data.

7. The time-series holographic database design method as described in claim 6, characterized in that: The event information report also includes an intelligent diagnostic suggestion module, which includes the following functions. Establish a risk assessment system based on event tags, and calculate equipment health status scores according to the type and number of event tags; The formula for calculating the equipment health status score is as follows: In the formula, H s Assess the health status of the device, where n is the number of event label types, and W is the number of device health status scores. i C represents the weight coefficient of the i-th type of event label. i Let be the number of times the i-th type of event label appears within the statistical period; Establish a fault risk level classification standard, when H s A value ≥80 is considered normal; when 60≤H s A warning state is activated when H < 80; a warning state is activated when 40 ≤ H s When H < 60, it is in alarm status; when H s A temperature below 40 indicates a dangerous situation. Personalized maintenance recommendations are generated based on the equipment's health status score.

8. A system for designing a time-series holographic database, employing a time-series holographic database design method as described in any one of claims 1 to 7, characterized in that, It includes a data collection module, a data storage module, a data analysis module, and a report generation module; The data collection module is responsible for collecting multi-source timing data of the transformer; The data storage module includes a data layer and an intermediate layer of a time-series holographic database; The data analysis module analyzes and processes the stored multi-source time-series data; The report generation module generates event information reports for transformer equipment status monitoring based on the prediction results of the event tagging system and the event prediction model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the time-series holographic database design method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the time-series holographic database design method according to any one of claims 1 to 7.

Images