A remote intelligent monitoring system, method and device for a power distribution room

Abstract

The application provides a remote intelligent monitoring system, method and equipment for a power distribution room, relates to the technical field of remote monitoring, and the system comprises a data stream generation module, a fault mode identification module, a risk analysis module and a fault tracing module.The data stream generation module generates a multimodal learning data stream through periodic sampling and migration learning of a multi-source sensor network.The fault mode identification module extracts an abnormal feature set and performs double-channel mode analysis on the power distribution room.The risk analysis module is used for risk early warning based on equipment fault degradation mode and propagation simulation based on power grid fault propagation mode.The fault tracing module is used for fault tracing according to a risk early warning signal and a fault propagation simulation path, and triggers a remote control instruction set to perform remote maintenance on the power distribution room.The application can solve the technical problem that fault abnormal tracing is difficult due to insufficient fault identification capability in the prior art, accurately judges the fault source through double-channel abnormal analysis, and improves the accuracy and comprehensiveness of power distribution room monitoring.

Description

Technical Field

[0001] This application relates to the field of remote monitoring technology, and in particular to a remote intelligent monitoring system, method and equipment for a power distribution room. Background Technology

[0002] As a crucial link in the power system responsible for power distribution and safety protection, the stability of the substation's operational status directly determines the reliable operation of downstream electrical equipment. Currently, the operation and maintenance monitoring of substations mainly relies on a traditional model combining manual inspections with periodic monitoring, supplemented by some monitoring equipment with basic data acquisition and uploading functions, forming a preliminary remote monitoring system. However, existing remote monitoring methods generally use static threshold comparison methods for fault status identification, triggering alarms when monitored parameters exceed preset ranges. Because the identification of the operating status of electrical equipment depends on data from a single or limited number of sensors, there is a lack of modeling and analysis capabilities for the correlation between multiple data sources, making it difficult to accurately identify complex degradative faults or latent anomalies. Furthermore, due to a lack of in-depth understanding of fault development paths and causes, the ability to trace anomalies is severely insufficient, easily overlooking long-term hidden dangers in equipment, further reducing the accuracy and timeliness of fault response, affecting the safety operation and maintenance efficiency of the substation, and lowering power supply reliability.

[0003] In summary, existing technologies suffer from technical problems such as insufficient fault identification capabilities, making it difficult to trace the source of fault anomalies, leading to false alarms and missed alarms, which further affect the safety operation and maintenance efficiency of power distribution rooms. Summary of the Invention

[0004] The purpose of this application is to provide a remote intelligent monitoring system, method and equipment for power distribution rooms, in order to solve the technical problems in the prior art that the lack of fault identification capability leads to difficulties in tracing the source of fault anomalies, and is prone to false alarms and missed alarms, which further affects the safety operation and maintenance efficiency of power distribution rooms.

[0005] In view of the above problems, this application provides a remote intelligent monitoring system, method and equipment for power distribution rooms.

[0006] In a first aspect, this application provides a remote intelligent monitoring system for a power distribution room, comprising: a data stream generation module, used to periodically sample power distribution equipment through a multi-source sensor network deployed in the power distribution room, obtain a dynamic monitoring dataset for transfer learning, and generate a multimodal learning data stream; a fault mode identification module, used to traverse the multimodal learning data stream to extract abnormal feature sets, perform dual-channel mode analysis on the power distribution room based on the abnormal feature sets, and identify dual modes, wherein the dual modes include equipment fault degradation mode and power grid fault propagation mode; a risk analysis module, used to provide risk warnings for the power distribution equipment based on the equipment fault degradation mode, generate risk warning signals, perform propagation simulation on the power distribution room based on the power grid fault propagation mode, and generate fault propagation simulation paths; and a fault tracing module, used to trace faults based on the risk warning signals and the fault propagation simulation paths, and trigger a remote control command set to remotely maintain the power distribution room based on the fault location results.

[0007] Optionally, a hierarchical sampling unit is used to perform hierarchical sampling of power distribution equipment through a multi-source sensor network to determine multiple key nodes; a periodic sampling unit is used to perform periodic sampling based on the multiple key nodes to obtain a dynamic monitoring dataset, the dynamic monitoring dataset containing a three-dimensional dynamic monitoring matrix; a spatial transfer learning unit is used to perform multi-dimensional analysis according to the three-dimensional dynamic monitoring matrix to obtain multi-dimensional monitoring features, and perform spatial transfer learning on the three-dimensional dynamic monitoring matrix according to the multi-dimensional monitoring features to obtain spatial modal data; a temporal transfer learning unit is used to perform temporal transfer learning on the three-dimensional dynamic monitoring matrix according to the multi-dimensional monitoring features to obtain temporal modal data; and a feature fusion unit is used to fuse the spatial modal data and the temporal modal data to construct the multi-modal learning data stream.

[0008] Optionally, the operation status analysis subunit is used to traverse the multiple key nodes to perform operation status analysis on the power distribution equipment in the power distribution room, and obtain a first equipment operation status parameter or a second equipment operation status parameter; the first sampling subunit is used to sample the multiple key nodes at a first sampling frequency when the power distribution equipment is at the first equipment operation status parameter, and obtain a first monitoring dataset; the second sampling subunit is used to automatically switch the first sampling frequency to the second sampling frequency to sample the multiple key nodes when the load rate of the power distribution equipment exceeds a preset threshold, and obtain a second monitoring dataset; the data identification subunit is used to identify data according to the three-dimensional space of the power distribution room based on the first monitoring dataset and the second monitoring dataset, and construct the three-dimensional dynamic monitoring matrix.

[0009] Optionally, the feature parsing unit is used to perform feature parsing based on the multimodal learning data stream to obtain a device feature set and a power grid feature set; the abnormal feature hierarchical extraction unit is used to perform abnormal feature set hierarchical extraction based on the device feature set and the power grid feature set to obtain an abnormal feature set, the abnormal feature set including device abnormal features and power grid abnormal features; the fault degradation analysis unit is used to activate a physical entity channel when the abnormal feature set is the device abnormal feature, synchronize the device abnormal feature to the physical entity channel for fault degradation analysis, and obtain the device fault degradation mode; the fault propagation analysis unit is used to activate a power grid virtual channel when the abnormal feature set is the power grid abnormal feature, synchronize the power grid abnormal feature to the power grid virtual channel for fault propagation analysis, and obtain the power grid fault propagation mode.

[0010] Optionally, a feature layer construction subunit is used to construct multiple abnormal feature layers, including a first abnormal feature layer and a second abnormal feature layer; an abnormal scanning subunit is used to scan the device feature set using a sliding time window through the first abnormal feature layer to capture device sudden pulse information; an abnormal location subunit is used to perform infrared temperature field topology analysis based on the device sudden pulse information to locate the coordinates of abnormal points, perform transient extraction based on the coordinates of the abnormal points to determine the transient abnormal features of the device, and output the transient abnormal features of the device as the device abnormal features; a distortion analysis subunit is used to perform distortion analysis on the power grid feature set through the second abnormal feature layer to mark the distortion accumulation interval; a fluctuation analysis subunit is used to perform power fluctuation analysis based on the distortion accumulation interval to construct a power fluctuation spectrum, perform oscillation analysis based on the power fluctuation spectrum to determine the steady-state abnormal features of the power grid, and output the steady-state abnormal features of the power grid as the power grid abnormal features.

[0011] Optionally, the degradation analysis unit is used to perform degradation analysis based on the equipment failure degradation mode, determine the degradation stage, perform operational impact analysis on the substation based on the degradation stage, and generate multiple stage impact coefficients; the early warning classification matching unit is used to perform early warning classification matching according to the multiple stage impact coefficients and construct the risk early warning signal; the fault propagation analysis unit is used to initialize the simulation environment to construct power grid simulation environment parameters, perform fault propagation analysis based on the power grid fault propagation mode, and determine the key propagation nodes; the propagation simulation unit is used to perform propagation simulation on the substation based on the key propagation nodes and generate the fault propagation simulation path.

[0012] Optionally, a signal path matching unit is used to match the risk warning signal with the fault propagation simulation path to obtain a signal-path matching result; an overlap analysis unit is used to perform overlap analysis based on the signal-path matching result, and determine direct and indirect correlation features based on the overlap data; a feature weighting calculation unit is used to perform feature weighting calculation based on the direct and indirect correlation features to determine multiple feature quantification indicators; a reverse tracing analysis unit is used to retrieve the equipment topology relationship and perform reverse tracing analysis according to the multiple feature quantification indicators to obtain a fault location result, the fault location result including the initial fault point; and a dynamic permission allocation unit is used to perform multi-level confidence verification based on the initial fault point to obtain a confidence verification result, and dynamically allocate permissions according to the confidence verification result and the initial fault point in conjunction with the power distribution room to generate the remote control instruction set.

[0013] Optionally, the spatial analysis subunit is used to perform spatial analysis based on the signal-path matching result and extract spatial coordinate data; the temporal analysis subunit is used to perform temporal analysis based on the signal-path matching result and extract time series labels; the first overlap subunit is used to perform three-dimensional spatial overlap analysis based on the spatial coordinate data and the time series labels to obtain a first overlap degree, and to identify the direct correlation features based on the first overlap degree; the second overlap subunit is used to perform temporal causal chain analysis based on the spatial coordinate data and the time series labels to obtain a second overlap degree, and to identify the indirect correlation features based on the second overlap degree.

[0014] Secondly, this application also provides a remote intelligent monitoring method for a power distribution room, wherein the remote intelligent monitoring method for a power distribution room includes: periodically sampling the power distribution equipment through a multi-source sensor network deployed in the power distribution room to obtain a dynamic monitoring dataset for transfer learning, generating a multimodal learning data stream; traversing the multimodal learning data stream to extract anomaly feature sets, performing dual-channel mode analysis on the power distribution room based on the anomaly feature sets, identifying dual modes, the dual modes including equipment failure degradation mode and power grid failure propagation mode; performing risk warning on the power distribution equipment based on the equipment failure degradation mode, generating a risk warning signal, performing propagation simulation on the power distribution room based on the power grid failure propagation mode, generating a fault propagation simulation path; performing fault tracing based on the risk warning signal and the fault propagation simulation path, and triggering a remote control command set to remotely maintain the power distribution room based on the fault location result.

[0015] Thirdly, this application also provides an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the functions of a remote intelligent monitoring system for a power distribution room as described in any of the first aspects above.

[0016] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0017] The system employs a data stream generation module to periodically sample power distribution equipment using a multi-source sensor network deployed in the power distribution room, obtain dynamic monitoring datasets, perform transfer learning, and generate a multimodal learning data stream. A fault mode identification module iterates through the multimodal learning data stream to extract abnormal feature sets, performs dual-channel mode analysis on the power distribution room based on these abnormal feature sets, and identifies two modes: equipment fault degradation mode and power grid fault propagation mode. A risk analysis module provides risk warnings for the power distribution equipment based on the equipment fault degradation mode, generates risk warning signals, and performs propagation simulation on the power distribution room based on the power grid fault propagation mode to generate a fault propagation simulation path. A fault tracing module performs fault tracing based on the risk warning signals and the fault propagation simulation path, and triggers a remote control command set to remotely maintain the power distribution room based on the fault location results. In other words, by collecting equipment operation data through a multi-source sensor network and performing transfer learning to extract abnormal feature sets, fault source tracing is performed through dual-channel fault mode analysis, and remote maintenance of the power distribution room is carried out based on the fault location results. This achieves accurate classification and source tracing of fault modes, improving the accuracy and comprehensiveness of power distribution room monitoring.

[0018] The above description is merely an overview of the technical solution of this application. To better understand the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the structure of a remote intelligent monitoring system for a power distribution room according to this application.

[0021] Figure 2 This is a flowchart illustrating a remote intelligent monitoring method for a power distribution room according to this application.

[0022] Figure 3 This is a schematic diagram of the structure of an exemplary electronic device of this application.

[0023] Figure labeling: Data stream generation module 11, fault mode identification module 12, risk analysis module 13, fault tracing module 14. Detailed Implementation

[0024] This application provides a remote intelligent monitoring system, method, and equipment for power distribution rooms, solving the technical problems in existing technologies where insufficient fault identification capabilities lead to difficulties in tracing fault anomalies, resulting in false alarms and missed alarms, further affecting the safe operation and maintenance efficiency of power distribution rooms. By collecting equipment operation data through a multi-source sensor network and performing transfer learning to extract anomaly feature sets, fault tracing is performed through dual-channel fault mode analysis. Based on the fault location results, remote maintenance of the power distribution room is carried out, achieving accurate classification and tracing of fault modes, thus improving the accuracy and comprehensiveness of power distribution room monitoring.

[0025] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. It should be understood that this application is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. It should also be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not all of them.

[0026] Example 1, please refer to the appendix. Figure 1 This application provides a remote intelligent monitoring system for a power distribution room, wherein the remote intelligent monitoring system for a power distribution room is used to implement a remote intelligent monitoring method for a power distribution room, and the remote intelligent monitoring system for a power distribution room includes:

[0027] The data stream generation module 11 is used to periodically sample the power distribution equipment through a multi-source sensor network deployed in the power distribution room, obtain a dynamic monitoring dataset, perform transfer learning, and generate a multimodal learning data stream.

[0028] Furthermore, the data stream generation module 11 in the remote intelligent monitoring system for a power distribution room is also used for: a hierarchical sampling unit, used for hierarchical sampling of power distribution equipment through a multi-source sensor network to determine multiple key nodes; a periodic sampling unit, used for periodic sampling based on the multiple key nodes to obtain a dynamic monitoring dataset, the dynamic monitoring dataset containing a three-dimensional dynamic monitoring matrix; a spatial transfer learning unit, used for multidimensional analysis according to the three-dimensional dynamic monitoring matrix to obtain multidimensional monitoring features, and for spatial transfer learning of the three-dimensional dynamic monitoring matrix according to the multidimensional monitoring features to obtain spatial modal data; a temporal transfer learning unit, used for temporal transfer learning of the three-dimensional dynamic monitoring matrix according to the multidimensional monitoring features to obtain temporal modal data; and a feature fusion unit, used for feature fusion of the spatial modal data and the temporal modal data to construct the multimodal learning data stream.

[0029] Specifically, based on the specific layout and equipment conditions of the power distribution room, various types of sensors are selected and installed, such as temperature sensors (monitoring equipment temperature), humidity sensors (monitoring ambient humidity), voltage / current sensors (monitoring electrical parameters), vibration sensors (monitoring equipment mechanical condition), infrared thermal imagers (monitoring equipment surface temperature distribution), and gas sensors (monitoring gas leaks). A multi-source sensor network composed of multiple types of sensors is deployed in the power distribution room to collect various data related to the operating status of the power distribution equipment, including equipment operating data and environmental condition data. The power distribution equipment is sampled in a tiered manner through the multi-source sensor network. Sampling frequencies are differentiated according to the importance of the equipment, the probability of failure, or the rate of data change, identifying several representative key nodes, such as load concentration points and areas with severe heat loss. Key nodes refer to equipment or locations where changes in status have a significant impact, the probability of failure is high, or the data changes rapidly; these are the key targets for monitoring.

[0030] At multiple key nodes, sampling is performed at fixed time intervals to construct a three-dimensional dynamic monitoring matrix containing spatial, temporal, and parameter three-dimensional structures. This organizes the dynamic monitoring dataset, reflecting the changes in the operating status of power distribution equipment over a period of time. For example, 12 key nodes were identified (including 2 main incoming switches, 4 busbar sections, and 6 outgoing circuit breakers). Each node is equipped with 5 types of sensors: current, voltage, temperature, humidity, and partial discharge. The 12 key nodes are divided into two levels: the sampling frequency of the first-level nodes (such as incoming switches) is 1 second / time, and the sampling frequency of the second-level nodes (such as outgoing circuits) is 10 seconds / time, with a sampling duration of 60 minutes. After hierarchical sampling, 36,000 data points are obtained for the first-level nodes and 18,000 data points are obtained for the second-level nodes, constructing a three-dimensional dynamic monitoring matrix of 12 nodes × 3,600 time steps (1 hour) × 5 feature dimensions.

[0031] Multidimensional analysis is performed on the three-dimensional dynamic monitoring matrix, which involves analyzing it from multiple dimensions to extract multidimensional monitoring features, such as the trend of data changes over time, the correlation between data from different sensors, and the correlation between different parameters. Based on the multidimensional monitoring features, spatial transfer learning is performed on the three-dimensional dynamic monitoring matrix. This means extracting spatially distributed monitoring features from the three-dimensional dynamic monitoring matrix, such as the correlation coefficient between node temperature and the current of adjacent nodes, the thermal response slope of a node under different loads, and the spatial distribution gradient of partial discharge signal under humidity conditions. Spatial transfer learning refers to transferring spatial features (such as temperature distribution and load correlation) learned on one equipment node (such as an incoming switch) to other nodes (such as other substations or equipment) to achieve model adaptation and generalization across equipment or regions. For example, taking nodes numbered 3 (incoming bus) and 11 (feeder switch) as an example, the labeled data of node 3 is known, while node 11 is a newly unlabeled node. A tensor is constructed using data from the most recent hour, representing 3600 time points and 6 feature parameters (such as voltage, current, temperature, partial discharge, humidity, and smoke). Node embedding vectors are obtained through training using a graph neural network, reflecting their distribution weights and operational states within the spatial structure. Adjacency matrices represent node connections, and a transfer inference formula is applied to the unlabeled node 11 to output its spatial modal features, i.e., spatial modal data. This spatial modal data, obtained through spatial transfer learning, focuses on reflecting the spatial relationships and commonalities between devices or nodes. Spatial transfer learning is particularly suitable for scenarios such as power distribution rooms, where different device nodes exhibit certain similarities in their operational modes. The state of unseen nodes can be inferred by transferring features from existing nodes.

[0032] Similarly, time-series transfer learning is applied to a 3D dynamic monitoring matrix to transfer the pattern of a time series data from one time period (source time period) to another time period (target time period) for understanding and identifying long-term and short-term trends in the time series. The goal of time-series transfer learning is to identify and predict the current state of a time period, especially predicting future trends, by transferring existing time series data (such as load changes and temperature trends in historical data). Historical data is collected from the sensor network as source time period data. Long-term and short-term trend features are extracted from the time series, and the time-series features learned from the source time period are transferred to the current time period. The trend patterns of the source time period are used to predict and identify trends such as load changes and temperature fluctuations in the current time period. Even without complete labels, it can accurately identify anomalies or potential faults. After time-series transfer learning, the resulting data is time-series modal data, reflecting the operating trend of the power distribution equipment in the current time period, such as current change trends and temperature changes in the next few minutes. For example, assuming a 35kV distribution room has 5 key nodes (e.g., transformers, incoming switches, outgoing circuits, etc.); 3 monitoring parameters for each node (e.g., current, voltage, temperature); and a data sampling frequency of once per second for 1 hour (3600 seconds), the resulting three-dimensional dynamic monitoring matrix is ​​3600×5×3. The peak load period of the past week (e.g., 8:00-10:00 AM daily) is selected as the source time period data, representing the changes in current and temperature during periods of high load. Features are extracted from the time series to obtain the current rise slope during peak load periods (e.g., an increase of 1A per minute) and the lag characteristics of temperature rise (e.g., temperature increases by 0.5℃ per minute after current increases). The time series features learned from the source time period are transferred to the current time period (e.g., today's peak load period), and temperature changes and load fluctuations are predicted by comparing the current data of the current time period (e.g., the trend of current changes when the load increases). Based on the time-series characteristics after migration, the predicted trends of current, voltage and temperature changes in the next 5 minutes are as follows: the current may increase by 3A and the temperature may rise by 2℃ in the next 5 minutes, approaching the safe threshold.

[0033] The generated spatial and temporal modal data are fused using a simple concatenation method to obtain a multimodal learning data stream. This stream includes both the spatial correlation patterns of device states and the dynamic characteristics of these states over time, forming a continuous, multi-dimensional data sequence. Through transfer learning of spatial and temporal features, the state of new nodes can be accurately identified even without complete labeled data. Spatial and temporal transfer learning reduce the need for labeled data for new nodes, saving time on data preparation and model training.

[0034] Furthermore, the data stream generation module 11 in the remote intelligent monitoring system for a power distribution room is also configured to: a running status analysis subunit, configured to traverse the multiple key nodes to perform running status analysis on the power distribution equipment in the power distribution room, and obtain a first equipment running status parameter or a second equipment running status parameter; a first sampling subunit, configured to execute a first sampling frequency to sample the multiple key nodes when the power distribution equipment is in the first equipment running status parameter, and obtain a first monitoring dataset; a second sampling subunit, configured to automatically switch the first sampling frequency to the second sampling frequency to sample the multiple key nodes when the load rate of the power distribution equipment exceeds a preset threshold, and obtain a second monitoring dataset; and a data identification subunit, configured to identify data according to the three-dimensional space of the power distribution room based on the first monitoring dataset and the second monitoring dataset, and construct the three-dimensional dynamic monitoring matrix.

[0035] Specifically, the operational status of the power distribution equipment in the substation is analyzed by traversing multiple key nodes. Real-time monitoring of indicators such as current, load rate, and temperature is used to determine the equipment's operating status. Load rate is the primary criterion. Based on the ratio of the current load to the rated capacity at each key node, a load rate of 80% or less is considered normal, indicating normal equipment operation, and is classified as the first equipment operating status parameter. A load rate greater than 80% is classified as the second equipment operating status parameter, indicating that the equipment has entered a heavy load zone and is at risk of overload. For example, if a line has a rated current of 100A and the current current is 75A, the load rate is 75%, which is classified as the first equipment operating status parameter; if the current current reaches 90A and the load rate is 90%, it is classified as the second equipment operating status parameter.

[0036] When the power distribution equipment operates under the first equipment operating status parameter, routine monitoring is performed at the first sampling frequency (e.g., once every 60 seconds) to obtain the first monitoring dataset. Once it switches to the second equipment operating status parameter, the sampling frequency is immediately switched to the second sampling frequency for intensive monitoring (e.g., once every 5 seconds) to ensure real-time capture of rapid fluctuations and abnormal evolutions, resulting in the second monitoring dataset. The first equipment operating status parameter indicates that the power distribution equipment is operating within the normal load range (i.e., the load rate is below a preset threshold), for example, the transformer load rate is within 70% of the rated capacity. The second equipment operating status parameter indicates that the power distribution equipment is operating under a high load state (i.e., the load rate exceeds the threshold), which is usually considered a potential risk period, such as the load rate exceeding 85% of the rated capacity. The first sampling frequency is used for data acquisition under normal conditions (low frequency), such as once every 60 seconds; the second sampling frequency is used for data acquisition under high load conditions (high frequency), such as once every 5 seconds, for more intensive monitoring. A higher sampling frequency results in denser data, capturing more subtle changes, but may generate a large amount of data, increasing the processing burden; a lower sampling frequency results in sparser data, reducing the amount of data, but may miss rapidly changing information.

[0037] The first and second monitoring datasets refer to the data sets obtained by sampling multiple key nodes at the first / second sampling frequency when the power distribution equipment is in the first or second equipment operating state, respectively. These datasets contain sensor readings of each key node within a specific time interval. Based on the three-dimensional spatial layout of the power distribution room, each data point in the first and second monitoring datasets is labeled with its spatial location (which node, which parameter). This data, carrying spatiotemporal information, is then organized into a dynamically updated three-dimensional matrix. This matrix not only records the state of each node at each time point but also implicitly contains their spatial distribution relationships. The matrix dimensions of the three-dimensional dynamic monitoring matrix are: time step × node position × parameter dimension.

[0038] For example, the power distribution room has 5 key nodes: main transformer, incoming cabinet A, outgoing cabinet B, feeder 1, and feeder 2; each node collects 3 parameters: temperature (°C), current (A), and voltage (V); the load rate threshold is set to 80%. The rated current of feeder 1 is 100A; between 8:00 and 8:30, the current fluctuates between 60A and 75A, the load rate is ≤75%, and the first frequency (sampling every 60 seconds) is used; after 8:30, the current rises to 92A (load rate 92%), triggering the second state, and the sampling frequency is switched to once every 5 seconds. The first monitoring dataset: 30 data points were collected from 8:00 to 8:30, a total of 30 minutes; the second monitoring dataset: 180 data points were collected from 8:30 to 8:45, a total of 15 minutes; each data point recorded the node location, sampling time, current, voltage, and temperature; a three-dimensional matrix was constructed: finally, T=210 samples × N=5 nodes × P=3 parameters were collected, forming a dynamic monitoring tensor with a size of 210×5×3.

[0039] By implementing a load-driven sampling frequency switching mechanism, the monitoring granularity is automatically adjusted according to the actual load of the equipment, avoiding data redundancy or omission, improving monitoring efficiency, and ensuring the accuracy of monitoring under normal conditions. When the equipment enters a high-load, high-risk operating range, the density and timeliness of data collection are automatically increased, avoiding the generation of too much redundant data during stable periods, while ensuring that subtle and rapid changes that may indicate faults are captured during critical periods.

[0040] The fault mode identification module 12 is used to traverse the multimodal learning data stream to extract abnormal feature sets, perform dual-channel mode analysis on the power distribution room based on the abnormal feature sets, and identify dual modes, including equipment fault degradation mode and power grid fault propagation mode.

[0041] Furthermore, the fault mode identification module 12 in the remote intelligent monitoring system for a power distribution room is also configured to: a feature parsing unit, configured to perform feature parsing based on the multimodal learning data stream to obtain a device feature set and a power grid feature set; an abnormal feature hierarchical extraction unit, configured to perform abnormal feature set hierarchical extraction based on the device feature set and the power grid feature set to obtain an abnormal feature set, wherein the abnormal feature set includes device abnormal features and power grid abnormal features; a fault degradation analysis unit, configured to, when the abnormal feature set is the device abnormal feature, activate a physical entity channel to synchronize the device abnormal feature to the physical entity channel for fault degradation analysis to obtain the device fault degradation mode; and a fault propagation analysis unit, configured to, when the abnormal feature set is the power grid abnormal feature, activate a power grid virtual channel to synchronize the power grid abnormal feature to the power grid virtual channel for fault propagation analysis to obtain the power grid fault propagation mode.

[0042] Furthermore, the fault mode identification module 12 in the remote intelligent monitoring system for a power distribution room is also configured to: a feature layer construction subunit, configured to construct multiple abnormal feature layers, the multiple abnormal feature layers including a first abnormal feature layer and a second abnormal feature layer; an abnormal scanning subunit, configured to scan the equipment feature set using a sliding time window through the first abnormal feature layer to capture equipment sudden pulse information; an abnormal location subunit, configured to perform infrared temperature field topology analysis based on the equipment sudden pulse information to locate the coordinates of abnormal points, perform transient extraction based on the coordinates of the abnormal points to determine the transient abnormal features of the equipment, and output the transient abnormal features of the equipment as the abnormal features of the equipment; a distortion analysis subunit, configured to perform distortion analysis on the power grid feature set through the second abnormal feature layer to mark the distortion accumulation interval; and a fluctuation analysis subunit, configured to perform power fluctuation analysis based on the distortion accumulation interval to construct a power fluctuation spectrum, perform oscillation analysis based on the power fluctuation spectrum to determine the steady-state abnormal features of the power grid, and output the steady-state abnormal features of the power grid as the abnormal features of the power grid.

[0043] Specifically, feature parsing is performed on the multimodal learning data stream to obtain equipment feature sets and power grid feature sets. This involves extracting structured features at two granularities—equipment layer and power grid layer—from the spatially and temporally interwoven multimodal learning data stream. The equipment feature set refers to the set of features directly related to the state, behavior, or evolution of individual power distribution equipment, such as temperature rise rate, current imbalance, and insulation temperature difference. The power grid feature set refers to the set of features related to the overall operating state or regional stability of the power distribution network, such as node load balance, power flow stability, and low-frequency oscillation trends.

[0044] Hierarchical extraction of anomaly features refers to performing anomaly detection at both the equipment and power grid levels, constructing multiple anomaly feature layers, including a first anomaly feature layer and a second anomaly feature layer, used for anomaly extraction from the equipment feature set and the power grid feature set, respectively. The first anomaly feature layer scans the equipment feature set for anomalies, identifying short-term sudden changes (such as temperature rises or current pulses). A sliding time window is set, which continuously extracts time series data on the time axis with a fixed window length, such as 10 seconds or 30 seconds, with a step size of 1 second. For example, using a 10-second sliding time window with a 1-second step size, the parameter fluctuations within the past 10 seconds are checked every second.

[0045] The first anomaly feature layer employs a sliding time window to efficiently capture abrupt change pulse information from the equipment feature set. This pulse signals indicate a rapid and drastic change in the equipment's state within a short period, potentially caused by transient faults, poor contact, short circuits, or external impacts. The first derivative (rate of change), standard deviation fluctuation coefficient, and maximum and minimum amplitude differences of each parameter are calculated. If the rate of change of a parameter exceeds a set threshold (e.g., current fluctuation greater than ±20% / s, temperature rise rate greater than 1.5℃ / s), the window is marked as an abrupt change pulse window. After detecting an abrupt change, the infrared thermal image of the corresponding equipment within that time period is retrieved, and infrared temperature field topology analysis is performed using image segmentation and boundary extraction algorithms. For example, the thermal image is divided into a 32×32 pixel grid, and the average temperature and four-neighbor gradient are calculated for each grid. If a grid shows a local temperature rise >12℃ and a stable gradient direction (e.g., in the terminal block area), the center point of that area is designated as the anomaly point coordinate.

[0046] Infrared temperature field topology analysis refers to using images of the surface temperature distribution (i.e., temperature field) of equipment captured by an infrared thermal imager. By analyzing temperature gradients, hot spot distribution, and hot zone shapes within the image, it is possible to infer the thermal state and potential problems inside or on the surface of the equipment. For example, localized overheating may indicate poor contact or abnormal resistance at that location. Infrared temperature field topology analysis can pinpoint the precise location of temperature anomaly areas in the image, i.e., the coordinates of the anomaly points.

[0047] Based on the coordinates of the anomaly point, the transient anomaly characteristics at that point are extracted, i.e., the specific manifestations or characteristics when a sudden change occurs, such as the transient temperature rise rate, the duration of the current spike, and the magnitude of the voltage drop. For example, if a sudden change in temperature and current is detected between 1020 and 1030 seconds, with a temperature change rate of +13.2℃ and a current surge of +28.7A, the coordinates of the anomaly point are determined to be (x=0.42m, y=1.12m), ΔT is +13.2℃, the current pulse is +28.7A, the FFT change is +21%, and the vibration increase is 14%. The transient anomaly characteristics of the equipment are then output as the equipment anomaly characteristics.

[0048] In the second anomaly feature layer, distortion analysis is performed on the power grid feature set to identify waveform distortion phenomena caused by various reasons, such as high-frequency noise in the current waveform, harmonic components in the voltage waveform, and abnormal load fluctuations in the power flow. Distortion refers to certain parameters in the power grid feature set deviating from their normal, ideal waveform or numerical range. Distortion analysis is the process of detecting and quantifying this deviation. For example, it analyzes whether the voltage or current waveform exhibits non-ideal shapes such as harmonics, spikes, gaps, or zero-crossing offsets, or analyzes whether certain indicators (such as power factor and imbalance) continuously deviate from the standard value.

[0049] Distortion analysis not only identifies the existence of distortions but also tracks whether these distortions are continuous or sporadic. A distortion accumulation interval refers to the time period during which distortion phenomena persist or gradually intensify. In other words, it's an abnormal interval formed by the gradual accumulation of certain distortion phenomena (such as current fluctuations and voltage harmonics) over a continuous period during power grid operation, reflecting the gradual deterioration of power grid stability. For example, if the voltage harmonic content consistently exceeds the standard for a certain period and shows a gradually increasing trend, then this period is marked as a distortion accumulation interval.

[0050] For the marked distortion accumulation interval, power fluctuation analysis is performed. The power grid power data within this interval is analyzed to construct a power fluctuation spectrum, revealing potential low-frequency or high-frequency fluctuations in grid operation. For example, the spectrum may contain low-frequency oscillations (e.g., 0.5Hz) or high-frequency fluctuations (e.g., 50Hz harmonic peaks). By performing a Fourier transform on the grid power data (e.g., current, voltage), the resulting frequency domain analysis is the power fluctuation spectrum, revealing the frequency components of oscillations in the grid, especially low-frequency and high-frequency oscillation modes.

[0051] Based on the power fluctuation spectrum, oscillation analysis is performed to identify abnormal fluctuations in the spectrum (such as abnormal changes in frequency and amplitude) to determine whether there are unstable phenomena in the power grid. For example, certain low-frequency oscillations (such as 1Hz or 0.5Hz) may indicate a relatively serious stability problem in the power grid, leading to voltage or power imbalance. The normal state of the power grid is determined by characteristics such as frequency range, amplitude variation, and periodicity. If abnormal oscillation components are found, they are marked as steady-state abnormal characteristics of the power grid, including specific harmonic pollution levels, persistent three-phase imbalance, and power oscillations at specific frequencies. These steady-state abnormal characteristics are output as power grid abnormal characteristics, reflecting faults, disturbances, or potential problems that occur during power grid operation. For example, a 10kV / 35kV distribution network includes 10 distribution nodes, 4 transformers, 3 switchgear, and 2 main feeders; the data sampling frequency is current, voltage, and power sampled once per second; the distortion analysis thresholds are current THD greater than 5% and voltage deviation greater than ±3%; the oscillation analysis covers a frequency range of 0.1Hz to 10Hz, with an energy threshold of 10%. The load fluctuation at grid node 3 showed significant instability between the 12th and 14th minutes, with the current THD exceeding 5%. Spectral analysis of the current waveform showed a significant increase (over 20%) in the amplitude of a low-frequency oscillation at approximately 0.8 Hz, and the voltage fluctuation range also increased during this period. The power fluctuation spectrum showed a wide-band high-amplitude fluctuation near 0.8 Hz, suggesting possible unstable load coupling in the grid. The output grid steady-state anomaly characteristics were: a low-frequency oscillation with a frequency of 0.8 Hz, an amplitude of 20%, a duration of 2 minutes, and a voltage fluctuation range increased to ±6%.

[0052] When the anomaly feature set matches equipment anomaly characteristics, the physical entity channel is activated. The physical entity channel is an analysis channel used to handle equipment-level anomalies. It contains equipment models provided by the equipment manufacturer, historical operating data of the equipment, specific physical parameters of the equipment (such as materials and structure), and simulation algorithms used to simulate the internal physical processes of the equipment. It focuses on the physical state of the power distribution equipment itself and the fault development process. Equipment anomaly features are input into the physical entity channel for fault degradation analysis. Combining the physical characteristics of the power distribution equipment, historical data, and possible fault mechanisms, it simulates possible changes within the power distribution equipment and predicts its performance degradation trend. Fault degradation analysis simulates or predicts how the performance of the equipment gradually declines (degrades) after an anomaly occurs. By analyzing equipment anomaly features and combining them with the equipment model and data in the physical entity channel, it is possible to infer possible changes within the equipment, the type of fault, and the expected degradation rate. For example, if the temperature rise rate is 1°C / minute and the current imbalance increases, the fault degradation mode of the power distribution equipment can be inferred, such as the gradual aging of transformer winding insulation, and the future fault risks of the power distribution equipment (such as possible short circuits or power outages) can be predicted. Fault degradation analysis yields equipment failure degradation patterns, describing the specific ways and paths in which the performance of a particular piece of equipment degrades under the influence of current abnormal characteristics. Equipment failure degradation patterns refer to the typical characteristic patterns exhibited by power distribution equipment during the degradation process, such as gradually increasing temperature and gradually increasing load imbalance, used to identify potential faults in power distribution equipment in advance. For example, the contact resistance of phase C contacts in switchgear A increases at a rate of 5% per month due to overheating, and is expected to lead to poor contact within 3 months.

[0053] When the anomaly feature set matches the characteristics of a power grid anomalies, the power grid virtual channel is activated. The power grid virtual channel is an analysis channel used to handle power grid-level anomalies. It contains information such as the power grid topology model, load model, generation model, line parameters, protection configuration, and simulation software for simulating the dynamic behavior of the power grid. The power grid anomaly features are input into the power grid virtual channel for fault propagation analysis, analyzing how faults in the power grid propagate from the source node to other nodes. This typically involves cascading effects within the power grid, assessing fault interactions between different devices and their impact range. Based on the power grid topology and power flow distribution, the power grid virtual channel begins fault propagation analysis, simulating the propagation path of faults from the source node and generating a power grid fault propagation model. The power grid fault propagation model describes the specific path, speed, and impact range of power grid anomaly propagation, involving the fault source node and other affected nodes, used to determine the most likely affected power grid areas. For example, if a transformer fails, the simulation shows how the fault propagates through the busbar and switchgear to other distribution areas, such as current imbalance from the transformer to the switchgear, causing a voltage drop at node 4.

[0054] By extracting and processing the characteristics of equipment anomalies and power grid anomalies in a hierarchical manner, the different natures of equipment faults and power grid anomalies can be accurately identified, avoiding false alarms and missed alarms. Through the linkage analysis of physical channels and virtual power grid channels, not only can equipment and power grid anomalies be identified in advance, but the evolution process and propagation path of faults can also be predicted.

[0055] The risk analysis module 13 is used to provide risk warnings for power distribution equipment based on the equipment failure degradation mode, generate risk warning signals, and simulate the propagation of power distribution rooms based on the power grid failure propagation mode to generate failure propagation simulation paths.

[0056] Furthermore, the risk analysis module 13 in the remote intelligent monitoring system for a power distribution room is also used for: a degradation analysis unit, used for performing degradation analysis according to the equipment failure degradation mode, determining the degradation stage, performing operational impact analysis on the power distribution room based on the degradation stage, and generating multiple stage impact coefficients; an early warning classification matching unit, used for performing early warning classification matching according to the multiple stage impact coefficients, and constructing the risk early warning signal; a fault propagation analysis unit, used for initializing the simulation environment to construct power grid simulation environment parameters, performing fault propagation analysis according to the power grid fault propagation mode, and determining key propagation nodes; and a propagation simulation unit, used for performing propagation simulation on the power distribution room based on the key propagation nodes, and generating the fault propagation simulation path.

[0057] Specifically, degradation analysis is performed based on equipment failure degradation patterns, including feature extraction and analysis of each stage before and after equipment failure, to identify the current degradation stage of the power distribution equipment. Degradation stages refer to different phases in the process of power distribution equipment from normal operation to failure, such as minor degradation, severe degradation, and the critical stage before failure. Based on the equipment's operating status and health information, the failure degradation pattern of the equipment is analyzed in depth to determine its current degradation stage. For example, a high temperature rise rate of a transformer indicates that it has entered the moderate degradation stage. Based on the equipment's degradation pattern, the current state of the equipment is analyzed to determine which degradation stage it belongs to. The initial degradation stage indicates that the power distribution equipment can still operate normally, but signs such as increased load and rising temperature appear; the severe degradation stage indicates a significant decline in the performance of the power distribution equipment, with frequent warning signals and an increased risk of failure; the critical degradation stage indicates that the power distribution equipment is about to fail and requires immediate repair or replacement.

[0058] By analyzing the impact of degradation stages on the overall operation of the substation, stage impact coefficients are determined. These coefficients represent the influence of power distribution equipment at different degradation stages on the substation's stability, power transmission, and load distribution. A stage impact coefficient is calculated for each stage. For example, in a severe degradation stage, the impact coefficient might be 0.8, indicating that equipment failure could have an 80% impact on the substation's operation. The stage impact coefficient quantifies the degree of impact of equipment on the substation's operation at different degradation stages. Each stage impact coefficient is typically represented by a numerical value to indicate the severity of the impact.

[0059] Based on the impact coefficients of multiple stages, the overall risk of the power distribution room is classified, and corresponding risk warning signals are generated to indicate the risk level of the power distribution room. For example, when a transformer enters a severe degradation stage with an impact coefficient of 0.7, a high-risk warning signal is generated, indicating the need for immediate inspection and maintenance. The risk of the power distribution room is classified into three levels—low, medium, and high—based on the magnitude of the stage impact coefficients. Warning levels are then matched to determine the corresponding warning grades. Based on the determined warning grades, risk warning signals of varying severity are generated so that maintenance personnel can prioritize addressing the most urgent issues.

[0060] Based on parameters such as the distribution network topology, power flow, and voltage level, a power grid simulation environment is initialized, and parameters are constructed to simulate the behavior and propagation path of the power grid during a fault. According to the determined power grid fault propagation mode, fault propagation analysis is performed in the power grid simulation environment. This involves simulating how a fault propagates from one node (e.g., a transformer) to other nodes, as well as the speed and extent of the propagation, identifying key nodes that play a crucial role in the fault propagation process (e.g., busbars connecting multiple branches, load-concentrated area entrances, etc.), thus obtaining critical propagation nodes. Based on these critical nodes, the fault signal is simulated in detail in the power grid simulation environment to determine how it propagates step by step, ultimately generating a clear fault propagation simulation path. Critical propagation nodes are the most critical nodes in the fault propagation process; they are the starting point or intermediate point of fault propagation. The fault propagation simulation path, based on the propagation simulation results of the critical nodes, determines the path of fault propagation in the power grid, demonstrating how the fault spreads from the source node to other nodes and predicting potentially affected areas. For example, it might show that the fault signal starts from the incoming switch, passes through the busbar section, and primarily affects two specific outgoing switchgear.

[0061] By analyzing the impact coefficients based on degradation stages, potential equipment and power grid failures can be identified in advance, ensuring the accuracy and real-time nature of early warning signals. Through power grid fault propagation analysis, key nodes in fault propagation can be identified, allowing for early detection and prevention of equipment degradation and power grid fault propagation, which helps reduce power outages and improve the reliability and stability of substations.

[0062] The fault tracing module 14 is used to trace the fault source based on the risk warning signal and the fault propagation simulation path, and to trigger the remote control command set of the power distribution room to perform remote maintenance based on the fault location result.

[0063] Furthermore, the fault tracing module 14 in the remote intelligent monitoring system for a power distribution room is also used for: a signal path matching unit, used to match the risk warning signal with the fault propagation simulation path to obtain a signal-path matching result; an overlap analysis unit, used to perform overlap analysis based on the signal-path matching result, and determine direct and indirect correlation features based on the overlap data; a feature weighting calculation unit, used to perform feature weighting calculation based on the direct and indirect correlation features to determine multiple feature quantification indicators; a reverse tracing analysis unit, used to retrieve the equipment topology relationship and perform reverse tracing analysis according to the multiple feature quantification indicators to obtain a fault location result, the fault location result including the initial fault point; and a dynamic permission allocation unit, used to perform multi-level confidence verification based on the initial fault point to obtain a confidence verification result, and dynamically allocate permissions according to the confidence verification result and the initial fault point in conjunction with the power distribution room to generate the remote control instruction set.

[0064] Furthermore, the fault tracing module 14 in the remote intelligent monitoring system for a power distribution room is also configured to: a spatial analysis subunit, used to perform spatial analysis based on the signal-path matching result and extract spatial coordinate data; a time analysis subunit, used to perform time analysis based on the signal-path matching result and extract time series labels; a first overlap subunit, used to perform three-dimensional spatial overlap analysis based on the spatial coordinate data and the time series labels to obtain a first overlap degree, and to identify the direct correlation feature based on the first overlap degree; and a second overlap subunit, used to perform temporal causal chain analysis based on the spatial coordinate data and the time series labels to obtain a second overlap degree, and to identify the indirect correlation feature based on the second overlap degree.

[0065] Specifically, the generated risk warning signals are compared one by one with the simulated fault propagation paths. Information such as whether the device corresponding to the warning signal is on the propagation path, its propagation time period, and the propagation time lag is extracted. A signal-path matching result table is output. For example, device T001 has a high warning level, its path location is node 3, its overlap time is 1200ms, and its path overlap is 0.95. The signal-path matching result matches the device location indicated in the risk warning signal with the nodes included in the fault propagation path, determining the correlation between the two. This includes information such as the location where the warning signal was issued, the coordinates of the nodes involved in the fault path, and the propagation time.

[0066] Spatial analysis is performed on the signal-path matching results to extract spatial coordinate data, including the coordinates of the warning signal (e.g., the spatial coordinates of device T005 are x=10m, y=5m, z=2m) and the spatial coordinates of each fault node in the propagation path (e.g., T008 is x=8m, y=5m, z=2m, T003 is x=9m, y=5m, z=2m). Temporal analysis is then performed on the signal-path matching results to extract time series labels, representing the occurrence times of the warning signal and propagation path events on the time axis, such as the delay time required for propagation to a certain node or the signal trigger time.

[0067] A three-dimensional spatial overlap analysis is performed on spatial coordinate data and time series labels. This combines spatial coordinate data and time series labels to examine the overlap of signals and paths within a three-dimensional framework (spatial X, spatial Y, time T). This involves not only checking for spatial overlap but also whether they occur simultaneously or close to each other temporally. For example, an anomaly in switchgear A (spatial coordinates: X=5.2, Y=3.8) indicated by a warning signal not only occurs at the warning time (time label: T=2:15) but also very close to the time when the fault signal shown in the simulated path is expected to arrive at the switchgear (T=2:18). If this degree of spatial and temporal overlap is high (first overlap close to 1), then the features related to switchgear A are identified as directly associated features.

[0068] A temporal causal chain analysis is performed on spatial coordinate data and time series labels to attempt to construct a causal chain from the warning signal to the fault path. For example, although bus segment B itself does not directly issue a warning, it is connected to the warning device switchgear A (spatial relationship), and the abnormal time of bus segment B (time label: T=2:17) occurs before the abnormal time of switchgear A (T=2:18), and the simulated path shows that the fault signal propagates from bus segment B to switchgear A. If the strength of this causal chain is high (second overlap close to 1), then the features related to bus segment B will be identified as indirect correlation features.

[0069] Direct correlation features are those where both spatial location and time point are highly matched, indicating that the warning source is highly likely to be a key node in the propagation chain. Indirect correlation features are those where spatial location or time point is not perfectly matched, but a causal relationship or time-delayed propagation path exists, such as abnormal responses from adjacent nodes or at similar times. For example, suppose a risk warning signal indicates that switchgear C (spatial coordinates: X=7.5, Y=2.1) has an overheating risk (time label: T=14:00). The fault propagation simulation path shows that due to poor contact at incoming switch D (spatial coordinates: X=1.0, Y=1.0), the fault signal is expected to propagate to bus segment E (spatial coordinates: X=4.5, Y=1.5) at T=14:05, and then to switchgear C at T=14:10. After spatial and temporal analysis, the spatial coordinates and key time points of these devices are obtained. Three-dimensional spatial overlap analysis revealed that while the warning time of switchgear C (T=14:00) and the arrival time of the fault signal shown in the simulated path (T=14:10) were not completely identical, they were within an acceptable error range (e.g., less than 5 minutes), and their spatial locations matched perfectly. Therefore, the first overlap degree was calculated to be 0.85, identifying the feature related to the overheating of switchgear C as a directly associated feature. Temporal causal chain analysis revealed that the poor contact of incoming switch D (although it may not have a direct warning, it could be a known weak point) occurred at the beginning of the path (before T=14:00), and the abnormal time of busbar segment E (T=14:05) occurred before the abnormal time of switchgear C (T=14:10), and they are spatially connected via the busbar. The simulated path also clearly showed the fault propagation sequence from D->E->C. Therefore, the second overlap degree was calculated to be 0.9, identifying the feature related to the poor contact of incoming switch D and the abnormality of busbar segment E as an indirectly associated feature.

[0070] Feature weighting calculations are performed based on directly and indirectly related characteristics, fusing multiple influencing factors to obtain comparable quantitative values, i.e., multiple feature quantification indicators, used to comprehensively assess the probability or scope of impact of a node failure. For example, the weight of a directly related characteristic is calculated as 0.8 × warning level coefficient + 0.2 × path overlap; the warning level coefficient ranges from 1 to 5, with higher numbers indicating more severe failures. The weight of an indirectly related characteristic is calculated as 0.5 × equipment impact factor + 0.5 × propagation time reduction ratio; the equipment impact factor reflects the node's criticality within the power grid. The feature quantification indicators are the result of the weighted feature calculations and represent the combined importance or influence of directly and indirectly related characteristics.

[0071] Retrieving equipment topology relationships—that is, the physical and electrical connections between various devices in the power distribution room (such as switchgear, transformers, busbars, etc.)—can typically be represented graphically as a topology diagram, clearly showing the current flow path and the interdependencies of the devices. Using multiple quantifiable indicators, reverse tracing analysis is performed on the equipment topology diagram. Instead of tracing the impact from the fault source, it starts from identified features (direct and indirect related features) and, combined with the equipment topology relationships, inversely infers the initial fault point most likely to trigger these features. The initial fault point is the earliest fault-prone equipment node that causes subsequent anomaly propagation.

[0072] Multi-layered confidence verification is performed on the initial fault point, checking its reliability from multiple perspectives (such as data consistency, causal chain verification, and simulation reproduction), and providing a confidence score, i.e., the confidence verification result. For example, it checks whether the initial fault point has experienced thermal runaway or short-term tripping in historical data (e.g., three instances of temperature spikes within 30 seconds in the past two months), verifies whether the difference between the current data and the modeled value at the initial fault point exceeds 15%, and examines whether abnormal image frames such as overheating or flashing appear at the location of the initial fault point in the video surveillance system. If the verification confidence score is higher than 80%, the initial fault point is identified as a valid initial fault point.

[0073] By combining the confidence verification results with the initial fault point, and coordinating with the power distribution room, the operating permissions for the power distribution room are adjusted to ensure that only appropriate personnel can perform subsequent operations. For example, this includes automatically cutting off power to certain power distribution sections, switching power supply paths, and retrieving video transmission permissions. Based on the fault location results and the dynamic allocation of permissions, a remote control command set is generated for remotely controlling equipment within the power distribution room, such as isolating the switchgear where the initial fault point is located, switching to backup power, and adjusting relevant protection settings.

[0074] By combining reverse tracing analysis with equipment topology relationships, the accuracy of locating the initial fault point is greatly improved, avoiding misidentification of symptom points as the source. Multi-layer confidence verification further enhances the reliability of the location results and reduces the risk of misoperation. Dynamic permission allocation ensures the security and effectiveness of subsequent control commands. The final generated remote control command set targets the root cause (initial fault point), rather than merely addressing surface symptoms. This improves control effectiveness and operational efficiency, significantly enhances the automation and intelligence level of power distribution room operation and maintenance, reduces reliance on manual experience judgment, lowers operational risks and costs, and ensures power supply reliability.

[0075] In summary, the remote intelligent monitoring system for a power distribution room provided in this application has the following technical advantages:

[0076] The system employs a data stream generation module to periodically sample power distribution equipment using a multi-source sensor network deployed in the power distribution room, obtain dynamic monitoring datasets, perform transfer learning, and generate a multimodal learning data stream. A fault mode identification module iterates through the multimodal learning data stream to extract abnormal feature sets, performs dual-channel mode analysis on the power distribution room based on these abnormal feature sets, and identifies two modes: equipment fault degradation mode and power grid fault propagation mode. A risk analysis module provides risk warnings for the power distribution equipment based on the equipment fault degradation mode, generates risk warning signals, and performs propagation simulation on the power distribution room based on the power grid fault propagation mode to generate a fault propagation simulation path. A fault tracing module performs fault tracing based on the risk warning signals and the fault propagation simulation path, and triggers a remote control command set to remotely maintain the power distribution room based on the fault location results. In other words, by collecting equipment operation data through a multi-source sensor network and performing transfer learning to extract abnormal feature sets, fault source tracing is performed through dual-channel fault mode analysis, and remote maintenance of the power distribution room is carried out based on the fault location results. This achieves accurate classification and source tracing of fault modes, improving the accuracy and comprehensiveness of power distribution room monitoring.

[0077] Example 2: Based on the same inventive concept as the remote intelligent monitoring system for a power distribution room in Example 1, this application also provides a remote intelligent monitoring method for a power distribution room. Please refer to the appendix. Figure 2 The remote intelligent monitoring method for a power distribution room includes:

[0078] A multi-source sensor network deployed in the power distribution room periodically samples the power distribution equipment to obtain a dynamic monitoring dataset, which is then used for transfer learning to generate a multimodal learning data stream. The multimodal learning data stream is traversed to extract anomaly feature sets. Based on these anomaly feature sets, dual-channel pattern analysis is performed on the power distribution room to identify two modes: equipment failure degradation mode and power grid failure propagation mode. Risk warnings are generated for the power distribution equipment based on the equipment failure degradation mode, and a failure propagation simulation is performed on the power distribution room based on the power grid failure propagation mode to generate a failure propagation simulation path. Fault tracing is performed based on the risk warning signal and the failure propagation simulation path. Based on the fault location results, a remote control command set is triggered to remotely maintain the power distribution room.

[0079] Furthermore, the step of periodically sampling the power distribution equipment through a multi-source sensor network deployed in the power distribution room to obtain a dynamic monitoring dataset for transfer learning and generating a multimodal learning data stream includes: performing hierarchical sampling of the power distribution equipment through the multi-source sensor network to determine multiple key nodes; performing periodic sampling based on the multiple key nodes to obtain a dynamic monitoring dataset, the dynamic monitoring dataset containing a three-dimensional dynamic monitoring matrix; performing multidimensional analysis according to the three-dimensional dynamic monitoring matrix to obtain multidimensional monitoring features; performing spatial transfer learning on the three-dimensional dynamic monitoring matrix according to the multidimensional monitoring features to obtain spatial modal data; performing temporal transfer learning on the three-dimensional dynamic monitoring matrix according to the multidimensional monitoring features to obtain temporal modal data; and fusing the spatial modal data and the temporal modal data to construct the multimodal learning data stream.

[0080] Furthermore, the step of periodically sampling based on the multiple key nodes to obtain a dynamic monitoring dataset includes: traversing the multiple key nodes to analyze the operating status of the power distribution equipment in the power distribution room, obtaining a first equipment operating status parameter or a second equipment operating status parameter; when the power distribution equipment is the first equipment operating status parameter, executing a first sampling frequency to sample the multiple key nodes to obtain a first monitoring dataset; when the load rate of the power distribution equipment exceeds a preset threshold, the power distribution equipment is the second equipment operating status parameter, and automatically switching the first sampling frequency to the second sampling frequency to sample the multiple key nodes to obtain a second monitoring dataset; and based on the first monitoring dataset and the second monitoring dataset, data identification is performed according to the three-dimensional space of the power distribution room to construct the three-dimensional dynamic monitoring matrix.

[0081] Furthermore, the step of traversing the multimodal learning data stream to extract anomaly feature sets, and performing dual-channel mode analysis on the power distribution room based on the anomaly feature sets to identify dual modes, including equipment fault degradation mode and power grid fault propagation mode, includes: performing feature parsing based on the multimodal learning data stream to obtain equipment feature sets and power grid feature sets; performing hierarchical extraction of anomaly feature sets based on the equipment feature sets and the power grid feature sets to obtain anomaly feature sets, which include equipment anomaly features and power grid anomaly features; when the anomaly feature set is the equipment anomaly feature, activating the physical entity channel to synchronize the equipment anomaly feature to the physical entity channel for fault degradation analysis to obtain the equipment fault degradation mode; when the anomaly feature set is the power grid anomaly feature, activating the power grid virtual channel to synchronize the power grid anomaly feature to the power grid virtual channel for fault propagation analysis to obtain the power grid fault propagation mode.

[0082] Further, the step of performing hierarchical extraction of abnormal features based on the equipment feature set and the power grid feature set to obtain an abnormal feature set, which includes equipment abnormal features and power grid abnormal features, includes: constructing multiple abnormal feature layers, including a first abnormal feature layer and a second abnormal feature layer; scanning the equipment feature set using a sliding time window through the first abnormal feature layer to capture equipment sudden pulse information; performing infrared temperature field topology analysis based on the equipment sudden pulse information to locate the coordinates of abnormal points; performing transient extraction based on the coordinates of the abnormal points to determine the transient abnormal features of the equipment; and outputting the transient abnormal features of the equipment as the equipment abnormal features; performing distortion analysis on the power grid feature set through the second abnormal feature layer to mark the distortion accumulation interval; performing power fluctuation analysis based on the distortion accumulation interval to construct a power fluctuation spectrum; performing oscillation analysis based on the power fluctuation spectrum to determine the steady-state abnormal features of the power grid; and outputting the steady-state abnormal features of the power grid as the power grid abnormal features.

[0083] Furthermore, the step of performing risk warning for power distribution equipment based on the equipment failure degradation mode and generating a risk warning signal, and performing propagation simulation for the power distribution room based on the power grid failure propagation mode and generating a fault propagation simulation path, includes: performing degradation analysis based on the equipment failure degradation mode to determine the degradation stage; performing operational impact analysis on the power distribution room based on the degradation stage and generating multiple stage impact coefficients; performing warning classification matching according to the multiple stage impact coefficients to construct the risk warning signal; initializing the simulation environment to construct power grid simulation environment parameters; performing fault propagation analysis based on the power grid failure propagation mode and determining key propagation nodes; and performing propagation simulation on the power distribution room based on the key propagation nodes to generate the fault propagation simulation path.

[0084] Furthermore, the step of tracing the fault source based on the risk warning signal and the fault propagation simulation path, and triggering a remote control command set by linking the power distribution room according to the fault location result, includes: matching the risk warning signal with the fault propagation simulation path to obtain a signal-path matching result; performing overlap analysis based on the signal-path matching result, and determining direct and indirect correlation features based on the overlap data; performing feature weighting calculation based on the direct and indirect correlation features to determine multiple feature quantification indicators; retrieving the equipment topology relationship and performing reverse tracing analysis according to the multiple feature quantification indicators to obtain a fault location result, the fault location result including an initial fault point; performing multi-level confidence verification based on the initial fault point to obtain a confidence verification result; and dynamically allocating permissions according to the confidence verification result and the initial fault point in conjunction with the power distribution room to generate the remote control command set.

[0085] Furthermore, the overlap analysis based on the signal-path matching results, and the determination of direct and indirect correlation features based on the overlap data, includes: performing spatial analysis based on the signal-path matching results to extract spatial coordinate data; performing temporal analysis based on the signal-path matching results to extract time series labels; performing three-dimensional spatial overlap analysis based on the spatial coordinate data and the time series labels to obtain a first degree of overlap, and identifying the direct correlation features based on the first degree of overlap; performing temporal causal chain analysis based on the spatial coordinate data and the time series labels to obtain a second degree of overlap, and identifying the indirect correlation features based on the second degree of overlap.

[0086] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Figure 1 The remote intelligent monitoring system and specific example of a power distribution room in Embodiment 1 are also applicable to the remote intelligent monitoring method of a power distribution room in this embodiment. Through the foregoing detailed description of the remote intelligent monitoring system of a power distribution room, those skilled in the art can clearly understand the remote intelligent monitoring method of a power distribution room in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.

[0087] Example 3: Based on the same inventive concept as the remote intelligent monitoring system for a power distribution room in Example 1 above, this application also provides an electronic device, including: at least one processor; a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to implement the function of the remote intelligent monitoring system for a power distribution room as described in any one of Examples 1 above.

[0088] Appendix Figure 3 This is a schematic diagram of the structure of an exemplary electronic device of this application. Figure 3In this document, the bus architecture is represented by bus 300. Bus 300 may include any number of interconnected buses and bridges, and bus 300 connects various circuits including one or more processors represented by processor 302 and memory represented by memory 304. Bus 300 may also connect various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. Bus interface 305 provides an interface between bus 300 and receiver 301 and transmitter 303. Receiver 301 and transmitter 303 may be the same element, i.e., a transceiver, providing a unit for communicating with various other devices over a transmission medium. Processor 302 is responsible for managing bus 300 and general processing, while memory 304 can be used to store data used by processor 302 during operation.

[0089] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0090] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application also intends to include such modifications and variations.

Claims

1. A remote intelligent monitoring system for a power distribution room, characterized in that, The method comprises the following steps: A data stream generation module is used to periodically sample power distribution equipment through a multi-source sensor network deployed in a power distribution room to obtain a dynamic monitoring data set for transfer learning and generate a multi-modal learning data stream; A fault mode identification module is used to traverse the multi-modal learning data stream to extract an abnormal feature set, perform double-channel mode analysis on the power distribution room based on the abnormal feature set, and identify a double mode, which includes an equipment fault degradation mode and a power grid fault propagation mode; A risk analysis module is used to perform risk early warning on the power distribution equipment based on the equipment fault degradation mode, generate a risk early warning signal, and perform propagation simulation on the power distribution room based on the power grid fault propagation mode to generate a fault propagation simulation path; A fault tracing module is used to trace faults based on the risk early warning signal and the fault propagation simulation path, and perform remote maintenance on the power distribution room based on the fault positioning result by triggering a set of remote control instructions; The data stream generation module comprises: A hierarchical sampling unit is used to hierarchically sample the power distribution equipment through a multi-source sensor network to determine a plurality of key nodes; A periodic sampling unit is used to periodically sample based on the plurality of key nodes to obtain a dynamic monitoring data set, which includes a three-dimensional dynamic monitoring matrix; A spatial transfer learning unit is used to perform multi-dimensional analysis according to the three-dimensional dynamic monitoring matrix to obtain multi-dimensional monitoring features, and perform spatial transfer learning on the three-dimensional dynamic monitoring matrix based on the multi-dimensional monitoring features to obtain spatial modal data; A time series transfer learning unit is used to perform time series transfer learning on the three-dimensional dynamic monitoring matrix based on the multi-dimensional monitoring features to obtain time series modal data; A feature fusion unit is used to fuse the spatial modal data and the time series modal data to construct the multi-modal learning data stream.

2. The remote intelligent monitoring system for an electrical distribution room of claim 1, wherein, The periodic sampling unit comprises: An operating state analysis subunit is used to analyze the operating state of the power distribution equipment in the power distribution room by traversing the plurality of key nodes to obtain first or second equipment operating state parameters; A first sampling subunit is used to sample the plurality of key nodes at a first sampling frequency to obtain a first monitoring data set when the power distribution equipment is in the first equipment operating state; A second sampling subunit is used to automatically switch the first sampling frequency to a second sampling frequency to sample the plurality of key nodes to obtain a second monitoring data set when the load rate of the power distribution equipment exceeds a preset threshold, which is a second equipment operating state parameter; A data identification subunit is used to identify data in a three-dimensional space of the power distribution room based on the first and second monitoring data sets to construct the three-dimensional dynamic monitoring matrix.

3. The remote intelligent monitoring system for an electrical distribution room of claim 1, wherein, The fault mode identification module comprises: A feature analysis unit is used to analyze features based on the multi-modal learning data stream to obtain an equipment feature set and a power grid feature set; The abnormal feature hierarchical extraction unit is configured to perform abnormal feature set hierarchical extraction based on the device feature set and the power grid feature set, and obtain an abnormal feature set, which includes device abnormal features and power grid abnormal features. The fault degradation analysis unit is configured to, when the abnormal feature set is the device abnormal feature, start a physical entity channel, synchronize the device abnormal feature to the physical entity channel for fault degradation analysis, and obtain the device fault degradation mode. The fault propagation analysis unit is configured to, when the abnormal feature set is the power grid abnormal feature, start a power grid virtual channel, synchronize the power grid abnormal feature to the power grid virtual channel for fault propagation analysis, and obtain the power grid fault propagation mode.

4. The remote intelligent monitoring system for an electrical distribution room of claim 3, wherein, The abnormal feature hierarchical extraction unit includes: The feature layer construction subunit is configured to construct a plurality of abnormal feature layers, including a first abnormal feature layer and a second abnormal feature layer. The abnormal scanning subunit is configured to scan the device feature set through the first abnormal feature layer using a sliding time window to capture device mutation pulse information. The abnormal positioning subunit is configured to perform infrared temperature field topology analysis based on the device mutation pulse information, locate abnormal point coordinates, perform transient extraction based on the abnormal point coordinates, determine device transient abnormal features, and output the device transient abnormal features as the device abnormal features. The distortion analysis subunit is configured to perform distortion analysis on the power grid feature set through the second abnormal feature layer, and mark a distortion accumulation interval. The fluctuation analysis subunit is configured to perform power fluctuation analysis based on the distortion accumulation interval, construct a power fluctuation frequency spectrum, perform oscillation analysis based on the power fluctuation frequency spectrum, determine power grid steady-state abnormal features, and output the power grid steady-state abnormal features as the power grid abnormal features.

5. The remote intelligent monitoring system for an electrical distribution room of claim 1, wherein, The risk analysis module includes: The degradation analysis unit is configured to perform degradation analysis based on the device fault degradation mode, determine a degradation stage, perform operation impact analysis on the power distribution room based on the degradation stage, and generate a plurality of stage impact coefficients. The early warning classification matching unit is configured to perform early warning classification matching according to the plurality of stage impact coefficients, and construct the risk early warning signal. The fault diffusion analysis unit is configured to initialize a simulation environment, construct power grid simulation environment parameters, perform fault diffusion analysis based on the power grid fault propagation mode, and determine diffusion key nodes. The propagation simulation unit is configured to perform propagation simulation on the power distribution room based on the diffusion key nodes, and generate the fault propagation simulation path.

6. The remote intelligent monitoring system for an electrical distribution room of claim 1, wherein, The fault tracing module includes: The signal-path matching unit is configured to match the risk early warning signal with the fault propagation simulation path, and obtain a signal-path matching result. The coincidence analysis unit is configured to perform coincidence analysis based on the signal-path matching result, and determine direct correlation features and indirect correlation features based on coincidence data. The feature weighting calculation unit is configured to perform feature weighting calculation based on the direct correlation features and the indirect correlation features, and determine a plurality of feature quantization indicators. The reverse trace analysis unit is configured to perform reverse trace analysis on the device topology relationship according to the plurality of characteristic quantization indexes, and obtain a fault positioning result, wherein the fault positioning result comprises an initial fault point. The permission dynamic allocation unit is configured to perform multi-layer confidence verification based on the initial fault point, obtain a confidence verification result, and perform permission dynamic allocation in the power distribution room according to the confidence verification result and the initial fault point, and generate the remote control instruction set.

7. A remote intelligent monitoring system for an electrical distribution room as claimed in claim 6 wherein, The coincidence analysis unit comprises: The spatial resolution subunit is configured to perform spatial resolution based on the signal-path matching result, and extract spatial coordinate data; The time resolution subunit is configured to perform time resolution based on the signal-path matching result, and extract time sequence labels; The first coincidence subunit is configured to perform three-dimensional spatial coincidence analysis according to the spatial coordinate data and the time sequence labels, obtain a first coincidence degree, and identify the direct correlation characteristic according to the first coincidence degree; The second coincidence subunit is configured to perform time sequence causal chain analysis according to the spatial coordinate data and the time sequence labels, obtain a second coincidence degree, and identify the indirect correlation characteristic according to the second coincidence degree.

8. A remote intelligent monitoring method of a power distribution room, characterized in that, The remote intelligent monitoring method of the power distribution room according to any one of claims 1 to 7 comprises: Periodically sampling the power distribution equipment by the multi-source sensor network deployed in the power distribution room, obtaining a dynamic monitoring data set for transfer learning, and generating a multi-modal learning data stream; Extracting an abnormal feature set by traversing the multi-modal learning data stream, performing double-channel mode analysis on the power distribution room according to the abnormal feature set, and identifying a double mode, wherein the double mode comprises a device fault degradation mode and a power grid fault propagation mode; Performing risk warning on the power distribution equipment based on the device fault degradation mode, generating a risk warning signal, and performing propagation simulation on the power distribution room based on the power grid fault propagation mode, and generating a fault propagation simulation path; Performing fault tracing according to the risk warning signal and the fault propagation simulation path, and triggering a remote control instruction set for the power distribution room according to a fault positioning result to perform remote maintenance on the power distribution room.

9. An electronic device, comprising: Comprise: At least one processor; A memory in communication connection with the at least one processor; The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the functions of the remote intelligent monitoring system of the power distribution room according to any one of claims 1 to 7.

Images