Online monitoring method and system for equipment state of phase modifier based on intelligent sensor

Abstract

The invention discloses a phase modifier equipment state on-line monitoring method and system based on an intelligent sensor. The method comprises the following steps of 1, collecting and standardizing original operation data of phase modifier equipment; 2, obtaining a phase modifier heterogeneous graph according to a spatial layout construction edge of the sensor; step 3, spatial feature learning is carried out through a Siamese network; 4, sorting the spatial feature vector set according to stamps and carrying out differential processing; 5, performing spatial-temporal feature analysis through an improved GraphSAGE model to obtain a spatial-temporal feature vector cluster; step 6, performing anomaly detection through statistical analysis based on the spatio-temporal feature vector cluster; and step 7, evaluating the state of the phase modifier equipment, generating fault early warning information, and adjusting an equipment monitoring strategy. According to the invention, through the Siamese network and the improved GraphSAGE model, intelligent state monitoring and fault early warning of the phase modifier equipment are realized.

Description

Technical Field

[0001] This invention relates to the field of equipment condition monitoring technology, and in particular to a method and system for online monitoring of the condition of a synchronous condenser based on intelligent sensors. Background Technology

[0002] With the continuous development of industrial automation, synchronous condensers, as important equipment in power systems, undertake the crucial task of regulating power system stability and power supply quality. The reliability of synchronous condensers is paramount to the operation of the power system; therefore, condition monitoring and fault diagnosis of synchronous condensers have become key to ensuring stable system operation. However, traditional methods for monitoring the condition of synchronous condensers mainly rely on manual inspections and simple sensor data collection, lacking real-time capabilities and intelligence. This makes it difficult to detect potential faults in a timely manner, easily leading to unforeseen failures and impacting the power system.

[0003] In existing technologies, many synchronous condenser equipment monitoring methods rely on traditional single sensors for data acquisition, often focusing only on a single type of equipment status data (such as temperature, pressure, vibration, etc.), ignoring the interrelationships and spatiotemporal characteristics between multiple sensor data. While this monitoring method can provide some fault indication, its limited and singular data processing lacks comprehensive analysis of complex equipment states, resulting in low accuracy and timeliness of monitoring results. Therefore, traditional methods fail to effectively utilize multimodal sensor data and cannot provide real-time early warning and fault prediction for equipment status changes, thus hindering intelligent equipment management. Existing technologies mostly employ traditional fault detection algorithms, which generally rely on rule settings or models based on historical data, making them unable to adapt to the diversity and complexity of equipment status changes. Traditional methods typically detect anomalies based on fixed thresholds. While simple, this approach struggles to address the diversity of different equipment and environmental conditions. Once equipment status changes exceed the set rule range, false alarms or missed alarms may occur, failing to detect potential faults in a timely and accurate manner. Therefore, existing fault detection methods lack sufficient flexibility and intelligence, failing to consider complex factors such as temporal variations and spatial dependencies, severely restricting the online monitoring and fault prediction capabilities of synchronous condenser equipment.

[0004] Therefore, how to provide a method and system for online monitoring of the status of synchronous condenser equipment based on intelligent sensors is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose an online monitoring method and system for the status of synchronous condenser equipment based on intelligent sensors. This invention utilizes multi-sensor data fusion and graph convolutional networks, combined with Siamese networks and an improved GraphSAGE model, to accurately capture the spatiotemporal characteristics of the synchronous condenser equipment, achieving intelligent monitoring and fault prediction of the equipment status. Through spatiotemporal feature analysis and statistical anomaly detection, fault warnings are automatically generated, and equipment monitoring strategies are dynamically adjusted, significantly improving the accuracy of fault identification, monitoring efficiency, and system intelligence level, ensuring the safe and stable operation of the equipment.

[0006] According to an embodiment of the present invention, an online monitoring method for the status of a synchronous condenser device based on a smart sensor includes the following steps:

[0007] Step 1: Collect raw operating data of the synchronous condenser equipment through multiple sensors, and standardize the raw operating data to obtain the synchronous condenser operating dataset;

[0008] Step 2: Based on the camera condenser's running dataset, each sensor is treated as a node, and connection edges are constructed according to the spatial layout of the sensors to obtain a heterogeneous graph of the camera condenser;

[0009] Step 3: Input the heterogeneous image of the camera condenser into the Siamese network for spatial feature learning to obtain the spatial feature vector set of the camera condenser equipment;

[0010] Step 4: Sort the spatial feature vector set by stamp and perform differential processing to obtain the detrended time series dataset of the camera device;

[0011] Step 5: Input the detrended time series dataset from the camera adjustment device into the improved GraphSAGE model, and perform spatiotemporal feature analysis through the spatial information encoding module, the neighbor sampling update module, and the spatiotemporal dependency aggregation module to obtain spatiotemporal feature vector clusters;

[0012] Step 6: Based on the spatiotemporal feature vector cluster, perform anomaly detection through statistical analysis to obtain the fault detection results of the synchronous condenser equipment;

[0013] Step 7: Based on the fault detection results of the synchronous condenser equipment, assess the status of the synchronous condenser equipment, generate fault warning information, and adjust the equipment monitoring strategy.

[0014] Optionally, the various sensors include temperature sensors, pressure sensors, vibration sensors, audio sensors, infrared imaging sensors, and chemical sensors; the standardization processing steps include outlier removal, missing value imputation, timestamp alignment, and normalization processing of different types of sensor data to obtain a unified structured camera condenser operation dataset.

[0015] Optionally, step two specifically includes:

[0016] Each sensor of the synchronous condenser equipment is treated as a node in the synchronous condenser heterogeneous graph, and the node features of each sensor are composed of the operational data collected by the corresponding sensor in the synchronous condenser operational dataset.

[0017] Based on the actual spatial layout of the synchronous condenser equipment, the spatial distance between each sensor is determined by the installation coordinates of each sensor;

[0018] An edge connection is constructed based on the spatial distance between the sensors to obtain a heterogeneous graph of the camera, where the edge weight of the edge connection is the spatial distance between the sensors.

[0019] Optionally, step three specifically includes:

[0020] The feature vector of each sensor node in the heterogeneous graph of the camera condenser is input to the input layer of the Siamese network. Each pair of adjacent sensor nodes is combined into a pair of input data to obtain the input feature pair. The feature vector of the sensor node includes the corresponding sensor data, timestamp and spatial location in the camera condenser running dataset.

[0021] The input feature pairs are fed into a sub-network of the Siamese network to obtain a feature vector set of the input data pairs. The sub-network consists of a multi-layer fully connected neural network, which extracts the feature representation of the input feature pairs through the activation functions of multiple neurons.

[0022] The input data pairs feature vector sets are input to the similarity calculation layer. The cosine similarity of each pair of input data pairs feature vectors is calculated. The calculated cosine similarity is used as the spatial similarity score to obtain the spatial similarity score set. The cosine similarity represents the spatial relationship between sensor data.

[0023] The spatial similarity score set is input into the spatial feature fusion layer, and the similarity score is used as a weight to weight the feature vectors of the input data to obtain the spatial feature vector set of the camera adjustment device.

[0024] Optionally, step four specifically includes:

[0025] The spatial feature vector set of the camera condenser is used as input to represent the spatial features of the state of each sensor of the camera condenser, and sorted according to timestamps to obtain the time series data of each sensor.

[0026] The time-series data of each sensor are differentially processed, which involves calculating the difference between the data at the current time and the data at the previous time.

[0027] The time series data of each sensor is detrended by using a moving average;

[0028] Calculate the standard deviation of each sensor data within the sliding time window. When the standard deviation exceeds the preset normal threshold range, it is marked as an abnormal state.

[0029] The detrended data is subjected to weighted linear regression within a sliding window to predict data values ​​at several future times, thus obtaining the detrended time series dataset of the camera adjustment device.

[0030] Optionally, the improved GraphSAGE model is specifically as follows:

[0031] The detrended time series dataset of the camera adjustment device is input into the spatial information encoding module, the spatial location of each sensor node is converted into a vector representation, and concatenated with the spatial feature vector to obtain a spatiotemporal feature vector set;

[0032] The Manhattan distance between every two spatiotemporal feature vectors in the spatiotemporal feature vector set is calculated. If the Manhattan distance is less than a preset distance threshold, an edge connection is established between the corresponding sensor nodes to obtain the spatiotemporal feature map of the sensor. The edge weight of the edge connection is the corresponding Manhattan distance.

[0033] The spatiotemporal feature map is input to the neighbor sampling and update module. Based on the connection relationship between the spatiotemporal feature maps of the sensor node and the neighbor node, the feature map is sampled, and the feature map of the neighbor node is transmitted to the target node and fused to form the updated node feature map.

[0034] The fusion is to multiply the spatiotemporal feature vector corresponding to the neighbor node by the weight of the connection edge between the neighbor node and the target node, and then add it to the spatiotemporal feature vector corresponding to the target node.

[0035] The fused feature map of neighbor nodes is input into the spatiotemporal dependency aggregation module, and several vectors are randomly selected from the fused feature map of neighbor nodes as initial cluster centers.

[0036] Calculate the Euclidean distance between the spatiotemporal feature vector corresponding to each node in the neighbor node fusion feature map and the initial cluster center, and assign each node to the initial cluster center with the closest Euclidean distance;

[0037] Based on the mean of the spatiotemporal feature vectors corresponding to all nodes in each initial cluster center, the position of the cluster center is recalculated, and the cluster center is iteratively updated repeatedly until the number of changes in the cluster center is less than a preset threshold or the preset maximum number of iterations is reached, thus obtaining a final number of cluster centers.

[0038] Based on the clustering results, the spatiotemporal feature vectors corresponding to each node are divided into several clusters, resulting in spatiotemporal feature vector clusters.

[0039] Optionally, step six specifically includes:

[0040] By performing statistical analysis on all spatiotemporal feature vectors in the spatiotemporal feature vector cluster, the statistical analysis steps include:

[0041] Calculate the mean and standard deviation of each feature in all spatiotemporal feature vectors;

[0042] The normal range for each feature is set based on the mean and standard deviation. The normal range is defined as the range of the feature value mean plus or minus a certain number of standard deviations. Features that exceed the normal range are considered potential anomalies, thus obtaining the fault detection results of the synchronous condenser equipment.

[0043] Optionally, step seven specifically includes:

[0044] Based on the fault detection results of the synchronous condenser equipment, the status of the synchronous condenser equipment is evaluated. The evaluation steps are as follows:

[0045] Based on the fault detection results, identify the type of fault and determine the location or part of the equipment where the fault occurred;

[0046] The severity of the fault is assessed based on the fault detection results that exceed the normal range. Faults that exceed the normal range by more than a preset threshold are assessed as high-risk faults, while faults that exceed the preset threshold are assessed as low-risk faults.

[0047] Based on the comparison results of fault type, severity and threshold, fault warning information is generated and the fault warning is notified to relevant operators through multiple channels.

[0048] The monitoring strategy of the equipment is adjusted according to the generated fault warning information. The monitoring strategy includes increasing the monitoring frequency of relevant components and increasing the data acquisition frequency of sensors for equipment with high-risk faults or equipment that has already failed.

[0049] An online monitoring system for the status of a synchronous condenser based on a smart sensor, according to an embodiment of the present invention, includes the following modules:

[0050] The data acquisition and standardization module is used to acquire raw operating data of the synchronous condenser equipment through multiple sensors, and to standardize the raw operating data to obtain the synchronous condenser operating dataset.

[0051] The heterogeneous graph construction module is used to take each sensor as a node based on the camera adjustment dataset and construct edges based on the spatial layout of the sensors to obtain a heterogeneous graph of the camera adjustment.

[0052] The spatial feature learning module is used to input the heterogeneous map of the camera condenser into the Siamese network for spatial feature learning, and obtain the spatial feature vector set of the camera condenser equipment.

[0053] The detrending analysis module is used to sort the spatial feature vector set by stamp and perform differential processing to obtain the detrended time series dataset of the camera adjustment device.

[0054] The spatiotemporal feature analysis module is used to input the detrended time-series dataset of the camera condenser into the improved GraphSAGE model. Through the spatial information encoding module, the neighbor sampling update module, and the spatiotemporal dependency aggregation module, spatiotemporal feature analysis is performed to obtain spatiotemporal feature vector clusters.

[0055] Anomaly detection module is used to perform anomaly detection based on the spatiotemporal feature vector cluster through statistical analysis, and obtain the fault detection result of the synchronous condenser equipment;

[0056] The fault early warning module is used to assess the status of the synchronous condenser equipment based on the fault detection results, generate fault early warning information, and adjust the equipment monitoring strategy.

[0057] The beneficial effects of this invention are:

[0058] This invention provides an online monitoring method for the status of synchronous condenser equipment based on intelligent sensors, achieving significant technological breakthroughs and beneficial effects compared to existing technologies. By combining multimodal sensor data acquisition, spatial feature learning, spatiotemporal feature analysis, and fault detection, this method not only improves the accuracy of equipment status prediction but also greatly enhances the intelligence level of fault early warning and monitoring strategy adjustment. In traditional methods, equipment monitoring relies on single sensor data or monitoring methods based on fixed rules, lacking comprehensive analysis and flexible processing of equipment status, resulting in the inability to detect potential faults in a timely and accurate manner. This invention, however, obtains a unified structured synchronous condenser operation dataset through standardized processing of multiple sensor data. Based on this, it utilizes graph convolutional networks and deep learning algorithms to effectively capture the spatiotemporal dependencies between sensor data, achieving comprehensive monitoring of the synchronous condenser equipment status.

[0059] By using a Siamese network for spatial feature learning and an improved GraphSAGE model for spatiotemporal feature analysis, this invention can deeply mine the similarities between sensor data, optimize equipment status prediction and anomaly detection, and not only identify the current operating status of the equipment, but also predict possible future fault states based on historical data and the spatial relationships of the equipment. The innovation of this method lies in its ability to dynamically discover potential equipment faults through statistical analysis and anomaly detection of spatiotemporal feature vector clusters, and automatically generate fault warnings based on real-time equipment operation data. This fault detection mechanism based on real-time data and intelligent analysis overcomes the dependence on threshold settings in traditional methods, and can more accurately and flexibly address fault risks under different types of equipment and different environmental conditions.

[0060] In terms of equipment monitoring strategies, the technical solution of this invention also demonstrates significant advantages. When equipment malfunctions or has a potential malfunction, the system not only generates detailed fault warning information but also automatically adjusts the monitoring strategy, performing targeted monitoring based on the severity of different faults. This automatic adjustment capability ensures that high-risk faults are handled promptly while reducing resource waste on low-risk faults, thereby improving the efficiency and intelligence level of equipment monitoring. In summary, this invention achieves data acquisition and real-time analysis based on intelligent sensors, providing more accurate and efficient status monitoring and fault warnings for synchronous condenser equipment, significantly improving the safety and stability of equipment operation, and possessing significant technical value and application prospects. Attached Figure Description

[0061] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0062] Figure 1 This is an overall flowchart of an online monitoring method for the status of a synchronous condenser based on intelligent sensors, as proposed in this invention.

[0063] Figure 2 This is a schematic diagram of the structure of an online monitoring system for the status of a synchronous condenser based on intelligent sensors, as proposed in this invention. Detailed Implementation

[0064] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0065] refer to Figure 1 A method for online monitoring of the status of a synchronous condenser based on intelligent sensors, comprising:

[0066] Step 1: Collect raw operating data of the synchronous condenser equipment through multiple sensors, and standardize the raw operating data to obtain the synchronous condenser operating dataset;

[0067] Step 2: Based on the camera condenser's running dataset, each sensor is treated as a node, and connection edges are constructed according to the spatial layout of the sensors to obtain a heterogeneous graph of the camera condenser;

[0068] Step 3: Input the heterogeneous image of the camera condenser into the Siamese network for spatial feature learning to obtain the spatial feature vector set of the camera condenser equipment;

[0069] Step 4: Sort the spatial feature vector set by stamp and perform differential processing to obtain the detrended time series dataset of the camera device;

[0070] Step 5: Input the detrended time series dataset from the camera adjustment device into the improved GraphSAGE model, and perform spatiotemporal feature analysis through the spatial information encoding module, the neighbor sampling update module, and the spatiotemporal dependency aggregation module to obtain spatiotemporal feature vector clusters;

[0071] Step 6: Based on the spatiotemporal feature vector cluster, perform anomaly detection through statistical analysis to obtain the fault detection results of the synchronous condenser equipment;

[0072] Step 7: Based on the fault detection results of the synchronous condenser equipment, assess the status of the synchronous condenser equipment, generate fault warning information, and adjust the equipment monitoring strategy.

[0073] In this embodiment, the various sensors include temperature sensors, pressure sensors, vibration sensors, audio sensors, infrared imaging sensors, and chemical sensors. These sensors work together to monitor the real-time status of the synchronous condenser equipment, collecting data such as temperature, pressure, vibration, audio signals, infrared images, and chemical composition during equipment operation. The different data types and measurement ranges of each sensor ensure comprehensive monitoring and multi-faceted analysis of the equipment's status.

[0074] The standardization processing steps include performing the following processing on different types of sensor data respectively:

[0075] During data acquisition, some sensors may generate abnormal data due to equipment malfunction, environmental interference, or other reasons. To ensure data quality, statistical methods (3σ rule) are used to remove data that exceeds the normal range. For example, for temperature sensors, if the measured temperature value exceeds the normal operating temperature range of the equipment, the data will be considered an outlier and discarded.

[0076] During the data acquisition process, data loss may occur at certain times. To address the missing data, a linear interpolation method is used to fill in the missing data.

[0077] Because the sampling frequencies of different sensors may differ, the timestamps of the raw data collected may not be entirely consistent. To perform effective time series analysis, it is necessary to align the time series data from each sensor. Specifically, this involves unifying the timestamps of all sensors onto a standard time series, using the closest timestamp alignment method to synchronize the data from different sensors to the same point in time. For example, if the pressure sensor samples once per minute and the temperature sensor samples once per second, the system averages the data from each second to meet the overall time series analysis requirements of the equipment.

[0078] Data collected by various sensors may differ in units and ranges. To enable comparison and fusion of different data types, normalization is necessary. Common normalization methods include min-max normalization and Z-score standardization. For temperature data, the normal range is -20°C to 80°C. After min-max normalization, the data is compressed to the range [0,1]. For pressure sensor data, the normal operating pressure may be between 0 and 1000 Pa. After normalization, all pressure data are standardized to the same units, facilitating subsequent unified analysis. Finally, the sensor data processed in the above way will form a unified structured synchronous condenser operation dataset. This dataset contains standardized data from all sensors, and all data has a unified timestamp and format, providing high-quality basic data for subsequent data analysis, feature extraction, and fault detection.

[0079] In this embodiment, step two specifically includes:

[0080] Each sensor in the synchronous condenser equipment is treated as a node in the heterogeneous graph of the synchronous condenser. Within the equipment, each sensor monitors and collects different types of operational data, such as temperature, pressure, vibration, audio signals, infrared imaging data, or chemical data. The characteristics of each sensor node consist of the raw operational data collected by the corresponding sensor in the synchronous condenser's operational dataset, including but not limited to sensor measurements, timestamps, and sensor type. The characteristics of each node describe the sensor's status and operational condition within the equipment, providing foundational data for subsequent equipment status analysis and fault detection.

[0081] Based on the actual spatial layout of the synchronous condenser equipment, the spatial relationships between each sensor are determined. The installation position of each sensor is typically based on the equipment's coordinate system, using three-dimensional coordinates (X, Y, Z coordinates) to represent the sensor's spatial location. Using the installation coordinates of each sensor, the spatial distance between them can be calculated. Spatial distance calculations typically use standard measurement methods such as Euclidean distance or Manhattan distance to ensure accurate reflection of the physical positional relationships between the sensors within the equipment. For example, if a temperature sensor and a pressure sensor are installed in different parts of the equipment, the spatial distance between them can be obtained by calculating the difference in their three-dimensional coordinates.

[0082] Based on the spatial distance between sensors, edge connections are constructed in the heterogeneous graph of the phase condenser. The heterogeneous graph of the phase condenser is a graph structure composed of sensor nodes and edges, where each sensor is a node, and nodes are connected by edges to form the graph structure. The construction of edges not only reflects the connection relationships between sensors but also reveals their spatial dependencies and mutual influences. Edge connections between sensor nodes are established based on their spatial distance. Specifically, when the distance between two nodes is less than a preset maximum connection distance, a connection edge is established between them. For example, if the distance between a temperature sensor and a pressure sensor is less than a set maximum threshold (5 meters), an edge connection will be established between them.

[0083] While establishing edge connections, a weight is assigned to each edge, which represents the spatial distance between the sensors. The method for setting the edge weights can be determined according to actual needs. For example, the reciprocal of the Euclidean distance can be used as the edge weight, making the connection between sensors that are closer together more compact, thus reflecting a stronger mutual influence in the graph structure. In this way, the spatial relationship between sensors is intuitively represented as edges in the graph, and their spatial dependence is reflected by the edge weights. After these steps, a complete heterogeneous graph of the camera is obtained, which consists of each sensor as a node, connected by edges reflecting spatial relationships. The weight of each edge represents the spatial distance between the sensors, providing a basic graph structure for subsequent spatial feature analysis, state prediction, and fault detection.

[0084] In this embodiment, step three specifically includes:

[0085] The feature vector of each sensor node in the heterogeneous graph of the camera condenser is input to the input layer of the Siamese network. Each pair of adjacent sensor nodes is combined into a pair of input data to obtain the input feature pair. The feature vector of the sensor node includes the corresponding sensor data, timestamp and spatial location in the camera condenser running dataset.

[0086] The input feature pairs are fed into a sub-network of the Siamese network to obtain a feature vector set of the input data pairs. The sub-network consists of a multi-layer fully connected neural network, which extracts the feature representation of the input feature pairs through the activation functions of multiple neurons.

[0087] The input data pairs feature vector sets are input to the similarity calculation layer. The cosine similarity of each pair of input data pairs feature vectors is calculated. The calculated cosine similarity is used as the spatial similarity score to obtain the spatial similarity score set. The cosine similarity represents the spatial relationship between sensor data.

[0088] The spatial similarity score set is input into the spatial feature fusion layer, and the similarity score is used as a weight to weight the feature vectors of the input data to obtain the spatial feature vector set of the camera adjustment device.

[0089] This step, by combining Siamese networks and spatial information encoding, successfully improves the accuracy and intelligence level of synchronous condenser equipment status monitoring. Through the fusion of multiple sensor data and deep learning of spatiotemporal features, the method can capture the complex spatial dependencies between sensors and accurately identify changes in equipment status. Cosine similarity calculation and spatial feature fusion effectively integrate the spatial and temporal information of the sensors, providing more accurate data support for equipment fault prediction and anomaly detection. By automatically learning the similarity between sensors, this method not only improves the accuracy of equipment status prediction but also monitors abnormal equipment status in real time and provides early warnings of potential faults. Furthermore, the optimized spatial feature representation makes the equipment monitoring strategy more intelligent, reduces manual intervention, improves monitoring efficiency and equipment maintenance response speed, and significantly enhances the safety and reliability of synchronous condenser equipment.

[0090] In this embodiment, step four specifically includes:

[0091] The spatial feature vector set of the synchronous condenser equipment is used as input to represent the state-space characteristics of each sensor in the equipment. Each sensor's spatiotemporal feature vector contains information such as measurement data, timestamps, and spatial location. These feature vectors are derived from the raw data collected by the sensors through the aforementioned steps and further processed in the spatial information encoding module to obtain the complete state characteristics of each sensor. Next, these spatial feature vector sets are sorted according to their timestamps to obtain the time-series data for each sensor. In this way, the sensor's time-series data is organized into a time-series format, ensuring that subsequent processing can effectively analyze the data in chronological order.

[0092] Differential processing is performed on the time-series data for each sensor. The purpose of differential processing is to remove long-term trends from the time-series data, helping to identify short-term fluctuations and potential anomalies. In this step, differential processing specifically calculates the difference between the current data and the data from the previous time. This processing method effectively eliminates potential trend changes in the data, making short-term changes in the device's state more significant. For example, for temperature data, if the current temperature is 50°C and the previous temperature was 48°C, the calculated difference is 2°C, representing the temperature change over a short period.

[0093] Trend removal is performed on the time-series data of each sensor using a moving average method. The moving average method smooths out fluctuations in the data, removing long-term trends or periodic changes, and focusing more on short-term variations. It works by calculating the average of the data over a specific time window and subtracting this average value, thus removing long-term trends. Specifically, a moving average window (e.g., 5 time units) is set, and the time-series data of each sensor is processed to obtain detrended data. This process effectively eliminates seasonal variations or long-term periodic fluctuations in equipment status, allowing for better focus on short-term changes in the equipment.

[0094] Calculate the standard deviation of each sensor's data within a sliding time window. The standard deviation quantifies the volatility of the data, reflecting its dispersion. The standard deviation of each sensor's time-series data within the sliding window reveals the volatility of the device's state, helping to identify abnormal changes. For example, if the temperature changes significantly within a certain period, the standard deviation will increase significantly, indicating that the device may be experiencing abnormal fluctuations. When the standard deviation exceeds a preset normal threshold range, the sensor's state is marked as abnormal. The preset normal threshold can be set based on the volatility of historical data or the device's operating range. For example, if the temperature standard deviation exceeds a set 0.5°C, it may indicate that the device's temperature fluctuations are outside the normal range, and the system will mark it as abnormal for subsequent fault diagnosis.

[0095] The detrended data is subjected to weighted linear regression within a sliding window to predict data values ​​at future time points. The purpose of weighted linear regression is to predict future equipment status changes based on known time-series data. Weighted regression differs from ordinary linear regression in that it dynamically adjusts the weight of each data point based on historical data volatility or other factors. Data points with larger weights have a greater impact on the regression results, while data points with smaller volatility have a smaller impact. Through weighted linear regression, the future trend of equipment status changes can be predicted, thus better anticipating potential faults or anomalies. Through these processing steps, a detrended time-series dataset of the synchronous condenser equipment is finally obtained. This dataset not only removes long-term trends and periodic fluctuations but also retains the short-term change characteristics of the equipment status, making it suitable for subsequent fault detection, equipment health management, and anomaly prediction, providing a more accurate data foundation for online monitoring of synchronous condenser equipment.

[0096] In this embodiment, the improved GraphSAGE model is specifically as follows:

[0097] The detrended time series dataset of the camera adjustment device is input into the spatial information encoding module, the spatial location of each sensor node is converted into a vector representation, and concatenated with the spatial feature vector to obtain a spatiotemporal feature vector set;

[0098] The Manhattan distance between every two spatiotemporal feature vectors in the spatiotemporal feature vector set is calculated. If the Manhattan distance is less than a preset distance threshold, an edge connection is established between the corresponding sensor nodes to obtain the spatiotemporal feature map of the sensor. The edge weight of the edge connection is the corresponding Manhattan distance.

[0099] The spatiotemporal feature map is input to the neighbor sampling and update module. Based on the connection relationship between the spatiotemporal feature maps of the sensor node and the neighbor node, the feature map is sampled, and the feature map of the neighbor node is transmitted to the target node and fused to form the updated node feature map.

[0100] The fusion is to multiply the spatiotemporal feature vector corresponding to the neighbor node by the weight of the connection edge between the neighbor node and the target node, and then add it to the spatiotemporal feature vector corresponding to the target node.

[0101] The fused feature map of neighbor nodes is input into the spatiotemporal dependency aggregation module, and several vectors are randomly selected from the fused feature map of neighbor nodes as initial cluster centers.

[0102] Calculate the Euclidean distance between the spatiotemporal feature vector corresponding to each node in the neighbor node fusion feature map and the initial cluster center, and assign each node to the initial cluster center with the closest Euclidean distance;

[0103] Based on the mean of the spatiotemporal feature vectors corresponding to all nodes in each initial cluster center, the position of the cluster center is recalculated, and the cluster center is iteratively updated repeatedly until the number of changes in the cluster center is less than a preset threshold or the preset maximum number of iterations is reached, thus obtaining a final number of cluster centers.

[0104] Based on the clustering results, the spatiotemporal feature vectors corresponding to each node are divided into several clusters, resulting in spatiotemporal feature vector clusters.

[0105] This step improves the accuracy and intelligence of phase-shifting equipment status monitoring through an enhanced GraphSAGE model. By performing deep learning and analysis on the spatiotemporal characteristics of each sensor in the equipment, this method can accurately capture the spatial dependencies between sensors, thereby optimizing equipment status prediction. The Manhattan distance calculation and spatiotemporal feature fusion strategy enable the system to establish a more accurate sensor relationship graph, fully reflecting the spatial connection strength between sensors, thus improving the accuracy of fault detection. The spatiotemporal dependency aggregation module further optimizes the feature vector clustering process, allowing for more effective classification and analysis of the status of each part of the equipment, reducing false alarms and missed alarms. Simultaneously, this invention can perform real-time monitoring and fault early warning based on spatiotemporal feature vector clusters, intelligently adjusting equipment monitoring strategies to ensure that high-risk equipment receives timely attention, greatly improving equipment management efficiency and reliability.

[0106] In this embodiment, step six specifically includes:

[0107] Statistical analysis is performed on all spatiotemporal feature vectors in the spatiotemporal feature vector cluster to assess equipment status and detect anomalies. The statistical analysis step includes a comprehensive statistical analysis of all data in the spatiotemporal feature vector cluster to determine which features are outside the normal range, thereby identifying potential equipment faults. The specific operation is as follows:

[0108] Calculate the mean and standard deviation of each feature in all spatiotemporal feature vectors. These features may include various equipment status data such as temperature, pressure, vibration amplitude, and chemical concentration. The mean represents the central position of each feature in the entire dataset, while the standard deviation reflects the range of data fluctuation. For example, for temperature sensor data, calculate the mean and standard deviation of the temperature values ​​at all time points. The mean provides a temperature reference when the equipment is operating normally, while the standard deviation indicates the fluctuation range of the equipment under normal operating conditions.

[0109] Based on the calculated mean and standard deviation, a normal range is further set for each feature. The normal range can be determined using the following formula: the normal range of a feature value is the mean plus or minus a certain number of standard deviations. Typically, three times the standard deviation is chosen as the threshold for the normal range; that is, feature values ​​falling within three times the standard deviation of the mean are considered normal. Specifically, for temperature sensor data, if the mean temperature is 50°C and the standard deviation is 2°C, then the normal temperature range is defined as 50°C ± 6°C, meaning temperature values ​​between 44°C and 56°C are considered normal. If the temperature exceeds this range, it is considered abnormal, potentially indicating a device malfunction or abnormal fluctuation.

[0110] Based on the established normal range, each feature in each spatiotemporal feature vector is evaluated. If the value of a feature exceeds its normal range, that feature is considered a potential anomaly. These anomalous data represent deviations from the device's state and may be precursors to malfunctions or potential problems. For example, if a temperature sensor measures a temperature of 62°C at a certain moment, exceeding the normal range of 56°C, the temperature sensor's data will be marked as anomalous, indicating a possible overheating problem in the device.

[0111] Based on the above statistical analysis and anomaly detection, the final fault detection results for the synchronous condenser equipment are obtained. These results include all detected abnormal features and their corresponding locations, indicating potential faults or instabilities in the equipment during operation. The fault detection results can be used for subsequent fault diagnosis and equipment health assessment, and provide timely early warning information to equipment maintenance personnel. Through this series of statistical analyses and anomaly detection processes, this invention can effectively identify potential faults in synchronous condenser equipment during operation, ensuring real-time monitoring and accurate diagnosis of equipment status, and improving equipment safety and reliability.

[0112] Based on the fault detection results of the synchronous condenser equipment, the status of the synchronous condenser equipment is assessed to ensure its health and provide timely warnings of potential faults. The assessment steps include:

[0113] Based on the fault detection results, the system identifies the type of fault and determines its location or component within the equipment. Specifically, the system uses abnormal data from different sensors to determine whether the fault occurs in terms of temperature, pressure, vibration, chemical composition, etc. Then, by analyzing the spatial location of the sensors, the system pinpoints the specific area where the fault occurs. For example, if a temperature sensor shows an over-temperature anomaly, the system will further confirm which area of ​​the equipment the fault occurs in, such as the cooling system, heat exchanger, or other components.

[0114] Based on the abnormal data in the fault detection results, the severity of the fault is assessed. Specifically, this is done by comparing the sensor measurements with preset normal ranges to determine the fault's risk level. If the detected fault data exceeds the preset normal range, and the extent of the exceedance is greater than a set preset threshold, the fault is assessed as a high-risk fault. For example, if the equipment's temperature sensor data exceeds the set maximum temperature threshold (60°C), and this exceedance is more than 2°C, the temperature problem is considered high-risk and may lead to equipment damage or malfunction. However, if the exceedance is less than the preset threshold (within 1°C), the fault is assessed as a low-risk fault and is unlikely to cause serious problems. The specific value of the preset threshold can be set based on the equipment's operating specifications and historical operating data, typically using the standard deviation range of the equipment's long-term operating status.

[0115] Based on the comparison results of fault type, severity, and threshold, a fault warning message is generated. The warning message will describe in detail the fault type, location, severity, and potential impact on equipment operation. The system will promptly deliver the warning message to relevant operators or maintenance personnel through various channels (such as SMS, email, and mobile push notifications) to ensure they are aware of the abnormal equipment status and can take appropriate countermeasures.

[0116] Based on the generated fault warning information, the system will automatically adjust the equipment monitoring strategy to ensure more accurate monitoring of high-risk parts of the equipment. Specifically, for high-risk faults or equipment that has already failed, the system will automatically increase the monitoring frequency of relevant components and the data acquisition frequency of sensors. For example, if a temperature sensor detects an abnormal temperature and is assessed as high-risk, the system will automatically increase the data acquisition frequency of the temperature sensor, possibly increasing the sampling rate, to record temperature changes in the equipment in real time and prevent the fault from worsening; for low-risk faults, the monitoring frequency may remain within the normal range.

[0117] This invention can automatically assess equipment status based on fault detection results and promptly notify operators through an intelligent early warning system, ensuring that equipment faults are detected and addressed as early as possible. Simultaneously, the system can flexibly adjust monitoring strategies to more accurately monitor critical equipment components, significantly improving operational safety and reliability and preventing major losses due to untimely fault detection.

[0118] refer to Figure 2 An online monitoring system for the status of a synchronous condenser based on intelligent sensors includes the following modules:

[0119] The data acquisition and standardization module is used to acquire raw operating data of the synchronous condenser equipment through multiple sensors, and to standardize the raw operating data to obtain the synchronous condenser operating dataset.

[0120] The heterogeneous graph construction module is used to take each sensor as a node based on the camera adjustment dataset and construct edges based on the spatial layout of the sensors to obtain a heterogeneous graph of the camera adjustment.

[0121] The spatial feature learning module is used to input the heterogeneous map of the camera condenser into the Siamese network for spatial feature learning, and obtain the spatial feature vector set of the camera condenser equipment.

[0122] The detrending analysis module is used to sort the spatial feature vector set by stamp and perform differential processing to obtain the detrended time series dataset of the camera adjustment device.

[0123] The spatiotemporal feature analysis module is used to input the detrended time-series dataset of the camera condenser into the improved GraphSAGE model. Through the spatial information encoding module, the neighbor sampling update module, and the spatiotemporal dependency aggregation module, spatiotemporal feature analysis is performed to obtain spatiotemporal feature vector clusters.

[0124] Anomaly detection module is used to perform anomaly detection based on the spatiotemporal feature vector cluster through statistical analysis, and obtain the fault detection result of the synchronous condenser equipment;

[0125] The fault early warning module is used to assess the status of the synchronous condenser equipment based on the fault detection results, generate fault early warning information, and adjust the equipment monitoring strategy.

[0126] Example 1:

[0127] To verify the feasibility of this invention in practice, it was applied to a large power company. Synchronous condensers are widely used to regulate grid voltage and power system stability. The company adopted the online monitoring method for synchronous condenser status based on intelligent sensors, as described in this invention, to improve equipment operating efficiency and reduce the probability of failures. During equipment operation, various sensors, including temperature, pressure, vibration, audio, infrared imaging, and chemical sensors, are widely used to comprehensively monitor the working status of the synchronous condenser. These sensors collect physical and chemical state data from different parts of the equipment during operation. Since synchronous condensers typically operate in high-load, high-pressure environments, various equipment failures (such as excessive temperature, excessive pressure, or abnormal vibration) can lead to serious power system failures or even power outages. Therefore, real-time and accurate monitoring of equipment status and timely detection of potential failures have become crucial tasks in equipment management. When applying the technical solution of this invention, firstly, multiple sensors are installed at various parts of the equipment and collect data in real time. For example, temperature sensors monitor the internal temperature of the equipment, pressure sensors monitor the internal pressure, vibration sensors monitor the vibration amplitude, audio sensors monitor sound fluctuations during equipment operation, infrared imaging sensors detect thermal anomalies, and chemical sensors detect the composition of gases surrounding the equipment. The system first performs standardization processing on the collected sensor data, including outlier removal, missing value imputation, timestamp alignment, and normalization, ensuring that all sensor data can be analyzed under the same standards. This not only guarantees data quality but also facilitates subsequent comprehensive analysis of the equipment's condition.

[0128] The standardized data was used to construct a heterogeneous graph of the synchronous condenser equipment. Each sensor was treated as a node in the graph, and the node features consisted of relevant operational data collected by that sensor. The spatial layout was calculated using the installation coordinates of each sensor, thus constructing spatial relationship edges between the sensors. These spatial relationships helped the system capture the interdependencies between various parts of the equipment. To enhance spatial feature learning, the system used a Siamese network to learn spatiotemporal features, further optimizing the equipment status prediction and fault detection capabilities by calculating the similarity between adjacent sensor nodes.

[0129] This method allows for equipment status prediction to move beyond relying solely on single sensor data. Instead, it integrates multi-dimensional data from various sensors, reflecting the overall operational status of the equipment. For instance, during the operation of a synchronous condenser, a temperature sensor detected the equipment's temperature rising to 82°C. The system automatically generated a high-temperature warning based on a pre-set threshold (60°C ± 2°C). At this point, the system, through real-time assessment of the equipment status, not only identified sensor nodes with abnormal temperatures but also discovered significant fluctuations in pressure and vibration sensors on one side of the equipment through spatiotemporal feature analysis, further confirming a potential equipment failure risk. The system automatically increases the monitoring frequency of these critical components, ensuring that potential equipment damage is detected promptly through further data collection and analysis.

[0130] To verify the beneficial effects of this invention, the following is a comparative data table of experimental results from the company's actual application:

[0131] Table 1. Analysis of Fault Detection and Early Warning of Synchronous Condenser Equipment

[0132] Device status Fault detection rate (%) Fault warning response time (minutes) Fault accuracy (%) Temperature too high 95 10 92 Pressure anomaly 90 12 88 Vibration too large 93 8 90 Other faults 85 15 85

[0133] Analysis of the data in Table 1 shows that this invention has significant advantages in fault detection, achieving accuracy rates of 95%, 90%, and 93%, respectively. These figures indicate that the method of this invention can more accurately capture potential equipment faults, improving the safety of equipment operation. Regarding fault warning response time, this invention provides a 10-minute warning response time for overheating faults; 12 minutes for abnormal pressure faults; and 8 minutes for excessive vibration faults. The rapid warning response provides operators with more time to take countermeasures, reducing losses caused by faults. The fault accuracy of this invention is high. The accuracy for overheating faults is 92%, for abnormal pressure faults 88%, and for excessive vibration faults 90%. These high accuracy values ​​indicate that the method of this invention can effectively reduce false alarms and missed alarms, ensuring that only genuine faults are detected and warned, thereby improving the efficiency of equipment management. By combining data from multiple sensors and intelligent algorithms, this invention significantly improves the accuracy and response speed of equipment monitoring in fault detection and warning, providing a strong guarantee for the safe operation of equipment.

[0134] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for online monitoring of the status of a synchronous condenser based on intelligent sensors, characterized in that, include: Step 1: Collect raw operating data of the synchronous condenser equipment through multiple sensors, and standardize the raw operating data to obtain the synchronous condenser operating dataset; Step 2: Based on the camera condenser's running dataset, each sensor is treated as a node, and connection edges are constructed according to the spatial layout of the sensors to obtain a heterogeneous graph of the camera condenser; Step 3: Input the heterogeneous image of the camera condenser into the Siamese network for spatial feature learning to obtain the spatial feature vector set of the camera condenser equipment; Step 4: Sort the spatial feature vector set by stamp and perform differential processing to obtain the detrended time series dataset of the camera device; Step 5: Input the detrended time series dataset from the camera adjustment device into the improved GraphSAGE model, and perform spatiotemporal feature analysis through the spatial information encoding module, the neighbor sampling update module, and the spatiotemporal dependency aggregation module to obtain spatiotemporal feature vector clusters; Step 6: Based on the spatiotemporal feature vector cluster, perform anomaly detection through statistical analysis to obtain the fault detection results of the synchronous condenser equipment; Step 7: Based on the fault detection results of the synchronous condenser equipment, assess the status of the synchronous condenser equipment, generate fault warning information, and adjust the equipment monitoring strategy.

2. The method for online monitoring of the status of a synchronous condenser based on intelligent sensors according to claim 1, characterized in that, The various sensors include temperature sensors, pressure sensors, vibration sensors, audio sensors, infrared imaging sensors, and chemical sensors; the standardization processing steps include outlier removal, missing value imputation, timestamp alignment, and normalization of different types of sensor data to obtain a unified structured camera condenser operation dataset.

3. The method for online monitoring of the status of a synchronous condenser based on intelligent sensors according to claim 1, characterized in that, Step two specifically involves: Each sensor of the synchronous condenser equipment is treated as a node in the synchronous condenser heterogeneous graph, and the node features of each sensor are composed of the operational data collected by the corresponding sensor in the synchronous condenser operational dataset. Based on the actual spatial layout of the synchronous condenser equipment, the spatial distance between each sensor is determined by the installation coordinates of each sensor; An edge connection is constructed based on the spatial distance between the sensors to obtain a heterogeneous graph of the camera, where the edge weight of the edge connection is the spatial distance between the sensors.

4. The method for online monitoring of the status of a synchronous condenser based on intelligent sensors according to claim 1, characterized in that, Step three specifically involves: The feature vector of each sensor node in the heterogeneous graph of the camera condenser is input to the input layer of the Siamese network. Each pair of adjacent sensor nodes is combined into a pair of input data to obtain the input feature pair. The feature vector of the sensor node includes the corresponding sensor data, timestamp and spatial location in the camera condenser running dataset. The input feature pairs are fed into a sub-network of the Siamese network to obtain a feature vector set of the input data pairs. The sub-network consists of a multi-layer fully connected neural network, which extracts the feature representation of the input feature pairs through the activation functions of multiple neurons. The input data pairs feature vector sets are input to the similarity calculation layer. The cosine similarity of each pair of input data pairs feature vectors is calculated. The calculated cosine similarity is used as the spatial similarity score to obtain the spatial similarity score set. The cosine similarity represents the spatial relationship between sensor data. The spatial similarity score set is input into the spatial feature fusion layer, and the similarity score is used as a weight to weight the feature vectors of the input data to obtain the spatial feature vector set of the camera adjustment device.

5. The method for online monitoring of the status of a synchronous condenser based on intelligent sensors according to claim 1, characterized in that, Step four specifically involves: The spatial feature vector set of the camera condenser is used as input to represent the spatial features of the state of each sensor of the camera condenser, and sorted according to timestamps to obtain the time series data of each sensor. The time-series data of each sensor are differentially processed, which involves calculating the difference between the data at the current time and the data at the previous time. The time series data of each sensor is detrended by using a moving average; Calculate the standard deviation of each sensor data within the sliding time window. When the standard deviation exceeds the preset normal threshold range, it is marked as an abnormal state. The detrended data is subjected to weighted linear regression within a sliding window to predict data values ​​for several future time points, thus obtaining the detrended time series dataset of the camera adjustment device.

6. The method for online monitoring of the status of a synchronous condenser based on intelligent sensors according to claim 1, characterized in that, The improved GraphSAGE model is specifically as follows: The detrended time series dataset of the camera adjustment device is input into the spatial information encoding module, the spatial location of each sensor node is converted into a vector representation, and concatenated with the spatial feature vector to obtain a spatiotemporal feature vector set; The Manhattan distance between every two spatiotemporal feature vectors in the spatiotemporal feature vector set is calculated. If the Manhattan distance is less than a preset distance threshold, an edge connection is established between the corresponding sensor nodes to obtain the spatiotemporal feature map of the sensor. The edge weight of the edge connection is the corresponding Manhattan distance. The spatiotemporal feature map is input to the neighbor sampling and update module. Based on the connection relationship between the spatiotemporal feature maps of the sensor node and the neighbor node, the feature map is sampled, and the feature map of the neighbor node is transmitted to the target node and fused to form the updated node feature map. The fusion is to multiply the spatiotemporal feature vector corresponding to the neighbor node by the weight of the connection edge between the neighbor node and the target node, and then add it to the spatiotemporal feature vector corresponding to the target node. The fused feature map of neighbor nodes is input into the spatiotemporal dependency aggregation module, and several vectors are randomly selected from the fused feature map of neighbor nodes as initial cluster centers. Calculate the Euclidean distance between the spatiotemporal feature vector corresponding to each node in the neighbor node fusion feature map and the initial cluster center, and assign each node to the initial cluster center with the closest Euclidean distance; Based on the mean of the spatiotemporal feature vectors corresponding to all nodes in each initial cluster center, the position of the cluster center is recalculated, and the cluster center is iteratively updated repeatedly until the number of changes in the cluster center is less than a preset threshold or the preset maximum number of iterations is reached, thus obtaining a final number of cluster centers. Based on the clustering results, the spatiotemporal feature vectors corresponding to each node are divided into several clusters, resulting in spatiotemporal feature vector clusters.

7. The method for online monitoring of the status of a synchronous condenser based on intelligent sensors according to claim 1, characterized in that, Step six specifically involves: By performing statistical analysis on all spatiotemporal feature vectors in the spatiotemporal feature vector cluster, the statistical analysis steps include: Calculate the mean and standard deviation of each feature in all spatiotemporal feature vectors; The normal range for each feature is set based on the mean and standard deviation. The normal range is defined as the range of the feature value mean plus or minus a certain number of standard deviations. Features that exceed the normal range are considered potential anomalies, thus obtaining the fault detection results of the synchronous condenser equipment.

8. The method for online monitoring of the status of a synchronous condenser based on intelligent sensors according to claim 1, characterized in that, Step seven specifically involves: Based on the fault detection results of the synchronous condenser equipment, the status of the synchronous condenser equipment is evaluated. The evaluation steps are as follows: Based on the fault detection results, identify the type of fault and determine the location or part of the equipment where the fault occurred; The severity of the fault is assessed based on the fault detection results that exceed the normal range. Faults that exceed the normal range by more than a preset threshold are assessed as high-risk faults, while faults that exceed the preset threshold are assessed as low-risk faults. Based on the comparison results of fault type, severity and threshold, fault warning information is generated and the fault warning is notified to relevant operators through multiple channels. The monitoring strategy of the equipment is adjusted according to the generated fault warning information. The monitoring strategy includes increasing the monitoring frequency of relevant components and increasing the data acquisition frequency of sensors for equipment with high-risk faults or equipment that has already failed.

9. A system for online monitoring of the status of a synchronous condenser based on intelligent sensors, comprising executing the method for online monitoring of the status of a synchronous condenser based on intelligent sensors as described in any one of claims 1 to 8, characterized in that, Includes the following modules: The data acquisition and standardization module is used to acquire raw operating data of the synchronous condenser equipment through multiple sensors, and to standardize the raw operating data to obtain the synchronous condenser operating dataset. The heterogeneous graph construction module is used to take each sensor as a node based on the camera adjustment dataset and construct edges based on the spatial layout of the sensors to obtain a heterogeneous graph of the camera adjustment. The spatial feature learning module is used to input the heterogeneous map of the camera condenser into the Siamese network for spatial feature learning, and obtain the spatial feature vector set of the camera condenser equipment. The detrending analysis module is used to sort the spatial feature vector set by stamp and perform differential processing to obtain the detrended time series dataset of the camera adjustment device. The spatiotemporal feature analysis module is used to input the detrended time series dataset of the camera condenser into the improved GraphSAGE model. It performs spatiotemporal feature analysis through the spatial information encoding module, the neighbor sampling update module, and the spatiotemporal dependency aggregation module to obtain a spatiotemporal feature vector cluster. Anomaly detection module is used to perform anomaly detection based on the spatiotemporal feature vector cluster through statistical analysis, and obtain the fault detection result of the synchronous condenser equipment; The fault early warning module is used to assess the status of the synchronous condenser equipment based on the fault detection results, generate fault early warning information, and adjust the equipment monitoring strategy.

Images