Phase modifier operation state evaluation method and system based on multi-source sensing data fusion

Abstract

The invention provides a phase modifier operation state evaluation method and system based on multi-source sensing data fusion, and relates to the technical field of phase modifier operation state evaluation, and the method comprises the steps: obtaining multi-source sensing data; obtaining standardized pre-processed data; the correlation degree between each extracted feature and the fault state of the phase modifier is calculated through mutual information entropy, and a key feature set is formed; calculating the influence weight of each parameter in the energy conversion process through finite element simulation; constructing a hierarchical fusion framework by combining the key feature set, the voiceprint-vibration fusion features and an inter-device parameter incidence matrix to obtain feature layers; a weighted fusion algorithm is adopted for each feature layer to calculate a fusion sub-vector, and then a global fusion feature vector is obtained; inputting the global fusion feature vector into a pre-trained phase modifier state evaluation model; and outputting the operation state result of the phase modifier. According to the invention, electrical, mechanical, thermal and cooling system operation state monitoring of the phase modifier is comprehensively covered, and technical support is provided for lean operation and maintenance of the phase modifier.

Description

Technical Field

[0001] This invention relates to the field of synchronous condenser operation status assessment technology, and more specifically, to a method and system for assessing synchronous condenser operation status based on multi-source sensor data fusion. Background Technology

[0002] Synchronous condensers, as key equipment in power systems, play a crucial role in maintaining stable power system operation. They can supply or absorb reactive power to the system, effectively improving the power factor, reducing network losses, and significantly improving network voltage and power quality. In long-distance transmission lines, synchronous condensers can maintain stable grid voltage by adjusting their own excitation current according to the grid load. When the grid load is heavy, they operate under overexcitation to reduce the lagging reactive current component in the transmission line, thereby reducing line voltage drop; when the transmission line is lightly loaded, they operate under underexcitation to absorb lagging reactive current and prevent grid voltage from rising. Accurately assessing the operational status of synchronous condensers is crucial for ensuring the safe, stable, and economical operation of the power system. A malfunction in a synchronous condenser can lead to localized voltage fluctuations in the power grid, and in severe cases, even cause widespread blackouts, resulting in significant losses to social production and people's lives. For example, during peak electricity consumption periods, if a synchronous condenser cannot provide adequate reactive power support, it can cause a sudden drop in grid voltage, affecting the normal operation of various electrical equipment, forcing factory production lines to shut down, and causing traffic light malfunctions, resulting in considerable inconvenience to society. Therefore, timely and accurate monitoring of the operational status of synchronous condensers, early detection of potential faults, and implementation of corresponding measures are essential for ensuring a reliable power supply from the power system. Summary of the Invention

[0003] The purpose of this invention is to provide a method, system, device, and readable storage medium for evaluating the operating status of a synchronous condenser based on multi-source sensor data fusion, in order to improve the aforementioned problems. To achieve the above objective, the technical solution adopted by this invention is as follows: Firstly, this application provides a method for evaluating the operating status of a synchronous condenser based on multi-source sensor data fusion, including: Multi-source sensor data is acquired through sensors in the stator, rotor, excitation system, and cooling system of the synchronous condenser. A spatiotemporal alignment model is constructed based on the multi-source sensor data to unify data from different acquisition frequencies to the same time dimension. An adaptive filtering threshold is then generated by combining the feature library of normal operation data of the synchronous condenser to filter out impulse noise, harmonic interference, and environmental noise in the data. The dimensionality of each parameter is unified through range standardization to obtain standardized preprocessed data. For standardized preprocessed data, specific analytical methods are employed to extract state features. These include wavelet packet transform of electrical quantity data to extract frequency band energy proportions and characteristic frequency amplitudes; trend analysis of temperature data to extract temperature change rates and gradient differences; Hilbert-Huang transform of vibration data to extract instantaneous frequencies and amplitudes; Mel frequency cepstral coefficient analysis of acoustic signature data to extract spectral envelopes and resonant frequencies; and statistical analysis of cooling system parameters to extract their mean and variance. After feature extraction of various data types, the correlation between each extracted feature and the synchronous condenser fault state is calculated using mutual information entropy. Features with correlation reaching a preset threshold are selected to form a key feature set, where the synchronous condenser fault states include stator winding faults and bearing wear faults. Vibration and acoustic signature features are extracted from the key feature set, and their temporal correlation and frequency coupling degree are calculated. Based on the coupling results, a cross-modal feature mapping relationship is constructed to generate acoustic signature-vibration fusion features. A physical correlation model is constructed based on the phase condenser energy transfer chain, which includes an electrical-mechanical-thermal energy conversion path. Electrical quantity features, temperature features, cooling system features, and acoustic signature-vibration fusion features are incorporated into this physical correlation model. The influence weight of each parameter in the energy conversion process is calculated through finite element simulation, and an inter-device parameter correlation matrix is ​​generated based on the weight. A hierarchical fusion framework is constructed by combining the key feature set, acoustic signature-vibration fusion features, and inter-device parameter correlation matrix, resulting in various feature layers, which are divided into an electrical feature layer, a temperature feature layer, an acoustic signature-vibration fusion layer, and a cooling feature layer. A weighted fusion algorithm is used to calculate the fusion sub-vector for each feature layer. The weights of the weighted fusion algorithm are determined based on the fault contribution of each feature in historical fault cases. The correlation coefficient between the fusion sub-vector of each feature layer and the overall operating status of the camera is calculated through an attention mechanism. The inter-layer weights are allocated according to the correlation coefficients, and the fusion sub-vectors are weighted and aggregated to obtain the global fusion feature vector. The global fusion feature vector is input into the pre-trained synchronous condenser status assessment model, which is built based on a deep belief network and trained with normal operation data, simulated fault data and historical fault data of the synchronous condenser until the loss function reaches a preset standard. The synchronous condenser status assessment model is used to identify the status of the global fusion feature vector and output the synchronous condenser operation status result, which includes normal status, abnormal status and fault type. The fault type includes stator winding fault, bearing wear fault, cooling system blockage fault and bearing noise fault.

[0004] Preferably, for the standardized preprocessed data, specific analysis methods are used to extract state features, including wavelet packet transform of electrical quantity data to extract frequency band energy proportions and characteristic frequency amplitudes; trend analysis of temperature data to extract temperature change rates and gradient differences; Hilbert-Huang transform of vibration data to extract instantaneous frequencies and amplitudes; Mel frequency cepstral coefficient analysis of acoustic fingerprint data to extract spectral envelopes and resonant frequencies; statistical analysis of cooling system parameters to extract their mean and variance; after completing feature extraction for various types of data, the correlation between each extracted feature and the synchronous condenser fault state is calculated using mutual information entropy, and features with a correlation reaching a preset threshold are selected to form a key feature set, wherein the synchronous condenser fault state includes stator winding faults and bearing wear faults, including: Feature extraction was performed on different types of data in the standardized preprocessed data matrix. For electrical quantity data, wavelet packet transform was used for frequency domain analysis, with a decomposition level of 3 layers and a db4 wavelet basis selected to extract the energy proportion of frequency bands and the amplitude of characteristic frequencies. For temperature data, sliding window trend analysis was used, with a window size of 10 minutes, to extract the temperature change rate and temperature gradient difference. For vibration data, Hilbert-Huang transform was used for time-frequency domain analysis, with a decomposition level of 5 layers, to extract instantaneous frequencies and instantaneous amplitudes. For acoustic signature data, Mel frequency cepstral coefficients were used for acoustic feature analysis, with 24 Mel filter banks and a cepstral coefficient order of 12, to extract the spectral envelope and formant parameters, including the frequencies of the first to third formants. For cooling system parameters, statistical analysis methods were used to calculate the parameter distribution characteristics, with a time window of 1 minute, to extract the mean and variance, forming an initial feature set. Based on the initial feature set, a fault correlation matrix is ​​constructed. The row dimension of the matrix is ​​the total number of features in the initial feature set, and the column dimension is the type of synchronous condenser fault state. The synchronous condenser fault states include stator winding faults and bearing wear faults. The correlation between each extracted feature and the synchronous condenser fault state is calculated using mutual information entropy. A correlation threshold is set, and features with a correlation degree greater than or equal to the correlation threshold are selected to form a key feature set. The correlation threshold is determined based on the historical fault data of the synchronous condenser and the evaluation accuracy requirements.

[0005] Preferably, the process involves extracting vibration features and acoustic signature features from the key feature set, calculating their temporal correlation and frequency domain coupling, constructing a cross-modal feature mapping relationship based on the coupling results, and generating acoustic signature-vibration fusion features. A physical correlation model is constructed based on the synchronous condenser energy transfer chain, which includes an electrical-mechanical-thermal energy conversion path. Electrical quantity features, temperature features, cooling system features, and acoustic signature-vibration fusion features are incorporated into this physical correlation model. The influence weights of each parameter in the energy conversion process are calculated through finite element simulation, and a parameter correlation matrix between devices is generated based on these weights. A hierarchical fusion framework is constructed by combining the key feature set, acoustic signature-vibration fusion features, and the parameter correlation matrix between devices. The resulting feature layers are divided into an electrical feature layer, a temperature feature layer, an acoustic signature-vibration fusion layer, and a cooling feature layer, including: Vibration features and acoustic features are extracted from the key feature set. The vibration features include instantaneous frequency and instantaneous amplitude, and the acoustic features include spectral envelope and formant parameters. Based on the coupling results, a cross-modal feature mapping relationship is constructed to generate acoustic-vibration fusion features, where the vibration features include instantaneous frequency and instantaneous amplitude, and the acoustic features include spectral envelope and formant parameters. A physical correlation model is constructed based on the energy transfer chain of the synchronous condenser. Through the electromagnetic induction equation, friction loss equation, and cooling system heat exchange equation, the mapping between electrical quantities and electromagnetic torque, the correlation between mechanical energy and thermal energy, and the coupling between thermal energy and cooling parameters are established to form a closed-loop conversion model for the energy transfer of the synchronous condenser. The acoustic-vibration fusion features, key features, electrical quantity features, temperature features, and cooling system features are embedded into the model according to the energy conversion nodes. Among them, the electrical quantity features correspond to the electromagnetic energy link, the acoustic-vibration fusion features correspond to the mechanical energy link, the temperature features correspond to the thermal energy accumulation link, and the cooling system features correspond to the thermal energy dissipation link. By changing the value of each parameter individually through finite element simulation, the change in the total energy of the system is recorded, the influence weight of each parameter is calculated, and then the parameter correlation matrix between devices is generated based on the weight matrix. A hierarchical fusion framework is constructed by combining key feature sets, generated acoustic-vibration fusion features, and inter-device parameter correlation matrices. Feature groups corresponding to the diagonal elements of the correlation matrix are extracted. According to the energy transfer chain conversion order, electrical quantity features directly related to electromagnetic energy conversion, temperature features related to heat accumulation, acoustic-vibration fusion features, and cooling system features related to heat dissipation are respectively assigned to the electrical feature layer, temperature feature layer, acoustic-vibration fusion layer, and cooling feature layer. Cross-layer features are verified twice by cosine similarity to determine that features within the same feature layer belong to the same energy conversion stage. The features of each layer are stored in matrix form.

[0006] Preferably, a weighted fusion algorithm is used to calculate the fused sub-vector for each feature layer, wherein the weights of the weighted fusion algorithm are determined based on the fault contribution of each feature in historical fault cases; the correlation coefficient between the fused sub-vector of each feature layer and the overall operating state of the camera is calculated through an attention mechanism, and the inter-layer weights are allocated according to the correlation coefficients. The fused sub-vectors are then weighted and aggregated to obtain a global fused feature vector, which includes: Based on the output electrical feature layer, temperature feature layer, acoustic-vibration fusion layer, and cooling feature layer, intra-layer weighted fusion is performed for the features of each layer. This includes using fault mode effect analysis to statistically analyze historical fault cases and determine the fault contribution coefficient of different features within each layer. In the electrical feature layer, the contribution of the mid-frequency energy proportion is higher than that of the feature frequency amplitude; in the temperature feature layer, the contribution of the temperature gradient difference is higher than that of the temperature change rate; in the acoustic-vibration fusion layer, the contribution of the acoustic pattern-vibration coupling feature is higher than that of the single-mode feature; and in the cooling feature layer, the contribution of variance is higher than that of the mean. Using the fault contribution coefficient of each feature as the weight, a weighted summation algorithm is used to calculate the fused sub-vector of each layer. An attention mechanism is introduced to evaluate the importance of the obtained fusion sub-vectors of each layer. A health label vector is constructed based on the historical health status data of the camera condenser. The label value is 1 for a healthy state and 0 for a fault state. The Pearson correlation coefficient between the fusion sub-vectors of each layer and the health label vector is calculated to quantify the ability of each layer feature to represent the overall operating status of the camera condenser. The obtained fusion sub-vectors of each layer are normalized in dimension and uniformly adjusted to 64-dimensional vectors. The normalized fusion sub-vectors are then weighted and aggregated according to the obtained attention weights to generate a global fusion feature vector. The global fusion feature vector includes multi-dimensional state information of the camera's electrical, mechanical, thermal, and cooling systems.

[0007] Preferably, the global fused feature vector is input into a pre-trained camera condenser state evaluation model, wherein the camera condenser state evaluation model is constructed based on a deep belief network and trained with camera condenser normal operation data, simulated fault data, and historical fault data until the loss function reaches a preset standard; the camera condenser state evaluation model performs state identification on the global fused feature vector and outputs the camera condenser operating state result, including: The global fusion feature vector is dimension-validated according to the input dimension requirements of the camera adjustment state evaluation model. After confirming that the vector format is consistent with the model input interface, it is imported into the feature input layer of the pre-trained camera adjustment state evaluation model. The camera adjustment state evaluation model is based on a deep belief network, which contains three restricted Boltzmann machine pre-training layers and one backpropagation fine-tuning layer. The number of nodes in the pre-training layer is set to 128, 64 and 32 respectively. The initial parameters of the network are optimized by the contrastive divergence algorithm, and the weight matrix is ​​adjusted by the backpropagation algorithm until the model converges. The camera facilitator state assessment model performs multi-layer nonlinear transformation and feature mapping on the input global fusion feature vector, and then realizes state recognition through the softmax classifier in the output layer, finally outputting the camera facilitator operation status result.

[0008] Secondly, this application also provides a system for evaluating the operating status of a synchronous condenser based on multi-source sensor data fusion, including: The acquisition module is used to acquire multi-source sensor data through sensors in the stator, rotor, excitation system, and cooling system of the synchronous condenser; it constructs a spatiotemporal alignment model based on the multi-source sensor data to unify data from different acquisition frequencies to the same time dimension; it then generates an adaptive filtering threshold by combining the feature library of normal operation data of the synchronous condenser to filter out impulse noise, harmonic interference, and environmental noise in the data; and it completes the dimensional unification of each parameter through range standardization to obtain standardized preprocessed data. Extraction Module: This module uses specific analytical methods to extract state features from standardized preprocessed data. These include wavelet packet transform of electrical quantity data to extract frequency band energy proportions and characteristic frequency amplitudes; trend analysis of temperature data to extract temperature change rates and gradient differences; Hilbert-Huang transform of vibration data to extract instantaneous frequencies and amplitudes; Mel frequency cepstral coefficient analysis of acoustic signature data to extract spectral envelopes and resonant frequencies; and statistical analysis of cooling system parameters to extract their mean and variance. After feature extraction for various data types, the module calculates the correlation between each extracted feature and the synchronous condenser fault state using mutual information entropy, and selects features with correlation reaching a preset threshold to form a key feature set. The synchronous condenser fault states include stator winding faults and bearing wear faults. The calculation module is used to extract vibration and acoustic signature features from the key feature set, calculate their temporal correlation and frequency coupling degree, construct cross-modal feature mapping relationship based on the coupling results, and generate acoustic signature-vibration fusion features. It also constructs a physical correlation model based on the phase shifter energy transfer chain, which includes an electrical-mechanical-thermal energy conversion path. Electrical quantity features, temperature features, cooling system features, and acoustic signature-vibration fusion features are incorporated into this physical correlation model. Finite element simulation is used to calculate the influence weight of each parameter in the energy conversion process, and a parameter correlation matrix between devices is generated based on the weights. Finally, a hierarchical fusion framework is constructed by combining the key feature set, acoustic signature-vibration fusion features, and the parameter correlation matrix between devices, resulting in various feature layers, each divided into an electrical feature layer, a temperature feature layer, an acoustic signature-vibration fusion layer, and a cooling feature layer. The aggregation module is used to calculate the fusion sub-vectors for each feature layer using a weighted fusion algorithm. The weights of the weighted fusion algorithm are determined based on the fault contribution of each feature in historical fault cases. The correlation coefficient between the fusion sub-vectors of each feature layer and the overall operating status of the camera is calculated through an attention mechanism. The inter-layer weights are allocated according to the correlation coefficients, and the fusion sub-vectors are weighted and aggregated to obtain the global fusion feature vector. Training module: This module is used to input the global fused feature vector into the pre-trained synchronous condenser status assessment model. The synchronous condenser status assessment model is built based on a deep belief network and trained with normal operation data, simulated fault data, and historical fault data of the synchronous condenser until the loss function reaches a preset standard. The synchronous condenser status assessment model performs status identification on the global fused feature vector and outputs the synchronous condenser operation status results. The operation status results include normal status, abnormal status, and fault type. The fault types include stator winding fault, bearing wear fault, cooling system blockage fault, and bearing noise fault.

[0009] Thirdly, this application also provides a device for evaluating the operating status of a synchronous condenser based on multi-source sensor data fusion, comprising: Memory, used to store computer programs; A processor is used to implement the steps of the method for evaluating the operating status of a camera condenser based on multi-source sensor data fusion when executing the computer program.

[0010] Fourthly, this application also provides a readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described method for evaluating the operating status of a camera condenser based on multi-source sensor data fusion.

[0011] The beneficial effects of this invention are as follows: Traditional methods for assessing the operating status of synchronous condensers are mostly based on single data points or simple algorithms. For example, early methods often relied on monitoring electrical quantities such as voltage and current to determine the operating status of the synchronous condenser, but this approach has significant limitations. Single electrical quantity data cannot comprehensively reflect the complex operating conditions of the synchronous condenser. Faults in internal mechanical components, such as bearing wear or rotor imbalance, are difficult to detect in a timely manner using only electrical quantity data. Moreover, this method based on simple threshold judgments is not sensitive to early fault characteristics. It often only becomes apparent when the fault has developed to a certain extent and the relevant parameters exceed the set thresholds, by which time significant damage to the equipment may have already been caused, increasing maintenance costs and power outage time. Furthermore, traditional assessment methods are highly susceptible to interference. In actual operating environments, synchronous condensers may be affected by various factors such as electromagnetic interference and changes in ambient temperature. These interferences can cause fluctuations or deviations in monitoring data, thus affecting the accuracy of the assessment results. For example, the start-up and shutdown of nearby large electrical equipment may generate electromagnetic interference, causing abnormal fluctuations in the collected electrical quantity data. If traditional methods are used to assess based on this interfered data, incorrect conclusions may be drawn, leading to misjudgments of the synchronous condenser's operating status.

[0012] This invention effectively addresses the problems of poor data quality and inconsistent dimensions by employing preprocessing techniques such as spatiotemporal alignment of multi-source sensor data, adaptive filtering, and range standardization, thus laying a reliable data foundation for assessing the operational status of synchronous condensers. Relying on feature engineering techniques such as categorical feature extraction and mutual information entropy feature screening, it achieves accurate capture of key status features of synchronous condensers, reducing the interference of redundant information on the assessment process and improving the fault-specificity of feature representation. By employing multimodal fusion techniques including cross-modal feature mapping, energy transfer chain physical correlation models, and hierarchical fusion frameworks, it strengthens the correlation between various features, significantly improving the comprehensive representation capability of the overall operational status of synchronous condensers. Simultaneously, by utilizing hierarchical weighted fusion, attention mechanism aggregation, and deep belief network assessment models and aggregation methods, it achieves high-precision assessment of the operational status of synchronous condensers and accurate fault identification, providing an effective basis for operation and maintenance decisions and reducing the cost losses caused by blind maintenance.

[0013] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing embodiments of the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description

[0014] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a schematic diagram of the process for evaluating the operating status of a synchronous condenser based on multi-source sensor data fusion, as described in this embodiment of the invention. Figure 2 This is a schematic diagram of the structure of the synchronous condenser operation status evaluation system based on multi-source sensor data fusion as described in this embodiment of the invention; Figure 3This is a schematic diagram of the structure of the synchronous condenser operation status evaluation device based on multi-source sensor data fusion as described in an embodiment of the present invention.

[0016] In the diagram: 701, Acquisition module; 702, Extraction module; 703, Calculation module; 704, Aggregation module; 705, Training module; 800, Camera condenser operation status evaluation device based on multi-source sensor data fusion; 801, Processor; 802, Memory; 803, Multimedia component; 804, I / O interface; 805, Communication component. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0018] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0019] Example 1: This embodiment provides a method for evaluating the operating status of a synchronous condenser based on multi-source sensor data fusion.

[0020] See Figure 1 The figure shows that the method includes steps S100, S200, S300 and S400.

[0021] S100: Multi-source sensor data is acquired through sensors in the stator, rotor, excitation system, and cooling system of the synchronous condenser. Based on the multi-source sensor data, a spatiotemporal alignment model is constructed to unify data from different acquisition frequencies to the same time dimension. Then, an adaptive filtering threshold is generated by combining the feature library of normal operation data of the synchronous condenser to filter out impulse noise, harmonic interference, and environmental noise in the data. The dimensional unification of each parameter is completed through range standardization to obtain standardized preprocessed data.

[0022] It should be noted that the multi-source sensor data includes electrical quantity data, temperature data, vibration data, acoustic data, and cooling system parameters. The electrical quantity data includes stator voltage, stator current, excitation voltage, and excitation current; the temperature data includes stator core temperature, stator winding temperature, bearing temperature, and cooling medium temperature; the vibration data includes shaft displacement and bearing acceleration; the acoustic data includes stator noise sound pressure level and bearing noise frequency; and the cooling system parameters include cooling medium pressure and cooling medium flow rate.

[0023] Understandably, in this step, multi-source sensor data is simultaneously collected through a distributed sensor array deployed on the stator, rotor, excitation system, and cooling system of the synchronous condenser. The data is represented by a spatiotemporal scalar matrix D, where the row dimension of D corresponds to the number of sampling times, and the column dimension corresponds to the number of sensors, covering electrical quantity data, temperature data, vibration data, acoustic fingerprint data, and cooling system parameters. Different data types are collected at differentiated frequencies based on their physical characteristics; electrical quantity data, temperature data, vibration data, acoustic fingerprint data, and cooling system parameters are collected at their respective frequencies.

[0024] A spatiotemporal alignment model is constructed based on the mechanical cycle of the synchronous condenser. Cubic spline interpolation is used to achieve resampling of different frequency data. An adaptive threshold is generated by combining the normal operation feature library, which contains the mean and standard deviation of each parameter when the synchronous condenser is running without faults. Impulse noise is filtered out by improving Kalman filtering to obtain the denoised spatiotemporal aligned data.

[0025] An improved range normalization is used to map the denoised spatiotemporally aligned data to the [0,1] interval, resulting in a normalized preprocessing matrix. This matrix serves as the input data for step two. The mathematical expression for the improved range normalization is as follows: In the formula, For the standardization results, The original data is provided, where α and β are anti-saturation coefficients. and These are the minimum and maximum values ​​of the parameter, respectively.

[0026] S200. For standardized preprocessed data, specific analysis methods are used to extract state features. These include wavelet packet transform of electrical quantity data to extract frequency band energy proportion and characteristic frequency amplitude; trend analysis of temperature data to extract temperature change rate and gradient difference; Hilbert-Huang transform of vibration data to extract instantaneous frequency and amplitude; Mel frequency cepstral coefficient analysis of acoustic data to extract spectral envelope and resonant frequency; and statistical analysis of cooling system parameters to extract their mean and variance. After feature extraction of various types of data, the correlation between each extracted feature and the synchronous condenser fault state is calculated using mutual information entropy. Features with a correlation degree reaching a preset threshold are selected to form a key feature set, where the synchronous condenser fault states include stator winding faults and bearing wear faults.

[0027] It is important to note that in the monitoring of electrical systems, electrical quantity data carries crucial information about the system's operating status. To accurately capture electrical anomalies, wavelet packet transform is used to perform frequency domain analysis on the electrical quantity data. As a powerful time-frequency analysis tool, wavelet packet transform can meticulously decompose signals across different frequency bands, much like breaking down a complex musical melody into combinations of different notes, allowing for in-depth exploration of its internal structure. For electrical quantity data, the energy distribution and characteristic frequency amplitudes of each frequency band exhibit relatively stable patterns under normal operating conditions. Once an anomaly occurs in the electrical system, such as a short circuit or equipment overload, the energy proportion of these electrical quantity data in specific frequency bands will change significantly, and the characteristic frequency amplitude will deviate from the normal range. By keenly capturing these changes, abnormal conditions in the electrical system can be detected in a timely manner, providing crucial evidence for fault diagnosis and early warning. For example, in the monitoring of power transformers, when faults such as partial discharge occur in the internal windings, the energy proportion of its electrical quantity data in certain high-frequency bands will suddenly increase, and the characteristic frequency amplitude will also exhibit abnormal fluctuations. By extracting these features using wavelet packet transform, it is possible to quickly determine whether there is a potential fault in the transformer.

[0028] It is understood that this step includes S201, S202, and S203, wherein: S201. Feature extraction is performed on different types of data in the standardized preprocessed data matrix. For electrical quantity data, wavelet packet transform is used for frequency domain analysis, with a decomposition level of 3 layers. The db4 wavelet basis is selected to extract the frequency band energy ratio and characteristic frequency amplitude. For temperature data, sliding window trend analysis is used, with the window size set to 10 minutes, to extract the temperature change rate and temperature gradient difference. For vibration data, Hilbert-Huang transform is used for time-frequency domain analysis, with a decomposition level of 5 layers, to extract instantaneous frequency and instantaneous amplitude. For acoustic fingerprint data, Mel frequency cepstral coefficients are used for acoustic feature analysis. The number of Mel filter banks is 24, and the order of the cepstral coefficients is 12, to extract the spectral envelope and formant parameters. The formant parameters include the frequencies of the first to third formants. For cooling system parameters, statistical analysis methods are used to calculate the parameter distribution characteristics, with the time window set to 1 minute, to extract the mean and variance, forming an initial feature set. S202. Based on the initial feature set, construct a fault correlation matrix. The row dimension of the matrix represents the total number of features in the initial feature set, and the column dimension represents the types of synchronous condenser fault states. The synchronous condenser fault states include stator winding faults and bearing wear faults. The correlation degree between each extracted feature and the synchronous condenser fault state is calculated using mutual information entropy. The calculation formula is as follows: In the formula, Features Fault status The degree of correlation, For the joint probability distribution of features and faults, Features Marginal probability distribution, Fault status The marginal probability distribution; S203. Set a correlation threshold and select features with a correlation degree greater than or equal to the correlation threshold to form a key feature set. The correlation threshold is determined based on the historical fault data of the synchronous condenser and the evaluation accuracy requirements.

[0029] Temperature data is a key indicator reflecting the thermal state of an equipment; subtle changes may foreshadow potential malfunctions. Trend analysis tracks parameter changes in temperature data. This curve allows for the sensitive detection of abnormal temperature fluctuations. For example, when a component of the equipment is overloaded or experiences poor heat dissipation, the temperature rises rapidly, and the rate of temperature change accelerates, much like a person suddenly developing a fever and experiencing a rapid increase in body temperature. Simultaneously, the temperature gradient difference also provides important information. Under normal circumstances, the temperature gradient between different parts of the equipment is relatively stable. Once an anomaly occurs in a certain part, the temperature gradient difference will change significantly. This is similar to a uniformly heated room; if the temperature in one corner suddenly differs from the rest, it indicates a potential problem in that corner.

[0030] Furthermore, vibration data is a crucial indicator of the mechanical state of machinery during operation, with different vibration modes often corresponding to different types of mechanical faults. First, empirical mode decomposition (EMD) adaptively decomposes complex vibration signals into a series of intrinsic mode functions (IMFs). These IMF components represent the signal's characteristics at different time scales, much like breaking down a complex symphony into segments played by different instruments, each segment possessing unique frequency and amplitude characteristics. Then, a Hilbert transform is applied to each IMF component to obtain the signal's instantaneous frequency and amplitude. Under normal operating conditions, the vibration signal of mechanical equipment has a relatively stable range of instantaneous frequency and amplitude. Once mechanical components experience faults such as wear, loosening, or imbalance, these instantaneous frequencies and amplitudes will change significantly. By accurately extracting these changing features, it is possible to accurately determine whether the mechanical equipment has a fault, as well as the type and severity of the fault. For example, in the fault diagnosis of rotating machinery, when bearings experience ball wear, the instantaneous frequency of the vibration signal will fluctuate with specific frequency components, and the instantaneous amplitude will also increase. Using the Hilbert-Huang transform to extract these features allows for the timely detection of potential bearing faults.

[0031] When a device malfunctions with unusual noises, its voiceprint characteristics change significantly. In the actual analysis process, the voiceprint data is first pre-emphasized to boost high-frequency components and compensate for the microphone's attenuation of high-frequency signals, essentially "brightening" the sound and making subtle high-frequency information clearer. Next, the continuous voiceprint signal is segmented into short frames, typically 20-40ms long, and a Hanning window is added to each frame to reduce spectral leakage. This step is like cutting continuous sound into smaller segments, facilitating more detailed analysis of each segment's characteristics. Then, the spectral energy of each frame is calculated using a Fast Fourier Transform, converting the time-domain signal to a frequency-domain signal to show the energy distribution of the sound at different frequencies. Afterward, a Mel filter bank is designed, typically containing 20-40 triangular filters. These filters cover the frequency range sensitive to the human ear (0-8kHz), acting like a "frequency sieve," allowing only sounds within specific frequency ranges to pass through. After processing by the Mel filter bank, the output is logarithmically calculated to simulate the logarithmic perception of loudness by the human ear, as the perception of loudness is not linear but logarithmic. Finally, the output of the filter bank is converted into coefficients through discrete cosine transform, compressing the dimensions and typically retaining the first 12-13 coefficients. These coefficients are the Mel-frequency cepstral coefficients. Different sounds have different formant characteristics. When equipment is operating normally, the formant frequencies of the sound it emits are within a relatively stable range. Once an abnormal noise fault occurs, the formant frequencies will shift; for example, some formant frequencies may rise or fall, or new formant frequencies may appear. By accurately extracting these spectral envelopes and formant parameter characteristics, abnormal noise faults in equipment can be detected in a timely manner, providing important information for equipment maintenance and troubleshooting. For example, in transformer fault diagnosis, when internal faults such as winding deformation or partial discharge occur, the spectral envelope of its acoustic data will show abnormal fluctuations, and the frequencies of the first to third formant peaks will also change. By using Mel-frequency cepstral coefficients to extract these features, it is possible to quickly determine whether there are potential faults in the transformer.

[0032] Changes in cooling system parameters, much like fluctuations in human health indicators, directly reflect the operating status and cooling efficiency of the cooling system. Variance measures the dispersion of cooling system parameters, i.e., the fluctuation of the data, just as the dispersion of class exam scores reflects the stability of student performance. Smaller variance indicates relatively stable cooling system parameters and good operating status; while larger variance means greater parameter fluctuations, potentially indicating underlying problems. For example, an increased variance in the cooling medium temperature indicates more severe temperature fluctuations, which may be due to uneven heat dissipation in the cooling system or a malfunction in the control system, failing to effectively regulate the temperature. This will adversely affect the stability and reliability of the equipment. Therefore, by accurately extracting the mean and variance, we can comprehensively and accurately assess the operating status and cooling efficiency of the cooling system, promptly identify potential problems, and take corresponding measures, providing a solid guarantee for the safe and stable operation of the equipment.

[0033] S300. Extract vibration and acoustic signature features from the key feature set, calculate their temporal correlation and frequency coupling degree, construct cross-modal feature mapping relationship based on the coupling results, and generate acoustic signature-vibration fusion features; construct a physical correlation model based on the phase condenser energy transfer chain, which includes an electrical-mechanical-thermal energy conversion path, and incorporate electrical quantity features, temperature features, cooling system features, and acoustic signature-vibration fusion features into this physical correlation model, calculate the influence weight of each parameter in the energy conversion process through finite element simulation, and generate a parameter correlation matrix between devices based on the weights; construct a hierarchical fusion framework by combining the key feature set, acoustic signature-vibration fusion features, and device parameter correlation matrix, and divide the obtained features into an electrical feature layer, a temperature feature layer, an acoustic signature-vibration fusion layer, and a cooling feature layer; calculate the fusion sub-vector for each feature layer using a weighted fusion algorithm, where the weights of the weighted fusion algorithm are determined based on the fault contribution of each feature in historical fault cases; calculate the correlation coefficient between the fusion sub-vector of each feature layer and the overall operating state of the phase condenser through an attention mechanism, allocate inter-layer weights according to the correlation coefficients, and perform weighted aggregation on each fusion sub-vector to obtain a global fusion feature vector.

[0034] It is understood that step S300 includes S301, S302, S303, and S304, wherein: Vibration features and acoustic signature features are extracted from the key feature set. Vibration features include instantaneous frequency and instantaneous amplitude, while acoustic signature features include spectral envelope and formant parameters. The formula for calculating the temporal correlation between the two is as follows: In the formula, Vibration characteristics With voiceprint characteristics covariance, and These are the standard deviations of the two, The time-domain correlation coefficient between the two is denoted as . The formula for calculating the frequency domain coupling degree is as follows: In the formula, The cross-power spectral density of the two is... and Each of them has its own power spectral density. Frequency domain coupling degree; Based on the coupling results, a cross-modal feature mapping relationship is constructed to generate acoustic-vibration fusion features, where the vibration features include instantaneous frequency and instantaneous amplitude, and the acoustic features include spectral envelope and formant parameters. It should be noted that, based on the temporal correlation coefficient and frequency domain coupling degree, a cross-modal feature mapping relationship is constructed, and a mapping matrix T=diag(ρ,C(f)) is established to achieve feature dimension matching. Vibration features and acoustic signature features are substituted into the mapping matrix to generate acoustic signature-vibration fusion features.

[0035] A physical correlation model is constructed based on the energy transfer chain of the synchronous condenser. Through the electromagnetic induction equation, friction loss equation, and cooling system heat exchange equation, the mapping between electrical quantities and electromagnetic torque, the correlation between mechanical energy and thermal energy, and the coupling between thermal energy and cooling parameters are established to form a closed-loop conversion model for the energy transfer of the synchronous condenser. The acoustic-vibration fusion features, key features, electrical quantity features, temperature features, and cooling system features are embedded into the model according to the energy conversion nodes. Among them, the electrical quantity features correspond to the electromagnetic energy link, the acoustic-vibration fusion features correspond to the mechanical energy link, the temperature features correspond to the thermal energy accumulation link, and the cooling system features correspond to the thermal energy dissipation link. By changing the value of each parameter individually through finite element simulation, the change in the total energy of the system is recorded, the influence weight of each parameter is calculated, and then the parameter correlation matrix between devices is generated based on the weight matrix. A hierarchical fusion framework is constructed by combining key feature sets, generated acoustic-vibration fusion features, and inter-device parameter correlation matrices. Feature groups corresponding to the diagonal elements of the correlation matrix are extracted. According to the energy transfer chain conversion order, electrical quantity features directly related to electromagnetic energy conversion, temperature features related to heat accumulation, acoustic-vibration fusion features, and cooling system features related to heat dissipation are respectively assigned to the electrical feature layer, temperature feature layer, acoustic-vibration fusion layer, and cooling feature layer. Cross-layer features are verified twice by cosine similarity to determine that features within the same feature layer belong to the same energy conversion stage. The features of each layer are stored in matrix form.

[0036] S400. A weighted fusion algorithm is used to calculate the fusion sub-vector for each feature layer. The weights of the weighted fusion algorithm are determined based on the fault contribution of each feature in historical fault cases. The correlation coefficient between the fusion sub-vector of each feature layer and the overall operating status of the camera is calculated through an attention mechanism. The inter-layer weights are allocated according to the correlation coefficients, and the fusion sub-vectors are weighted and aggregated to obtain the global fusion feature vector.

[0037] It is understood that step S400 includes S401, S402, and S403, wherein: S401. Based on the output electrical feature layer, temperature feature layer, acoustic-vibration fusion layer, and cooling feature layer, weighted fusion is performed within each layer. This includes using Fault Mode and Effects Analysis (FEMA) to statistically analyze historical fault cases and determine the fault contribution coefficients of different features within each layer. In the electrical feature layer, the contribution of mid-frequency energy proportion is higher than that of the feature frequency amplitude; in the temperature feature layer, the contribution of temperature gradient difference is higher than that of temperature change rate; in the acoustic-vibration fusion layer, the contribution of acoustic signature-vibration coupling features is higher than that of single-mode features; and in the cooling feature layer, the contribution of variance is higher than that of the mean. Using the fault contribution coefficients of each feature as weights, a weighted summation algorithm is used to calculate the fused sub-vector of each layer. The calculation formula is as follows: In the formula, For the first Layer fusion sub-vectors, These correspond to the electrical, temperature, acoustic-vibration fusion, and cooling feature layers, respectively. For the first The fault contribution coefficient of the k-th feature of layer and satisfying , For the first The k-th feature vector of layer, For the first Number of layer features; S402. An attention mechanism is introduced to evaluate the importance of the obtained fusion sub-vectors of each layer. A health label vector is constructed based on the historical health status data of the camera condenser. The label value is 1 for a healthy state and 0 for a fault state. The Pearson correlation coefficient between the fusion sub-vectors of each layer and the health label vector is calculated to quantify the ability of each layer feature to represent the overall operating status of the camera condenser. It should be noted that the quantitative feature layer's ability to represent the overall operating state of the camera is as follows: the acoustic-vibration fusion layer has the highest correlation with mechanical fault states, the electrical feature layer has the highest correlation with electrical fault states, the temperature feature layer has the highest correlation with thermal fault states, and the cooling feature layer has the highest correlation with cooling system fault states. Attention weight allocation is performed based on these correlation coefficients to ensure that feature layers with a greater impact on state assessment receive higher weights.

[0038] S403. Perform dimension normalization on the obtained fusion sub-vectors of each layer and uniformly adjust them to 64-dimensional vectors. Then, perform weighted aggregation on the normalized fusion sub-vectors according to the obtained attention weights to generate a global fusion feature vector. The global fusion feature vector includes multi-dimensional state information of the camera's electrical, mechanical, thermal, and cooling systems.

[0039] It should be noted that a weighted fusion algorithm is used to calculate the fusion sub-vector for each feature layer. The weights of the weighted fusion algorithm are determined based on the fault contribution of each feature in historical fault cases. The fault contribution is obtained by statistically analyzing the influence of each feature on the occurrence of the fault in a large number of historical fault cases. The fusion sub-vector is calculated by summing the features within that layer according to their corresponding weights, resulting in the fusion sub-vectors for the electrical feature layer, temperature feature layer, acoustic-vibration fusion layer, and cooling feature layer.

[0040] The correlation coefficient between the fused sub-vectors of each feature layer and the overall operating state of the camera is calculated using an attention mechanism. This correlation coefficient reflects the ability of each fused sub-vector to represent the overall operating state of the camera. Inter-layer weights are assigned according to the correlation coefficients, and these weights are obtained by normalizing the correlation coefficients using a softmax function, ensuring the sum of the weights is 1, thus reflecting the importance of each layer's features in the overall evaluation. The fused sub-vectors are then weighted and aggregated according to their respective inter-layer weights to obtain the global fused feature vector G. This vector serves as the input data for step five, comprehensively reflecting various aspects of the camera's operating state.

[0041] S500. Input the global fusion feature vector into the pre-trained synchronous condenser status evaluation model. The synchronous condenser status evaluation model is constructed based on a deep belief network and trained with normal operation data, simulated fault data and historical fault data of the synchronous condenser until the loss function reaches a preset standard. The synchronous condenser status evaluation model performs status identification on the global fusion feature vector and outputs the synchronous condenser operation status result. The operation status result includes normal status, abnormal status and fault type. The fault type includes stator winding fault, bearing wear fault, cooling system blockage fault and bearing abnormal noise fault. It is understood that step S500 includes S501 and S502, wherein: S501. The global fusion feature vector is dimension-validated according to the input dimension requirements of the camera adjustment state evaluation model. After confirming that the vector format is consistent with the model input interface, it is imported into the feature input layer of the pre-trained camera adjustment state evaluation model. The camera adjustment state evaluation model is built based on a deep belief network. This network contains three restricted Boltzmann machine pre-training layers and one backpropagation fine-tuning layer. The number of nodes in the pre-training layer is set to 128, 64 and 32 respectively. The initial parameters of the network are optimized by the contrastive divergence algorithm, and the weight matrix is ​​adjusted by the backpropagation algorithm until the model converges. The S502 camera facilitator state assessment model performs multi-layer nonlinear transformation and feature mapping on the input global fusion feature vector, and then realizes state recognition through the softmax classifier of the output layer, finally outputting the camera facilitator operation status result.

[0042] It should be noted that the camera state evaluation model processes the input vector through multi-layer nonlinear transformations of DBN to achieve "step-by-step feature abstraction": The first layer of RBM (128 nodes): The input features are non-linearly mapped by the sigmoid activation function, which transforms continuous features (such as a temperature gradient difference of 2.5℃) into response values ​​in the range of 0-1 (such as 0.78), highlighting abnormal information in the features (such as when the response value of electrical quantity features is >0.8, there is a high probability of electrical anomalies). The second layer RBM (64 nodes): Gauss-Bernoulli transformation is used to map the 128-dimensional features of the first layer into a 64-dimensional "feature mask" (features with a response value of 1 are retained, and those with 0 are removed), to achieve redundant feature filtering (such as if the mean and variance of cooling system parameters are highly correlated, only the variance feature with the higher response value is retained). The third layer RBM (32 nodes): Through cross-modal feature association algorithms (such as weight allocation of attention mechanism), the cross-modal information in the second layer features (such as voiceprint-vibration fusion features and temperature features) is coupled and mapped to generate 32-dimensional "global association features" (such as feature value 0.92 representing the coupling state of voiceprint-vibration anomaly and temperature rise). Backpropagation (BP) fine-tuning layer: The hidden layer uses the ReLU activation function to avoid gradient vanishing and maps the 32-dimensional global correlation features into a 6-dimensional "state score vector", which corresponds to the highest score for the abnormal mechanical state.

[0043] In this step, the softmax classifier of the output layer needs to convert the 6-dimensional state score vector of the BP layer into a probability distribution. In practical applications, the probability threshold needs to be optimized in conjunction with the operation and maintenance requirements of the synchronous condenser: if the probability of a certain state is >0.8 (e.g., mechanical anomaly probability 0.95), the single state is directly output; if there are two states with probabilities in the range of 0.3-0.8 (e.g., electrical anomaly 0.45, thermal anomaly 0.4), the "composite fault" state is output (refer to the judgment criteria for composite faults in the "Technical Specification for Fault Diagnosis of Power Equipment"); if the probability of all states is <0.3, the "to be observed" state is output (to avoid misjudgment caused by temporary sensor interference, secondary verification is required based on the next sampling data).

[0044] Therefore, multi-layer nonlinear transformation can improve the sensitivity of feature anomaly recognition by 40% (compared to linear transformation), and the probability output mechanism of the softmax classifier can reduce the false judgment rate to below 2%, providing accurate status basis for the operation and maintenance decision of the synchronous condenser.

[0045] Example 2: like Figure 2 As shown, this embodiment provides a system for evaluating the operating status of a synchronous condenser based on multi-source sensor data fusion. See [link to documentation]. Figure 2 The system includes: Acquisition module 701: is used to acquire multi-source sensor data through sensors in the stator, rotor, excitation system and cooling system of the synchronous condenser; construct a spatiotemporal alignment model based on the multi-source sensor data, unify data from different acquisition frequencies to the same time dimension, and then generate an adaptive filtering threshold by combining the feature library of normal operation data of the synchronous condenser to filter out impulse noise, harmonic interference and environmental noise in the data. The dimensional unification of each parameter is completed through range standardization to obtain standardized preprocessed data. Extraction module 702: Used to extract state features from standardized preprocessed data using specific analysis methods. These include performing wavelet packet transform on electrical quantity data to extract frequency band energy proportions and characteristic frequency amplitudes; performing trend analysis on temperature data to extract temperature change rates and gradient differences; performing Hilbert-Huang transform on vibration data to extract instantaneous frequencies and amplitudes; performing Mel frequency cepstral coefficient analysis on acoustic fingerprint data to extract spectral envelopes and resonant frequencies; performing statistical analysis on cooling system parameters to extract their mean and variance; after completing feature extraction for various types of data, calculating the correlation between each extracted feature and the synchronous condenser fault state using mutual information entropy, and selecting features with a correlation degree reaching a preset threshold to form a key feature set. The synchronous condenser fault states include stator winding faults and bearing wear faults. Calculation module 703: Extracts vibration and acoustic signature features from the key feature set, calculates their temporal correlation and frequency coupling, constructs a cross-modal feature mapping relationship based on the coupling results, and generates acoustic signature-vibration fusion features; constructs a physical correlation model based on the phase shifter energy transfer chain, which includes an electrical-mechanical-thermal energy conversion path, incorporates electrical quantity features, temperature features, cooling system features, and acoustic signature-vibration fusion features into this physical correlation model, calculates the influence weight of each parameter in the energy conversion process through finite element simulation, and generates an inter-device parameter correlation matrix based on the weights; combines the key feature set, acoustic signature-vibration fusion features, and inter-device parameter correlation matrix to construct a hierarchical fusion framework, resulting in various feature layers, each divided into an electrical feature layer, a temperature feature layer, an acoustic signature-vibration fusion layer, and a cooling feature layer; Aggregation module 704: is used to calculate the fusion sub-vector of each feature layer using a weighted fusion algorithm. The weights of the weighted fusion algorithm are determined based on the fault contribution of each feature in historical fault cases. The correlation coefficient between the fusion sub-vector of each feature layer and the overall operating status of the camera is calculated through an attention mechanism. The inter-layer weights are allocated according to the correlation coefficients, and the fusion sub-vectors are weighted and aggregated to obtain the global fusion feature vector. Training module 705: This module is used to input the global fused feature vector into the pre-trained synchronous condenser status evaluation model. The synchronous condenser status evaluation model is constructed based on a deep belief network and trained with normal operation data, simulated fault data, and historical fault data of the synchronous condenser until the loss function reaches a preset standard. The synchronous condenser status evaluation model is used to identify the status of the global fused feature vector and output the synchronous condenser operation status result. The operation status result includes normal status, abnormal status, and fault type. The fault type includes stator winding fault, bearing wear fault, cooling system blockage fault, and bearing abnormal noise fault.

[0046] Specifically, the extraction module 702 includes: The analytical extraction unit is used to extract features from different types of data in the standardized preprocessed data matrix. For electrical quantity data, wavelet packet transform is used for frequency domain analysis, with a decomposition level of 3 layers and a db4 wavelet basis, extracting the energy proportion of frequency bands and the amplitude of characteristic frequencies. For temperature data, sliding window trend analysis is used, with a window size of 10 minutes, extracting the rate of temperature change and the temperature gradient difference. For vibration data, Hilbert-Huang transform is used for time-frequency domain analysis, with a decomposition level of 5 layers, extracting instantaneous frequencies and instantaneous amplitudes. For acoustic signature data, Mel frequency cepstral coefficients are used for acoustic feature analysis, with 24 Mel filter banks and a cepstral coefficient order of 12, extracting the spectral envelope and formant parameters, including the frequencies of the first to third formants. For cooling system parameters, statistical analysis methods are used to calculate the parameter distribution characteristics, with a time window of 1 minute, extracting the mean and variance to form an initial feature set. Construction Unit: Used to construct a fault correlation matrix based on the initial feature set. The matrix's row dimension represents the total number of features in the initial feature set, and the column dimension represents the types of synchronous condenser fault states. Synchronous condenser fault states include stator winding faults and bearing wear faults. The correlation degree between each extracted feature and the synchronous condenser fault state is calculated using mutual information entropy, and the calculation formula is as follows: In the formula, Features Fault status The degree of correlation, For the joint probability distribution of features and faults, Features Marginal probability distribution, Fault status The marginal probability distribution; Filtering unit: Used to set the correlation threshold and filter features with a correlation degree greater than or equal to the correlation threshold to form a key feature set. The correlation threshold is determined based on the synchronous condenser's historical fault data and the evaluation accuracy requirements.

[0047] Specifically, the computing module 703 includes: Extraction Unit: Used to extract vibration features and acoustic signature features from the key feature set. Vibration features include instantaneous frequency and instantaneous amplitude, while acoustic signature features include spectral envelope and formant parameters. The formula for calculating the temporal correlation between the two is as follows: In the formula, Vibration characteristics With voiceprint characteristics covariance, and These are the standard deviations of the two, The time-domain correlation coefficient between the two is denoted as . The formula for calculating the frequency domain coupling degree is as follows: In the formula, The cross-power spectral density of the two is... and Each of them has its own power spectral density. Frequency domain coupling degree; Generation unit: used to construct cross-modal feature mapping relationship based on coupling results, and generate acoustic-vibration fusion features, where vibration features include instantaneous frequency and instantaneous amplitude, and acoustic features include spectral envelope and formant parameters; The unit is used to construct a physical correlation model based on the energy transfer chain of the synchronous condenser. Through the electromagnetic induction equation, friction loss equation, and cooling system heat exchange equation, it establishes the mapping between electrical quantities and electromagnetic torque, the correlation between mechanical energy and thermal energy, and the coupling between thermal energy and cooling parameters, forming a closed-loop conversion model for the energy transfer of the synchronous condenser. The acoustic-vibration fusion features, key features, electrical quantity features, temperature features, and cooling system features are embedded into the model according to the energy conversion nodes. Among them, the electrical quantity features correspond to the electromagnetic energy link, the acoustic-vibration fusion features correspond to the mechanical energy link, the temperature features correspond to the thermal energy accumulation link, and the cooling system features correspond to the thermal energy dissipation link. By changing the value of each parameter individually through finite element simulation, the change in the total energy of the system is recorded, the influence weight of each parameter is calculated, and then the parameter correlation matrix between devices is generated based on the weight matrix. The defining unit is used to construct a hierarchical fusion framework by combining key feature sets, generated acoustic-vibration fusion features, and inter-device parameter correlation matrices. Feature groups corresponding to the diagonal elements of the correlation matrix are extracted. According to the energy transfer chain conversion order, electrical quantity features directly related to electromagnetic energy conversion, temperature features related to heat accumulation, acoustic-vibration fusion features, and cooling system features related to heat dissipation are respectively assigned to the electrical feature layer, temperature feature layer, acoustic-vibration fusion layer, and cooling feature layer. Cross-layer features are verified twice by cosine similarity to determine that features within the same feature layer belong to the same energy conversion stage. The features of each layer are stored in matrix form.

[0048] Specifically, the aggregation module 704 includes: The fusion unit is used to perform intra-layer weighted fusion based on the output electrical feature layer, temperature feature layer, acoustic-vibration fusion layer, and cooling feature layer. This includes using Fault Mode and Effects Analysis (FEMA) to statistically analyze historical fault cases and determine the fault contribution coefficients of different features within each layer. In the electrical feature layer, the contribution of mid-frequency energy proportion is higher than that of the feature frequency amplitude; in the temperature feature layer, the contribution of temperature gradient difference is higher than that of temperature change rate; in the acoustic-vibration fusion layer, the contribution of acoustic signature-vibration coupling features is higher than that of single-mode features; and in the cooling feature layer, the contribution of variance is higher than that of the mean. Using the fault contribution coefficients of each feature as weights, a weighted summation algorithm is used to calculate the fused sub-vector of each layer. The calculation formula is as follows: In the formula, For the first Layer fusion sub-vectors, These correspond to the electrical, temperature, acoustic-vibration fusion, and cooling feature layers, respectively. For the first The fault contribution coefficient of the k-th feature of layer and satisfying , For the first The k-th feature vector of layer, For the first Number of layer features; Evaluation Unit: This unit introduces an attention mechanism to evaluate the importance of the obtained fusion sub-vectors of each layer. It constructs a health label vector based on the historical health status data of the camera condenser, where a healthy state corresponds to a label value of 1 and a fault state corresponds to a label value of 0. It calculates the Pearson correlation coefficient between each layer fusion sub-vector and the health label vector to quantify the ability of each layer's features to represent the overall operating status of the camera condenser. The generation unit is used to normalize the dimensions of the obtained fusion sub-vectors of each layer and uniformly adjust them to 64-dimensional vectors. It then performs weighted aggregation on the normalized fusion sub-vectors according to the obtained attention weights to generate a global fusion feature vector. The global fusion feature vector includes multi-dimensional state information of the camera's electrical, mechanical, thermal, and cooling systems.

[0049] Specifically, the training module 705 includes: Verification Unit: This unit verifies the dimension of the global fusion feature vector according to the input dimension requirements of the camera state evaluation model. After confirming that the vector format is consistent with the model input interface, it imports the vector into the feature input layer of the pre-trained camera state evaluation model. The camera state evaluation model is built based on a deep belief network, which contains three restricted Boltzmann machine pre-training layers and one backpropagation fine-tuning layer. The number of nodes in the pre-training layer is set to 128, 64 and 32 respectively. The initial parameters of the network are optimized by the contrastive divergence algorithm, and the weight matrix is ​​adjusted by the backpropagation algorithm until the model converges. Output unit: The camera state assessment model performs multi-layer nonlinear transformation and feature mapping on the input global fusion feature vector, and then realizes state recognition through the softmax classifier of the output layer, and finally outputs the camera state result.

[0050] It should be noted that the specific methods by which each module performs operations in the system described in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0051] Example 3: Corresponding to the above method embodiments, this embodiment also provides a synchronous condenser operation status evaluation device based on multi-source sensor data fusion. The synchronous condenser operation status evaluation device based on multi-source sensor data fusion described below and the synchronous condenser operation status evaluation method based on multi-source sensor data fusion described above can be referred to in correspondence.

[0052] Figure 3 This is a block diagram illustrating a phase shifter operation status evaluation device 800 based on multi-source sensor data fusion, according to an exemplary embodiment. Figure 3 As shown, the synchronous condenser operation status evaluation device 800 based on multi-source sensor data fusion includes a processor 801 and a memory 802. The synchronous condenser operation status evaluation device 800 also includes one or more of a multimedia component 803, an I / O interface 804, and a communication component 805.

[0053] The processor 801 controls the overall operation of the multi-source sensor data fusion-based synchronous condenser operation status assessment device 800 to complete all or part of the steps in the aforementioned multi-source sensor data fusion-based synchronous condenser operation status assessment method. The memory 802 stores various types of data to support the operation of the multi-source sensor data fusion-based synchronous condenser operation status assessment device 800. This data may include, for example, instructions for any application or method operating on the multi-source sensor data fusion-based synchronous condenser operation status assessment device 800, as well as application-related data, such as contact data, sent and received messages, images, audio, video, etc. The memory 802 can be implemented using any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The multimedia component 803 may include a screen and an audio component. The screen may be, for example, a touchscreen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signals may be further stored in the memory 802 or transmitted via the communication component 805. The audio component also includes at least one speaker for outputting audio signals. I / O interface 804 provides an interface between processor 801 and other interface modules, such as keyboards, mice, or buttons. These buttons can be virtual or physical. Communication component 805 is used for wired or wireless communication between the multi-source sensor data fusion-based camera operation status evaluation device 800 and other devices. Wireless communication includes, for example, Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, or 4G, or a combination thereof. Therefore, the corresponding communication component 805 may include a Wi-Fi module, a Bluetooth module, or an NFC module.

[0054] In an exemplary embodiment, the synchronous condenser operation status assessment device 800 based on multi-source sensor data fusion can be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the aforementioned synchronous condenser operation status assessment method based on multi-source sensor data fusion.

[0055] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided. When executed by a processor, these program instructions implement the steps of the above-described method for evaluating the operating status of a camera condenser based on multi-source sensor data fusion. For example, the computer-readable storage medium may be the memory 802 including the program instructions, which may be executed by the processor 801 of the camera condenser operating status evaluation device 800 based on multi-source sensor data fusion to complete the above-described method for evaluating the operating status of a camera condenser based on multi-source sensor data fusion.

[0056] Example 4: Corresponding to the above method embodiments, this embodiment also provides a readable storage medium. The readable storage medium described below can be referred to in conjunction with the above-described method for evaluating the operating status of a condenser camera based on multi-source sensor data fusion.

[0057] A computer program is stored on a readable storage medium, and when the computer program is executed by a processor, it implements the steps of the method embodiment described above for evaluating the operating status of a camera condenser based on multi-source sensor data fusion.

[0058] Specifically, the readable storage medium can be a USB flash drive, external hard drive, read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk, or any other readable storage medium capable of storing program code.

[0059] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0060] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for evaluating the operating status of a synchronous condenser based on multi-source sensor data fusion, characterized in that, include: Multi-source sensor data is acquired by adjusting the sensors in the stator, rotor, excitation system, and cooling system of the phase-shifting camera. A spatiotemporal alignment model is constructed based on multi-source sensor data to unify data from different acquisition frequencies to the same time dimension. Then, an adaptive filtering threshold is generated by combining the feature library of normal operation data of the synchronous condenser to filter out impulse noise, harmonic interference and environmental noise in the data. The dimensionality of each parameter is unified through range standardization to obtain standardized preprocessed data. For standardized preprocessed data, specific analytical methods are employed to extract state features. These include wavelet packet transform of electrical quantity data to extract frequency band energy proportions and characteristic frequency amplitudes; trend analysis of temperature data to extract temperature change rates and gradient differences; Hilbert-Huang transform of vibration data to extract instantaneous frequencies and amplitudes; Mel frequency cepstral coefficient analysis of acoustic signature data to extract spectral envelopes and resonant frequencies; and statistical analysis of cooling system parameters to extract their mean and variance. After feature extraction of various data types, the correlation between each extracted feature and the synchronous condenser fault state is calculated using mutual information entropy. Features with correlation reaching a preset threshold are selected to form a key feature set, where the synchronous condenser fault states include stator winding faults and bearing wear faults. Vibration features and acoustic features are extracted from the key feature set, and the temporal correlation and frequency coupling degree of the two are calculated. Based on the coupling results, a cross-modal feature mapping relationship is constructed to generate acoustic-vibration fusion features. A physical correlation model is constructed based on the energy transfer chain of a synchronous condenser. This energy transfer chain includes an electrical-mechanical-thermal energy conversion path. Electrical quantity characteristics, temperature characteristics, cooling system characteristics, and acoustic-vibration fusion characteristics are incorporated into this physical correlation model. The influence weight of each parameter in the energy conversion process is calculated through finite element simulation, and a parameter correlation matrix between devices is generated based on the weight. A hierarchical fusion framework is constructed by combining the key feature set, acoustic-vibration fusion characteristics, and the parameter correlation matrix between devices. The resulting feature layers are divided into an electrical feature layer, a temperature feature layer, an acoustic-vibration fusion layer, and a cooling feature layer. A weighted fusion algorithm is used to calculate the fusion sub-vector for each feature layer. The weights of the weighted fusion algorithm are determined based on the fault contribution of each feature in historical fault cases. The correlation coefficient between the fusion sub-vector of each feature layer and the overall operating status of the camera is calculated through an attention mechanism. The inter-layer weights are allocated according to the correlation coefficients, and the fusion sub-vectors are weighted and aggregated to obtain the global fusion feature vector. The global fusion feature vector is input into the pre-trained synchronous condenser status assessment model, which is built based on a deep belief network and trained with normal operation data, simulated fault data and historical fault data of the synchronous condenser until the loss function reaches a preset standard. The synchronous condenser status assessment model is used to identify the status of the global fusion feature vector and output the synchronous condenser operation status result, which includes normal status, abnormal status and fault type. The fault type includes stator winding fault, bearing wear fault, cooling system blockage fault and bearing noise fault.

2. The method for evaluating the operating status of a synchronous condenser based on multi-source sensor data fusion according to claim 1, characterized in that, For the standardized preprocessed data, specific analytical methods are employed to extract state features. These include: wavelet packet transform of electrical quantity data to extract frequency band energy proportions and characteristic frequency amplitudes; trend analysis of temperature data to extract temperature change rates and gradient differences; Hilbert-Huang transform of vibration data to extract instantaneous frequencies and amplitudes; Mel frequency cepstral coefficient analysis of acoustic signature data to extract spectral envelopes and resonant frequencies; and statistical analysis of cooling system parameters to extract their mean and variance. After feature extraction for various data types, the correlation between each extracted feature and the synchronous condenser fault state is calculated using mutual information entropy. Features with a correlation reaching a preset threshold are selected to form a key feature set. The synchronous condenser fault states include stator winding faults and bearing wear faults, including: Feature extraction was performed on different types of data in the standardized preprocessed data matrix. For electrical quantity data, wavelet packet transform was used for frequency domain analysis, with a decomposition level of 3 layers and a db4 wavelet basis selected to extract the energy proportion of frequency bands and the amplitude of characteristic frequencies. For temperature data, sliding window trend analysis was used, with a window size of 10 minutes, to extract the temperature change rate and temperature gradient difference. For vibration data, Hilbert-Huang transform was used for time-frequency domain analysis, with a decomposition level of 5 layers, to extract instantaneous frequencies and instantaneous amplitudes. For acoustic signature data, Mel frequency cepstral coefficients were used for acoustic feature analysis, with 24 Mel filter banks and a cepstral coefficient order of 12, to extract the spectral envelope and formant parameters, including the frequencies of the first to third formants. For cooling system parameters, statistical analysis methods were used to calculate the parameter distribution characteristics, with a time window of 1 minute, to extract the mean and variance, forming an initial feature set. Based on the initial feature set, a fault correlation matrix is ​​constructed. The row dimension of the matrix represents the total number of features in the initial feature set, and the column dimension represents the types of synchronous condenser fault states. The synchronous condenser fault states include stator winding faults and bearing wear faults. The correlation degree between each extracted feature and the synchronous condenser fault state is calculated using mutual information entropy. The calculation formula is as follows: In the formula, Features Fault status The degree of correlation, For the joint probability distribution of features and faults, Features Marginal probability distribution, Fault status The marginal probability distribution; A correlation threshold is set, and features with a correlation degree greater than or equal to the correlation threshold are selected to form a key feature set. The correlation threshold is determined based on the historical fault data of the synchronous condenser and the evaluation accuracy requirements.

3. The method for evaluating the operating status of a synchronous condenser based on multi-source sensor data fusion according to claim 1, characterized in that, The process involves extracting vibration features and acoustic signature features from a set of key features, calculating their temporal correlation and frequency domain coupling, constructing a cross-modal feature mapping relationship based on the coupling results, and generating acoustic signature-vibration fusion features. A physical correlation model is constructed based on the energy transfer chain of a synchronous condenser, which includes an electrical-mechanical-thermal energy conversion path. Electrical quantity characteristics, temperature characteristics, cooling system characteristics, and acoustic-vibration fusion characteristics are incorporated into this physical correlation model. The influence weights of each parameter in the energy conversion process are calculated through finite element simulation, and an inter-device parameter correlation matrix is ​​generated based on these weights. A hierarchical fusion framework is constructed by combining the key feature set, acoustic-vibration fusion characteristics, and the inter-device parameter correlation matrix, resulting in various feature layers. Each feature layer is divided into an electrical feature layer, a temperature feature layer, an acoustic-vibration fusion layer, and a cooling feature layer, including: Vibration features and acoustic signature features are extracted from the key feature set. Vibration features include instantaneous frequency and instantaneous amplitude, while acoustic signature features include spectral envelope and formant parameters. The formula for calculating the temporal correlation between the two is as follows: In the formula, Vibration characteristics With voiceprint characteristics covariance, and These are the standard deviations of the two, The time-domain correlation coefficient between the two is denoted as . The formula for calculating the frequency domain coupling degree is as follows: In the formula, The cross-power spectral density of the two is... and Each of them has its own power spectral density. Frequency domain coupling degree; Based on the coupling results, a cross-modal feature mapping relationship is constructed to generate acoustic-vibration fusion features, where the vibration features include instantaneous frequency and instantaneous amplitude, and the acoustic features include spectral envelope and formant parameters. A physical correlation model is constructed based on the energy transfer chain of the synchronous condenser. Through the electromagnetic induction equation, friction loss equation, and cooling system heat exchange equation, the mapping between electrical quantities and electromagnetic torque, the correlation between mechanical energy and thermal energy, and the coupling between thermal energy and cooling parameters are established to form a closed-loop conversion model for the energy transfer of the synchronous condenser. The acoustic-vibration fusion features, key features, electrical quantity features, temperature features, and cooling system features are embedded into the model according to the energy conversion nodes. Among them, the electrical quantity features correspond to the electromagnetic energy link, the acoustic-vibration fusion features correspond to the mechanical energy link, the temperature features correspond to the thermal energy accumulation link, and the cooling system features correspond to the thermal energy dissipation link. By changing the value of each parameter individually through finite element simulation, the change in the total energy of the system is recorded, the influence weight of each parameter is calculated, and then the parameter correlation matrix between devices is generated based on the weight matrix. A hierarchical fusion framework is constructed by combining key feature sets, generated acoustic-vibration fusion features, and inter-device parameter correlation matrices. Feature groups corresponding to the diagonal elements of the correlation matrix are extracted. According to the energy transfer chain conversion order, electrical quantity features directly related to electromagnetic energy conversion, temperature features related to heat accumulation, acoustic-vibration fusion features, and cooling system features related to heat dissipation are respectively assigned to the electrical feature layer, temperature feature layer, acoustic-vibration fusion layer, and cooling feature layer. Cross-layer features are verified twice by cosine similarity to determine that features within the same feature layer belong to the same energy conversion stage. The features of each layer are stored in matrix form.

4. The method for evaluating the operating status of a synchronous condenser based on multi-source sensor data fusion according to claim 1, characterized in that, The weighted fusion algorithm is used to calculate the fused sub-vector for each feature layer. The weights of the weighted fusion algorithm are determined based on the fault contribution of each feature in historical fault cases. An attention mechanism is used to calculate the correlation coefficient between the fused sub-vector of each feature layer and the overall operating state of the camera condenser. Inter-layer weights are allocated according to the correlation coefficient, and the fused sub-vectors are weighted and aggregated to obtain the global fused feature vector, which includes: Based on the output electrical feature layer, temperature feature layer, acoustic-vibration fusion layer, and cooling feature layer, intra-layer weighted fusion is performed for each layer's features. This includes using Fault Mode and Effects Analysis (FEMA) to statistically analyze historical fault cases and determine the fault contribution coefficients of different features within each layer. In the electrical feature layer, the contribution of mid-frequency energy proportion is higher than that of the feature frequency amplitude; in the temperature feature layer, the contribution of temperature gradient difference is higher than that of temperature change rate; in the acoustic-vibration fusion layer, the contribution of acoustic signature-vibration coupling features is higher than that of single-mode features; and in the cooling feature layer, the contribution of variance is higher than that of the mean. Using the fault contribution coefficients of each feature as weights, a weighted summation algorithm is used to calculate the fused sub-vectors of each layer. The calculation formula is as follows: In the formula, For the first Layer fusion sub-vectors, These correspond to the electrical, temperature, acoustic-vibration fusion, and cooling feature layers, respectively. For the first The fault contribution coefficient of the k-th feature of layer and satisfying , For the first The k-th feature vector of layer, For the first Number of layer features; An attention mechanism is introduced to evaluate the importance of the obtained fusion sub-vectors of each layer. A health label vector is constructed based on the historical health status data of the camera condenser. The label value is 1 for a healthy state and 0 for a fault state. The Pearson correlation coefficient between the fusion sub-vectors of each layer and the health label vector is calculated to quantify the ability of each layer feature to represent the overall operating status of the camera condenser. The obtained fusion sub-vectors of each layer are normalized in dimension and uniformly adjusted to 64-dimensional vectors. The normalized fusion sub-vectors are then weighted and aggregated according to the obtained attention weights to generate a global fusion feature vector. The global fusion feature vector includes multi-dimensional state information of the camera's electrical, mechanical, thermal, and cooling systems.

5. The method for evaluating the operating status of a synchronous condenser based on multi-source sensor data fusion according to claim 1, characterized in that, The global fused feature vector is input into a pre-trained camera condenser state evaluation model, which is built based on a deep belief network and trained with camera condenser normal operation data, simulated fault data, and historical fault data until the loss function reaches a preset standard. The camera condenser state evaluation model performs state identification on the global fused feature vector and outputs the camera condenser operating state result, including: The global fusion feature vector is dimension-validated according to the input dimension requirements of the camera adjustment state evaluation model. After confirming that the vector format is consistent with the model input interface, it is imported into the feature input layer of the pre-trained camera adjustment state evaluation model. The camera adjustment state evaluation model is based on a deep belief network, which contains three restricted Boltzmann machine pre-training layers and one backpropagation fine-tuning layer. The number of nodes in the pre-training layer is set to 128, 64 and 32 respectively. The initial parameters of the network are optimized by the contrastive divergence algorithm, and the weight matrix is ​​adjusted by the backpropagation algorithm until the model converges. The camera facilitator state assessment model performs multi-layer nonlinear transformation and feature mapping on the input global fusion feature vector, and then realizes state recognition through the softmax classifier in the output layer, finally outputting the camera facilitator operation status result.

6. A synchronous condenser operation status evaluation system based on multi-source sensor data fusion, based on the synchronous condenser operation status evaluation method based on multi-source sensor data fusion as described in claim 1, characterized in that, include: Acquisition module: used to acquire multi-source sensor data through sensors in the stator, rotor, excitation system and cooling system of the synchronous condenser; A spatiotemporal alignment model is constructed based on multi-source sensor data to unify data from different acquisition frequencies to the same time dimension. Then, an adaptive filtering threshold is generated by combining the feature library of normal operation data of the synchronous condenser to filter out impulse noise, harmonic interference and environmental noise in the data. The dimensionality of each parameter is unified through range standardization to obtain standardized preprocessed data. Extraction Module: This module uses specific analytical methods to extract state features from standardized preprocessed data. These include wavelet packet transform of electrical quantity data to extract frequency band energy proportions and characteristic frequency amplitudes; trend analysis of temperature data to extract temperature change rates and gradient differences; Hilbert-Huang transform of vibration data to extract instantaneous frequencies and amplitudes; Mel frequency cepstral coefficient analysis of acoustic signature data to extract spectral envelopes and resonant frequencies; and statistical analysis of cooling system parameters to extract their mean and variance. After feature extraction for various data types, the module calculates the correlation between each extracted feature and the synchronous condenser fault state using mutual information entropy, and selects features with correlation reaching a preset threshold to form a key feature set. The synchronous condenser fault states include stator winding faults and bearing wear faults. The calculation module is used to extract vibration features and acoustic signature features from the key feature set, calculate the temporal correlation and frequency domain coupling degree between the two, construct cross-modal feature mapping relationship based on the coupling results, and generate acoustic signature-vibration fusion features. A physical correlation model is constructed based on the energy transfer chain of a synchronous condenser. This energy transfer chain includes an electrical-mechanical-thermal energy conversion path. Electrical quantity characteristics, temperature characteristics, cooling system characteristics, and acoustic-vibration fusion characteristics are incorporated into this physical correlation model. The influence weight of each parameter in the energy conversion process is calculated through finite element simulation, and a parameter correlation matrix between devices is generated based on the weight. A hierarchical fusion framework is constructed by combining the key feature set, acoustic-vibration fusion characteristics, and the parameter correlation matrix between devices. The resulting feature layers are divided into an electrical feature layer, a temperature feature layer, an acoustic-vibration fusion layer, and a cooling feature layer. The aggregation module is used to calculate the fusion sub-vectors for each feature layer using a weighted fusion algorithm. The weights of the weighted fusion algorithm are determined based on the fault contribution of each feature in historical fault cases. The correlation coefficient between the fusion sub-vectors of each feature layer and the overall operating status of the camera is calculated through an attention mechanism. The inter-layer weights are allocated according to the correlation coefficients, and the fusion sub-vectors are weighted and aggregated to obtain the global fusion feature vector. Training module: This module is used to input the global fused feature vector into the pre-trained synchronous condenser status assessment model. The synchronous condenser status assessment model is built based on a deep belief network and trained with normal operation data, simulated fault data, and historical fault data of the synchronous condenser until the loss function reaches a preset standard. The synchronous condenser status assessment model performs status identification on the global fused feature vector and outputs the synchronous condenser operation status results. The operation status results include normal status, abnormal status, and fault type. The fault types include stator winding fault, bearing wear fault, cooling system blockage fault, and bearing noise fault.

7. The synchronous condenser operation status evaluation system based on multi-source sensor data fusion according to claim 6, characterized in that, The extraction module includes: The analytical extraction unit is used to extract features from different types of data in the standardized preprocessed data matrix. For electrical quantity data, wavelet packet transform is used for frequency domain analysis, with a decomposition level of 3 layers and a db4 wavelet basis, extracting the energy proportion of frequency bands and the amplitude of characteristic frequencies. For temperature data, sliding window trend analysis is used, with a window size of 10 minutes, extracting the rate of temperature change and the temperature gradient difference. For vibration data, Hilbert-Huang transform is used for time-frequency domain analysis, with a decomposition level of 5 layers, extracting instantaneous frequencies and instantaneous amplitudes. For acoustic signature data, Mel frequency cepstral coefficients are used for acoustic feature analysis, with 24 Mel filter banks and a cepstral coefficient order of 12, extracting the spectral envelope and formant parameters, including the frequencies of the first to third formants. For cooling system parameters, statistical analysis methods are used to calculate the parameter distribution characteristics, with a time window of 1 minute, extracting the mean and variance to form an initial feature set. Construction Unit: Used to construct a fault correlation matrix based on the initial feature set. The matrix's row dimension represents the total number of features in the initial feature set, and the column dimension represents the types of synchronous condenser fault states. Synchronous condenser fault states include stator winding faults and bearing wear faults. The correlation degree between each extracted feature and the synchronous condenser fault state is calculated using mutual information entropy, and the calculation formula is as follows: In the formula, Features Fault status The degree of correlation, For the joint probability distribution of features and faults, Features Marginal probability distribution, Fault status The marginal probability distribution; Filtering unit: Used to set the correlation threshold and filter features with a correlation degree greater than or equal to the correlation threshold to form a key feature set. The correlation threshold is determined based on the synchronous condenser's historical fault data and the evaluation accuracy requirements.

8. The synchronous condenser operation status evaluation system based on multi-source sensor data fusion according to claim 6, characterized in that, The computing module includes: Extraction Unit: Used to extract vibration features and acoustic signature features from the key feature set. Vibration features include instantaneous frequency and instantaneous amplitude, while acoustic signature features include spectral envelope and formant parameters. The formula for calculating the temporal correlation between the two is as follows: In the formula, Vibration characteristics With voiceprint characteristics covariance, and These are the standard deviations of the two, The time-domain correlation coefficient between the two is denoted as . The formula for calculating the frequency domain coupling degree is as follows: In the formula, The cross-power spectral density of the two is... and Each of them has its own power spectral density. Frequency domain coupling degree; Generation unit: used to construct cross-modal feature mapping relationship based on coupling results, and generate acoustic-vibration fusion features, where vibration features include instantaneous frequency and instantaneous amplitude, and acoustic features include spectral envelope and formant parameters; The unit is used to construct a physical correlation model based on the energy transfer chain of the synchronous condenser. Through the electromagnetic induction equation, friction loss equation, and cooling system heat exchange equation, it establishes the mapping between electrical quantities and electromagnetic torque, the correlation between mechanical energy and thermal energy, and the coupling between thermal energy and cooling parameters, forming a closed-loop conversion model for the energy transfer of the synchronous condenser. The acoustic-vibration fusion features, key features, electrical quantity features, temperature features, and cooling system features are embedded into the model according to the energy conversion nodes. Among them, the electrical quantity features correspond to the electromagnetic energy link, the acoustic-vibration fusion features correspond to the mechanical energy link, the temperature features correspond to the thermal energy accumulation link, and the cooling system features correspond to the thermal energy dissipation link. By changing the value of each parameter individually through finite element simulation, the change in the total energy of the system is recorded, the influence weight of each parameter is calculated, and then the parameter correlation matrix between devices is generated based on the weight matrix. The defining unit is used to construct a hierarchical fusion framework by combining key feature sets, generated acoustic-vibration fusion features, and inter-device parameter correlation matrices. Feature groups corresponding to the diagonal elements of the correlation matrix are extracted. According to the energy transfer chain conversion order, electrical quantity features directly related to electromagnetic energy conversion, temperature features related to heat accumulation, acoustic-vibration fusion features, and cooling system features related to heat dissipation are respectively assigned to the electrical feature layer, temperature feature layer, acoustic-vibration fusion layer, and cooling feature layer. Cross-layer features are verified twice by cosine similarity to determine that features within the same feature layer belong to the same energy conversion stage. The features of each layer are stored in matrix form.

9. The synchronous condenser operation status evaluation system based on multi-source sensor data fusion according to claim 6, characterized in that, The aggregation module includes: The fusion unit is used to perform intra-layer weighted fusion based on the output electrical feature layer, temperature feature layer, acoustic-vibration fusion layer, and cooling feature layer. This includes using Fault Mode and Effects Analysis (FEMA) to statistically analyze historical fault cases and determine the fault contribution coefficients of different features within each layer. In the electrical feature layer, the contribution of mid-frequency energy proportion is higher than that of the feature frequency amplitude; in the temperature feature layer, the contribution of temperature gradient difference is higher than that of temperature change rate; in the acoustic-vibration fusion layer, the contribution of acoustic signature-vibration coupling features is higher than that of single-mode features; and in the cooling feature layer, the contribution of variance is higher than that of the mean. Using the fault contribution coefficients of each feature as weights, a weighted summation algorithm is used to calculate the fused sub-vector of each layer. The calculation formula is as follows: In the formula, For the first Layer fusion sub-vectors, These correspond to the electrical, temperature, acoustic-vibration fusion, and cooling feature layers, respectively. For the first The fault contribution coefficient of the k-th feature of layer and satisfying , For the first The k-th feature vector of layer, For the first Number of layer features; Evaluation Unit: This unit introduces an attention mechanism to evaluate the importance of the obtained fusion sub-vectors of each layer. It constructs a health label vector based on the historical health status data of the camera condenser, where a healthy state corresponds to a label value of 1 and a fault state corresponds to a label value of 0. It calculates the Pearson correlation coefficient between each layer fusion sub-vector and the health label vector to quantify the ability of each layer's features to represent the overall operating status of the camera condenser. The generation unit is used to normalize the dimensions of the obtained fusion sub-vectors of each layer and uniformly adjust them to 64-dimensional vectors. It then performs weighted aggregation on the normalized fusion sub-vectors according to the obtained attention weights to generate a global fusion feature vector. The global fusion feature vector includes multi-dimensional state information of the camera's electrical, mechanical, thermal, and cooling systems.

10. The synchronous condenser operation status evaluation system based on multi-source sensor data fusion according to claim 6, characterized in that, The training module includes: Verification Unit: This unit verifies the dimension of the global fusion feature vector according to the input dimension requirements of the camera state evaluation model. After confirming that the vector format is consistent with the model input interface, it imports the vector into the feature input layer of the pre-trained camera state evaluation model. The camera state evaluation model is built based on a deep belief network, which contains three restricted Boltzmann machine pre-training layers and one backpropagation fine-tuning layer. The number of nodes in the pre-training layer is set to 128, 64 and 32 respectively. The initial parameters of the network are optimized by the contrastive divergence algorithm, and the weight matrix is ​​adjusted by the backpropagation algorithm until the model converges. Output unit: The camera state assessment model performs multi-layer nonlinear transformation and feature mapping on the input global fusion feature vector, and then realizes state recognition through the softmax classifier of the output layer, and finally outputs the camera state result.

Images