Continuous operation risk assessment system and method based on main network operation scene

Abstract

The invention discloses a continuous operation risk assessment system and method based on a main network operation scene. The method comprises the steps that a mobile terminal selects or newly builds an operation scene template, generates a SceneID after instantiation, pulls risk factors and weights, and subscribes related real-time data streams; the edge node collects and processes sensor data, inputs the sensor data into the lightweight dynamic Bayesian network model edge reasoning to calculate a risk value, and judges whether risk mutation is triggered or not. And during triggering, the cloud center pulls full-amount data, a management and control measure is generated through high-precision dynamic Bayesian network recalculation and reinforcement learning and pushed to the terminal, and the terminal writes back or issues and executes the management and control measure after confirmation. And after the operation is finished, packaging data and uploading the data to the cloud center, updating the model by the cloud center through federal learning, and pushing a new weight to the edge node through OTA.

Description

Technical Field

[0001] This invention relates to the field of power grid safety technology, and more specifically, to a system and method for continuous operation risk assessment based on main grid operation scenarios. Background Technology

[0002] Maintenance, technical upgrades, and infrastructure construction in the main power grid area are becoming increasingly intensive, exhibiting complex characteristics such as high voltage, large spans, multiple professional intersections, and short working windows. Against this backdrop, on-site safety risk management faces unprecedented challenges. Currently, risk assessments in the power industry are generally guided by the "Power Safety Work Regulations" and the "Power Grid Enterprise Safety Production Risk Management System," employing a static model of "one-time pre-operation assessment and periodic in-operation review." This traditional model has revealed many limitations in practice and is no longer suitable for the dynamic and complex modern power grid operating environment, specifically manifested in the following ways: First, the risk assessment scenarios are too coarse-grained and lack specificity. Most companies still roughly categorize main grid operations into broad categories such as "transmission, substation, and distribution," failing to further subdivide them into specific operational scenarios such as "live crossing of ±800 kV DC lines on the same tower" or "500 kV GIS expansion and uninterrupted busbar connection." This results in assessment templates that are too general and lack specificity, making it impossible to accurately identify and address unique risk combinations in specific scenarios.

[0003] Secondly, risk assessments are static and cannot respond to dynamic changes in real time. Assessment results from traditional methods are mostly fixed in paper or spreadsheet form, remaining largely unchanged throughout the entire operational cycle. Their assessment system fails to effectively couple with multi-source real-time data streams from SCADA (Supervisory Control and Data Acquisition), lightning location, micro-meteorology, personnel location, and smart wearable devices. This results in the inability to perceive and quantify the real-time impact of dynamic factors such as sudden increases in wind speed, sudden increases in load, and accumulated personnel fatigue on the risk level.

[0004] Furthermore, the risk information chain throughout the entire operation is fragmented. From site survey, plan preparation, work permitting, on-site implementation to final acceptance, risk assessment information at each stage is isolated, failing to form a coherent data flow. Subsequent stages cannot automatically inherit and make decisions based on the latest risk status of the previous stage, resulting in repetitive assessment work, redundant data, and a high risk of significant discrepancies between the actual on-site implementation and the original safety plan.

[0005] Furthermore, existing technological solutions have limited support capabilities and low levels of intelligent decision-making. Current mainstream solutions, such as work ticket risk assessment based on expert scoring methods (e.g., LEC method or risk matrix), are limited by the subjective differences in expert experience, and the assessment results cannot be updated once generated. BIM+GIS-based 3D scene simulation technology, due to its long model update cycle, high computing power requirements, and lack of deep integration with real-time data, is difficult to apply to temporary or urgent maintenance operations. Widely deployed IoT single-factor monitoring and early warning systems can only achieve independent threshold alarms; the fragmented information from various sensors cannot be integrated to generate a scene-level comprehensive risk index, let alone drive the automatic updating of work ticket risk levels and the intelligent upgrading or downgrading of control measures.

[0006] Chinese invention application No. 202311262974.2 discloses "A Dynamic Cyclic Rating System and Method for Power Grid Safety Risk Assessment". The method includes: S1, determining risk factors affecting power grid safety and collecting risk factor data; S2, quantitatively analyzing the risk factor data corresponding to each risk factor; S3, determining the weight of each risk factor; S4, constructing a comprehensive risk assessment model; and training, verifying and optimizing it; S5, acquiring risk factors of power operations and inputting them into the comprehensive risk characteristic assessment model, outputting the safety risk assessment rating of the power operations; S6, matching corresponding risk control measures from the risk measure library according to the safety risk assessment rating; S7, dynamically collecting risk factors and repeating steps S5-S6, continuously cyclically rating until the operation is completed. Summary of the Invention

[0007] To address the problem that existing risk assessment methods cannot update risk quantification results in real time with the progress of operations and dynamic on-site factors, this invention provides a continuous operation risk assessment system and method based on mainnet operation scenarios. The technical solution adopted by this invention is as follows: The first aspect of this invention provides a continuous operation risk assessment system based on mainnet operation scenarios, the system comprising: The mobile terminal is used to select or create a new work scenario template to initiate a risk assessment, generate a scenario identifier (SceneID), and receive and display real-time risk level and control measures prompts. Edge nodes, deployed at the work site, are used to collect multi-source sensor data in real time and perform standardized processing. They perform edge inference based on a lightweight dynamic Bayesian network model, calculate the current risk value R(t), and trigger reporting when a risk mutation event is detected. Cloud center, used for: The system receives risk mutation events reported by the edge nodes, runs a high-precision dynamic Bayesian network recalculation model based on the full data of the recent period to recalculate the risk, obtains a high-precision risk value R′(t), and uses a reinforcement learning strategy network to generate a control measure ActionList. The system receives packaged data from the entire process, updates the global model using FedAvg federated learning, and pushes the new weights to edge nodes via OTA.

[0008] The packaging and sending end is used to package the entire process data and upload it to the cloud center. The entire process data includes the scenario, risk factors, and actual events. The system confirms the execution of the control measures through the mobile terminal and writes them back to the work ticket system or the interlocking device, thereby realizing the closed-loop execution of risk assessment and control.

[0009] As a preferred embodiment, the mobile terminal is an explosion-proof tablet or a smart safety helmet, running an application that supports offline caching; the mobile terminal is also used to subscribe to real-time data streams from external systems via a RESTful interface.

[0010] As a preferred embodiment, the edge node is an industrial-grade edge gateway integrating multiple communication modules, used to collect sensor data packets at fixed intervals, perform Z-Score normalization on continuous variable data, and perform gridded encoding on personnel location data.

[0011] As a preferred embodiment, the edge nodes use the TensorRT-FP16 engine to run a lightweight dynamic Bayesian network model, while the cloud center uses PyTorch and GPUs to run a high-precision dynamic Bayesian network recalculation model.

[0012] As a preferred embodiment, the risk mutation event is triggered by the edge node when it detects that the absolute difference between the current risk value R(t) and the risk value R(t-Δt) of the previous period exceeds a preset threshold δ.

[0013] As a preferred embodiment, the reinforcement learning policy network takes the high-precision risk value R′(t), the operation stage, and the remaining working hours as state inputs and outputs action a. The set of actions a includes maintaining the status quo, upgrading control, and suspending the operation.

[0014] As a preferred option, the control measures ActionList include increasing the number of monitoring personnel, shortening the continuous operation time, or suspending the operation and triggering a lockout when R′(t) is greater than a first threshold and the environmental parameters exceed a second threshold.

[0015] As a preferred embodiment, the system also includes a scenario knowledge base for storing scenario templates, risk factors and their weights in JSON-LD format, and supports versioned hot updates.

[0016] As a preferred solution, the system uses a Kafka message queue partitioned by job ID to ensure the ordering of multi-source data for the same job.

[0017] A second aspect of the present invention provides a method for continuous operation risk assessment based on a mainnet operation scenario, the method comprising: Step S1: Select or create a new job scenario template via mobile terminal, instantiate the job scenario, generate SceneID, retrieve the corresponding risk factor list and initial weights, and subscribe to relevant real-time data streams; Step S2: Edge nodes periodically collect sensor data and perform standardization processing, input the data into a lightweight dynamic Bayesian network model for edge inference, calculate the current risk value R(t), and determine whether a risk mutation event has been triggered. Step S3: When a risk mutation event is triggered, the cloud center pulls the full amount of data from the recent period, runs a high-precision dynamic Bayesian network to recalculate the model to obtain a high-precision risk value R′(t), and uses a reinforcement learning strategy network to generate a control measure ActionList; Step S4: Push the control measures ActionList to the mobile terminal for display, and after confirmation, write it back to the work ticket system or send it to the interlocking device for execution to complete the closed loop; Step S5: After the job is completed, package the entire process data and upload it to the cloud center. Step S6: The cloud center receives the packaged full-process data, updates the global model using FedAvg federated learning, and pushes the new weights to the edge nodes via OTA.

[0018] Compared with the prior art, the beneficial effects of this invention are: This invention constructs a dynamic Bayesian network reasoning and reinforcement learning decision-making mechanism based on "edge-cloud collaboration," achieving end-to-end closed-loop automatic updates from multi-source data collection and risk value calculation to the generation of control measures. This completely eliminates the potential for accidents caused by a severe disconnect between assessment results and on-site conditions.

[0019] This invention constructs a hot-updateable scenario knowledge base, subdividing mainnet operations into more than a hundred specific scenarios that can be instantiated, which significantly improves the reusability of evaluation templates and greatly reduces evaluation preparation time and manpower costs.

[0020] This invention enables plug-and-play access to heterogeneous systems by employing an edge gateway with a built-in multi-protocol conversion module and organized using a Kafka data bus, which greatly shortens the system integration cycle and reduces additional hardware and licensing costs.

[0021] This invention achieves millisecond-level high-speed risk calculation and low-power operation by deploying a lightweight inference engine at edge nodes, ensuring that the on-site explosion-proof mobile terminal can smoothly and in real time display risk changes, meeting the practical requirements of on-the-spot operation and assessment.

[0022] This invention automatically connects the risk assessment system with work permits and intelligent interlocking devices, enabling one-click confirmation and automatic execution of control measures. This effectively reduces human intervention and significantly lowers approval time and the risk of human error. Attached Figure Description

[0023] Figure 1 This embodiment provides a flowchart of a continuous operation risk assessment method based on a mainnet operation scenario. Detailed Implementation The accompanying drawings are for illustrative purposes only and should not be construed as limiting the invention. It should be understood that the described embodiments are merely some, not all, of the embodiments of this application. 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 the embodiments of this application.

[0024] The terminology used in the embodiments of this application is for the purpose of describing particular embodiments only and is not intended to limit the embodiments of this application. The singular forms “a,” “the,” and “the” used in the embodiments of this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0025] In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims. In the description of this application, it should be understood that the terms "first," "second," "third," etc., are used only to distinguish similar objects and are not necessarily used to describe a specific order or sequence, nor should they be construed as indicating or implying relative importance. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0026] Furthermore, in the description of this application, unless otherwise stated, "multiple" means two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. The invention will be further described below with reference to the accompanying drawings and embodiments.

[0027] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0028] Example 1 This embodiment provides a continuous operation risk assessment system based on mainnet operation scenarios. The system includes: The mobile terminal is used to select or create a new work scenario template to initiate a risk assessment, generate a scenario identifier (SceneID), and receive and display real-time risk level and control measures prompts. Edge nodes, deployed at the work site, are used to collect multi-source sensor data in real time and perform standardized processing. They perform edge inference based on a lightweight dynamic Bayesian network model, calculate the current risk value R(t), and trigger reporting when a risk mutation event is detected. Cloud center, used for: The system receives risk mutation events reported by the edge nodes, runs a high-precision dynamic Bayesian network recalculation model based on the full data of the recent period to recalculate the risk, obtains a high-precision risk value R′(t), and uses a reinforcement learning strategy network to generate a control measure ActionList. The system receives packaged data from the entire process, updates the global model using FedAvg federated learning, and pushes the new weights to edge nodes via OTA.

[0029] The packaging and sending end is used to package the entire process data and upload it to the cloud center. The entire process data includes the scenario, risk factors, and actual events. The system confirms the execution of the control measures through the mobile terminal and writes them back to the work ticket system or the interlocking device, thereby realizing the closed-loop execution of risk assessment and control.

[0030] In one specific embodiment, the mobile terminal is an explosion-proof tablet or a smart safety helmet, running an application that supports offline caching; the mobile terminal is also used to subscribe to real-time data streams from external systems via a RESTful interface; Specifically, the mobile terminal runs a Flutter APP and displays real-time risk levels and control prompts.

[0031] In one specific embodiment, the edge node is an industrial-grade edge gateway integrating multiple communication modules, used to collect sensor data packets at fixed intervals, perform Z-Score normalization on continuous variable data, and perform gridded encoding on personnel location data; Specifically, the industrial-grade edge gateway integrates multiple communication modules such as 4G / 5G, LoRa, Wi-SUN, and BLE, and is responsible for local sensor access, protocol conversion, and lightweight model inference.

[0032] In one specific embodiment, the edge nodes use the TensorRT-FP16 engine to run a lightweight dynamic Bayesian network model, while the cloud center uses PyTorch and GPU to run a high-precision dynamic Bayesian network recalculation model. Specifically, the lightweight dynamic Bayesian network model is constructed with five-dimensional nodes of "human, machine, material, method, and environment", and the edge weights change with the operation stage; the TensorRT-FP16 engine runs at the edge and a single inference time is <120 ms.

[0033] In one specific embodiment, the risk mutation event is triggered by the edge node when it detects that the absolute difference between the current risk value R(t) and the risk value R(t-Δt) of the previous period exceeds a preset threshold δ.

[0034] In one specific embodiment, the reinforcement learning policy network takes the high-precision risk value R′(t), the job stage, and the remaining working hours as state inputs and outputs action a. The set of actions a includes maintaining the status quo, upgrading control, and suspending the job.

[0035] In one specific embodiment, the control measures ActionList include increasing the number of monitoring personnel, shortening the continuous operation time, or suspending the operation and triggering a lockout when R′(t) is greater than a first threshold and the environmental parameters exceed a second threshold.

[0036] In one specific embodiment, the cloud center also includes a scenario knowledge base for storing scenario templates, risk factors and their weights in JSON-LD format, and supporting versioned hot updates; Specifically, the cloud center adopts a Kubernetes elastic cluster and has a built-in scene knowledge base, deep learning risk model, and BIM / GIS services.

[0037] In one specific embodiment, the system uses a Kafka message queue partitioned by job ID to ensure the ordering of multi-source data for the same job.

[0038] Example 2 Please refer to Figure 1This embodiment provides a method for continuous operation risk assessment based on the mainnet operation scenario, using the system described in Embodiment 1. The method includes: Step S1: Select or create a new job scenario template via mobile terminal, instantiate the job scenario, generate SceneID, retrieve the corresponding risk factor list and initial weights, and subscribe to relevant real-time data streams; Specifically, in step S1 (at time t0): The person in charge of the operation selects or creates a "operation scenario template" on the mobile terminal, and the system automatically generates a SceneID; calls the Scene-KB API to pull the risk factor list F={f1,f2,…,fn} and initial weight W0={w1,w2,…,wn} for the scenario; and subscribes to real-time data streams from SCADA, lightning positioning, micro-meteorology and other systems through the RESTful interface, and returns the subscription handle H.

[0039] Step S2: Edge nodes periodically collect sensor data and perform standardization processing, input the data into a lightweight dynamic Bayesian network model for edge inference, calculate the current risk value R(t), and determine whether a risk mutation event has been triggered. Specifically, the edge gateway collects sensor data packets D={d1,d2,…,dm} every 5 seconds; data standardization: Z-scores are applied to continuous variables such as wind speed, temperature, and current, and gridded encoding is applied to personnel locations; D is input into the DBN model, and the hidden state probability is updated using the Forward-Backward algorithm, outputting the current risk value R(t); if |R(t)-R(t-Δt)|>δ (default δ=0.15), a "risk mutation event" EventE is triggered and immediately pushed to the mobile terminal and the cloud.

[0040] Step S3: When a risk mutation event is triggered, the cloud center pulls the full amount of data from the recent period, runs a high-precision dynamic Bayesian network to recalculate the model to obtain a high-precision risk value R′(t), and uses a reinforcement learning strategy network to generate a control measure ActionList; Specifically, the cloud center retrieves nearly 30 minutes of the full data for the task, re-runs the high-precision dynamic Bayesian network to recalculate the model (hidden nodes × 10), and obtains R′(t); a reinforcement learning policy network π(a|s) is adopted, where the state s = {R′(t), task stage, remaining time}, and the action a ∈ {maintain, escalate control, suspend task}; the output control measure list ActionList includes: Examples of upgrade measures: adding a second guardian, shortening the continuous duration of a single climbing session to ≤90 min; Suspension conditions: R′(t)≥0.85 and wind speed>15 m / s.

[0041] Step S4: Push the control measures ActionList to the mobile terminal for display, and after confirmation, write it back to the work ticket system or send it to the interlocking device for execution to complete the closed loop; Specifically, the mobile terminal displays the R′(t) level in real time using red / orange / yellow / green lights at the UI layer; after clicking "Confirm Execution", ActionList is written to the work ticket system via MQTT, automatically generating a supplementary safety measures column; if the action is "Pause Work", it is sent to the intelligent interlocking device via IEC 61850 GOOSE message to realize the forced interlocking of the circuit breaker / grounding switch.

[0042] Step S5: After the job is completed, package the entire process data and upload it to the cloud center. Step S6: The cloud center receives the packaged full-process data, updates the global model using FedAvg federated learning, and pushes the new weights to the edge nodes via OTA; Specifically, FedAvg federated learning is used to update the global model, generating new weights Wnew every 24 hours; these weights are pushed to edge nodes via OTA to achieve "learning on the go" without leaking the original data.

[0043] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A continuous operation risk assessment system based on mainnet operation scenarios, characterized in that, The system includes: The mobile terminal is used to select or create a new work scenario template to initiate a risk assessment, generate a scenario identifier (SceneID), and receive and display real-time risk level and control measures prompts. Edge nodes, deployed at the work site, are used to collect multi-source sensor data in real time and perform standardized processing. They perform edge inference based on a lightweight dynamic Bayesian network model, calculate the current risk value R(t), and trigger reporting when a risk mutation event is detected. Cloud center, used for: The system receives risk mutation events reported by the edge nodes, runs a high-precision dynamic Bayesian network recalculation model based on the full data of the recent period to recalculate the risk, obtains a high-precision risk value R′(t), and uses a reinforcement learning strategy network to generate a control measure ActionList. The system receives packaged data from the entire process, updates the global model using FedAvg federated learning, and pushes the new weights to edge nodes via OTA. The packaging and sending end is used to package the entire process data and upload it to the cloud center, where the entire process data includes the scenario, risk factors, and actual events; The system confirms the execution of the control measures through the mobile terminal and writes them back to the work ticket system or the interlocking device, thereby realizing the closed-loop execution of risk assessment and control.

2. The continuous operation risk assessment system based on mainnet operation scenarios according to claim 1, characterized in that, The mobile terminal is an explosion-proof tablet or a smart safety helmet, running an application that supports offline caching; the mobile terminal is also used to subscribe to real-time data streams from external systems via a RESTful interface.

3. The continuous operation risk assessment system based on mainnet operation scenarios according to claim 1, characterized in that, The edge node is an industrial-grade edge gateway integrating multiple communication modules, used to collect sensor data packets at fixed intervals, perform Z-Score normalization on continuous variable data, and perform gridded encoding on personnel location data.

4. The continuous operation risk assessment system based on mainnet operation scenarios according to claim 1, characterized in that, The edge nodes use the TensorRT-FP16 engine to run lightweight dynamic Bayesian network models, while the cloud center uses PyTorch and GPUs to run high-precision dynamic Bayesian network recalculation models.

5. The continuous operation risk assessment system based on mainnet operation scenarios according to claim 1, characterized in that, The risk mutation event is triggered by the edge node when it detects that the absolute difference between the current risk value R(t) and the risk value R(t-Δt) of the previous period exceeds a preset threshold δ.

6. The continuous operation risk assessment system based on mainnet operation scenarios according to claim 1, characterized in that, The reinforcement learning policy network takes the high-precision risk value R′(t), the operation stage, and the remaining working hours as state inputs and outputs action a. The set of actions a includes maintaining the status quo, upgrading control, and suspending the operation.

7. The continuous operation risk assessment system based on mainnet operation scenarios according to claim 6, characterized in that, The control measures ActionList include increasing the number of monitoring personnel, shortening the continuous operation time, or suspending the operation and triggering a lockout when R′(t) is greater than the first threshold and the environmental parameters exceed the second threshold.

8. The continuous operation risk assessment system based on mainnet operation scenarios according to claim 1, characterized in that, The cloud center also includes a scenario knowledge base, which stores scenario templates, risk factors and their weights in JSON-LD format and supports versioned hot updates.

9. A continuous operation risk assessment system based on a mainnet operation scenario as described in claim 1, characterized in that, The system uses a Kafka message queue partitioned by job ID to ensure the orderliness of multi-source data for the same job.

10. A method for continuous operation risk assessment based on a mainnet operation scenario, using the system described in any one of claims 1-9, characterized in that, The method includes: Step S1: Select or create a new job scenario template via mobile terminal, instantiate the job scenario, generate SceneID, retrieve the corresponding risk factor list and initial weights, and subscribe to relevant real-time data streams; Step S2: Edge nodes periodically collect sensor data and perform standardization processing, input the data into a lightweight dynamic Bayesian network model for edge inference, calculate the current risk value R(t), and determine whether a risk mutation event has been triggered. Step S3: When a risk mutation event is triggered, the cloud center pulls the full amount of data from the recent period, runs a high-precision dynamic Bayesian network to recalculate the model to obtain a high-precision risk value R′(t), and uses a reinforcement learning strategy network to generate a control measure ActionList; Step S4: Push the control measures ActionList to the mobile terminal for display, and after confirmation, write it back to the work ticket system or send it to the interlocking device for execution to complete the closed loop; Step S5: After the job is completed, package the entire process data and upload it to the cloud center. Step S6: The cloud center receives the packaged full-process data, updates the global model using FedAvg federated learning, and pushes the new weights to the edge nodes via OTA.

Images