A method for evaluating the operating state of a phase modifier based on multi-source information fusion

Abstract

The application discloses a method for evaluating the operation state of a phase modifier based on multi-source information fusion, which comprises the following steps: obtaining multi-source operation data of the phase modifier, identifying the current operation condition according to the operation condition parameters; extracting a plurality of features representing the operation state from the multi-source operation data; calling the corresponding condition baseline for the current operation condition, comparing the extracted features with the condition baseline to obtain the deviation degree of each feature; and performing fusion processing on the deviation degrees of the features to generate an evaluation result representing the overall operation state of the phase modifier. The application effectively reduces the false alarm rate by dynamically matching the condition baseline, decouples internal and external abnormalities by identifying power grid disturbance events, and improves the accuracy of the evaluation result and the adaptive ability in the whole life cycle by combining the dynamic weight distribution and conflict processing mechanism for hierarchical fusion and introducing a learning mechanism based on maintenance feedback.

Description

Technical Field

[0001] This invention relates to the field of power system equipment condition monitoring technology, and in particular to a method for evaluating the operating status of synchronous condensers based on multi-source information fusion. Background Technology

[0002] With the large-scale construction of ultra-high voltage direct current (UHVDC) transmission projects and the continuous increase in installed capacity of new energy power generation in my country, synchronous condensers, as core equipment for dynamic reactive power compensation and voltage support in the power grid, directly affect the voltage stability and safe power supply capacity of inter-regional power grids. Traditional synchronous condenser operation and maintenance models mainly rely on periodic preventative testing and manual inspections, which suffer from drawbacks such as fixed maintenance cycles, delayed fault detection, and a coexistence of over-maintenance and under-maintenance, making them unsuitable for the stringent requirements of modern power grids for lean equipment operation and maintenance. In recent years, condition assessment technology based on online monitoring has developed rapidly, gradually evolving from monitoring single electrical parameters to multi-dimensional data fusion. By integrating multi-source operating data such as electrical, mechanical, and thermal data, preliminary assessment of equipment status can be achieved. However, existing technologies still have some limitations in terms of operating condition adaptability and assessment accuracy.

[0003] Existing methods for assessing the operational status of synchronous condensers generally employ a uniform health baseline standard, failing to consider the normal fluctuations in equipment characteristic parameters under various operating conditions such as different reactive power levels, start-up and shutdown phases, cooling status, and grid disturbances. This results in a persistently high false alarm rate for anomaly detection. Furthermore, most methods fail to effectively distinguish between instantaneous deviations in characteristics caused by external grid disturbances and persistent anomalies caused by internal equipment degradation, easily leading to unnecessary maintenance interventions. In addition, existing multi-source information fusion methods often employ fixed weight allocation strategies, failing to comprehensively consider sensor quality, data sampling integrity, and the sensitivity of the correlation between features and fault types. They also lack dynamic handling mechanisms for conflicts in multi-source evidence, making it difficult to guarantee the accuracy and robustness of the assessment results. While some methods introduce multi-source data fusion, they lack a model self-updating mechanism based on maintenance feedback, failing to adapt to the performance evolution patterns throughout the equipment's lifecycle. Summary of the Invention

[0004] The purpose of this section is to outline some aspects of embodiments of the present invention and to briefly describe some preferred embodiments. Simplifications or omissions may be made in this section, as well as in the abstract and title of this application, to avoid obscuring the purpose of these documents; however, such simplifications or omissions should not be construed as limiting the scope of the invention.

[0005] In view of the aforementioned existing problems, this invention is proposed. Therefore, this invention provides a method for evaluating the operating status of a synchronous condenser based on multi-source information fusion, to solve the problems mentioned in the background art.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a method for evaluating the operating status of a synchronous condenser based on multi-source information fusion, comprising: Acquire multi-source operating data of the synchronous condenser, and identify the current operating condition of the synchronous condenser based on the operating condition parameters in the multi-source operating data; Extract multiple features characterizing the operating state of the synchronous condenser from the multi-source operating data; For the current operating condition of the synchronous condenser, a preset operating condition baseline corresponding to the operating condition is invoked, and multiple features of the operating state of the synchronous condenser are compared with the operating condition baseline to obtain the deviation of each feature; The deviation of multiple features of the synchronous condenser's operating state is fused to generate an evaluation result characterizing the overall operating state of the synchronous condenser.

[0007] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention abandons the traditional global fixed alarm threshold, and identifies the current operating condition of the synchronous condenser through multi-dimensional feature matching, and calls the corresponding operating condition baseline to calculate the deviation. This dynamic baseline comparison mechanism, which adapts to different conditions, fully considers the normal fluctuation characteristics of the characteristic parameters of the synchronous condenser under different loads and start-up / shutdown phases, effectively avoiding false alarms under heavy load conditions and missed alarms under light load conditions.

[0008] 2. This invention addresses the physical characteristics of synchronous condensers providing dynamic reactive power support by introducing a power grid disturbance event identification mechanism. By tagging data within a transient time window with an external disturbance response label, it logically distinguishes between normal transient changes in equipment response to external power grid demands and actual internal degradation faults, thus eliminating misjudgments of synchronous condenser status and invalid maintenance dispatches caused by power grid faults.

[0009] 3. In addition, the present invention also adopts a dynamic weight allocation and conflict resolution mechanism in the multi-source data fusion stage. It comprehensively considers sensor quality, data integrity and fault correlation sensitivity to allocate initial weights, and introduces conflict detection and dynamic weight reduction rules, which can effectively shield the interference of individual sensor hardware failures or communication anomalies on the overall evaluation results, prevent key fault signals from being overwhelmed by a large number of normal signals, and improve the robustness of the algorithm in harsh industrial environments. Attached Figure Description

[0010] 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 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. Wherein: Figure 1 This is a flowchart illustrating the overall process of a synchronous condenser operation status evaluation method based on multi-source information fusion, as described in one embodiment of the present invention. Detailed Implementation

[0011] 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.

[0012] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0013] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0014] Furthermore, in the description of this invention, it should be noted that the terms "upper," "lower," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are used solely for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. In addition, the terms "first," "second," or "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0015] Example 1

[0016] Reference Figure 1 This is the first embodiment of the present invention, which provides a method for evaluating the operating status of a synchronous condenser based on multi-source information fusion, including: S1. Obtain multi-source operating data of the synchronous condenser, and identify the current operating condition of the synchronous condenser based on the operating condition parameters in the multi-source operating data.

[0017] It should be noted that synchronous condensers, as core rotating equipment providing dynamic reactive power support in the power grid, typically involve complex electromagnetic, mechanical, and thermal coupling relationships. Single-dimensional monitoring often only reflects localized characteristics of the equipment's condition, failing to comprehensively describe its overall health level. Therefore, this invention first utilizes a sensor network deployed on each subsystem of the synchronous condenser to acquire multi-source operational data in real time.

[0018] In one specific embodiment, these multi-source operating data include, but are not limited to, at least two of the following: electrical information (e.g., stator voltage, stator current, terminal power factor, etc.), mechanical information (e.g., shaft vibration, bearing elevation, lubricating oil pressure, etc.), thermal information (e.g., stator winding temperature, core temperature, bearing bearing temperature, etc.), excitation information (e.g., excitation current, excitation voltage, rotor insulation resistance, etc.), cooling information (e.g., cooling water flow rate, cold / hot air temperature, hydrogen pressure and purity, etc.), and insulation information (e.g., partial discharge quantity, grounding current, etc.).

[0019] Furthermore, after acquiring multi-source operational data, the synchronous condenser system needs to extract operational parameters that can characterize the current operating environment and task status.

[0020] In one specific embodiment, these operating condition parameters include at least one of the following: reactive power, speed, start-stop phase, cooling status, and grid disturbance events.

[0021] It should be clarified that the operating characteristic parameters of a synchronous condenser (such as vibration amplitude and winding temperature) typically fluctuate significantly with changes in operating conditions. For example, the heating mechanism of a synchronous condenser during full-load leading-phase operation differs from that during full-load lagging-phase operation; the mechanical vibration amplitude during the acceleration and start-up phase is usually much higher than the steady-state vibration amplitude at rated speed. If a uniform evaluation baseline is used without distinguishing operating conditions, false alarms or missed alarms may occur. Therefore, to achieve high-precision operating condition identification, this embodiment constructs an operating condition identification logic based on multi-dimensional feature matching. This logic includes constructing a real-time operating condition feature set, matching typical operating condition templates, and outputting the operating condition identification results.

[0022] Furthermore, a real-time operating condition feature set is first constructed.

[0023] Specifically, the extracted continuous operating parameters (such as reactive power, speed, and cooling medium flow rate) are synchronized, aligned, and normalized, and combined with discrete state indicators (such as start-up and shutdown stages like turning gears, speed-up, grid connection, and disconnection) to generate a real-time operating feature set for the current moment. It should be noted that the purpose of normalization is to eliminate differences in dimensions and orders of magnitude between different physical quantities.

[0024] Next, we will match templates for typical working conditions.

[0025] Specifically, the system pre-generates multiple typical operating condition templates from a historical database using clustering analysis algorithms. Each template represents a common synchronous condenser operating state (e.g., full-load delayed phase operation at rated speed, no-load operation at rated speed, etc.) and includes a set of standard operating condition parameter center values. The system calculates the multi-dimensional spatial distance (i.e., similarity) between the current real-time operating condition feature set and each typical operating condition template. When calculating this spatial distance, the system performs a weighted processing. Specifically, the system assigns different weights to each operating condition parameter based on their importance in distinguishing operating conditions (e.g., reactive power typically has a higher weight than cooling water temperature), thereby eliminating the influence of correlations between parameters of different dimensions.

[0026] In one specific embodiment, the clustering analysis algorithm employs the weighted K-means algorithm. This is used to calculate the real-time operating condition feature set at the current moment. With the A typical working condition template center Multidimensional spatial distance between When using the weighted Euclidean distance formula: ,in For the dimensions of operating condition parameters, For the preset first The weighting coefficients for each operating condition parameter (e.g., the weighting coefficient for reactive power is set to 0.5, and the weighting coefficient for cooling water temperature is set to 0.1).

[0027] Furthermore, the final output is the working condition identification result.

[0028] Specifically, by comparing the calculation results, the typical operating condition template with the smallest spatial distance (i.e., the highest similarity) is selected as the current operating condition identification result of the synchronous condenser. This result will be directly output and used to select the baseline that matches the current operating condition from multiple preset operating condition baselines, thereby achieving dynamic baseline evaluation that is tailored to each specific condition.

[0029] Furthermore, this embodiment also incorporates a special identification and handling mechanism for grid disturbance events. Specifically, the core function of a synchronous condenser is to instantly provide massive reactive power to support the grid voltage when a grid fault occurs (e.g., UHVDC blocking, AC system short circuit, etc.). During this transient process, parameters such as the stator current and shaft torsional vibration of the synchronous condenser will undergo drastic changes. These drastic changes are normal physical manifestations of the equipment responding to external grid demands, rather than indicating internal degradation or failure. Therefore, the voltage drop at the generator terminals and the rate of change of reactive power can be monitored in real time. When any of the following triggering conditions are met, the system identifies a grid disturbance event: Condition 1: The voltage drop at the terminal exceeds the preset voltage drop threshold.

[0030] Condition 2: The absolute value of the reactive power change rate is greater than the preset abrupt change rate threshold.

[0031] Furthermore, once a power grid disturbance event is identified, the system will automatically open a preset transient time window (the length of which is usually set according to the transient response time constant of the synchronous condenser, such as 3 to 5 seconds). Within this time window, any feature deviations extracted from the multi-source operating data of the synchronous condenser will be forcibly labeled with the external disturbance response by the system.

[0032] It's important to explain that the benefit of adding this tag is that, when generating subsequent evaluation results, it's possible to clearly distinguish between transient characteristic deviations caused by external power grid disturbances and persistent characteristic deviations caused by internal equipment degradation (such as insulation aging and bearing wear). For data tagged with "external disturbance response," the system will pause triggering regular equipment health degradation alarms and instead redirect it to the transient performance evaluation module, thereby eliminating misjudgments of synchronous condenser status and invalid maintenance dispatches caused by power grid faults.

[0033] S2. Extract multiple features representing the operating status of the synchronous condenser from multi-source operating data.

[0034] Furthermore, due to the significant differences in the sampling frequencies of the sensors in each subsystem of the synchronous condenser (e.g., the sampling frequency of shaft vibration signals is typically in the kHz range, while the sampling frequency of stator winding temperature is typically in the Hz range or even lower), this invention first employs resampling techniques (e.g., cubic spline interpolation) to unify all multi-source operating data to a preset reference time resolution. Subsequently, a sliding time window is used to extract data, and the data within the window is standardized.

[0035] Specifically, the standardization process involves calculating the mean and standard deviation of the data for each sensor within the current sliding time window, subtracting the mean from the original sampled value, and then dividing by the standard deviation. It's important to explain that the purpose of standardization is to eliminate dimensional differences between different physical quantities (e.g., temperature in degrees Celsius versus vibration in millimeters per second), mapping them uniformly to a dimensionless standard scale, thereby preventing features with large absolute values ​​from dominating subsequent calculations.

[0036] Furthermore, for different types of operational data, based on the physical operating mechanism of the synchronous condenser, time-domain and frequency-domain features with clear physical meaning are extracted.

[0037] Specifically, for mechanical information (e.g., shaft vibration), the system extracts the root mean square (RMS) value and kurtosis in the time domain. The RMS value, obtained by taking the square root of the average of the squares of the amplitudes at all sampling points within the time window, characterizes the overall energy of the vibration signal, reflecting the severity of mechanical faults such as rotor imbalance or misalignment. Kurtosis, on the other hand, characterizes the steepness of the signal waveform; due to its sensitivity to impact signals, it is often used to characterize minor defects such as early localized spalling or pitting in bearings. For electrical and thermal information, features such as steady-state mean, peak-to-peak value, and rate of change are extracted to characterize the symmetry of the stator current and the heat dissipation efficiency of the cooling system.

[0038] Furthermore, traditional domain feature extraction often severs the intrinsic connections between the various subsystems of a synchronous condenser (e.g., an increase in reactive power is inevitably accompanied by an increase in stator current and a rise in winding temperature). To capture this cross-subsystem coupling relationship, this embodiment introduces a spatiotemporal attention autoencoder model to extract these features.

[0039] Specifically, the model's input is a multivariate time series matrix formed after the aforementioned standardization process. The rows of this matrix represent the time step within a sliding time window, and the columns represent the total number of multi-source sensor channels (covering electrical, mechanical, and thermal systems). Specifically, the standardized sampled values ​​from the sensors of each subsystem (electrical, mechanical, thermal, etc.) after time synchronization and alignment are arranged sequentially according to a preset channel order as the matrix's column vectors. This ensures that each row represents a snapshot of the global multiphysics state at the same sampling time, thus transforming the multi-source heterogeneous data into a two-dimensional tensor input that the model can process. The model's output is a hidden layer feature vector obtained after dimensionality reduction and compression by the encoder. This vector has a dimension much smaller than the dimension of the input multivariate time series matrix and contains core features representing the inherent coupling relationships of the multi-source data.

[0040] Specifically, the model includes an encoder and a decoder. The encoder comprises a one-dimensional convolutional layer and a multi-head self-attention layer. First, the one-dimensional convolutional layer performs sliding convolution along the sensor channel dimension to extract local spatial correlations between different sensor data at the same time. Subsequently, the multi-head self-attention layer captures long-distance evolutionary dependencies in the temporal dimension. Specifically, the input data is mapped to query vectors, key vectors, and value vectors through an attention mechanism. The similarity between the query vector at the current time step and the key vectors at all historical time steps is calculated to obtain attention weights. These weights are then used to perform a weighted summation of the value vectors at all historical time steps. It is particularly important to emphasize that the network structure of the model decoder is symmetrical to that of the encoder. Its role is to remap and expand the low-dimensional hidden layer feature vectors output by the encoder, attempting to reconstruct a matrix with the same dimension as the original input.

[0041] Specifically, the model employs unsupervised learning for offline training. A large amount of multi-source data from the camera adjustment system during its historical normal operation is collected as the training set. The training objective is to minimize the reconstruction error, i.e., by continuously adjusting the network parameters through backpropagation to minimize the difference (e.g., mean squared error) between the decoded matrix and the original input matrix.

[0042] It should be noted that by using a training method that first compresses and then reconstructs, the model is forced to filter out random noise in the intermediate hidden layer (i.e., the bottleneck layer), retaining only the core information that best represents the normal operation of the camera adjuster and the coupling relationship of the subsystems. Furthermore, during the online evaluation phase, only real-time data needs to be input into the trained encoder to directly extract the hidden layer output as the coupling features.

[0043] It should be noted that the dimension of the multivariate time series matrix is ​​the product of the time step within the sliding time window and the total number of multi-source sensor channels.

[0044] In a specific embodiment, the encoder in this invention sequentially includes: a one-dimensional convolutional layer (with a kernel size of 3 and a stride of 1, used to extract local features), a positional encoding layer (used to preserve the sequential information of the time series), a multi-head self-attention layer (with 4 attention heads, used to capture global temporal dependencies), a flattening layer, and a fully connected bottleneck layer. The output of the fully connected bottleneck layer is the dimensionality-reduced hidden layer feature vector, whose dimension is much smaller than that of the multivariate time series matrix. The decoder structure is symmetrical to the encoder, and sequentially includes: a fully connected layer (used to map the hidden layer vector back to a higher dimension), a reshaping layer (used to restore the time step and channel dimensions), a multi-head self-attention layer, and a one-dimensional transposed convolutional layer, ultimately outputting a reconstructed matrix with the same dimension as the input. During the offline training phase of the model, a multi-source data matrix under historical normal operating conditions is used as the training set. Mean squared error is used as the loss function to calculate the difference between the original input matrix and the reconstructed matrix. The Adam optimizer is used to update the network parameters. The initial learning rate of the model is set to 0.001, the batch size is set to 64, and iterative training is performed until the loss function converges or the preset maximum number of iterations is reached.

[0045] Furthermore, the extracted time-domain and frequency-domain features are concatenated with the extracted coupling features to obtain the initial feature set. Simultaneously, to avoid the curse of dimensionality causing a surge in computational complexity during subsequent baseline comparison, this embodiment also employs Principal Component Analysis (PCA) to reduce the dimensionality of the initial feature set. Specifically, the eigenvalues ​​and eigenvectors of the initial feature set covariance matrix are calculated. The eigenvalues ​​are sorted from largest to smallest, and the top K principal components whose cumulative variance contribution rate reaches a preset threshold (e.g., 90%~95%) are selected. The low-dimensional vector formed by these K principal components constitutes the final output features representing the operating state of the camera.

[0046] Specifically, the dimensionality reduction logic uses orthogonal transformation to convert the original high-dimensional correlated variables into a set of linearly uncorrelated low-dimensional variables (i.e., principal components). While retaining the maximum variance (i.e., main information) of the original data, redundant and highly collinear feature components are eliminated.

[0047] It should be noted that the initial feature set output after dimensionality reduction consists of multiple features characterizing the operating state of the synchronous condenser. This feature set not only contains intuitive physical state indications but also implies the coupling rules of the synchronous condenser subsystem, and can be directly used as the input basis for calculating the deviation from the operating baseline.

[0048] S3. For the current operating condition of the synchronous condenser, call the preset operating condition baseline corresponding to the operating condition, and compare multiple features of the synchronous condenser's operating status with the operating condition baseline to obtain the deviation of each feature.

[0049] It should be noted that traditional equipment condition assessment methods often use globally fixed alarm thresholds (e.g., specifying that an alarm is triggered when bearing vibration exceeds a certain fixed value). However, under different operating conditions (e.g., full-load phase advance and no-load operation), the electromagnetic force distribution, thermal field distribution, and mechanical stress state of a synchronous condenser are fundamentally different, leading to significant changes in the normal baseline values ​​and fluctuation ranges of various characteristic parameters. If a fixed threshold is used, false alarms may occur under heavy-load conditions, or missed alarms may occur under light-load conditions. Based on this, the present invention employs a dynamic baseline comparison mechanism.

[0050] Furthermore, for the current operating conditions of the synchronous condenser, the baseline expected value and baseline fluctuation range corresponding to each feature are obtained from the operating condition baseline.

[0051] Specifically, a baseline database of operating conditions is pre-built and stored within the system. This database is generated based on massive amounts of data accumulated over a long period of operation of the synchronous condenser under historical healthy conditions, and is categorized and statistically analyzed according to the aforementioned typical operating condition templates. When S1 outputs the operating condition identification result of the current synchronous condenser, the system uses this as an index to retrieve the baseline model matching that operating condition from the database. Simultaneously, for each feature extracted in S2 (whether physical features (time domain and frequency domain) or coupling features), the system extracts the following two key statistical benchmark parameters from the retrieved baseline operating conditions: The first parameter is the baseline expectation value (i.e. the statistical mean of historical health data under this operating condition). This expectation value represents the point at which the characteristic parameters of the synchronous condenser are most ideal and typical under a specific operating condition.

[0052] The second parameter is the baseline fluctuation range (i.e., the standard deviation or variance of historical health data under this operating condition). This fluctuation range represents the normal degree of dispersion of characteristic parameters under the current operating condition due to environmental noise, measurement errors, or permissible minor operating condition fluctuations, i.e., the reasonable fluctuation boundary allowed by the system.

[0053] Furthermore, the absolute difference between the current real-time value of each feature and the baseline expected value is calculated, and the absolute difference is normalized using the baseline fluctuation range to calculate the deviation of each feature.

[0054] Specifically, after obtaining the baseline parameters, it is also necessary to quantify the severity of the current real-time feature's deviation from the normal state. First, calculate the absolute difference between the current real-time value of the feature extracted by S2 and the corresponding expected baseline value. The absolute difference is used because regardless of whether the feature parameter is abnormally high (e.g., excessively high winding temperature) or abnormally low (e.g., insufficient cooling water flow), as long as it deviates from the expected value, it means that the equipment condition may deteriorate. Therefore, taking the absolute value can uniformly measure the magnitude of the deviation, without being affected by the direction of the deviation. Second, divide the calculated absolute difference by the corresponding baseline fluctuation range to complete the normalization process. The quotient of this division operation is the deviation degree of the feature. It is important to note that the normalization process here is different from the one mentioned above. Its significance lies in the fact that different feature parameters have vastly different sensitivities to anomalies and their own stability. For example, the temperature characteristic of a certain cooling water pipe is usually extremely stable (the baseline fluctuation range is very small), and even a deviation of only 2°C may indicate a serious blockage fault; while the stator current characteristic is affected by small fluctuations in the power grid, and its fluctuation is usually large (the baseline fluctuation range is large), so a deviation of 2A may be completely normal. Looking only at the absolute difference is clearly insufficient to accurately assess the severity of a fault. By dividing the absolute difference by the respective baseline fluctuation range, the absolute physical deviation is effectively transformed into a relative statistical anomaly, thereby eliminating the incomparability between different characteristics caused by differences in physical dimensions (e.g., temperature in °C, vibration in mm / s, current in A), orders of magnitude, and inherent fluctuation characteristics.

[0055] Furthermore, the calculated deviation is mapped onto a dimensionless standard scale. The larger the deviation value, the greater the extent to which the feature deviates from the normal fluctuation range, and the higher the probability of potential defects in the equipment.

[0056] Furthermore, combined with the special processing mechanism in S1, if the data within the current time window carries an external disturbance response label (i.e., a power grid disturbance event is identified), the system will still calculate the deviation according to the above logic. However, the high deviation calculated at this time will not be judged as internal equipment degradation, but will be stored separately as a quantitative evaluation indicator of the transient support capability of the synchronous condenser, thereby logically decoupling the assessment of the internal health status of the equipment from the assessment of the external transient response performance.

[0057] S4. The deviation of multiple features of the camera condenser's operating status is fused to generate an evaluation result that characterizes the overall operating status of the camera condenser.

[0058] It should be noted that in S3, the system has already acquired the deviation of each independent feature. However, a synchronous condenser is a highly complex electromechanical-thermal coupled system, and the deviation of a single feature is often insufficient to fully diagnose the true health status of the equipment. For example, a simple increase in bearing temperature could be due to a blockage in the lubrication circuit (mechanical fault), heat conduction from the stator windings (electrical / thermal fault), or even simply an increase in ambient temperature or drift of the temperature sensor itself. Therefore, the deviations of multiple features must be fused to eliminate the one-sidedness and uncertainty of single information.

[0059] Furthermore, before performing data fusion, it must be recognized that the reliability and importance of different sensors and features in characterizing device status vary significantly. If a simple arithmetic average is used for fusion, critical fault signals will be overwhelmed by a large number of normal signals. Therefore, this embodiment constructs a dynamic weight allocation mechanism.

[0060] Specifically, the initial evidence weights for each feature are determined by comprehensively evaluating them from the following three dimensions: Sensor quality dimensions are determined based on the sensor's historical calibration records, service life, and hardware accuracy calibration. A sensor that has just undergone calibration will naturally have higher data reliability than a sensor that has exceeded its service life, and therefore will be assigned a higher quality score.

[0061] Data sampling integrity dimension: determined by the ratio of the number of valid sampling points to the theoretically required number of sampling points within the current time window. If a sensor experiences severe packet loss or data interruption during communication, its integrity score will be significantly reduced, thus weakening the influence of this unreliable data in the final evaluation.

[0062] The correlation sensitivity dimension is derived from a pre-built historical fault mechanism analysis library and characterizes the sensitivity of specific features to common synchronous condenser faults. For example, the rotor grounding current characteristic is extremely sensitive to rotor insulation faults, and therefore is assigned an extremely high sensitivity score.

[0063] Furthermore, by weighting and summing the scores of the above three dimensions according to a preset proportional coefficient, the comprehensive evaluation score of each feature can be obtained; then, the proportion of the comprehensive evaluation score of each feature in the total score of all features is calculated, and this proportion is used as the initial evidence weight of that feature.

[0064] Furthermore, during weighted fusion, it detects whether there is a pre-defined conflict relationship between the deviations of different features. When a conflict relationship is detected, the evidence weights of each feature with a conflict relationship are dynamically adjusted according to the pre-defined conflict handling rules.

[0065] Specifically, in actual industrial settings, conflicting evidence frequently arises between multiple sensor sources. For example, a vibration sensor for a bearing might show extremely high deviation (alarm status), while the deviations of the temperature sensor, acoustic emission sensor, and lubricating oil pressure sensor at the same location are all close to zero (normal status). This high degree of conflict often indicates a hardware failure in the sensor displaying the abnormal reading (e.g., a loose probe), rather than a genuine equipment malfunction. Directly fusing these sensors without addressing the conflict can lead to the overall health index being dragged down by a single erroneous signal, triggering false alarms. Therefore, this invention employs the following logic for conflict detection and handling: First, the system compares the deviations of each feature within the same subsystem pairwise to calculate the degree of conflict between them. This degree of conflict is directly proportional to the difference in deviation between the two features and inversely proportional to their initial evidence weights. It's important to note that the greater the difference in deviation between two features and the lower their initial weights (i.e., the less reliable they are), the higher the probability of a substantial conflict between them. When the system detects that the degree of conflict between a feature and most other features within the same subsystem exceeds a preset conflict threshold, a conflict is identified. At this point, the conflict resolution rule is triggered: the initial evidence weights of these conflicting features are extracted and compared. For features with lower weights (i.e., relatively unreliable features), the system determines that they are likely abnormal interference items and reduces their initial evidence weight by multiplying it by a discount factor less than 1. This conflict resolution method, which respects the authority of high-weight features while respecting the majority, effectively shields the overall evaluation results from the interference of individual sensor failures, improving the algorithm's robustness in harsh industrial environments.

[0066] It should be added that the features With features The degree of conflict between The calculation formula is as follows:

[0067] in, For feature deviation, As the initial evidence weight, To prevent extremely small constants with denominators of zero, a value of 10 can be used. -5 Or 10 -6 A conflict is determined to exist when the conflict level exceeds a preset conflict threshold.

[0068] Furthermore, the deviation of each feature is stratified and weighted according to the adjusted evidence weights to generate an evaluation result that includes the overall health index and the subsystem health index.

[0069] Specifically, to provide maintenance personnel with accurate troubleshooting guidance, this embodiment is not limited to outputting a single overall system status. Furthermore, before performing deviation fusion calculations, it first checks the label attributes of the feature data within the current time window. For feature deviation data with external disturbance response labels, the system removes them from the current health index fusion calculation (or forces their deviation to zero), and only sends them to the transient performance evaluation module. By shielding external disturbance data at the source of the fusion calculation, interference from power grid transient processes on the internal health status evaluation of the equipment is completely avoided.

[0070] Furthermore, firstly, a subsystem health index is generated. The synchronous condenser is divided into multiple subsystems, including electrical, mechanical, thermal, and excitation subsystems. The system sets the initial state of each subsystem to full score (representing absolute health). Then, the system extracts all features belonging to the same subsystem, multiplies the deviation of these features by their corresponding adjusted evidence weights, and calculates the deduction value by combining this with a preset penalty amplification coefficient. Subtracting the sum of the deduction values ​​of all features within the subsystem from the full score yields the subsystem's health index. Next, the overall system health index is generated. The system pre-sets the importance weight of each subsystem within the overall system (e.g., for older synchronous condensers, the importance weights of the insulation and mechanical subsystems are usually set higher). The system then weights and sums the health indices of each subsystem with their corresponding importance weights to obtain the final overall system health index.

[0071] It should be noted that the final output evaluation results not only include an intuitive overall machine health index (used to assist in decision-making on whether an emergency shutdown is required), but also a set of subsystem health index radar charts (used to indicate specific deteriorated parts, such as clearly indicating that the mechanical subsystem health is low and the shaft system needs to be investigated).

[0072] It should be noted that the subsystem health index The calculation formula is as follows:

[0073] in, This represents the total number of features within the subsystem. For the adjusted weight of evidence, The preset penalty amplification factor is set according to the correlation between the feature and the serious fault. For example, the value is 1.0 for general features; for key fault features that directly characterize the damage of core components (such as rotor grounding current and severe shaft vibration), the value is greater than 1.5 to 2.0 to ensure that key defects can lower the health index and trigger high-level alarms.

[0074] Furthermore, the maintenance feedback information generated by the maintenance actions performed in response to the assessment results is obtained, and the maintenance feedback information is used to update the rules for determining the operating condition baseline or evidence weight.

[0075] It is important to emphasize that any evaluation model built based on historical data will experience model drift as the service life of the synchronous condenser increases and components naturally age. To enable the evaluation method to have full life-cycle adaptability, this embodiment introduces a self-learning mechanism based on maintenance feedback.

[0076] Specifically, when the system outputs a low health index and triggers a maintenance work order, on-site maintenance personnel will perform actual maintenance actions. After maintenance is completed, the maintenance personnel will enter the actual fault type, fault severity, and whether the system has any false alarms / missed alarms as maintenance feedback information into the system. When the system receives the feedback information, it compares it with historical assessment results: If the feedback indicates that the system has generated a false alarm (i.e., the system score is very low, but the actual equipment is functioning normally), it means that the model's tolerance for normal fluctuations is too low. In this case, the system will automatically widen the reasonable fluctuation range of the operating condition baseline for the corresponding feature in S3.

[0077] If the feedback indicates that the system has missed a report (i.e., the system score is normal, but a failure has actually occurred), it means that the model is not sensitive enough to key failure features. In this case, the system will automatically increase the correlation sensitivity score in S4 corresponding to the fault-sensitive feature, thereby increasing the evidence weight of that feature in future fusion calculations.

[0078] It should be noted that by iterating the above processing steps, the evaluation model of this invention can continuously self-correct and evolve in long-term operational practice, just like an experienced engineer, ensuring that the evaluation accuracy remains up-to-date.

[0079] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended 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 evaluating the operating status of a synchronous condenser based on multi-source information fusion, characterized in that, include: Acquire multi-source operating data of the synchronous condenser, and identify the current operating condition of the synchronous condenser based on the operating condition parameters in the multi-source operating data; Extract multiple features characterizing the operating state of the synchronous condenser from the multi-source operating data; For the current operating condition of the synchronous condenser, a preset operating condition baseline corresponding to the operating condition is invoked, and multiple features of the operating state of the synchronous condenser are compared with the operating condition baseline to obtain the deviation of each feature; The deviation of multiple features of the synchronous condenser's operating state is fused to generate an evaluation result characterizing the overall operating state of the synchronous condenser.

2. The method for evaluating the operating status of a synchronous condenser based on multi-source information fusion as described in claim 1, characterized in that, The multi-source operational data includes: At least two of the following categories: electrical information, mechanical information, thermal information, excitation information, cooling information, and insulation information.

3. The method for evaluating the operating status of a synchronous condenser based on multi-source information fusion as described in claim 1, characterized in that, The operating condition parameters include: At least one of reactive power, speed, start-stop phase, cooling status, and grid disturbance event; The identification result of the operating condition is used to select the operating condition baseline corresponding to the operating condition from multiple preset operating condition baselines.

4. The method for evaluating the operating status of a synchronous condenser based on multi-source information fusion as described in claim 3, characterized in that, When the operating condition parameters identify a power grid disturbance event, the characteristic deviation caused by the power grid disturbance event within a preset time window is marked as an external disturbance response, which is used to distinguish it from the deviation caused by changes in the internal state of the equipment when generating evaluation results.

5. The method for evaluating the operating status of a synchronous condenser based on multi-source information fusion as described in claim 1, characterized in that, The deviation of each feature is obtained, including: For the current operating conditions of the synchronous condenser, the baseline mean and baseline standard deviation corresponding to each feature are obtained from the operating condition baseline; Calculate the absolute difference between the current value of each feature and the baseline mean, and normalize the absolute difference using the baseline standard deviation to obtain the deviation of each feature.

6. The method for evaluating the operating status of a synchronous condenser based on multi-source information fusion as described in claim 1, characterized in that, Prior to the fusion process, the following is also included: Based on preset rules, determine the evidence weight corresponding to each feature; The fusion process includes weighted fusion of the deviation of each feature according to the evidence weights to generate an evaluation result of the overall operating status of the synchronous condenser.

7. The method for evaluating the operating status of a synchronous condenser based on multi-source information fusion as described in claim 1, characterized in that, The evaluation results include an overall machine health index and at least one subsystem health index; the subsystem health index is generated by fusing deviations corresponding to the characteristics of the subsystem, and the subsystem includes at least one of an electrical subsystem, a mechanical subsystem, a thermal subsystem, an excitation subsystem, an insulation subsystem, and a cooling subsystem.

8. The method for evaluating the operating status of a synchronous condenser based on multi-source information fusion as described in claim 6, characterized in that, The rules for determining the weight of evidence include: The weight value of each feature is determined by comprehensively considering the sensor quality of the data source corresponding to each feature, the integrity of data sampling, and the sensitivity of the feature to a specific fault type.

9. The method for evaluating the operating status of a synchronous condenser based on multi-source information fusion as described in claim 6, characterized in that, The method further includes: During weighted fusion, it is necessary to detect whether there is a pre-defined conflict relationship between the deviations of different features; When a conflict is detected, the evidence weights of each feature with the conflict are adjusted according to preset conflict handling rules. The conflict handling rules include discounting the weights of one or more features whose evidence weights are not the highest among the features with conflicting relationships, or reducing their contribution in the fusion calculation.

10. The method for evaluating the operating status of a synchronous condenser based on multi-source information fusion as described in claim 1 or 6, characterized in that, The method further includes: Obtain maintenance feedback information generated by maintenance actions performed in response to the evaluation results; The maintenance feedback information is used to update the rules for determining the operating condition baseline or the evidence weight.

Images