Risk early warning and decision support method and system for uninterrupted operation

Abstract

The application discloses a risk early warning and decision support method and system for live working, and the method comprises the following steps: acquiring real-time monitoring data and associated historical data in a target working area; quantitatively evaluating arc risk of current live working; constructing a risk propagation network model, simulating power flow transfer caused by current live working, and quantitatively evaluating cascading overload risk of current live working based on the simulation result; comprehensively evaluating by fusing the arc risk evaluation result, the cascading overload risk evaluation result and environmental data in the real-time monitoring data to determine a real-time risk level of current live working; when the real-time risk level exceeds a preset threshold, generating decision support information according to a risk dominant factor, which is used for optimizing a working scheme of current live working. The application better improves safety and decision scientificity of the working scheme.

Description

Technical Field

[0001] This invention relates to the field of power system safety operation and artificial intelligence application technology, specifically to a risk warning and decision support method and system for live-line operation, and in particular, a risk warning and decision support method and system for live-line operation based on artificial intelligence (AI). Background Technology

[0002] Live-line work (such as energized work and bypass work) is a key technology for ensuring reliable power supply from the power grid and improving the user's electricity experience. However, such work is carried out while the power equipment is energized or partially energized, facing multiple complex and coupled risks, including arc burns, equipment overload, insulation breakdown, and disruption of the power grid's operating mode. Traditional work safety management relies heavily on procedures and the experience of workers, lacking the ability to quantitatively perceive and proactively warn of dynamic and hidden risks.

[0003] Existing technologies contain some research on power operation safety or grid risk early warning, but they generally suffer from the following shortcomings: 1) Most assessments are based on static work tickets and grid models, resulting in static risk perception and a lack of early warning capabilities for dynamic risks that occur instantaneously during operations (such as sudden load transfers and transient overvoltages); 2) The risk assessment models used are relatively general and cannot accurately and timely assess the unique operational logic, physical processes (such as arc generation mechanisms and insulation gap changes) and risk evolution paths of live-line work; 3) Early warning results are often presented in a simple manner and cannot provide actionable optimization and adjustment suggestions that are strongly correlated with specific work steps, resulting in weak decision support. Summary of the Invention

[0004] To address the issues of insufficient dynamic risk perception, disconnect between early warning models and operational scenarios, and low level of decision support, this invention provides a risk early warning and decision support method and system for uninterrupted power supply operations. This solution enables proactive perception of dynamic risks, better alignment of early warning models with specific operational scenarios, and the transformation of early warning information into actionable safety decisions.

[0005] On the one hand, the present invention provides a risk warning and decision support method for uninterrupted power supply operations, comprising the following steps: S1: For the current live-line work, obtain real-time monitoring data and related historical data within the target work area, including historical arc fault waveform data; S2: Based on the current and / or voltage waveforms in the real-time monitoring data, a time-series anomaly detection model trained from the historical arc fault waveform data is used to dynamically identify the high-frequency transient distortion components of current and / or voltage that characterize potential arc features in the waveforms, so as to quantitatively assess the arc risk of the current uninterrupted power supply operation. S3: Based on the power grid topology of the target work area and the real-time monitoring data, construct a risk propagation network model, simulate the power flow transfer caused by the current uninterrupted power operation, and quantitatively assess the cascading overload risk of the current uninterrupted power operation based on the simulation results; S4: Integrate the results of the electric arc risk assessment, the results of the cascading overload risk assessment, and the environmental data in the real-time monitoring data to conduct a comprehensive assessment to determine the real-time risk level of the current uninterrupted power supply operation; S5: When the real-time risk level exceeds the preset threshold, decision support information is generated based on the risk-dominant factors to optimize the current uninterrupted power supply operation plan.

[0006] Furthermore, the dynamic identification of high-frequency transient distortion components of current and / or voltage in the waveform that characterize potential arc features in step S2 specifically includes: identifying harmonic abrupt changes or zero-crossing distortion features of current and voltage in a specific frequency band caused by arc-like transient processes.

[0007] Furthermore, the construction of the risk propagation network model in step S3 specifically includes: determining the key electrical associated equipment in the upstream, downstream and parallel tie lines of the working equipment based on the power grid topology connection relationship and real-time power flow sensitivity analysis; constructing a directed weighted graph as the local risk propagation network model with the working equipment as the root node, the key electrical associated equipment as the extension node, and the electrical connection relationship and power flow transfer relationship as the edges.

[0008] Furthermore, in step S2, the timing anomaly detection model is a hybrid model, which includes a convolutional layer for extracting local features from multi-channel current and voltage waveforms, and a recurrent neural network layer or attention layer for capturing the timing dependencies between the local features.

[0009] Furthermore, in step S3, the predicted cascading overload risk of related devices is obtained by inputting the topology, node load, and branch capacity of the local risk propagation network model into a graph neural network model.

[0010] Furthermore, in step S5, decision support information is generated based on the risk-dominant factors to form optimization suggestions for the current live-line work, specifically including: If the dominant risk factor is determined to be arc risk, the generated optimization suggestions will include at least adjusting the working phase, installing a transient overvoltage suppressor, or extending the insulation shielding range. If the dominant risk factor is determined to be the cascading overload risk of electrical equipment, the generated optimization suggestions will include at least adjusting the grid operation mode to transfer load, adding mobile energy storage vehicles as temporary power sources, or modifying the operation sequence to avoid load peaks.

[0011] Furthermore, after formulating the current optimization recommendations for live-line work, the following step S6 is also included: In a digital twin environment built based on a real power grid model, the work plan adjusted according to the optimization suggestions is loaded for simulation and deduction. During the simulation process, virtual operation steps are executed step by step, and the time-series anomaly detection model and local risk propagation network model are invoked in real time to re-predict and determine risks. Output the risk evolution trajectory and a list of key risk control points throughout the entire simulation process.

[0012] On the other hand, the present invention provides a risk warning and decision support system for uninterrupted power supply operations, used to implement the method described above, including: The data sensing and real-time access module is used to collect and access the multidimensional data, and to preprocess the real-time monitoring data. The feature extraction and risk calculation engine is used to run the time-series anomaly detection model and the graph neural network model to complete the identification and prediction of specific risks. The integrated risk assessment and decision generation module is used to integrate information from multiple risk sources and generate risk levels and optimization suggestions. The scheme simulation and verification module is used to simulate and verify the optimized operation scheme in a digital twin environment.

[0013] Furthermore, the feature extraction and risk calculation engine includes: An arc risk calculation unit is used to receive current and / or voltage waveforms in the unified data stream, call and run the timing anomaly detection model, and output arc risk assessment results. The overload risk calculation unit is used to construct a risk propagation network model based on the power grid topology of the target operating area and the preprocessed real-time monitoring data, call and run the graph neural network model, and output the cascading overload risk assessment results.

[0014] Furthermore, it also includes a human-computer interaction and visualization module, used for: The power grid equipment, operators, and operation processes in the digital twin environment are presented in a three-dimensional visualization format. The spatial distribution of real-time risks is overlaid on the three-dimensional visualization interface in the form of a heat map. It provides an interactive interface for users to adjust the parameters of the work plan and trigger a new round of risk assessment and simulation in real time.

[0015] Compared with the prior art, the present invention has at least the following beneficial effects: 1) By using a dedicated time-series anomaly detection model (such as 1D-CNN combined with Bi-LSTM), it is possible to automatically and dynamically extract and identify specific high-frequency transient distortion components from real-time three-phase current and voltage waveforms, thereby achieving forward-looking and accurate identification and early warning of instantaneous and weak arc risks, solving the problem of insufficient perception of dynamic risks by traditional methods.

[0016] 2) By combining real-time power flow sensitivity analysis (such as calculating LODF based on DC power flow model) to dynamically construct a local risk propagation network model with the operating equipment as the core, and using graph neural networks (such as GAT) to learn and infer the network, it is possible to simulate and quantify the risk of cascading overloads caused by power flow transfer due to operation disturbances to related equipment. This realizes topological and quantitative prediction of grid cascading overload risks, and extends risk assessment from single-point static judgment to network dynamic propagation, resulting in stronger early warning capabilities.

[0017] 3) It not only provides a comprehensive risk level integrating multi-source risk information, but also generates actionable optimization suggestions (such as phase adjustment and load transfer) that are strongly correlated with specific work steps based on the dominant risk factors. Furthermore, by simulating and re-verifying the optimized work plan in a digital twin environment, it achieves pre-emptive risk simulation and scheme comparison, upgrading decision support from passive early warning to proactive optimization and verification, forming an intelligent decision-making closed loop of "assessment-optimization-verification". This not only greatly improves the intelligence level of live-line work management, but also significantly enhances work safety and the scientific nature of decision-making. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention 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 only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the steps of the risk warning and decision support method for uninterrupted power supply operations according to the present invention.

[0020] Figure 2 This is a simplified flowchart of the risk warning and decision support method for uninterrupted power supply operations according to the present invention.

[0021] Figure 3 The present invention provides a system architecture diagram for a risk warning and decision support system for uninterrupted power supply operations. Detailed Implementation

[0022] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0023] It should be understood that the described embodiments are merely some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention. Method Implementation Examples

[0024] refer to Figure 1 ,like Figure 1 and Figure 2 As shown, the present invention provides a risk warning and decision support method for live-line working, comprising the following steps: S1: For the current live-line work, obtain real-time monitoring data and related historical data within the target work area. The historical data includes historical arc fault waveform data. S2: Based on the current and / or voltage waveforms in real-time monitoring data, a time-series anomaly detection model trained from historical arc fault waveform data is used to dynamically identify the high-frequency transient distortion components of current and / or voltage that characterize potential arc features in the waveforms, so as to quantitatively assess the arc risk of the current uninterrupted power supply operation. S3: Based on the power grid topology and real-time monitoring data of the target work area, construct a risk propagation network model, simulate the power flow transfer caused by the current uninterrupted power operation, and quantitatively assess the cascading overload risk of the current uninterrupted power operation based on the simulation results; S4: Integrate the results of the arc risk assessment, the results of the cascading overload risk assessment, and the environmental data in the real-time monitoring data to conduct a comprehensive assessment to determine the real-time risk level of the current live-line work; S5: When the real-time risk level exceeds the preset threshold, generate decision support information based on the risk-dominant factors to optimize the current work plan for uninterrupted power supply operations.

[0025] The core of step S1 is to achieve multi-dimensional real-time data acquisition and fusion (referring to preprocessing and correlation alignment). The following data is acquired and aggregated in real time via a data bus: 1) Operating Condition Data. Obtain three-phase current and voltage waveforms (sampling rate ≥ 4kHz) and real-time load data of the lines within the target operating area from SCADA / EMS (Supervisory Control And Data Acquisition / Energy Management System). For example, operating condition data can be obtained from a SCADA system using power standards such as IEC 104. The obtained operating condition data includes: high-sampling-rate instantaneous waveforms of three-phase current and voltage (e.g., sampling rate 4kHz) containing phase information, as well as switch status and active / reactive power.

[0026] 2) Equipment status data. This data includes, but is not limited to, the operational status data of relevant electrical equipment within the target work area, such as switches, transformers, circuit breakers, and lines. In specific implementation, temperature and partial discharge signals of relevant switches, transformers, and other equipment can be obtained from online monitoring systems. For example, online monitoring data such as temperature, partial discharge, and mechanical characteristics (such as opening and closing coil current) of key equipment can be obtained through Internet of Things protocols (such as the MQTT lightweight messaging protocol) to assess the equipment's "health."

[0027] 3) Environmental Data. Real-time wind speed, wind direction, humidity, and rainfall intensity data are obtained from micro-weather stations at the work site. For example, data from micro-weather sensors deployed on or near the work poles can be accessed to collect environmental data including wind speed, wind direction, humidity, temperature, and rainfall. Special attention should be paid to the real-time accuracy of wind speed and humidity, as these directly affect air insulation strength.

[0028] 4) Work Behavior Data. The location of workers is obtained through UWB (Ultra-Wideband) positioning systems, and the operational status is obtained through smart tool sensors. For example, workers are located using UWB or Bluetooth beacons, tool status (such as the angle and torque of insulating rods) is sensed through smart safety helmets or tool sensors, and the progress of standard operating procedures (SOPs) is recorded through a mobile terminal APP.

[0029] 5) Historical Data. Retrieve arc fault waveform files and accident analysis reports from the database under similar historical operating conditions. For example, retrieve and extract arc fault waveform files, accident analysis reports, and typical operation records for similar scenarios from historical databases (such as fault waveform databases and production management systems, PMS) by tags such as voltage level, operation type, equipment type, and weather conditions.

[0030] Step S1 above, which achieves the "fusion" of multidimensional data, refers to the preprocessing and correlation alignment of multi-source heterogeneous data, specifically including: I) Time synchronization. Data from different systems (such as SCADA, weather stations, and positioning systems) are given a unified time stamp to ensure that subsequent analyses are based on a "system snapshot" at the same time.

[0031] II) Spatial Association. This involves associating equipment status data, environmental data, and personnel location data with specific equipment or locations within the power grid topology.

[0032] III) Format standardization. Convert data from different protocols and formats into a unified data structure that the model can process.

[0033] Taking "live-line connection of current-carrying wire on a 10kV line" as an example, data acquisition is initiated 30 minutes before the operation: I) Obtain the A / B / C three-phase current and voltage waveform data (sampling rate 4kHz) of the target line (such as "XX line 912 switch") from the distribution automation master station, as well as the associated pole-mounted switch status.

[0034] II) Obtain the cable head temperature monitoring data (current value: 32°C) and partial discharge monitoring data (current value: 15pC) of the switch on the working tower.

[0035] III) Obtain data from the micro-weather station at the top of the tower: wind speed 3.5 m / s (gusts 5.1 m / s), humidity 65%, no rainfall.

[0036] IV) The person in charge of the operation selects the standard work order "10kV live connection of lead wire (insulated glove operation method)" through the mobile terminal, and the system begins to receive the location information (real-time coordinates) of the UWB tags worn by the two operators (ID:001, 002).

[0037] V) The system automatically retrieves waveform files and analysis conclusions of three "arc flashover" faults that occurred within the past year on the same line, under the same operation type and with similar wind speeds (2-5 m / s) from the historical database, and loads them into memory as background data.

[0038] Step S2 primarily involves risk identification of high-frequency transient arcs. See also... Figure 2The real-time three-phase current and voltage waveforms (as three-channel timing inputs) are input into a pre-trained timing anomaly detection model. This model employs a multi-channel one-dimensional convolutional module (containing multiple parallel convolutional kernels for extracting local features of each phase signal and inter-phase relationships) followed by a bidirectional long short-term memory network layer (for learning long-term temporal dependencies of features). The input is multi-channel timing data, i.e., real-time acquired three-phase current and voltage waveforms, typically input in a sliding window format with a window length of 100ms to 500ms. The output represents the probability value (e.g., between 0 and 1) of the presence of arc-like transient distortion within the current time window; for example, outputting a "high-frequency transient distortion risk probability value P_arc" characterizing the potential presence of an arc at the current moment. Local features refer to frequency or time domain features extracted from the waveform through the convolutional layers, such as high-frequency energy, zero-crossing distortion, and harmonic content. Temporal dependencies refer to the evolution of features over time, such as patterns like "high-frequency oscillation followed by current zero-crossing delay." In other words, this timing anomaly detection model uses a one-dimensional convolutional neural network to extract local features of the waveform, and then uses a recurrent neural network or attention mechanism to capture the temporal dependencies between features, ultimately outputting a probability value characterizing the risk of an electric arc. Preferably, dynamically identifying the high-frequency transient distortion components of current and / or voltage that characterize potential electric arc features specifically includes: identifying harmonic abrupt changes or current / voltage zero-crossing distortion features generated by early electric arcs in specific frequency bands. For example, preferably, the features identified by this timing anomaly detection model are specifically manifested as energy surges or abnormal distortions at current zero crossings within the frequency band of 5kHz to 500kHz.

[0039] Preferably, in step S2, the time-series anomaly detection model is a hybrid model based on deep learning, comprising convolutional layers for extracting local features from multi-channel current and voltage waveforms, and recurrent neural network layers or attention layers for capturing temporal dependencies between features. In practical applications, the preferred architecture of this time-series anomaly detection model is a hybrid model structure of "multi-channel one-dimensional convolutional neural network (1D-CNN) + bidirectional long short-term memory network (Bi-LSTM) + fully connected layers". The one-dimensional convolutional neural network (1D-CNN) uses multiple layers with different widths (e.g., 10ms). A convolutional kernel with a 50ms time interval is used to scan the three-phase waveform in parallel, extracting local features of each phase and inter-phase correlation features (e.g., the coordinated distortion of phase A and phase B waveforms at a specific moment). A bidirectional long short-term memory network (Bi-LSTM) receives the feature sequence extracted by the CNN and learns the long-term temporal dependencies of features from both forward and backward dimensions, capturing fixed patterns such as "high-frequency oscillation followed by a current zero-crossing delay". An optional attention mechanism is added after the Bi-LSTM network to allow the timing anomaly detection model to focus on the most abnormal time segment of the waveform. The input of this timing anomaly detection model is the real-time three-phase current and voltage waveform (as a three-channel timing input), and the final output is a value between 0 and 1 output by the Sigmoid function, representing the probability P_arc of the presence of arc-like high-frequency transient distortion within the current time window.

[0040] Furthermore, the training of this timing anomaly detection model utilizes a large amount of labeled historical waveform data. Positive samples are waveform recordings taken hundreds of milliseconds before and after an arc fault; negative samples are waveform data under other disturbances such as normal operation, lightning interference, and switching operations. Real-time sampled three-phase waveform data (e.g., a 200ms sliding window) is input into the timing anomaly detection model, which outputs in real-time the probability P_arc value (within the range of 0-1) of the presence of arc-induced high-frequency transient distortion. When the probability P_arc = 0.8 exceeds a preset threshold (e.g., a preset threshold of 0.7), an early warning is triggered. It is understandable that this preset threshold is set according to the actual working conditions; different working conditions and operational requirements will lead to different preset threshold settings.

[0041] Continuing with the example of "live-line connection of a 10kV conductor," the timing anomaly detection model continuously analyzes the real-time waveform during the operation. At the moment the worker is about to contact the conductor, the model detects a 300kHz high-frequency oscillation envelope with an amplitude only 5% of the power frequency current within 2ms of the B-phase current waveform, accompanied by a slight distortion at the zero-crossing point. Bi-LSTM identifies this pattern as highly similar to historical positive arc samples. The probability P_arc value of the timing anomaly detection model, indicating the presence of arc-related high-frequency transient distortion, rapidly jumps from 0.1 to 0.85. The system determines that a "potential arc risk has been detected" and marks the risk moment as the current time point.

[0042] Step S3 primarily involves predicting the propagation of cascading overload risks. See also... Figure 3 Based on the current power grid topology and operational plan, a local risk propagation network model is constructed. Preferably, constructing the local risk propagation network model specifically includes: determining key associated equipment upstream, downstream, and in parallel tie lines of the operational equipment based on the power grid topology connections and real-time power flow sensitivity analysis. Real-time power flow sensitivity analysis quantifies the proportional relationship of power flow redistribution in other branches when the power flow in a certain branch (such as downstream equipment) changes, i.e., the power flow transfer relationship. As an optional implementation, a DC power flow model can be used for rapid approximate calculation. The specific process includes: a) Based on the real-time power grid topology (circuit breaker and disconnector status) and the active power injected into each node, the node admittance matrix B′ under DC power flow is formed.

[0043] b) Calculate the Power Transfer Distribution Factor (PTDF) matrix, which describes the change in power flow in each branch when a unit power is transferred between any pair of source-sink nodes.

[0044] c) Based on the PTDF matrix, further calculate the Line Outage Distribution Factor (LODF). For a branch k (connecting nodes m and n) that is planned to be disconnected or affected by work, the LODF coefficient (LODF_{l,k}) for power flow transfer to any other operating branch l after disconnection can be estimated by the following formula: LODF_ {l,k} = (PTDF_ {l,m} - PTDF_ {l,n} ) / (1 - PTDF_ {k,m} + PTDF_ {k,n}); where PTDF_ {l,m}This represents the percentage change in power flow in branch l when unit power is injected from node m. Therefore, if the pre-disconnect power flow in branch k is P... _k Then, after it is disconnected, the power flow ΔP on branch l increases approximately. _l For: ΔP _l ≈ LODF_ {l,k} × P _k This ΔP _l This refers to the quantified power flow relationship, which can be used to construct edge weights in a risk propagation network model.

[0045] The determination of critical electrical interconnected equipment can be based on one or more of the following factors: electrical distance (e.g., electrical coupling coefficient); power flow transfer sensitivity coefficient; current load rate of the equipment; rated capacity and load margin of the equipment; and whether it is located upstream, downstream, or in a parallel connection path of the operating path. For example, real-time power flow sensitivity analysis can be used to calculate the power flow changes of each interconnected equipment after the operating equipment is disconnected or operated. If the power flow change of a certain equipment exceeds a certain percentage of its rated capacity (e.g., 10%), or if the current load rate of the equipment is already high (e.g., >80%), it is determined to be a critical electrical interconnected equipment.

[0046] Using the equipment within the target work area as the root node, key associated equipment with which it has an electrical connection as the extension node, and electrical connections and power flow transfer relationships as edges, a directed weighted graph is constructed as a local risk propagation network model. Preferably, the prediction of cascading overload risks of associated equipment is accomplished by inputting the topology, node load, and branch capacity of the local risk propagation network model into a graph neural network model. Node load refers to the real-time active / reactive power, current, or load factor of a node in the power grid (such as a substation busbar or distribution transformer). Branch capacity refers to the maximum current or power limit allowed to pass through branch equipment such as transmission lines and transformers. Node load data can be obtained in real time from the energy management system; branch capacity data can be retrieved from the equipment parameter database or ledger system. For example, taking the planned operation's "phase B conductor of tower A" as the root node, based on electrical connection relationships and power flow transfer coefficients, key associated equipment on the upstream power supply side, downstream load side, and parallel tie lines (such as "10kV outgoing switch of substation C," "distribution transformer T1," etc.) are automatically identified as extended nodes to form a directed weighted graph. This directed weighted topology, the real-time load rate of each node, and the capacity limit of each branch are input into the graph attention network model. This graph attention network model learns the risk propagation patterns in historical overload events and outputs the predicted load rate of each associated device and the "chain overload risk probability value P_overload_i" after the operation is executed.

[0047] Specifically, for example, the work plan involves isolating the section of the line downstream of "XX line 912 switch" for bypass work. The system automatically constructs a risk propagation diagram with "912 switch" as the root node. Extended nodes include: the upstream "10kV busbar of 110kV substation A", the downstream "distribution transformers T1, T2, T3", and the parallel feeder "YY line" connected via tie switch "912Z". Real-time data input: node load rate (busbar: 65%, T1: 80%, T2: 60%, T3: 75%, YY line: 50%). After analysis by the Graph Attention Network (GAT) model, it predicts that when 912 switch is opened, its downstream 30kW load will be completely transferred to the YY line. This will cause the load rate of the YY line to surge from 50% to 85%, and the load current of its upstream sectionalizing switch (rated current 400A) will reach 378A (predicted value), exceeding its long-term safe operation threshold (365A). Therefore, the model outputs P_overload_i = 0.92 (high risk) for this segmented switch.

[0048] Step S4 primarily involves a comprehensive dynamic risk assessment. It receives the high-frequency transient distortion risk probability value P_arc from step S2 and the cascading overload risk probability value P_overload_i from step S3. Weights W1 and W2 are set (e.g., for live-line work, W1=0.7 focuses on arcing; for complex bypass work, W2=0.6 focuses on grid operation mode), and the comprehensive risk value R is calculated as R=W1*P_arc+W2*max(P_overload_i). Simultaneously, real-time wind speed (to correct for air gap breakdown risk; for example, when gusts exceed level 5, the risk level of all outdoor electrical systems is automatically increased by one level) and humidity (high humidity >80% increases insulator surface leakage current and flashover probability) are introduced as correction factors to adjust the comprehensive risk value R. Based on the preset range into which the corrected comprehensive risk value R falls, for example, the preset range is: low risk (R<0.3), medium risk (0.3≤R<0.6), high risk (0.6≤R<0.8), and extremely high risk (R≥0.8); determine the final risk level (e.g., low, medium, high, extremely high).

[0049] For example, the system obtains: P_arc=0.85, max(P_overload_i) = 0.92. It sets W1=0.6 and W2=0.4.

[0050] Calculate the base risk value: R_base = 0.6 * 0.85 + 0.4 * 0.92 = 0.878.

[0051] Environmental correction: Current wind speed 5.1 m / s (gusts), risk level increased by one level according to rules (equivalent to an R value increase of 0.15). Humidity 65%, no additional correction required.

[0052] Final risk value: R = 0.878 + 0.15 = 1.028.

[0053] Level mapping: (R=1.028) ≥ 0.8, judged as "extremely high risk".

[0054] Step S5 primarily involves intelligently generating optimization suggestions for uninterrupted power supply (UPS) operations. Preferably, the generated optimization suggestions specifically include: if the dominant risk factor is determined to be arc risk, the generated optimization measures should at least include adjusting the operating phase, installing a transient overvoltage suppressor, or extending the insulation shielding range. For example, if the risk level is "high" or higher, the system initiates source tracing analysis. If the contribution of the high-frequency transient distortion risk probability value P_arc exceeds 70%, and the dominant risk is determined to be arc risk, then measures are retrieved from the knowledge base: "It is recommended to adjust the operating phase from the currently heavily loaded B phase to the lightly loaded A phase" or "It is recommended to install a transient overvoltage suppressor at the operating point."

[0055] Preferably, if the dominant risk factor is determined to be the overload risk of associated equipment, the generated optimization measures will include at least adjusting the grid operation mode to transfer the load, adding a mobile energy storage vehicle as a temporary power source, or modifying the operation sequence to avoid peak loads. For example, if a certain cascading overload risk probability value P_overload_i has the largest contribution, and the dominant risk is determined to be cascading overload, then the following suggestions will be generated: "It is recommended to transfer the XX load to the adjacent line through the feeder tie switch before operation" or "It is recommended to call a 500kW mobile energy storage vehicle as a temporary bypass power source".

[0056] For example, in the above case, system tracing shows that the contribution of the high-frequency transient distortion risk probability value (arc risk probability) P_arc is approximately (0.6*0.85) / 0.878 ≈ 58%, and the maximum contribution of the cascading overload risk probability value (P_overload_i) is approximately (0.4*0.92) / 0.878 ≈ 42%. Therefore, the dominant risk is determined to be arc risk.

[0057] Rule matching was performed, identifying electric arc risk as the dominant risk, coupled with high ambient wind speed. The knowledge base matching rule was: "IF high electric arc risk AND wind speed > 5 m / s, THEN recommends prioritizing 'adjusting the operation phase' or 'suspending the operation until the wind speed decreases'." Therefore, the system generated two optimization suggestions: Recommendation 1 (Recommended): Immediately suspend contact operations on phase B conductors. It is recommended to adjust the working phase to phase A (load rate 45%), which currently has a lighter load. This is expected to reduce the arc risk probability P_arc to below 0.3.

[0058] Recommendation 2 (Alternative): Wait for the wind speed to drop below 4 m / s before continuing operations. The estimated waiting time is approximately 30 minutes.

[0059] Optionally, after generating the optimization suggestions for uninterrupted power supply operations (step S5), the following steps S6 are also included: In a digital twin environment built based on a real power grid model, the operation plan adjusted by the optimization suggestions is loaded for simulation; during the simulation, virtual operation steps are executed step by step, and the time-series anomaly detection model and local risk propagation network model are called in real time to re-predict and determine the risks; the risk evolution trajectory and list of key risk control points of the entire simulation process are output.

[0060] During the simulation, after each virtual operation step, the system first reconstructs or updates the local risk propagation network model based on the updated power grid status (such as switch changes and power flow variations) in the digital twin environment. Subsequently, based on this updated model, the system calls the time-series anomaly detection model and the graph neural network model in real time to re-predict risks. This process ensures that risk prediction can dynamically evolve with the operation progress and the power grid status.

[0061] Therefore, step S6 mainly involves simulation and confirmation of the proposed solution. The new operational plan, after adopting the optimized suggestions, is imported into a digital twin environment built based on real power grid parameters. Furthermore, as the system progressively drives the virtual operational process, it synchronously calls the models from steps S2 and S3 at each step to perform real-time risk re-prediction and determination. Finally, it outputs a "risk evolution curve" from the start to the end of the operation, along with marked risk peak points, for decision-makers to confirm and make final decisions.

[0062] System Implementation Examples On the other hand, the present invention provides a risk warning and decision support system for uninterrupted power supply operations, used to implement the method described above. Figure 3 As shown, the system of the present invention mainly includes: a data perception and real-time access module, a feature extraction and risk calculation engine, a comprehensive risk assessment and decision generation module, and a scheme simulation and verification module.

[0063] The data perception and real-time access module, serving as the system's data hub, is used to collect and access multi-dimensional data. This module is built upon a high-performance message queue (such as Apache Kafka) and a stream processing platform. It is also configured with various adapters for: a) Communicate with the SCADA / EMS system of the provincial / prefecture-level dispatch center through power protocols such as IEC 104, IEC 61850, and DNP3 to subscribe to and obtain in real time the power flow, switch status, and high sampling rate waveform data of the target area.

[0064] b) Access various online monitoring devices (such as intelligent inspection robots, cable head temperature measurement, and partial discharge monitoring) deployed in substations and lines via protocols such as MQTT and HTTPS to obtain equipment status data.

[0065] c) Access refined forecast data from meteorological departments and real-time data from on-site micro-weather stations via standard APIs.

[0066] d) Receive precise location and status signals of workers, vehicles, and tools in real time through the local area network interface of the UWB positioning system base station and the Bluetooth beacon gateway.

[0067] e) Query and extract historical case data from the enterprise's production management system (PMS) and historical fault recording database as needed.

[0068] In addition, the data perception and real-time access module cleans, aligns (unifies the time stamp), and formats all accessed heterogeneous data, and publishes it to a unified data topic for upper-layer modules (such as the feature extraction and risk calculation engine) to subscribe to, access, and consume.

[0069] The feature extraction and risk calculation engine is used to run the timing anomaly detection model and risk propagation network model to complete specific risk identification and prediction. This module is the intelligent core of the system, adopting a microservice architecture and containing two independent but collaborative computing units: an arc risk calculation unit and an overload risk calculation unit. The arc risk calculation unit receives current and / or voltage waveforms from the unified data stream, calls and runs the timing anomaly detection model (such as one based on TensorFlow Serving or PyTorch Serve), and outputs arc risk assessment results. This unit continuously subscribes to real-time three-phase waveform data streams from the data bus, performs real-time inference in a sliding window manner, outputs P_arc values ​​multiple times per second, accompanied by timestamps of anomaly feature segments. The overload risk calculation unit constructs a risk propagation network model based on the power grid topology of the target operating area and preprocessed real-time monitoring data, calls and runs the graph neural network model (GAT), and outputs cascading overload risk assessment results. When a new work plan is input, this unit first calls the network analysis service based on the real-time topology to dynamically generate a "local risk propagation network model." Subsequently, the adjacency matrix, node feature matrix, and edge feature matrix of the subgraph are input into the GAT model for forward inference, outputting the P_overload_i set of all associated devices in one go. This unit is preferably designed to support batch computation and real-time fast recalculation.

[0070] The integrated risk assessment and decision generation module is a combination of a rules engine and decision logic. This module receives results from the risk calculation engine and executes the following logic: 1) Configurable Management. Provides a management interface that allows administrators to preset risk fusion weights (W1, W2), risk thresholds at each level, and environmental correction parameter tables according to the type of work (such as live-line work, bypass work).

[0071] 2) Dynamic assessment. The above configuration is invoked in real time to calculate the comprehensive risk value R and determine the risk level.

[0072] 3) Knowledge base matching. An embedded structured "risk-measure" knowledge graph. When recommendations need to be generated, based on the risk tracing (contribution analysis) results, the most relevant optimization measures are matched from the knowledge graph, and specific parameters (such as the recommended phase name and the recommended load transfer amount) are automatically filled in.

[0073] 4) Report Generation. Automatically generates a structured report that includes a risk overview, analysis details, optimization suggestions, and deductive conclusions.

[0074] The scheme simulation and verification module integrates a lightweight power grid digital twin simulation engine. This engine maintains a virtual model synchronized with the physical power grid. When a work scheme to be verified is received, it performs scheme analysis, interactive simulation, trajectory recording, and comparison. Scheme analysis includes parsing the steps described in natural language or standardized work tickets into a series of operational instructions (e.g., "disconnect switch A," "close bypass switch B") on the equipment status in the twin model. Interactive simulation involves executing instructions step-by-step; at each step, an internal interface is used to call the risk calculation engine to recalculate the risk based on the updated state of the twin. Trajectory recording and comparison involves completely recording the changes in all risk indicators during the simulation process, forming a "risk evolution trajectory," which can be compared with the risk trajectory of the original scheme to quantitatively demonstrate the optimization effect.

[0075] Preferably, the system of the present invention further includes a human-computer interaction and visualization module, which provides a web application accessible from multiple terminals. This module is specifically used to implement: 1) 3D Visualization Scenes. Presenting power grid equipment, personnel, and operational processes in a digital twin environment using 3D visualization. For example, based on WebGL technology, constructing 3D scenes of substations and transmission lines can intuitively display equipment status, risk heat maps (e.g., overloaded equipment is displayed in red), the dynamic location of personnel (virtual avatars), and safety electronic fences.

[0076] 2) Decision-Making Cockpit Function (Risk Distribution Visualization). The spatial distribution of real-time risks is displayed overlaid on a 3D visualization interface in the form of a heat map. For example, a dashboard can centrally display the overall risk level, key indicator trends, and warning lists.

[0077] 3) Provide an interactive interface (or interactive work sandbox) and graphical tools that allow users to drag and drop equipment and plan work steps directly on the twin model. The system can provide risk assessment feedback in real time, realize the "what you see is what you get" solution formulation, allow users to adjust work plan parameters and trigger a new round of risk assessment and simulation in real time.

[0078] 4) Multi-role view. Provides customized views with different information densities and focus points for work planners, site supervisors, and safety monitors.

[0079] It should be noted that in this paper, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply these relationships. There is no such actual relationship or order between entities or operations. Furthermore, the terms "including" and "package" do not apply. The word "comprise" or any other variation thereof is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprises a..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0080] In this embodiment of the invention, the term "and / or" describes the relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. The character " / " generally indicates that the preceding and following associated objects have an "or" relationship.

[0081] The various embodiments in this specification are described in a related manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0082] In particular, the device embodiments are basically similar to the method embodiments, so they are described in a simpler way. For relevant details, please refer to the description of the method embodiments.

[0083] For ease of description, the above apparatus is described by dividing it into various functional units / modules. Of course, in implementing this invention, the functions of each unit / module can be implemented in one or more software and / or hardware.

[0084] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0085] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for risk warning and decision support for live-line work, characterized in that, Includes the following steps: S1: For the current live-line work, obtain real-time monitoring data and related historical data within the target work area, including historical arc fault waveform data; S2: Based on the current and / or voltage waveforms in the real-time monitoring data, a time-series anomaly detection model trained from the historical arc fault waveform data is used to dynamically identify the high-frequency transient distortion components of current and / or voltage that characterize potential arc features in the waveforms, so as to quantitatively assess the arc risk of the current uninterrupted power supply operation. S3: Based on the power grid topology of the target work area and the real-time monitoring data, construct a risk propagation network model, simulate the power flow transfer caused by the current uninterrupted power operation, and quantitatively assess the cascading overload risk of the current uninterrupted power operation based on the simulation results; S4: Integrate the results of the electric arc risk assessment, the results of the cascading overload risk assessment, and the environmental data in the real-time monitoring data to conduct a comprehensive assessment to determine the real-time risk level of the current uninterrupted power supply operation; S5: When the real-time risk level exceeds the preset threshold, decision support information is generated based on the risk-dominant factors to optimize the current uninterrupted power supply operation plan.

2. The method according to claim 1, characterized in that, The step S2, which describes dynamically identifying the high-frequency transient distortion components of current and / or voltage in the waveform that characterize potential arcing features, specifically includes: identifying harmonic abrupt changes or zero-crossing distortion features of current and voltage in a specific frequency band caused by arcing transient processes.

3. The method according to claim 1, characterized in that, The construction of the risk propagation network model in step S3 specifically includes: determining the key electrical associated equipment in the upstream, downstream and parallel tie lines of the working equipment based on the power grid topology connection relationship and real-time power flow sensitivity analysis; constructing a directed weighted graph as the local risk propagation network model with the working equipment as the root node, the key electrical associated equipment as the extension node, and the electrical connection relationship and power flow transfer relationship as the edges.

4. The method according to claim 1, characterized in that, In step S2, the timing anomaly detection model is a hybrid model, which includes a convolutional layer for extracting local features from multi-channel current and voltage waveforms, and a recurrent neural network layer or attention layer for capturing the timing dependencies between the local features.

5. The method according to claim 1, characterized in that, In step S3, the predicted cascading overload risk of related devices is obtained by inputting the topology, node load, and branch capacity of the local risk propagation network model into a graph neural network model.

6. The method according to claim 1, characterized in that, In step S5, decision support information is generated based on the risk-dominant factors to form optimization suggestions for the current live-line work, specifically including: If the dominant risk factor is determined to be arc risk, the generated optimization suggestions will include at least adjusting the working phase, installing a transient overvoltage suppressor, or extending the insulation shielding range. If the dominant risk factor is determined to be the cascading overload risk of electrical equipment, the generated optimization suggestions will include at least adjusting the grid operation mode to transfer load, adding mobile energy storage vehicles as temporary power sources, or modifying the operation sequence to avoid load peaks.

7. The method according to claim 1, characterized in that, After formulating the optimization recommendations for the current live-line work, the following step S6 is also included: In a digital twin environment built based on a real power grid model, the work plan adjusted according to the optimization suggestions is loaded for simulation and deduction. During the simulation process, virtual operation steps are executed step by step, and the time-series anomaly detection model and local risk propagation network model are invoked in real time to re-predict and determine risks. Output the risk evolution trajectory and a list of key risk control points throughout the entire simulation process.

8. A risk early warning and decision support system for live-line work, used to implement the method according to any one of claims 1-7, characterized in that, include: The data sensing and real-time access module is used to collect and access the multidimensional data, and to preprocess the real-time monitoring data. The feature extraction and risk calculation engine is used to run the time-series anomaly detection model and the graph neural network model to complete the identification and prediction of specific risks. The integrated risk assessment and decision generation module is used to integrate information from multiple risk sources and generate risk levels and optimization suggestions. The scheme simulation and verification module is used to simulate and verify the optimized operation scheme in a digital twin environment.

9. The system according to claim 8, characterized in that, The feature extraction and risk calculation engine includes: An arc risk calculation unit is used to receive current and / or voltage waveforms in the unified data stream, call and run the timing anomaly detection model, and output arc risk assessment results. The overload risk calculation unit is used to construct a risk propagation network model based on the power grid topology of the target operating area and the preprocessed real-time monitoring data, call and run the graph neural network model, and output the cascading overload risk assessment results.

10. The system according to claim 9, characterized in that, It also includes a human-computer interaction and visualization module, used for: The power grid equipment, operators, and operation processes in the digital twin environment are presented in a three-dimensional visualization format. The spatial distribution of real-time risks is overlaid on the three-dimensional visualization interface in the form of a heat map. It provides an interactive interface for users to adjust the parameters of the work plan and trigger a new round of risk assessment and simulation in real time.

Images