Power grid asset maintenance decision-making method, device and equipment based on unmanned aerial vehicle inspection, and medium

Abstract

The invention relates to the technical field of artificial intelligence, and discloses a power grid asset maintenance decision-making method based on unmanned aerial vehicle inspection, and the method comprises the steps: obtaining visual image data of a target asset and weather forecast data of an environment where the target asset is located; performing asset health assessment according to the visual image data to obtain health state data of the target asset; performing meteorological risk assessment according to the meteorological forecast data to obtain environmental risk data of the target assets; performing data fusion on the health state data and the environmental risk data to obtain global collaborative risk data; generating a target maintenance instruction for the target asset based on the global collaborative risk data through the maintenance decision agent; the reward function of the maintenance decision agent is obtained according to the influence of a maintenance instruction generated by the maintenance decision agent on the target assets and the target power grid in the specified operation and maintenance dimension in the training process. The method has the beneficial effects that the pertinence of the target maintenance instruction is effectively improved, and the long-term operation capability of the target assets and the target power grid is improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method, device, equipment and medium for power grid asset maintenance decision-making based on drone inspection. Background Technology

[0002] As a crucial part of national infrastructure, the power grid bears the responsibility of ensuring energy supply and supporting socio-economic activities. Therefore, the safe operation of the power grid is vital to production and daily life. The power grid includes various types of assets, such as transmission lines, towers, and insulators. The operational status of these assets is affected by multiple factors. On the one hand, as the power grid assets themselves age, they develop inherent defects or equipment aging problems, thus impacting the power grid from within. On the other hand, power grid assets are also affected by the external environment, which can directly cause equipment failures or damage that accumulates over time. Therefore, to ensure the normal operation of power grid assets, timely maintenance is usually necessary to improve the reliability of both the assets and the power grid.

[0003] In related technologies, the foresight of technical solutions for maintaining power grid assets still needs to be improved. Summary of the Invention

[0004] This application provides a method, device, equipment, and medium for power grid asset maintenance decision-making based on UAV inspection. Based on the multi-dimensional impact of maintenance instructions on target assets and target power grids, the intelligent agent executing maintenance decisions is trained to generate target maintenance instructions by combining the multi-modal information of target assets. This effectively improves the relevance of target maintenance instructions and enhances the long-term operational capability of target assets and target power grids.

[0005] To achieve the above objectives, the main technical solutions adopted in this application include: In a first aspect, embodiments of this application provide a power grid asset maintenance decision-making method based on unmanned aerial vehicle (UAV) inspection, the method comprising: The system acquires visual image data of the target asset and weather forecast data of the environment in which the target asset is located; wherein the target asset belongs to the target power grid, and the visual image data is obtained through drone inspection. Based on the visual image data, an asset health assessment is performed on the target asset to obtain the health status data of the target asset; based on the weather forecast data, a meteorological risk assessment is performed on the target asset to obtain the environmental risk data of the target asset. The health status data and the environmental risk data are fused to obtain the global collaborative risk data of the target asset; The global collaborative risk data is input into a reinforcement learning-based maintenance decision agent, which then generates target maintenance instructions for the target asset based on the global collaborative risk data. The reward function of the maintenance decision agent is obtained based on the impact of the maintenance instructions generated by the maintenance decision agent during training on the target asset and the target power grid in a specified operation and maintenance dimension.

[0006] The power grid asset maintenance decision-making method based on UAV inspection proposed in this application acquires visual image data of the target asset and meteorological forecast data of its surrounding environment. It evaluates the target asset's health status based on the visual image data and its environmental risk based on the meteorological forecast data. The health status data and environmental risk data are fused, and maintenance instructions are generated based on the maintenance decision-making agent and the fused global collaborative risk data. Compared with related technologies, the reward function of the maintenance decision-making agent in this application is obtained based on the impact of maintenance instructions on the target asset and the target power grid in a specified operation and maintenance dimension during training. The maintenance decision-making agent is trained based on this reward function, incentivizing it to assess the long-term operational risks of the target asset and the target power grid after executing maintenance instructions. This generates globally optimal maintenance instructions that consider forward-looking operational risks, improving the long-term operational capabilities of the target asset and the target power grid.

[0007] Optionally, the maintenance instructions generated by the maintenance decision-making agent during training are recorded as training maintenance instructions; the reward function is obtained in the following manner: Service loss is assessed based on the load shedding status of the target power grid when the training and maintenance instructions are executed, resulting in a service impact penalty. A health depletion assessment is performed on the target asset based on its operating condition after the execution of the training and maintenance instructions, resulting in a lifecycle penalty item. Determine the operation safety penalty item; wherein, the operation safety penalty item includes an indicator function, the indicator function having a first value when the training and maintenance instruction meets the safety procedure requirements, and a second value when the training and maintenance instruction does not meet the safety procedure requirements, wherein the first value is less than the second value; Based on the simulation evaluation of the target power grid's response capability to equipment failures before and after executing the training and maintenance instructions, a long-term resilience penalty term is obtained; The reward function is obtained by performing a risk fusion assessment on the service impact penalty item, the lifecycle penalty item, the operational security penalty item, and the long-term resilience penalty item.

[0008] Optionally, the load shedding status includes the load shedding power and the load shedding time; the service loss assessment based on the load shedding status of the target power grid when the training and maintenance instructions are executed, to obtain a service impact penalty, includes: Based on the load shedding power and the load shedding time, the service value of the execution process of the training maintenance command is evaluated to obtain the service impact penalty item used to suppress the negative impact of the target maintenance command on the service capability of the target power grid.

[0009] Optionally, the step of assessing health depletion based on the operating condition of the target asset after executing the training and maintenance instructions to obtain a lifecycle penalty term includes: The operating conditions of the target asset after executing the training and maintenance instructions are simulated and predicted, and the health loss is calculated based on the simulation and prediction results to obtain the life cycle penalty item used to suppress the negative impact of the target maintenance instructions on the lifespan of the target asset.

[0010] Optionally, the step of simulating and evaluating the response capability of the target power grid to equipment faults before and after executing the training and maintenance instructions to obtain a long-term resilience penalty term includes: Before executing the training and maintenance instructions, the response capability of the target power grid to a preset fault is simulated to obtain a first response capability; After executing the training and maintenance instructions, the response capability of the target power grid to the preset fault is simulated to obtain a second response capability; A power grid resilience assessment is performed based on the first response capability and the second response capability to obtain a long-term resilience penalty term used to suppress the negative impact of the target maintenance command on the fault response capability of the target power grid.

[0011] Optionally, the risk fusion assessment of the service impact penalty, the lifecycle penalty, the operational security penalty, and the long-term resilience penalty to obtain the reward function includes: The state utility of each power grid asset in the target power grid is evaluated to obtain power grid stability reward items; Based on the operating status of the target asset after executing the training and maintenance instructions, risk prediction is performed on preset faults to obtain a forward-looking prevention reward item; The cost of the training and maintenance instructions is calculated to obtain a maintenance cost penalty item; The reward function is obtained by performing a risk fusion assessment on the service impact penalty, the life cycle penalty, the operational safety penalty, the long-term resilience penalty, the power grid stability reward, the forward prevention reward, and the maintenance cost penalty.

[0012] Optionally, the asset health assessment is performed by a hybrid identification agent; the hybrid identification agent is obtained through the following means: Construct an initial identification agent; wherein the initial identification agent includes an initial asset identification component, an initial defect identification component, and an initial result fusion component; The initial asset identification component is used to identify power grid assets in the training image data of the image training dataset to obtain the asset type identification result of the training image data; the parameters of the initial asset identification component are adjusted according to the asset type identification result and the asset type label corresponding to the training image data to obtain the target asset identification component. The initial defect identification component is used to identify defect features in the training image data to obtain defect type identification results for the training image data; the parameters of the initial defect identification component are adjusted according to the defect type identification results and the defect type labels corresponding to the training image data to obtain the target defect identification component. The parameters of the target asset identification component and the target defect identification component are locked; the initial result fusion component performs adaptive weight allocation on the asset type identification result and the defect type identification result, and fuses the asset type identification result and the defect type identification result according to the weight allocation result to obtain a hybrid identification result; the parameters of the initial result fusion component are adjusted according to the hybrid identification result and the image annotations corresponding to the training image data to obtain the target result fusion component; The hybrid recognition agent is constructed based on the target asset recognition component, the target defect recognition component, and the target result fusion component.

[0013] Secondly, embodiments of this application provide a power grid asset maintenance decision-making device based on unmanned aerial vehicle (UAV) inspection, the device comprising: The data acquisition module is used to acquire visual image data of the target asset and weather forecast data of the environment in which the target asset is located; wherein, the target asset belongs to the target power grid, and the visual image data is obtained through UAV inspection; The data assessment module is used to perform an asset health assessment on the target asset based on the visual image data to obtain the health status data of the target asset; and to perform a meteorological risk assessment on the target asset based on the meteorological forecast data to obtain the environmental risk data of the target asset. The data fusion module is used to fuse the health status data and the environmental risk data to obtain the global collaborative risk data of the target asset. The risk decision module is used to input the global collaborative risk data into a reinforcement learning-based maintenance decision agent, and the maintenance decision agent generates target maintenance instructions for the target asset based on the global collaborative risk data; wherein, the reward function of the maintenance decision agent is obtained according to the impact of the maintenance instructions generated by the maintenance decision agent during the training process on the target asset and the target power grid in a specified operation and maintenance dimension.

[0014] Thirdly, embodiments of this application provide a computer device, including: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the method described in any of the above embodiments.

[0015] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer instructions, which are used to cause a computer to perform the method described in any one of the above embodiments.

[0016] Fifthly, embodiments of this application provide a computer program product, including computer instructions, which are used to cause a computer to perform the method described in any of the above embodiments. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the specific embodiments of this application or the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating the steps of the power grid asset maintenance decision-making method based on unmanned aerial vehicle (UAV) inspection provided in this application embodiment; Figure 2 This is a flowchart of the power grid asset maintenance decision-making method based on UAV inspection in the embodiments of this application; Figure 3 This is a flowchart illustrating the steps involved in obtaining the reward function in an embodiment of this application. Figure 4 This is a flowchart illustrating the steps involved in obtaining the long-term toughness penalty term in an embodiment of this application. Figure 5 This is a flowchart illustrating the steps involved in obtaining the reward function in an embodiment of this application. Figure 6 This is a flowchart illustrating the steps involved in obtaining the hybrid recognition agent in the embodiments of this application. Figure 7This is a flowchart illustrating the training of the hybrid recognition agent in an embodiment of this application; Figure 8 A block diagram of a power grid asset maintenance decision-making device based on unmanned aerial vehicle (UAV) inspection provided in this application embodiment; Figure 9 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] As a crucial part of national infrastructure, the power grid bears the responsibility of ensuring energy supply and supporting socio-economic activities. Its safe operation is vital to production and daily life. The power grid includes various types of assets, such as transmission lines, towers, and insulators. The operational status of these assets is affected by multiple factors. On the one hand, the assets themselves develop inherent defects or age problems as they age, impacting the grid internally. On the other hand, they are also affected by the external environment, directly causing equipment failures or cumulative damage over time. Therefore, to ensure the normal operation of power grid assets, timely maintenance is usually necessary to improve the reliability of both the assets and the grid. However, the foresight of technical solutions for maintaining power grid assets still needs improvement.

[0021] To address the aforementioned issues, this application provides a power grid asset maintenance decision-making method based on unmanned aerial vehicle (UAV) inspection. The method involves acquiring visual image data of the target asset and meteorological forecast data of its surrounding environment; conducting an asset health assessment based on the visual image data to obtain the target asset's health status data; conducting a meteorological risk assessment based on the meteorological forecast data to obtain the target asset's environmental risk data; fusing the health status data and environmental risk data to obtain global collaborative risk data for the target asset; and generating target maintenance instructions for the target asset based on the global collaborative risk data using a maintenance decision-making agent. The reward function of the maintenance decision-making agent is obtained based on the impact of the maintenance instructions generated by the agent during training on the target asset and the target power grid in a specified operation and maintenance dimension.

[0022] The power grid asset maintenance decision-making method based on UAV inspection provided in this application acquires visual image data of the target asset and meteorological forecast data of the surrounding environment. It evaluates the target asset's health status data based on the visual image data and evaluates the target asset's environmental risk data based on the meteorological forecast data. It then fuses the health status data and environmental risk data and generates maintenance instructions based on the maintenance decision-making agent and the fused global collaborative risk data.

[0023] Compared with related technologies, the reward function of the maintenance decision agent in this application is obtained based on the impact of maintenance instructions on the target asset and the target power grid in a specified operation and maintenance dimension during the training process. The maintenance decision agent is trained based on the above reward function, which incentivizes the maintenance decision agent to assess the long-term operational risks of the target asset and the target power grid after executing maintenance instructions, thereby generating globally optimal maintenance instructions that take into account forward-looking operational risks and improving the long-term operational capabilities of the target asset and the target power grid.

[0024] The power grid asset maintenance decision-making method based on UAV inspection provided in this manual can be applied to fault identification and maintenance instruction generation for power grid assets, enabling timely maintenance. These power grid assets may include, but are not limited to, transmission lines, towers, and insulators. It is understood that, with adaptive modifications, this UAV-based power grid asset maintenance decision-making method can also be applied to generate corresponding maintenance instructions for other equipment besides power grid assets, for repair and maintenance of those other devices.

[0025] According to an embodiment of this application, a method for power grid asset maintenance decision-making based on UAV inspection is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0026] This embodiment provides a power grid asset maintenance decision-making method based on unmanned aerial vehicle (UAV) inspection, which can be applied to the aforementioned power grid assets. (Refer to...) Figure 1 As shown, the method includes: S100. Acquire visual image data of the target asset and weather forecast data of the environment in which the target asset is located; wherein, the target asset belongs to the target power grid, and the visual image data is obtained through drone inspection.

[0027] S200. Conduct an asset health assessment of the target asset based on visual image data to obtain the health status data of the target asset; conduct a meteorological risk assessment of the target asset based on meteorological forecast data to obtain the environmental risk data of the target asset.

[0028] S300. Data fusion of health status data and environmental risk data to obtain global collaborative risk data for the target asset.

[0029] S400. Input global collaborative risk data into a maintenance decision agent based on reinforcement learning, and generate target maintenance instructions for the target asset based on the global collaborative risk data; wherein, the reward function of the maintenance decision agent is obtained according to the impact of the maintenance instructions generated by the maintenance decision agent during the training process on the target asset and the target power grid in a specified operation and maintenance dimension.

[0030] Specifically, refer to Figure 2 As shown, any power grid asset within the target power grid is identified as the target asset. Visual image data of the target asset and weather forecast data of the environment in which the target asset is located are acquired. For example, the visual image data can be image data obtained by capturing images of the target asset's appearance. The visual image data can be acquired through methods such as drone patrols, ground personnel inspections, or satellite remote sensing. The weather forecast data can represent the weather changes in the environment of the target asset over a certain period of time. Weather forecast data can be acquired through on-site weather stations, mobile weather equipment, weather service interfaces, or satellite remote sensing data.

[0031] Further, image processing and feature extraction are performed on the visual image data to obtain asset image features representing the type of target asset and defect image features representing component defects appearing in the target asset. Based on the asset image features, power grid asset identification is performed on the target asset to determine its asset type. Based on the defect image features, defect feature identification is performed on the target asset to determine the type and severity of defects appearing in the target asset. Corresponding descriptive text is generated based on the asset type, defect type, and severity of the target asset. The obtained descriptive text is parsed to extract structured information, obtaining the intrinsic triplet information of the target asset, including its asset type, defect type, and severity. Vector mapping is performed on the intrinsic triplet information to obtain vector-form health status data, used to represent the internal health status of the target asset. In some embodiments, asset health assessment can be performed by a hybrid identification agent.

[0032] Furthermore, weather forecast data can be text data. Text processing and weather change identification are performed on the weather forecast data to extract the weather conditions contained within it. Based on pre-introduced knowledge of physical causality, a failure mechanism analysis is conducted on the impact of these weather conditions on the target asset to determine the possible failure modes and their risk probabilities under these weather conditions. Vector mapping is then applied to the failure modes and their risk probabilities to obtain vector-form environmental risk data, which represents the external risk status of the target asset. It should be noted that the knowledge of physical causality can refer to the causal relationship between weather conditions and power grid asset failures caused by physical changes or interactions. This is used to assess the impact of weather conditions on power grid assets from a physical perspective, improving the accuracy and scientific rigor of the environmental risk data. This, in turn, provides clear decision-making basis for the final maintenance instructions, facilitating the understanding of maintenance personnel.

[0033] In some embodiments, environmental risk data can be obtained through a causal reasoning agent, which can be pre-trained offline using a text training dataset. The text training dataset can include training meteorological data and corresponding fault mechanism data. The training meteorological data can be pre-defined specific meteorological conditions, including but not limited to strong winds, icing, high temperatures, heavy fog, or a time-series combination of multiple meteorological conditions. The fault mechanism data can be the specific mechanisms by which any meteorological condition in the training meteorological data affects power grid assets. These specific mechanisms can lead to specific types of fault risks in the power grid assets, including but not limited to conductor galloping, icing, and flashover. The process of training the causal reasoning agent can include: semantically parsing the fault mechanism data using the causal reasoning agent to obtain risk association information between the training meteorological data and the fault risks of power grid assets; iteratively adjusting the parameters of the causal reasoning agent based on the risk association information, and through multiple iterations, enabling the causal reasoning agent to learn the fault mechanism relationship between meteorological conditions and fault risks.

[0034] Furthermore, health status data and environmental risk data are fused to obtain global collaborative risk data for the target asset. This unifies and integrates the internal health status and external risk status of the target asset within the target power grid, thereby representing the global state of the target asset through a unified state representation, providing a complete and accurate data foundation for subsequent maintenance decisions. For example, when both health status data and environmental risk data are in vector form, the data fusion method can be vector concatenation, which can be represented as follows: in, This represents the overall collaborative risk data of the target asset; This is a vector concatenation operation; , and These are the health status data for each power grid asset; , and These are environmental risk data for each power grid asset; This represents the total number of power grid assets in the target power grid. It is understandable that by extracting features from visual image data and weather forecast data separately, and then fusing the extracted features as the data basis for maintenance decisions, multiple factors affecting the operational status of the target assets are considered, thereby improving the targeting of target maintenance instructions, optimizing the maintenance effect of target maintenance instructions, and ultimately enhancing the reliability of the target power grid.

[0035] Furthermore, global collaborative risk data is input into a reinforcement learning-based maintenance decision-making agent. This agent uses the global collaborative risk data as a data foundation to make maintenance decisions and generate target maintenance instructions for the target asset. The target maintenance instructions can be a task list including one or more maintenance tasks. By structuring the target maintenance instructions, the maintenance actions corresponding to each maintenance task can be obtained. Based on the execution priority and decision criteria of each maintenance task, the maintenance actions are planned to obtain structured maintenance instructions, which are then sent to the target power grid's operation and maintenance management system for execution.

[0036] It should be noted that the reward function of the maintenance decision-making agent is derived based on the impact of maintenance instructions generated by the agent during training on the target assets and the target power grid in a specified operational dimension. This specified operational dimension can include the immediate impact of the maintenance instructions generated during training on the target assets and the target power grid during execution, as well as the long-term impact of the instructions on the operational status of the target assets and the target power grid over a certain period after execution. Training the maintenance decision-making agent based on this reward function enables it to consider the multi-dimensional impact of maintenance instructions on the target assets and the target power grid, going beyond simply predicting fault risks. This allows for the generation of target maintenance instructions through a risk-sharing decision-making approach, effectively reducing the risk pressure borne by either party during operation and maintenance. This achieves intelligent management of the target power grid and its assets, thereby improving the long-term operational capabilities of the target assets and the target power grid.

[0037] The power grid asset maintenance decision-making method based on UAV inspection provided in this embodiment acquires visual image data of the target asset and meteorological forecast data of the surrounding environment. It evaluates the target asset's health status data based on the visual image data and evaluates the target asset's environmental risk data based on the meteorological forecast data. It then fuses the health status data and environmental risk data and generates maintenance instructions based on the maintenance decision-making agent and the fused global collaborative risk data.

[0038] Compared with related technologies, the reward function of the maintenance decision agent in this application is obtained based on the impact of maintenance instructions on the target asset and the target power grid in a specified operation and maintenance dimension during the training process. The maintenance decision agent is trained based on the above reward function, which incentivizes the maintenance decision agent to assess the long-term operational risks of the target asset and the target power grid after executing maintenance instructions, thereby generating globally optimal maintenance instructions that take into account forward-looking operational risks and improving the long-term operational capabilities of the target asset and the target power grid.

[0039] Reference Figure 3 As shown, in one embodiment of this application, the maintenance instructions generated by the maintenance decision agent during training are denoted as training maintenance instructions; the reward function is obtained in the following manner: S410. Based on the load shedding situation of the target power grid when executing training and maintenance instructions, perform service loss assessment and obtain service impact penalty items.

[0040] S420. Based on the operating conditions of the target asset after executing training and maintenance instructions, a health depletion assessment is performed to obtain a lifecycle penalty item.

[0041] S430. Determine the operation safety penalty item; wherein, the operation safety penalty item includes an instruction function, the instruction function having a first value when the training and maintenance instructions meet the safety procedure requirements, and a second value when the training and maintenance instructions do not meet the safety procedure requirements, and the first value is less than the second value.

[0042] S440. Based on the simulation evaluation of the target power grid's response capability to equipment failures before and after the execution of training and maintenance instructions, a long-term resilience penalty term is obtained.

[0043] S450. A risk fusion assessment is performed on the service impact penalty, lifecycle penalty, operational safety penalty, and long-term resilience penalty to obtain the reward function.

[0044] Specifically, the training process of the maintenance decision-making agent can be conducted in a simulation environment, which can be constructed based on historical or simulated data of the target power grid to simulate the fault risk situation of the target power grid in a real-world scenario. During the training process, the maintenance decision-making agent makes maintenance decisions using the power grid asset maintenance decision-making method based on UAV inspection provided in this specification. In each training round, corresponding training maintenance instructions are generated, and rewards are calculated for the training maintenance instructions through a reward function. Based on the reward results of the current training round, the maintenance decision-making agent is guided to adjust parameters, enabling the maintenance decision-making agent to gradually master a maintenance decision-making method that proactively considers the multi-dimensional operational impact of maintenance instructions, thereby improving the intelligence of the maintenance decision-making process.

[0045] It should be noted that maintenance instructions may pose risks to the target power grid and target assets across multiple operational dimensions. Executing maintenance instructions requires, on the one hand, partial load shedding of the target power grid to ensure its stable operation and the safe execution of the maintenance instructions; this load shedding impacts the load service capacity of the target power grid. On the other hand, the execution of maintenance instructions may cause target assets to operate under overload conditions, thereby exacerbating the deterioration of their health. After executing maintenance instructions, the topology of the target power grid may change, which may reduce the grid's fault response capability and affect its long-term system resilience. Furthermore, maintenance instructions must comply with safety procedures to reduce the probability of safety incidents.

[0046] For the reasons mentioned above, the reward function should include at least a service impact penalty, a lifecycle penalty, an operational safety penalty, and a long-term resilience penalty. The service impact penalty can be derived from the impact of training and maintenance instructions on the load service capacity of the target power grid. This is achieved by assessing service losses during load shedding in the target power grid during the execution of training and maintenance instructions to determine the negative impact of the instruction's execution on the power supply. The lifecycle penalty can be derived from the impact of training and maintenance instructions on the health status of any target asset in the target power grid to determine the additional health loss caused by the instructions. The operational safety penalty can be derived from the operational safety of the training and maintenance instructions to ensure their safe execution. The long-term resilience penalty can be derived from the impact of training and maintenance instructions on the long-term resilience of the target power grid to determine the negative impact of the instructions on the grid's stability and self-recovery capabilities.

[0047] In some embodiments, the operational safety penalty item includes a constant penalty value and an indicator function, wherein the indicator function is used to determine whether the training and maintenance instructions meet the safety procedure requirements. The indicator function takes a first value when the training and maintenance instructions meet the safety procedure requirements, and a second value when the training and maintenance instructions do not meet the safety procedure requirements. The first value is less than the second value, to ensure that the operational safety penalty item when the training and maintenance instructions do not meet the safety procedure requirements is greater than the operational safety penalty item when the training and maintenance instructions meet the safety procedure requirements. For example, the operational safety penalty item can be expressed as: in, This indicates operational safety penalties. To maintain instructions; The penalty value is a constant. This is a safety procedure check function that takes a true value if the maintenance instruction violates the provisions of the digital safety procedure library, and a false value otherwise. For indicator functions, in If the result is true, the second value (1) is used; otherwise, the first value (0) is used. It should be noted that this is a constant penalty value. It can be a constant value that is larger than other penalty and reward terms in the reward function, thereby completely excluding training and maintenance instructions that do not meet the requirements of the safety procedures and improving the operational safety of the maintenance process.

[0048] Furthermore, the above penalty terms are fused to obtain a composite risk penalty term representing the long-term operational impact of training maintenance instructions on the target power grid and target assets, which is then used to derive the reward function. For example, the composite risk penalty term can be expressed as: in, This indicates a compound risk penalty item; , , and These are the fusion weights for each penalty term, and all fusion weights are non-negative and satisfy the following conditions: ; The penalty items for the impact of normalized services; This is the normalized lifecycle penalty term; These are the normalized operational safety penalty items; This represents the normalized long-term resilience penalty term. Each penalty term can be normalized using the following formula: in, The minimum historical simulation value of the i-th penalty term; This represents the historical simulation maximum value of the i-th penalty item. It is understood that the fusion weight of each penalty item can be determined based on actual operation and maintenance strategies, expert experience, and simulation verification. For example, if the actual operation and maintenance strategy places greater emphasis on user impact, the fusion weight corresponding to the service impact penalty item can be appropriately increased.

[0049] As one embodiment of this application, the load shedding situation includes the load shedding power and the load shedding time; based on the load shedding situation of the target power grid when executing training and maintenance instructions, a service loss assessment is performed to obtain a service impact penalty, including: S412. Based on the load shedding power and load shedding time, perform a service value assessment on the execution process of the training maintenance command to obtain a service impact penalty item used to suppress the negative impact of the target maintenance command on the service capability of the target power grid.

[0050] Specifically, when executing training maintenance instructions, the target power grid may need to partially disconnect its load to ensure stable operation and the safe execution of the maintenance instructions. Load disconnection can include the load disconnection power and the load disconnection time. The load disconnection power can be the load power that must be disconnected when executing the training maintenance instructions, which can be obtained through power flow calculations; the load disconnection time can be the expected duration of the training maintenance instructions. Based on the load disconnection power and load disconnection time, a service value assessment is performed based on the economic loss caused by a unit load disconnection power per unit time, resulting in a service impact penalty term. This penalty is used to mitigate the negative impact of the target maintenance instructions on the service capacity of the target power grid.

[0051] For example, the service impact penalty can be represented as: in, This indicates the penalty items that affect the service; Value per unit of load shedding represents the economic loss caused by a unit of load shedding power per unit of time. To execute maintenance instructions The required load shedding power; To execute maintenance instructions The required load removal time.

[0052] As one embodiment of this application, a health depletion assessment is performed based on the operating condition of the target asset after executing training and maintenance instructions, resulting in a lifecycle penalty term, including: S422. Simulate and predict the operating conditions of the target asset after executing the training and maintenance instructions, and calculate the health loss based on the simulation and prediction results to obtain a lifecycle penalty item used to suppress the negative impact of the target maintenance instructions on the lifespan of the target asset.

[0053] Specifically, when executing training maintenance instructions, the operating conditions of the target asset may deviate from normal, negatively impacting its lifespan. Therefore, during the training process, after receiving the training maintenance instructions, a simulation environment is used to predict the operating conditions of the target asset after executing the instructions. Based on the predicted operating conditions, health depletion is calculated for the target asset, determining the additional aging degree under these conditions and deriving a lifespan penalty term to mitigate the negative impact of the target maintenance instructions on the asset's lifespan.

[0054] For example, the lifecycle penalty term can be represented as: in, Indicates a lifecycle penalty item; Let k be the replacement cost of the power grid asset. To execute maintenance instructions The health loss caused to power grid asset k can be obtained through the built-in State of Health (SOH) decay model in the simulation environment, which can be expressed as: in, To execute maintenance instructions The temperature of post-grid asset k, The normal temperature of power grid asset k; The duration of the temperature anomaly in power grid asset k; The rated design life of power grid asset k; The equivalent aging acceleration factor for power grid assets at temperature T can be expressed as: .

[0055] Reference Figure 4 As shown, in one embodiment of this application, a long-term resilience penalty term is obtained by simulating and evaluating the response capability of the target power grid to equipment faults before and after the execution of training and maintenance instructions, including: S442. Before executing the training and maintenance instructions, simulate the response capability of the target power grid to the preset fault to obtain the first response capability.

[0056] S444. After executing the training and maintenance instructions, the response capability of the target power grid to the preset fault is simulated to obtain the second response capability.

[0057] S446. Based on the first and second response capabilities, conduct a power grid resilience assessment to obtain a long-term resilience penalty term used to mitigate the negative impact of target maintenance instructions on the fault response capability of the target power grid.

[0058] Specifically, after executing training maintenance instructions, the topology of the target power grid may change, and this change may reduce the fault response capability of the target power grid, affecting its long-term system resilience. To assess the impact of training maintenance instructions on the long-term system resilience of the target power grid, the response capability of the target power grid to preset faults under different conditions is simulated in a simulation environment. This yields the initial first response capability of the target power grid before executing the training maintenance instructions, and the second response capability after executing the training maintenance instructions. By comparing the first and second response capabilities, a power grid resilience assessment is performed on the target power grid, resulting in a long-term resilience penalty term to suppress the negative impact of the target maintenance instructions on the fault response capability of the target power grid. In some embodiments, the process of obtaining the first and second response capabilities can employ a simplified Nk derivation method.

[0059] For example, the long-term resilience penalty term can be expressed as: in, This indicates a long-term resilience penalty. The monetization weighting coefficient for the resilience risk of the target power grid; The set of assets in the target power grid that have experienced a pre-set single-device failure; In order to execute maintenance instructions In the original topology G of the target power grid, the load shedding power required when the k-th grid asset fails; In order to execute maintenance instructions Temporary topology G of the target power grid ’ In the equation, the load shedding power required when the k-th grid asset fails is given.

[0060] Reference Figure 5 As shown, in one embodiment of this application, a risk fusion assessment is performed on service impact penalty, lifecycle penalty, operational safety penalty, and long-term resilience penalty to obtain a reward function, including: S452. Conduct state utility assessments on each power grid asset in the target power grid to obtain power grid stability rewards.

[0061] S454. Based on the operational status of the target asset after executing training and maintenance instructions, perform risk prediction on preset faults to obtain a forward-looking prevention reward item.

[0062] S456. Calculate the cost of training and maintenance instructions to obtain maintenance cost penalty items.

[0063] S458. A risk fusion assessment is performed on the service impact penalty, life cycle penalty, operational safety penalty, long-term resilience penalty, power grid stability reward, forward-looking prevention reward, and maintenance cost penalty to obtain the reward function.

[0064] Specifically, in addition to the aforementioned penalty term, the reward function may include multiple reward terms to positively incentivize the maintenance decision-making agent. The reward terms of the reward function may include a grid stability reward term and a forward-looking prevention reward term. The grid stability reward term may be obtained based on the topological integrity of the target grid after executing the trained maintenance instructions, and is used to incentivize the maintenance decision-making agent to generate target maintenance instructions that ensure the target grid has a complete topology, enabling stable operation of both the target grid asset and the target grid. For example, the grid stability reward term can be expressed as: in, This indicates a reward for grid stability. The importance weight of power grid asset k is determined based on the transmission power or capacity of power grid asset k; Let be the state utility function of power grid asset k at time t. It takes a positive value when power grid asset k is running stably and a negative value when power grid asset k experiences a fault. The asset set of the target power grid.

[0065] Furthermore, the forward-looking prevention reward term can be a function term that rewards proactive intervention in potential faults in the target power grid or target asset based on the trained maintenance instructions. This incentivizes the maintenance decision-making agent to generate target maintenance instructions that proactively prevent fault conditions, thereby improving the forward-looking nature of the maintenance instructions. For example, the forward-looking prevention reward term can be expressed as: in, This indicates a forward-looking prevention incentive program; The probability that the defect corresponding to the i-th maintenance action in the maintenance instruction will cause a failure without intervention; The average economic loss caused by a fault resulting from a defect corresponding to the i-th maintenance action in the maintenance instruction; An indicator function to determine whether the i-th maintenance action in a maintenance instruction is a preventative maintenance; This is the incentive coefficient.

[0066] Furthermore, the reward function may also include a maintenance cost penalty term, which can be obtained based on the execution cost required for each maintenance action in the maintenance instruction, to suppress the operational cost of the target maintenance instruction. For example, the maintenance cost penalty term can be expressed as: in, This indicates a penalty for maintenance costs; The cost of executing the i-th maintenance action in the maintenance instruction.

[0067] Furthermore, a risk fusion is performed based on a composite risk penalty term that includes service impact penalty, lifecycle penalty, operational safety penalty, and long-term resilience penalty, along with the aforementioned reward and penalty terms, to obtain a reward function for training the maintenance decision-making agent. For example, the reward function can be expressed as: .

[0068] Reference Figure 6 As shown, in one embodiment of this application, the asset health assessment is performed by a hybrid identification agent; the hybrid identification agent is obtained in the following way: S210. Construct an initial identification agent; wherein the initial identification agent includes an initial asset identification component, an initial defect identification component, and an initial result fusion component.

[0069] S220. The initial asset identification component is used to identify power grid assets in the training image data of the image training dataset to obtain the asset type identification result of the training image data; the parameters of the initial asset identification component are adjusted according to the asset type identification result and the asset type label corresponding to the training image data to obtain the target asset identification component.

[0070] S230. The initial defect identification component is used to identify defect features in the training image data to obtain the defect type identification result of the training image data; the parameters of the initial defect identification component are adjusted according to the defect type identification result and the defect type labeling corresponding to the training image data to obtain the target defect identification component.

[0071] S240. Parameter locking is performed on the target asset identification component and the target defect identification component; adaptive weight allocation is performed on the asset type identification result and the defect type identification result through the initial result fusion component, and the identification results of the asset type identification result and the defect type identification result are fused according to the weight allocation result to obtain the mixed identification result; the parameters of the initial result fusion component are adjusted according to the mixed identification result and the image annotations corresponding to the training image data to obtain the target result fusion component.

[0072] S250. Construct a hybrid identification agent based on the target asset identification component, the target defect identification component, and the target result fusion component.

[0073] Specifically, in the visual image data of power grid assets, the corresponding power grid assets are usually located in complex backgrounds and intertwined with other power grid components, making it difficult for a single recognition component to simultaneously identify the asset type and defect type in the visual image data. For these reasons, this embodiment constructs independent asset recognition and defect recognition components, and trains both separately to improve the recognition accuracy for complex image data.

[0074] Reference Figure 7 As shown, an initial recognition agent is constructed, comprising an initial asset recognition component, an initial defect recognition component, and an initial result fusion component. The initial asset recognition component and the initial defect recognition component can be constructed using Low-Rank Adaptation (LoRA) technology, and the initial result fusion component can be a gated network, including but not limited to a two-layer Multilayer Perceptron (MLP). Furthermore, an image training dataset is required for training the initial recognition agent. This dataset includes multiple training images, each with corresponding image annotations including asset type and defect type labels.

[0075] For the initial asset identification component, defective regions in the training image data are masked. The initial asset identification component is then used to identify power grid assets in the masked training image data, yielding asset type identification results. Based on this, the asset type identification results are compared with the corresponding asset type labels in the training image data to determine the accuracy of the initial asset identification component's asset type identification, and the parameters of the initial asset identification component are adjusted accordingly. After multiple training rounds, the target asset identification component is obtained.

[0076] For the initial defect recognition component, image slicing is performed on the defect regions in the training image data to obtain defect image slices containing only the defect regions. The initial defect recognition component is then used to identify defect features in these image slices, yielding the defect type identification results for the training image data. Based on this, the defect type identification results are compared with the corresponding defect type annotations in the training image data to determine the correctness of the initial defect recognition component's identification of defect types, and the parameters of the initial defect recognition component are adjusted accordingly. After multiple training epochs, the target defect recognition component is obtained.

[0077] After obtaining the target asset identification component and the target defect identification component, the component parameters of each component are locked. An initial result fusion component is used to adaptively assign weights to the asset type identification results and defect type identification results, and then the identification results are fused according to the weight assignment results to obtain a hybrid identification result. It is understood that the asset type identification results and defect type identification results used to train the result fusion component can be obtained during the training of the asset and defect identification components, or they can be pre-obtained training data. Based on this, the hybrid identification result is compared with the image annotations corresponding to the training image data to determine the correctness of the weight assignment of the initial result fusion component, and the parameters of the initial result fusion component are adjusted accordingly. After multiple training rounds, the target result fusion component is obtained.

[0078] Furthermore, a hybrid recognition agent is constructed based on the target asset identification component, the target defect identification component, and the target result fusion component to perform asset health assessment on the visual image data of the target asset. Understandably, the result fusion component, through adaptive weight allocation of asset type identification results and defect type identification results, can selectively determine the confidence levels of the asset identification component and the defect identification component to adaptively fuse multiple identification results, significantly improving the recognition agent's accuracy in complex scenes.

[0079] This application also provides a practical application scenario for a power grid asset maintenance decision-making method based on drone inspection. In this embodiment, the target asset is tower #35, with a voltage level of 220kV. Data is collected from tower #35 through drone inspection, obtaining high-resolution images of the vibration dampers as visual image data and uploading them. Furthermore, weather forecasts for the environment where tower #35 is located are obtained through a meteorological service API, serving as meteorological forecast data. In this embodiment, the meteorological forecast data includes a forecast of northerly winds of force 9 within the next 6 hours for the environment where tower #35 is located.

[0080] Visual image data is sent to a hybrid recognition agent to perform an asset health assessment on tower #35, obtaining its health status data. In this embodiment, the health status data indicates that the vibration damper exhibits corrosion slippage, corresponding to a defect level of 3. Simultaneously, weather forecast data is sent to a causal reasoning agent to perform a meteorological risk assessment on tower #35, obtaining its environmental risk data. In this embodiment, the environmental risk data indicates that under future weather conditions, tower #35 has an 85% probability of conductor galloping.

[0081] The health status data and environmental risk data of tower #35 are fused to obtain the global collaborative risk data for tower #35. This global collaborative risk data is then input into the maintenance decision-making agent to generate a target maintenance instruction for tower #35. In this embodiment, the maintenance task of "handling the vibration damper defect of tower #35" in the target maintenance instruction for tower #35 is given the highest priority due to its corresponding high forward-looking prevention reward. Simultaneously, based on the operational safety penalty, it is ensured that the maintenance instruction will not cause line tripping in the target power grid, thus ensuring the accuracy and safety of the maintenance instruction and effectively guiding maintenance personnel to perform maintenance work efficiently.

[0082] Accordingly, please refer to Figure 8 This application provides a power grid asset maintenance decision-making device based on unmanned aerial vehicle (UAV) inspection, the device comprising: The data acquisition module 810 is used to acquire visual image data of the target asset and weather forecast data of the environment in which the target asset is located; wherein, the target asset belongs to the target power grid, and the visual image data is obtained through drone inspection.

[0083] The data assessment module 820 is used to conduct an asset health assessment of the target asset based on visual image data to obtain the health status data of the target asset; and to conduct a meteorological risk assessment of the target asset based on meteorological forecast data to obtain the environmental risk data of the target asset.

[0084] The data fusion module 830 is used to fuse health status data and environmental risk data to obtain global collaborative risk data for the target asset.

[0085] The risk decision module 840 is used to input global collaborative risk data into a reinforcement learning-based maintenance decision agent, and generate target maintenance instructions for the target asset based on the global collaborative risk data. The reward function of the maintenance decision agent is obtained based on the impact of the maintenance instructions generated by the maintenance decision agent during the training process on the target asset and the target power grid in a specified operation and maintenance dimension.

[0086] In some alternative implementations, the risk decision module 840 includes: The service loss assessment unit is used to assess the service loss based on the load shedding of the target power grid when executing training and maintenance instructions, and to obtain the service impact penalty item.

[0087] The health depreciation assessment unit is used to assess the health depreciation of the target asset based on its operating conditions after the execution of training and maintenance instructions, and to obtain lifecycle penalty items.

[0088] An operation safety determination unit is used to determine operation safety penalty items; wherein, the operation safety penalty item includes an indication function, the indication function having a first value when the training and maintenance instructions meet the safety procedure requirements, and a second value when the training and maintenance instructions do not meet the safety procedure requirements, and the first value is less than the second value.

[0089] The response simulation evaluation unit is used to simulate and evaluate the response capability of the target power grid to equipment failures before and after the execution of training and maintenance instructions, and to obtain a long-term resilience penalty term.

[0090] The risk fusion assessment unit is used to perform risk fusion assessment on service impact penalties, lifecycle penalties, operational safety penalties, and long-term resilience penalties to obtain a reward function.

[0091] In some alternative implementations, the service loss assessment unit includes: The service value assessment subunit is used to assess the service value of the execution process of training maintenance instructions based on the load shedding power and load shedding time, and to obtain a service impact penalty item to suppress the negative impact of the target maintenance instructions on the service capability of the target power grid.

[0092] In some alternative implementations, the health depletion assessment unit includes: The operating condition simulation calculation subunit is used to simulate and predict the operating conditions of the target asset after executing training and maintenance instructions, and calculate the health loss based on the simulation prediction results to obtain a life cycle penalty term used to suppress the negative impact of the target maintenance instructions on the lifespan of the target asset.

[0093] In some optional implementations, the response simulation evaluation unit includes: The first response simulation subunit is used to simulate the response capability of the target power grid to preset faults before executing training and maintenance instructions, so as to obtain the first response capability.

[0094] The second response simulation subunit is used to simulate the response capability of the target power grid to preset faults after executing training and maintenance instructions, so as to obtain the second response capability.

[0095] The power grid resilience assessment subunit is used to assess the power grid resilience based on the first response capability and the second response capability, and to obtain a long-term resilience penalty term to mitigate the negative impact of the target maintenance command on the fault response capability of the target power grid.

[0096] In some optional implementations, the risk fusion assessment unit includes: The state utility assessment subunit is used to assess the state utility of each power grid asset in the target power grid and obtain power grid stability reward items.

[0097] The intervention risk prediction subunit is used to predict the risk of preset faults based on the operating status of the target asset after the execution of training and maintenance instructions, and to obtain a forward-looking prevention reward item.

[0098] The maintenance cost calculation subunit is used to calculate the cost of training maintenance instructions and obtain maintenance cost penalty items.

[0099] The global risk fusion subunit is used to perform risk fusion assessment on service impact penalty items, life cycle penalty items, operational safety penalty items, long-term resilience penalty items, power grid stability reward items, forward prevention reward items, and maintenance cost penalty items to obtain the reward function.

[0100] In some alternative implementations, the data evaluation module 820 includes: The initial intelligent agent construction unit is used to construct the initial identification intelligent agent; wherein, the initial identification intelligent agent includes an initial asset identification component, an initial defect identification component, and an initial result fusion component.

[0101] The asset recognition training unit is used to identify power grid assets by using the initial asset recognition component to identify training image data in the image training dataset, thereby obtaining the asset type recognition result of the training image data; and to adjust the parameters of the initial asset recognition component based on the asset type recognition result and the asset type label corresponding to the training image data to obtain the target asset recognition component.

[0102] The defect recognition training unit is used to identify defect features in training image data through the initial defect recognition component to obtain the defect type recognition result of the training image data; and to adjust the parameters of the initial defect recognition component according to the defect type recognition result and the defect type label corresponding to the training image data to obtain the target defect recognition component.

[0103] The weight allocation training unit is used to lock the parameters of the target asset recognition component and the target defect recognition component; the initial result fusion component performs adaptive weight allocation on the asset type recognition result and the defect type recognition result, and fuses the recognition results of the asset type recognition result and the defect type recognition result according to the weight allocation result to obtain the hybrid recognition result; the parameters of the initial result fusion component are adjusted according to the hybrid recognition result and the image annotations corresponding to the training image data to obtain the target result fusion component.

[0104] The target intelligent agent construction unit is used to construct a hybrid recognition intelligent agent based on the target asset recognition component, the target defect recognition component, and the target result fusion component.

[0105] Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0106] In this embodiment, the power grid asset maintenance decision-making device based on UAV inspection is presented in the form of a functional unit. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0107] Please see Figure 9 , Figure 9 This is a schematic diagram of a computer device according to an embodiment of this application. As shown in the figure, the computer device includes one or more processors 10, a memory 20, and interfaces for connecting the various components, including high-speed interfaces and low-speed interfaces. The various components communicate with each other using different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on an external input / output device (such as a display device coupled to the interface). In some alternative embodiments, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 9 Take a processor 10 as an example.

[0108] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GPA), or any combination thereof.

[0109] The memory 20 stores instructions executable by at least one processor 10 to cause the at least one processor 10 to perform the method shown in the above embodiments.

[0110] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0111] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0112] The computer device also includes a communication interface 30 for communicating with other devices or communication networks.

[0113] This application also provides a computer-readable storage medium. The methods described in this application can be implemented in hardware or firmware, or implemented as recordable on a storage medium, or implemented as computer code downloaded over a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and subsequently stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the methods shown in the above embodiments are implemented.

[0114] This application provides a computer program product including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the method of any embodiment of this application.

[0115] Although embodiments of this application have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of this application, and all such modifications and variations fall within the scope defined by the appended claims.

[0116] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0117] For ease of description, the above devices are described separately by function as various units. Of course, in implementing this application, the functions of each unit can be implemented in one or more software and / or hardware.

[0118] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0119] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0120] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1The function specified in one or more boxes.

[0121] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0122] It should also be noted that the terms "comprising," "including," or any other variations thereof are 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 limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0123] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0124] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

[0125] Although embodiments of this application have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of this application, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A power grid asset maintenance decision-making method based on unmanned aerial vehicle (UAV) inspection, characterized in that, The method includes: The system acquires visual image data of the target asset and weather forecast data of the environment in which the target asset is located; wherein the target asset belongs to the target power grid, and the visual image data is obtained through drone inspection. Based on the visual image data, an asset health assessment is performed on the target asset to obtain the health status data of the target asset; based on the weather forecast data, a meteorological risk assessment is performed on the target asset to obtain the environmental risk data of the target asset. The health status data and the environmental risk data are fused to obtain the global collaborative risk data of the target asset; The global collaborative risk data is input into a reinforcement learning-based maintenance decision agent, which then generates target maintenance instructions for the target asset based on the global collaborative risk data. The reward function of the maintenance decision agent is obtained based on the impact of the maintenance instructions generated by the maintenance decision agent during training on the target asset and the target power grid in a specified operation and maintenance dimension.

2. The method according to claim 1, characterized in that, The maintenance instructions generated by the maintenance decision-making agent during the training process are recorded as training maintenance instructions. The reward function is obtained in the following way: Service loss is assessed based on the load shedding status of the target power grid when the training and maintenance instructions are executed, resulting in a service impact penalty. A health depletion assessment is performed on the target asset based on its operating condition after the execution of the training and maintenance instructions, resulting in a lifecycle penalty item. Determine the operation safety penalty item; wherein, the operation safety penalty item includes an indicator function, the indicator function having a first value when the training and maintenance instruction meets the safety procedure requirements, and a second value when the training and maintenance instruction does not meet the safety procedure requirements, wherein the first value is less than the second value; Based on the simulation evaluation of the target power grid's response capability to equipment failures before and after executing the training and maintenance instructions, a long-term resilience penalty term is obtained; The reward function is obtained by performing a risk fusion assessment on the service impact penalty item, the lifecycle penalty item, the operational security penalty item, and the long-term resilience penalty item.

3. The method according to claim 2, characterized in that, The load shedding situation includes the load shedding power and the load shedding time; the service loss assessment based on the load shedding situation of the target power grid when executing the training and maintenance instructions, to obtain a service impact penalty, includes: Based on the load shedding power and the load shedding time, the service value of the execution process of the training maintenance command is evaluated to obtain the service impact penalty item used to suppress the negative impact of the target maintenance command on the service capability of the target power grid.

4. The method according to claim 2, characterized in that, The health depletion assessment based on the operating condition of the target asset after executing the training and maintenance instructions yields a lifecycle penalty item, including: The operating conditions of the target asset after executing the training and maintenance instructions are simulated and predicted, and the health loss is calculated based on the simulation and prediction results to obtain the life cycle penalty item used to suppress the negative impact of the target maintenance instructions on the lifespan of the target asset.

5. The method according to claim 2, characterized in that, The step of simulating and evaluating the response capability of the target power grid to equipment faults before and after executing the training and maintenance instructions to obtain a long-term resilience penalty term includes: Before executing the training and maintenance instructions, the response capability of the target power grid to a preset fault is simulated to obtain a first response capability; After executing the training and maintenance instructions, the response capability of the target power grid to the preset fault is simulated to obtain a second response capability; A power grid resilience assessment is performed based on the first response capability and the second response capability to obtain a long-term resilience penalty term used to suppress the negative impact of the target maintenance command on the fault response capability of the target power grid.

6. The method according to claim 2, characterized in that, The reward function is obtained by performing a risk fusion assessment on the service impact penalty, the lifecycle penalty, the operational security penalty, and the long-term resilience penalty, including: The state utility of each power grid asset in the target power grid is evaluated to obtain power grid stability reward items; Based on the operating status of the target asset after executing the training and maintenance instructions, risk prediction is performed on preset faults to obtain a forward-looking prevention reward item; The cost of the training and maintenance instructions is calculated to obtain a maintenance cost penalty item; The reward function is obtained by performing a risk fusion assessment on the service impact penalty, the life cycle penalty, the operational safety penalty, the long-term resilience penalty, the power grid stability reward, the forward prevention reward, and the maintenance cost penalty.

7. The method according to any one of claims 1 to 6, characterized in that, The asset health assessment is performed by a hybrid recognition agent; the hybrid recognition agent is obtained through the following methods: Construct an initial identification agent; wherein the initial identification agent includes an initial asset identification component, an initial defect identification component, and an initial result fusion component; The initial asset identification component is used to identify power grid assets in the training image data of the image training dataset to obtain the asset type identification result of the training image data; the parameters of the initial asset identification component are adjusted according to the asset type identification result and the asset type label corresponding to the training image data to obtain the target asset identification component. The initial defect identification component is used to identify defect features in the training image data to obtain defect type identification results for the training image data; the parameters of the initial defect identification component are adjusted according to the defect type identification results and the defect type labels corresponding to the training image data to obtain the target defect identification component. The parameters of the target asset identification component and the target defect identification component are locked; the initial result fusion component performs adaptive weight allocation on the asset type identification result and the defect type identification result, and fuses the asset type identification result and the defect type identification result according to the weight allocation result to obtain a hybrid identification result; the parameters of the initial result fusion component are adjusted according to the hybrid identification result and the image annotations corresponding to the training image data to obtain the target result fusion component; The hybrid recognition agent is constructed based on the target asset recognition component, the target defect recognition component, and the target result fusion component.

8. A power grid asset maintenance decision-making device based on unmanned aerial vehicle (UAV) inspection, characterized in that, The device includes: The data acquisition module is used to acquire visual image data of the target asset and weather forecast data of the environment in which the target asset is located; wherein, the target asset belongs to the target power grid, and the visual image data is obtained through UAV inspection; The data assessment module is used to perform an asset health assessment on the target asset based on the visual image data to obtain the health status data of the target asset; and to perform a meteorological risk assessment on the target asset based on the meteorological forecast data to obtain the environmental risk data of the target asset. The data fusion module is used to fuse the health status data and the environmental risk data to obtain the global collaborative risk data of the target asset. The risk decision module is used to input the global collaborative risk data into a reinforcement learning-based maintenance decision agent, and the maintenance decision agent generates target maintenance instructions for the target asset based on the global collaborative risk data; wherein, the reward function of the maintenance decision agent is obtained according to the impact of the maintenance instructions generated by the maintenance decision agent during the training process on the target asset and the target power grid in a specified operation and maintenance dimension.

9. A computer device, characterized in that, include: A memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, the processor executing the computer instructions to perform the method of any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to perform the method of any one of claims 1 to 7.

Images