Power transmission line state monitoring method and device based on flux-inductance fusion

Abstract

The invention provides a power transmission line state monitoring method and device based on flux-inductance fusion, and relates to the technical field of artificial intelligence, and the method comprises the steps: generating a plurality of different monitoring tasks based on the current state of a power transmission line; wherein the monitoring task comprises a data screening sub-task, a risk prediction sub-task and a comprehensive analysis sub-task; based on the calculation requirement of each monitoring task, determining a target model division strategy, a target resource allocation strategy and a target data acquisition signal; and based on the target model division strategy, the target resource allocation strategy and the target data acquisition signal, executing each monitoring task in parallel to obtain a state monitoring result of the power transmission line. According to the power transmission line state monitoring method based on the flux-inductance fusion, low-delay and high-efficiency intelligent accurate monitoring of the state of the power transmission line is realized, and the safety of a power grid is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and in particular to a power transmission line state monitoring method and device based on sensory fusion. BACKGROUND

[0002] The safety and stability of the power transmission network is of great importance. A large number of power transmission lines pass through remote areas with complex geographical environment and sparse population, and are subject to prominent risks such as line icing, dancing, forest fire, and geological disasters (landslide, debris flow).

[0003] At present, power transmission line state monitoring mainly relies on manual inspection, helicopters (or drones), and limited sensors (such as cameras, vibration sensors, etc.) installed at key locations. However, with the expansion of the power grid, the increasing complexity of the environment, and the frequent occurrence of extreme weather events, these traditional methods face great challenges in terms of monitoring range, real-time performance, accuracy, cost-effectiveness, and personnel safety. For example, manual inspection has the problems of long cycle, high cost, and high risk; unmanned aerial vehicle inspection is limited by endurance, weather conditions, and real-time data transmission; and conventional sensors are limited by power supply, communication, and maintenance difficulties, making it difficult to achieve effective coverage and reliable data transmission in the region.

[0004] Therefore, the prior art cannot accurately monitor the state of the power transmission line in a low-delay and high-efficiency manner, and cannot accurately predict the upgrade of the power transmission line fault, resulting in low safety of the power grid. SUMMARY

[0005] The present application provides a power transmission line state monitoring method and device based on sensory fusion, which solves the technical problem that the prior art cannot accurately monitor the state of the power transmission line in a low-delay and high-efficiency manner, and accurately predicts the upgrade of the power transmission line fault, thereby improving the safety of the power grid.

[0006] The present application provides a power transmission line state monitoring method based on sensory fusion, comprising the following steps: Based on the current state of the power transmission line, a plurality of different monitoring tasks are generated; the monitoring tasks include data filtering subtasks, risk prediction subtasks, and comprehensive analysis subtasks; Based on the computing requirements of each monitoring task, a target model division strategy, a target resource allocation strategy, and a target data acquisition signal are determined; Based on the target model division strategy, the target resource allocation strategy, and the target data acquisition signal, each monitoring task is executed in parallel to obtain the state monitoring result of the power transmission line.

[0007] According to the power transmission line state monitoring method based on sensory fusion provided by the present application, based on the current state of the power transmission line, a plurality of different monitoring tasks are generated, which comprises: Based on the current state of the transmission line, determine the potential risk category of the transmission line; Based on the potential risk categories, the multiple different monitoring tasks are generated.

[0008] According to the present invention, a transmission line condition monitoring method based on sensor fusion is provided, wherein determining the target model partitioning strategy, the target resource allocation strategy, and the target data acquisition signal based on the computational requirements of each monitoring task includes: Based on the computational requirements of each monitoring task, the preset iterative steps are repeatedly executed to update the intermediate probability parameters until the change range of the intermediate probability parameters is within the preset range. The intermediate probability parameters obtained in the last iteration are then used as the target probability parameters. Based on the target probability parameters, a target model partitioning strategy is generated for each monitoring task; Based on the target model partitioning strategy, determine the target resource allocation strategy for each monitoring task; Based on the target model partitioning strategy and the target resource allocation strategy, the target data acquisition signal corresponding to each monitoring task is determined.

[0009] According to the present invention, a transmission line condition monitoring method based on sensing fusion is provided, wherein the iterative steps include: Based on the intermediate probability parameters obtained in the previous iteration, the intermediate model partitioning strategy obtained in the previous iteration is updated to obtain the intermediate model partitioning strategy obtained in this iteration; wherein, each intermediate probability parameter corresponds to each monitoring task; each monitoring task corresponds to multiple intermediate model partitioning strategies; Based on the intermediate model partitioning strategy obtained in this iteration, an intermediate resource allocation strategy obtained in this iteration is generated. Based on the intermediate model partitioning strategy and intermediate resource allocation strategy obtained in this iteration, the intermediate data acquisition signal obtained in the previous iteration is updated to obtain the intermediate data acquisition signal obtained in this iteration. Based on the intermediate model partitioning strategy, the intermediate resource allocation strategy, and the intermediate data acquisition signals obtained in this iteration, each monitoring task is executed in parallel to obtain the corresponding task execution time. Based on the task execution time, a preset number of intermediate model partitioning strategies are selected from the intermediate model partitioning strategies obtained in this round of iteration as the first sample set; For each intermediate model partitioning strategy in the strategy sample set, a neighborhood search optimization is performed to obtain a second sample set; Based on the second sample set, the intermediate probability parameters obtained in the previous iteration are updated to obtain the intermediate probability parameters obtained in the current iteration. The intermediate probability parameters used in the first round of iterations are randomly generated based on the Bernoulli distribution; the intermediate data acquisition signals used in the first round of iterations are determined based on the data acquisition requirements and priorities of each monitoring task; and the priorities are determined based on preset rules.

[0010] According to the present invention, a transmission line condition monitoring method based on sensor fusion is provided, wherein the intermediate resource allocation strategy obtained in the current iteration is generated based on the intermediate model partitioning strategy obtained in the current iteration, including: Based on the intermediate model partitioning strategy obtained in this iteration, the deep learning model corresponding to each monitoring task is divided into a first intermediate sub-model, a second intermediate sub-model, and a third intermediate sub-model; the first intermediate sub-model is used to perform the data filtering sub-task; the second intermediate sub-model is used to perform the risk prediction sub-task; and the third intermediate sub-model is used to perform the comprehensive analysis sub-task. Based on the computational volume corresponding to the first intermediate sub-model, the second intermediate sub-model, and the third intermediate sub-model, the server computing resource requirements corresponding to the first intermediate sub-model, the second intermediate sub-model, and the third intermediate sub-model are determined. Based on the server's computing resource requirements, the intermediate resource allocation strategy obtained in this iteration is generated.

[0011] According to the present invention, a method for monitoring the condition of a transmission line based on sensor fusion, wherein each monitoring task is executed in parallel based on the target model partitioning strategy, the target resource allocation strategy, and the target data acquisition signal to obtain the condition monitoring results of the transmission line, including: Based on the target model partitioning strategy, the deep learning model corresponding to each monitoring task is divided into a first target sub-model, a second target sub-model, and a third target sub-model; the first target sub-model is used to perform the data filtering sub-task; the second target sub-model is used to perform the risk prediction sub-task; and the third target sub-model is used to perform the comprehensive analysis sub-task. Based on the target resource allocation strategy, corresponding first server computing resources, second server computing resources and third server computing resources are allocated to the first target sub-model, the second target sub-model and the third target sub-model, respectively. Based on the target data acquisition signal, the sensing data required for each monitoring task are collected respectively; The first target sub-model and the first server computing resources are used to filter the perceived data to obtain key features; Risk prediction is performed based on the key features using the second target sub-model and the second server computing resources to obtain prediction results; The prediction results are comprehensively analyzed using the third target sub-model and the third server computing resources to obtain the monitoring results corresponding to each monitoring task. The monitoring results corresponding to each monitoring task are used as the status monitoring results of the transmission line.

[0012] The present invention also provides a transmission line condition monitoring device based on sensing fusion, comprising the following modules: The generation module is used to generate multiple different monitoring tasks based on the current state of the transmission line; the monitoring tasks include data filtering sub-tasks, risk prediction sub-tasks, and comprehensive analysis sub-tasks. The allocation module is used to determine the target model partitioning strategy, target resource allocation strategy, and target data acquisition signals based on the computational requirements of each monitoring task. The monitoring module is used to execute each monitoring task in parallel based on the target model partitioning strategy, the target resource allocation strategy, and the target data acquisition signal to obtain the status monitoring results of the transmission line.

[0013] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the transmission line condition monitoring method based on sensing fusion as described above.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the transmission line condition monitoring method based on sensing fusion as described above.

[0015] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the transmission line condition monitoring method based on sensing fusion as described above.

[0016] The present invention provides a transmission line condition monitoring method and device based on sensor fusion. Based on the current state of the transmission line, it generates multiple different monitoring tasks, including data filtering sub-tasks, risk prediction sub-tasks, and comprehensive analysis sub-tasks. This achieves intelligent and differentiated task division from a fixed mode to a dynamic adaptive mode, making the monitoring tasks more targeted and efficient. Based on the computational requirements of each monitoring task, it determines the target model partitioning strategy, target resource allocation strategy, and target data acquisition signal. By pre-coordinating the segmentation method of the deep learning model and the sensory computing power (communication, sensing, and computing) resources, it provides a model and resource foundation for the subsequent parallel execution of distributed tasks. Based on the target model partitioning strategy, target resource allocation strategy, and target data acquisition signal, each monitoring task is executed in parallel to obtain the transmission line condition monitoring results. Parallel processing significantly shortens the total time from data acquisition to obtaining the final result, avoiding the time delay and resource waste of traditional time-division or frequency-division multi-task modes. This achieves low-latency, high-efficiency, intelligent, and accurate monitoring of the transmission line condition, thereby improving power grid security. Attached Figure Description

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

[0018] Figure 1 This is a flowchart illustrating the transmission line condition monitoring method based on sensor fusion provided by the present invention.

[0019] Figure 2 This is a schematic diagram of the "end-edge-cloud" three-layer collaborative reasoning framework provided in the embodiments of the present invention.

[0020] Figure 3 This is a schematic diagram of the simulation experiment results provided by the present invention.

[0021] Figure 4 This is a schematic diagram of intelligent energy allocation provided by the present invention.

[0022] Figure 5 This is a schematic diagram of the hierarchical collaborative DNN inference and joint optimization algorithm for minimizing task latency provided by the present invention.

[0023] Figure 6 This is a schematic diagram illustrating the convergence of the inner and outer layer joint optimization algorithm provided by this invention.

[0024] Figure 7This is a schematic diagram of the structure of the power transmission line condition monitoring device based on sensor fusion provided by the invention.

[0025] Figure 8 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0027] The following is combined Figures 1 to 8 This invention describes a method and apparatus for monitoring the condition of power transmission lines based on sensor fusion.

[0028] Figure 1 This is a flowchart illustrating the transmission line condition monitoring method based on sensor fusion provided by the present invention, as shown below. Figure 1 As shown, the method includes the following steps: Step 101: Based on the current state of the transmission line, generate multiple different monitoring tasks; the monitoring tasks include data filtering sub-tasks, risk prediction sub-tasks, and comprehensive analysis sub-tasks. Step 102: Based on the computational requirements of each monitoring task, determine the target model partitioning strategy, target resource allocation strategy, and target data acquisition signals; Step 103: Based on the target model partitioning strategy, the target resource allocation strategy, and the target data acquisition signal, execute each monitoring task in parallel to obtain the status monitoring results of the transmission line.

[0029] Specifically, integrated sensing and communication (ISAC) devices deployed on key tower sections of transmission lines continuously transmit signals, and lightweight artificial intelligence (AI) models on the device are used to analyze the echo signals in real time to initially screen out abnormal signals. The AI ​​model can be any deep learning model, such as a deep neural network (DNN), a long short-term memory network (LSTM), or a recurrent neural network (RNN).

[0030] Based on the type of abnormal signal, it is determined that the transmission line has entered an abnormal state (such as line icing, galloping, wildfire, or interference from foreign objects), and a multi-task concurrent mode is immediately activated to generate various monitoring tasks corresponding to the abnormal state. At the same time, the priority of each monitoring task is determined based on the rule base in the historical experience of the power grid.

[0031] Existing solutions typically face problems such as insufficient edge computing power, high overhead of raw data transmission, and high task processing latency when dealing with intelligent perception tasks that require complex AI models.

[0032] To address the aforementioned issues, this invention divides each monitoring task into data filtering sub-tasks, risk prediction sub-tasks, and comprehensive analysis sub-tasks. Then, edge computing nodes with integrated sensing capabilities are deployed on key tower sections or segments of dense transmission lines. Within this scenario, a three-layer collaborative inference framework is constructed: "End (ISAC device) - Edge (Multi-access Edge Computing (MEC) server) - Cloud (central control platform)". Within this framework, based on the computational requirements (including priority and complexity) of each monitoring task, a target model partitioning strategy is determined. According to this strategy, a pre-trained artificial intelligence model is dynamically divided into multiple computational fragments or sub-models, which are then deployed on the end, edge, and cloud layers to collaboratively execute different sub-tasks. This ensures that the communication link only transmits the "intermediate features" corresponding to sub-tasks with significantly reduced data volume, solving the communication bandwidth bottleneck and high transmission latency problems caused by excessively large original data volumes in traditional offloading schemes, and significantly reducing the demand for communication bandwidth and transmission latency.

[0033] Meanwhile, by utilizing multi-sensory task fusion technology on the ISAC device, intelligent nodes can simultaneously perform high-precision environmental perception and communication offloading tasks for intermediate features. Furthermore, a joint optimization method oriented towards minimizing total inference latency is proposed. A multi-task priority mechanism is introduced into the perception constraints. Under the premise of ensuring the performance of key perception tasks, perception, communication and computing resources are deeply integrated, effectively solving the latency and overhead problems caused by complex AI perception tasks.

[0034] Figure 2 This is a schematic diagram of the "end-edge-cloud" three-layer collaborative reasoning framework provided in an embodiment of the present invention, as shown below. Figure 2 As shown.

[0035] Intelligent ISAC devices (endpoints): These are edge computing nodes with integrated sensing and communication capabilities deployed in key tower sections or areas of dense power transmission lines. Each device is equipped with a Multiple-Input Multiple-Output (MIMO) antenna system and a high-performance signal processing and computing unit. Their core function is to transmit and receive ISAC signals, which simultaneously carry sensing waveforms and communication data streams. These nodes not only process data from traditional sensors (such as cameras and weather sensors, accessed through their standard interface modules), but also utilize their transmitted communication signals (such as millimeter-wave, terahertz, or specifically optimized 5G / 6G signals) to achieve high-precision, wide-range, and continuous dynamic sensing of the power line itself and its surrounding environment. Based on this, the communication data stream is primarily used to offload locally pre-processed sensing data (such as extracted feature parameters) or intermediate features output from AI models to the MEC server via a converged communication link. Therefore, the ISAC devices are responsible for performing the initial but critical sensing tasks of the environment, raw data acquisition, some intelligent processing, and initial communication of sensing results.

[0036] MEC Server (Edge): Deployed at the network edge, close to ISAC devices, it possesses strong computing and storage capabilities. It receives data offloaded from multiple ISAC devices via converged communication links, performs more complex AI model intermediate layer calculations, multi-source data aggregation and analysis, and can transmit risk prediction results or data requiring further processing to the cloud.

[0037] Central control platform (cloud): Possesses powerful computing and storage resources, responsible for performing the final layers of computation in AI models, global situational analysis, complex decision-making, model training and optimization, and long-term data storage.

[0038] This "edge-cloud" architecture organically combines sensing, computing, and communication resources through efficient communication links (converged communication), providing a basic platform for achieving accurate perception of complex power transmission line conditions.

[0039] Taking DNN as an example, according to the network hierarchy, DNN can be divided into three sub-models and deployed on the device, edge layer and cloud respectively. Among them, the sub-model on the device performs the data filtering sub-task to filter the sensing data collected by the ISAC device. Then, it is transmitted to the sub-model in the edge layer through optical fiber to perform the risk prediction sub-task. Finally, the risk prediction results are uploaded to the sub-model in the cloud to perform the comprehensive analysis sub-task to obtain the final transmission line status monitoring results.

[0040] Based on the computational cost of each network layer, the computational cost of each sub-model can be determined, thereby identifying the target resource allocation strategy and allocating appropriate server computing resources to each sub-model.

[0041] Based on the server computing resources and sub-model computing capabilities of the device, and combined with the data acquisition type requirements corresponding to the monitoring task, the amount and type of data that the device needs to collect can be determined, thereby determining the target data acquisition signal that the ISAC device needs to transmit.

[0042] Based on the target model partitioning strategy, target resource allocation strategy, and target data acquisition signals, each monitoring task is executed in parallel through an "edge-cloud" architecture to obtain the status monitoring results of the transmission line.

[0043] For example, in the scenario of real-time capture of wind-induced galloping of transmission lines and concurrent foreign object intrusion monitoring, in typical areas with abundant wind resources but harsh operating environments, transmission lines often experience low-frequency, large-amplitude galloping due to strong winds, seriously threatening power grid safety. Traditional inspection methods cannot effectively monitor and provide early warning for sudden, high-risk events such as galloping in real time. In this case, the steps for executing each monitoring task in parallel using the "end-edge-cloud" architecture provided in this embodiment of the invention are as follows: First, intelligent ISAC devices deployed on key tower sections continuously transmit signals and use lightweight DNN models on the device to analyze the echo signals in real time, initially screening out abnormal signals that meet the characteristics of galloping (such as low frequency and high energy). Once the vibration energy of the abnormal signal exceeds a preset threshold, it is determined that the line has entered a galloping state, and a "multi-task concurrent mode" is immediately activated. For example, a high-priority galloping fine analysis task and a low-priority concurrent foreign object intrusion wide-area scanning task issued by the central platform.

[0044] Then, based on the different priorities (high-priority gobbling analysis and low-priority foreign object scanning) and complexities of the two tasks (e.g., high-priority gobbling analysis requires complex model processing, while low-priority foreign object scanning only requires simple energy detection), the optimal DNN model partitioning strategy, the target data acquisition signals that can be taken into account for both tasks (e.g., the optimal ISAC beamforming matrix obtained by weighting the beammap signals corresponding to the two tasks), and the corresponding target resource allocation strategy are determined.

[0045] Finally, the device utilizes optimized resources to complete the initial calculations of the model, and then rapidly offloads intermediate features containing key timing information to the MEC server. The MEC server then performs precise analysis and risk prediction of the galloping parameters. Ultimately, the risk prediction results, including those related to galloping and foreign object intrusion, are reported to the cloud-based central control platform for comprehensive analysis or evaluation, yielding the transmission line status monitoring results.

[0046] Based on the above embodiments, Figure 3 This is a schematic diagram of the simulation experiment results provided by the present invention, such as... Figure 3As shown in the figure, the end-to-end latency comparison clearly illustrates the inference latency of state monitoring under different methods. The figure compares the latency performance of the three schemes: the latency of the traditional fully cloud-based processing scheme increases significantly with the amount of data; the latency of the present invention is relatively low before joint optimization (model partitioning according to the computational requirements of the monitoring task (including priority and complexity)) but still has room for improvement; after the joint optimization of the present invention, the end-to-end latency remains at an extremely low level, reducing the latency by 50-75% compared to the traditional cloud-based processing scheme. This proves that the present invention, through "end-edge-cloud" collaborative inference and resource optimization, can meet the application scenarios with extremely high real-time requirements such as dance monitoring.

[0047] This invention dynamically generates the most suitable combination of monitoring tasks based on actual environmental and risk changes, making monitoring more targeted. By adopting an "edge-cloud" collaborative reasoning framework, it jointly considers and optimizes model partitioning, resource allocation, and signal design. It only needs to transmit lightweight "intermediate features" instead of massive amounts of raw sensing data. Compared with traditional solutions such as directly transmitting high-definition video streams, the communication bandwidth requirement can be reduced by more than 90%, effectively solving the problem of scarce wireless communication resources in remote areas. It ensures the optimization of overall task execution efficiency under limited computing, communication, and spectrum resources. Through a multi-task parallel execution mechanism, it simultaneously meets the data acquisition needs of multiple different types of sensing tasks in a single target data acquisition signal transmission, avoiding the resource waste and time delays of traditional time-division or frequency-division multi-task modes. This wins valuable time for timely fault warning and handling, thereby improving power grid security.

[0048] This invention provides a transmission line condition monitoring method based on sensor fusion. Based on the current state of the transmission line, it generates multiple different monitoring tasks, including data filtering sub-tasks, risk prediction sub-tasks, and comprehensive analysis sub-tasks. This achieves intelligent and differentiated task division from a fixed mode to a dynamic adaptive mode, making the monitoring tasks more targeted and efficient. Based on the computational requirements of each monitoring task, it determines the target model partitioning strategy, target resource allocation strategy, and target data acquisition signal. By pre-coordinating the segmentation method of the deep learning model and sensory computing power (communication, sensing, and computing) resources, it provides a model and resource foundation for subsequent distributed task parallel execution. Based on the target model partitioning strategy, target resource allocation strategy, and target data acquisition signal, each monitoring task is executed in parallel to obtain the transmission line condition monitoring results. Parallel processing significantly shortens the total time from data acquisition to obtaining the final result, avoiding the time delay and resource waste of traditional time-division or frequency-division multi-task modes. This achieves low-latency, high-efficiency, intelligent, and accurate monitoring of the transmission line condition, thereby improving power grid security.

[0049] Furthermore, based on the current state of the transmission line, multiple different monitoring tasks are generated, including: Based on the current state of the transmission line, determine the potential risk category of the transmission line; Based on the potential risk categories, the multiple different monitoring tasks are generated.

[0050] Specifically, ISAC devices deployed on key tower sections of transmission lines continuously transmit signals, and lightweight AI models at the device end analyze the echo signals in real time to initially screen out abnormal signals. Based on the type of abnormal signal, it is determined that the transmission line has entered an abnormal state (e.g., line icing, galloping, wildfire, or interference from foreign objects). Based on historical power grid experience, the potential risk categories faced by the transmission line are identified, and a multi-task concurrent mode is immediately activated to generate various corresponding monitoring tasks. At the same time, the priority of each monitoring task is determined based on a rule base from historical power grid experience.

[0051] For example, in typical areas with abundant wind resources but harsh operating environments, transmission lines often experience low-frequency, high-amplitude galloping due to strong winds, seriously threatening power grid safety. Intelligent ISAC devices deployed on key tower sections continuously transmit signals and utilize lightweight DNN models at the device end to analyze the echo signals in real time, initially screening out abnormal signals that meet galloping characteristics (such as low frequency and high energy). Once the vibration energy of an abnormal signal exceeds a preset threshold, the line is determined to be in a galloping state, and the potential risk category can be galloping risk. At this point, a high-priority fine-grained galloping analysis task and a low-priority concurrent wide-area foreign object intrusion scanning task issued by the central platform can be generated.

[0052] By first identifying risk categories, this invention transforms monitoring behavior from passive response to proactive prediction, enabling the system to automatically and reliably generate reasonable combinations of monitoring tasks in different scenarios, thereby improving the efficiency, accuracy, and intelligence level of monitoring.

[0053] Furthermore, the determination of the target model partitioning strategy, target resource allocation strategy, and target data acquisition signals based on the computational requirements of each monitoring task includes: Based on the computational requirements of each monitoring task, the preset iterative steps are repeatedly executed to update the intermediate probability parameters until the change range of the intermediate probability parameters is within the preset range. The intermediate probability parameters obtained in the last iteration are then used as the target probability parameters. Based on the target probability parameters, a target model partitioning strategy is generated for each monitoring task; Based on the target model partitioning strategy, determine the target resource allocation strategy for each monitoring task; Based on the target model partitioning strategy and the target resource allocation strategy, the target data acquisition signal corresponding to each monitoring task is determined.

[0054] Furthermore, the iterative steps include: Based on the intermediate probability parameters obtained in the previous iteration, the intermediate model partitioning strategy obtained in the previous iteration is updated to obtain the intermediate model partitioning strategy obtained in this iteration; wherein, each intermediate probability parameter corresponds to each monitoring task; each monitoring task corresponds to multiple intermediate model partitioning strategies; Based on the intermediate model partitioning strategy obtained in this iteration, an intermediate resource allocation strategy obtained in this iteration is generated. Based on the intermediate model partitioning strategy and intermediate resource allocation strategy obtained in this iteration, the intermediate data acquisition signal obtained in the previous iteration is updated to obtain the intermediate data acquisition signal obtained in this iteration. Based on the intermediate model partitioning strategy, the intermediate resource allocation strategy, and the intermediate data acquisition signals obtained in this iteration, each monitoring task is executed in parallel to obtain the corresponding task execution time. Based on the task execution time, a preset number of intermediate model partitioning strategies are selected from the intermediate model partitioning strategies obtained in this round of iteration as the first sample set; For each intermediate model partitioning strategy in the strategy sample set, a neighborhood search optimization is performed to obtain a second sample set; Based on the second sample set, the intermediate probability parameters obtained in the previous iteration are updated to obtain the intermediate probability parameters obtained in the current iteration. The intermediate probability parameters used in the first round of iterations are randomly generated based on the Bernoulli distribution; the intermediate data acquisition signals used in the first round of iterations are determined based on the data acquisition requirements and priorities of each monitoring task; and the priorities are determined based on preset rules.

[0055] Furthermore, the intermediate resource allocation strategy generated from the intermediate model partitioning strategy obtained in this iteration includes: Based on the intermediate model partitioning strategy obtained in this iteration, the deep learning model corresponding to each monitoring task is divided into a first intermediate sub-model, a second intermediate sub-model, and a third intermediate sub-model; the first intermediate sub-model is used to perform the data filtering sub-task; the second intermediate sub-model is used to perform the risk prediction sub-task; and the third intermediate sub-model is used to perform the comprehensive analysis sub-task. Based on the computational volume corresponding to the first intermediate sub-model, the second intermediate sub-model, and the third intermediate sub-model, the server computing resource requirements corresponding to the first intermediate sub-model, the second intermediate sub-model, and the third intermediate sub-model are determined. Based on the server's computing resource requirements, the intermediate resource allocation strategy obtained in this iteration is generated.

[0056] Specifically, under the "end-edge-cloud" collaborative architecture, this invention proposes an adaptive high-precision multi-sensor task fusion method based on MIMO technology, which optimizes the design of the ISAC transmit beamforming matrix. This causes the actual transmitted signal covariance matrix to be generated. It can accurately match one or more desired beammap covariance matrices predefined according to the current monitoring task, enabling it to execute multiple monitoring tasks with different priorities simultaneously in a single signal transmission, achieving an upgrade in sensing mode from single focus to multi-point consideration and clear distinction between primary and secondary tasks.

[0057] Figure 4 This is a schematic diagram of intelligent energy allocation provided by the present invention, such as... Figure 4 As shown. The baseband equivalent ISAC signal transmitted by ISAC device k deployed on the transmission tower at time t. (in The number of transmitting antennas can be modeled as a linear superposition of communication signals and sensing signals. : In the formula, A sequence of symbols that carries communication data (e.g., intermediate features of DNN output). The corresponding communication beamforming matrix is ​​then used. For dedicated sensing waveform sequences, and These are the corresponding communication and sensing beamforming matrices, where and These represent the number of data streams for communication and sensing, respectively. and Shared spectrum and hardware for signals.

[0058] ISAC equipment Total transmitted signal covariance matrix It can be represented as: In the formula, Representing the conjugate transpose, the above single actual emission covariance matrix Must meet simultaneously Several concurrent monitoring tasks with different priorities. Each task Each has its specific data acquisition requirements (i.e., sensing requirements), which are determined by a desired beammap covariance matrix. and a normalized priority weight To define, where .

[0059] Desired beam pattern covariance matrix Adaptive focusing is achieved by pre-calculating and generating based on specific sensing requirements (e.g., main lobe width, gain requirements in a specific direction, or side lobe suppression level) or by dynamically adjusting based on the sensing results of the previous round.

[0060] The core objective of the optimization is to minimize the sum of the weighted differences between the actual transmitted signal covariance matrix and the desired covariance matrix, while simultaneously satisfying practical physical and performance constraints. This optimization problem can be expressed as: In the formula: Denotes the square of the Frobenius norm; The normalized priority weights are used to ensure that the matching error of high-priority tasks accounts for a larger proportion in the optimization and are therefore satisfied first. It is a weight matrix, used to assign different matching priorities to different elements of the covariance matrix (corresponding to different regions of the beam pattern). For example, high weight can be assigned to the main lobe direction and low weight to the side lobe direction, thereby achieving key optimization of the critical sensing area. symbol This represents the Hadamard product (element-wise product).

[0061] To constrain the total transmit power, ensure that the total power does not exceed [the specified limit]. ; This represents the sum of all elements on the main diagonal of the matrix.

[0062] To constrain communication service quality (QoS), ensure that the signal-to-interference-plus-noise ratio (SINR) of the communication link is not lower than a preset threshold. This ensures the reliable transmission of intermediate features and other data. From the equipment Channel vector to MEC It is noise power; Indicates the first Communication beamforming matrix for each ISAC device.

[0063] Optional beam pattern null constraints are used for specific direction sets. Each direction Null depressions are formed to suppress interference in specific areas or reduce clutter from specific directions. It is the corresponding guide vector The maximum permissible sidelobe level.

[0064] To solve the aforementioned non-convex optimization problem, this embodiment of the invention employs a semidefinite relaxation method. This is achieved by introducing a new positive semidefinite matrix variable. = The original problem is transformed into a standard semidefinite programming (SDP) problem. The objective function and all constraints (such as power, communication quality, and service quality) of the original optimization problem are reconstructed with respect to the new variables. It has a convex function form. Therefore, standard algorithms such as the interior-point method can be used to efficiently find its global optimum, and then high-quality feasible solutions can be recovered through methods such as Gaussian randomization.

[0065] To achieve truly adaptive high-precision sensing, the desired covariance matrix is... It is not fixed. This invention designs a closed-loop feedback mechanism to achieve dynamic adjustment.

[0066] It processes raw sensing data from ISAC devices in real time (such as echo signals and target scattering characteristics) as well as data from potentially connected external sensors (e.g., weather sensors and line condition monitoring sensors installed on the tower base). Its core function is to extract key feature parameters from multi-source data and output a state vector defining the current state. : In the formula, It contains information including the location, size, kinematic parameters, and other parameters of all monitored targets (such as ice accumulation and dancing points); This indicates information including environmental parameters such as wind speed, wind direction, temperature, and humidity. Channel quality information.

[0067] The core function of adaptive adjustment is to receive the state vector. And through a decision function Output new perception task instructions : This decision function is based on a pre-defined rule base. For example, when a high-risk target (such as ice thickness) is identified... Exceeding the preset safety threshold or the distance between trees and power lines Less than the safe distance When ), the focused scanning mode is triggered.

[0068] Sensing task instructions based on dynamic beamforming decision output It is responsible for calculating or selecting to generate a new expected covariance matrix. Based on the command type (such as focusing the main lobe, forming a null, etc.), a preset parameterized beam pattern template is invoked, and combined with information such as the target position and interference direction, the corresponding beam pattern is calculated. For example, in focused scanning mode, a narrower main lobe will be generated (e.g., the main lobe width is reduced from...). Reduce to ), higher energy gain (such as increased gain) dB), and precisely point to the target area. of .

[0069] By solving the aforementioned optimization problem, the shape and direction of the transmitted beam can be dynamically adjusted, focusing sensing energy more effectively on specific targets or areas while ensuring reliable transmission of communication data streams. This achieves deep integration and synergistic gain between sensing and communication functions. Through optimization algorithms, the beam pattern of a single transmitted ISAC signal can be "weighted approximated" to simultaneously meet the differentiated needs of multiple sensing tasks. This clear prioritization of sensing capabilities allows for intelligent allocation and focusing of sensing resources based on real-time risks and operational needs, enhancing the flexibility and efficiency of sensing.

[0070] On the other hand, complex AI models (described in this embodiment using DNN as an example) are dynamically divided into three layers for execution: edge, cloud, and endpoint. The ISAC device (edge) executes the first few layers of the DNN model, generating lightweight intermediate features. These intermediate features are then offloaded to the MEC server (edge) for intermediate layer processing via a converged communication link (i.e., utilizing the communication capabilities in the ISAC signal). The MEC server then processes the features and offloads the results or further intermediate features to the cloud platform for final processing. In this hierarchical collaborative processing mode, minimizing the end-to-end latency of the entire intelligent sensing task by optimizing the DNN partitioning strategy, the allocation of communication resources (especially the communication beamforming in the ISAC signal used to transmit intermediate features), and the allocation of computing resources at each layer is key to improving response speed and practicality.

[0071] To address the challenges of resource allocation and task scheduling inherently coupled within the "edge-cloud" collaborative architecture, this invention proposes a hierarchical collaborative DNN inference and joint optimization algorithm for minimizing task latency. This algorithm solves a Mixed-Integer Nonlinear Programming (MINLP) problem. Through an efficient two-layer optimization framework, it collaboratively optimizes the DNN model's partitioning strategy, ISAC beamforming, and the allocation of computational resources at each layer to minimize the end-to-end total latency of the intelligent sensing task.

[0072] Figure 5 This is a schematic diagram of the hierarchical collaborative DNN inference and joint optimization algorithm for minimizing task latency provided by the present invention, as shown below. Figure 5 As shown. The DNN model is partitioned and executed collaboratively in a three-layer architecture. Let the first layer be... The size of the intermediate feature data output by the layer is The end-to-end inference latency of the entire intelligent sensing task for device k It can be represented as: In the formula, These represent the computational cost of each network layer; , , The computation frequency of each network layer; , , These are the computational efficiency coefficients for each network layer; For equipment Unload intermediate features to MEC server converged communication rate (subject to communication beamforming) Influence); For the backhaul rate from MEC to the cloud, This indicates the computational latency at the ISAC device end. Indicates uplink transmission delay. This indicates the computational latency of the MEC server. This indicates a delay in the return transmission. This indicates cloud computing latency.

[0073] Meanwhile, considering the energy limitations of end-side devices, its local computing power consumption... This is a key constraint, which can be modeled as: In the formula: The partitioning strategy of a DNN model determines the computational cost of each layer. and the amount of data to be uninstalled ; The power consumption factor is determined by the device's hardware architecture; This refers to the number of floating-point operations per CPU cycle of the device. Indicates equipment Computational power when performing local calculations.

[0074] This invention's embodiments find the optimal DNN partitioning strategy. (i.e., target model partitioning strategy), optimal ISAC beamforming matrix (i.e., target data acquisition signal), and optimal allocation of computing resources. (i.e., the target resource allocation strategy), thereby minimizing the weighted total latency of all monitoring tasks. This joint optimization problem can be formulated as a MINLP problem: Subject to the following constraints: (C1) DNN Partitioning Strategy Binary constraints: (C2) Maximum transmit power constraint: (C3) Perceived quality hard constraints (emission covariance matrix matching): This constraint will target the covariance matrix. This serves as a hard prerequisite, ensuring that the preset perceived performance is not sacrificed while minimizing latency.

[0075] (C4) Quality of Service (QoS) Constraints: (C5) End-side device computing capacity constraints: (C6) Total computing power constraint of MEC server: (C7) Energy consumption budget constraints for end-side equipment: In the formula, Indicates allocation to device Comprehensive priority weight; This represents the total number of intelligent ISAC devices participating in collaborative optimization within the system. Indicates deployment on the device The total number of layers contained in the sub-model; This indicates the maximum computing frequency that the edge device can achieve. Indicates the first Total computing resources owned by each MEC server; This indicates the energy consumption budget / energy consumption limit set by the end-side device.

[0076] To minimize total or average inference latency The embodiments of the present invention employ a two-layer joint optimization algorithm to collaboratively optimize the DNN partitioning strategy. ISAC beamforming matrix (Including communication beams) and sensing beam ), and computing resource allocation and .

[0077] Since the MINLP problem is NP-hard, involving coupled discrete and continuous variables, it is difficult to solve directly. This invention utilizes a hierarchical structure of variables to vertically decompose the MINLP problem into two sub-problems: an outer optimization problem for discrete DNN partitioning strategies, and an inner optimization problem for continuous resource allocation variables.

[0078] Taking DNN as an example, given the DNN partitioning strategy β, the inner layer optimization problem focuses on solving continuous variables. and The optimal allocation can be expressed as: Obey constraints (C2) to (C7).

[0079] The outer optimization problem aims to find the optimal discrete partitioning strategy. , can be represented as: It conforms to constraint (C1). In the formula, This indicates that under a given model partitioning strategy The optimal (i.e. minimized) end-to-end inference latency that can be obtained through inner-layer optimization.

[0080] Outer layer optimization aims to solve problems concerning discrete variables. This invention addresses the combinatorial optimization problem. In this embodiment, a probabilistic learning algorithm based on Hybrid Cross-Entropy-Local Search (HCE-LS) is employed to guide the search. Compared to traditional cross-entropy methods, this algorithm combines global exploration and local mining capabilities, enabling it to find high-quality solutions more efficiently.

[0081] This algorithm divides the DNN into decision variables. The model is based on a Bernoulli distribution. First, probability parameters are randomly generated based on the Bernoulli distribution. Then, the probability parameters are updated iteratively to gradually learn the optimal or near-optimal DNN partitioning strategy. The preset iterative steps include: (1) Random sampling: based on intermediate probability parameters generate A sample of candidate partitioning strategies that satisfy constraint (C1) (i.e., intermediate model partitioning strategies). (2) Performance evaluation: For each sample, the inner optimization algorithm is called to solve for its corresponding minimum latency, which is used as the performance score of the sample; (3) Sample selection: Based on performance scores, the top-performing samples are selected. (Preset number) samples as the elite set (i.e., the first sample set); (4) Local search: For each selected elite sample A local search is performed within its neighborhood. (The neighborhood is defined as a small perturbation to the original partitioning strategy.) If a better partitioning strategy (i.e., lower total latency) is found within the neighborhood, the samples in the current elite set are replaced with the better strategy to obtain the second sample set. (5) Parameter Update: Update the probability parameter vector based on the second sample set to make it more likely to generate high-quality solutions. The update rule adopts smooth update: In the formula, It is a new probability parameter calculated from the elite set. The learning rate is used. The algorithm iterates until the change in the intermediate probability parameter is within a preset range, indicating that the probability distribution has converged. The intermediate probability parameter obtained at this point is then used as the target probability parameter.

[0082] The inner layer optimization algorithm includes: For any fixed DNN partitioning strategy given in the outer layer (such as the intermediate model partitioning strategy in outer layer optimization) Under the condition that it has been determined, the goal of inner-layer optimization is to find the best continuous resource allocation scheme (i.e., the corresponding intermediate resource allocation strategy) to minimize the task latency under this partition.

[0083] (1) Calculate resource allocation and For a fixed amount of computational task The computational latency minimization problem is a standard convex optimization problem. This invention derives the optimal solution using the Karush-Kuhn-Tucker (KKT) conditions. For local computing resources, the optimal resource allocation strategy... Determined by both local maximum computing power and the newly added energy consumption constraint (C7): In the formula, This indicates the computational efficiency of the edge device hardware, specifically the number of floating-point operations that the device can perform per CPU cycle.

[0084] (2) ISAC beamforming optimization Under the premise of satisfying the hard constraints of perception (C3) and the QoS constraints of communication (C4) (i.e., ensuring perception quality), optimize the communication beamforming. To maximize the transmission rate of intermediate features This minimizes the uninstallation delay. .

[0085] This invention employs an iterative algorithm based on the Majorization-Minimization (MM) framework, combined with the Weighted Minimum Mean Square Error (WMMSE) method, to transform the non-convex rate maximization problem into an iteratively solvable mean square error minimization problem. During the iteration process, perceptual constraints are efficiently handled, ultimately converging to a high-quality local optimum. This refers to the intermediate data acquisition signal corresponding to the intermediate model partitioning strategy and the intermediate resource allocation strategy.

[0086] Through iterative collaboration between inner and outer layer algorithms, complex MINLP problems can be solved efficiently, achieving tight coupling and global performance optimization of DNN model partitioning, inductive beamforming, and computational resource allocation, ultimately achieving accurate perception of transmission line status with minimal end-to-end delay.

[0087] Figure 6 This is a schematic diagram illustrating the convergence of the inner and outer layer joint optimization algorithm provided by this invention, as shown below. Figure 6 As shown in the embodiment of the above-mentioned scenario of real-time capture of wind-induced galloping of transmission lines and concurrent foreign object intrusion monitoring, the target model partitioning strategy, target resource allocation strategy, and target data acquisition signal are determined through preset iterative steps using the inner and outer layer joint optimization algorithm provided by this invention. The curve shows that after the algorithm starts iterating, the total delay decreases rapidly and converges to a stable low value within a relatively small number of iterations. This proves that through the above-mentioned preset iterative steps, the optimal target model partitioning strategy, target resource allocation strategy, and target data acquisition signal can be accurately and quickly determined, thereby reducing transmission delay and ensuring that galloping risks are captured in a timely manner.

[0088] This invention employs an iterative optimization method to treat the three core variable factors affecting latency (target model partitioning strategy, target resource allocation strategy, and target data acquisition signal) as a unified, coupled whole for collaborative optimization. Simultaneously, it transforms a complex joint optimization problem that is difficult to solve directly into a feasible problem that can iteratively approach the optimal solution. This allows for the search for the optimal solution within a global or near-global scope, rather than getting trapped in local optima. Therefore, it ensures that the final determined strategy combination minimizes task latency, achieving low-latency, high-efficiency, intelligent, and precise monitoring of transmission line status.

[0089] Furthermore, the step of executing each monitoring task in parallel based on the target model partitioning strategy, the target resource allocation strategy, and the target data acquisition signal to obtain the state monitoring results of the transmission line includes: Based on the target model partitioning strategy, the deep learning model corresponding to each monitoring task is divided into a first target sub-model, a second target sub-model, and a third target sub-model; the first target sub-model is used to perform the data filtering sub-task; the second target sub-model is used to perform the risk prediction sub-task; and the third target sub-model is used to perform the comprehensive analysis sub-task. Based on the target resource allocation strategy, corresponding first server computing resources, second server computing resources and third server computing resources are allocated to the first target sub-model, the second target sub-model and the third target sub-model, respectively. Based on the target data acquisition signal, the sensing data required for each monitoring task are collected respectively; The first target sub-model and the first server computing resources are used to filter the perceived data to obtain key features; Risk prediction is performed based on the key features using the second target sub-model and the second server computing resources to obtain prediction results; The prediction results are comprehensively analyzed using the third target sub-model and the third server computing resources to obtain the monitoring results corresponding to each monitoring task. The monitoring results corresponding to each monitoring task are used as the status monitoring results of the transmission line.

[0090] Specifically, after the iterative optimization was completed, the target model partitioning strategy, the target resource allocation strategy, and the target data acquisition signal were obtained.

[0091] In actual transmission line condition monitoring, the deep learning model is first divided into three target sub-models according to the target model partitioning strategy. These sub-models are then deployed at the edge, cloud, and terminal levels to collaboratively execute data filtering, risk prediction, and comprehensive analysis sub-tasks. Simultaneously, server computing resources are allocated to the edge, cloud, and the three sub-models according to the target resource allocation strategy. Next, target data acquisition signals are transmitted via ISAC devices, simultaneously acquiring the sensing data required for each monitoring task with a single signal transmission. The first target sub-model then extracts and filters features from the sensing data to obtain key features. These key features are then transmitted to the MEC server via the ISAC devices, where the second target sub-model performs risk prediction based on these key features, obtaining the prediction results. Finally, the prediction results are uploaded to the cloud, where the third target sub-model performs comprehensive analysis to output the transmission line condition monitoring results.

[0092] This invention achieves low-latency, high-precision, and intelligent monitoring of transmission line status through an end-to-end efficient workflow with minimal latency and optimal resource utilization.

[0093] The following describes the transmission line condition monitoring device based on sensor fusion provided by the present invention. The transmission line condition monitoring device based on sensor fusion described below and the transmission line condition monitoring method based on sensor fusion described above can be referred to in correspondence.

[0094] Based on any of the above embodiments Figure 7 This is a schematic diagram of the structure of the transmission line condition monitoring device based on sensor fusion provided by the present invention, as shown below. Figure 7 As shown. This embodiment of the invention provides a transmission line status monitoring device based on sensor fusion, including a generation module 701, a distribution module 702, and a monitoring module 703, wherein: The generation module 701 is used to generate multiple different monitoring tasks based on the current state of the transmission line; the monitoring tasks include data filtering sub-tasks, risk prediction sub-tasks, and comprehensive analysis sub-tasks. The allocation module 702 is used to determine the target model partitioning strategy, target resource allocation strategy, and target data acquisition signal based on the computational requirements of each monitoring task; The monitoring module 703 is used to execute each monitoring task in parallel based on the target model partitioning strategy, the target resource allocation strategy, and the target data acquisition signal to obtain the status monitoring results of the transmission line.

[0095] The transmission line condition monitoring device based on sensor fusion provided by this invention generates multiple different monitoring tasks based on the current state of the transmission line. These monitoring tasks include data filtering sub-tasks, risk prediction sub-tasks, and comprehensive analysis sub-tasks, thereby achieving intelligent and differentiated task division from a fixed mode to a dynamic adaptive mode, making the monitoring tasks more targeted and efficient. Based on the computational requirements of each monitoring task, the device determines the target model partitioning strategy, target resource allocation strategy, and target data acquisition signal. This pre-planning of the deep learning model's segmentation method and sensory computing power (communication, sensing, and computing) resources provides a model and resource foundation for subsequent distributed task parallel execution. Based on the target model partitioning strategy, target resource allocation strategy, and target data acquisition signal, each monitoring task is executed in parallel to obtain the transmission line condition monitoring results. Parallel processing significantly shortens the total time from data acquisition to obtaining the final result, avoiding the time delay and resource waste of traditional time-division or frequency-division multi-task modes. This achieves low-latency, high-efficiency, intelligent, and accurate monitoring of the transmission line condition, thereby improving power grid security.

[0096] Figure 8 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 8 As shown, the electronic device may include: a processor 810, a communications interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communications interface 820, and the memory 830 communicate with each other via the communication bus 840. The processor 810 can call logical instructions in the memory 830 to execute a transmission line condition monitoring method based on sensing fusion, the method including: Based on the current status of the transmission line, multiple different monitoring tasks are generated; the monitoring tasks include data filtering sub-tasks, risk prediction sub-tasks, and comprehensive analysis sub-tasks. Based on the computational requirements of each monitoring task, the target model partitioning strategy, target resource allocation strategy, and target data acquisition signals are determined. Based on the target model partitioning strategy, the target resource allocation strategy, and the target data acquisition signal, each monitoring task is executed in parallel to obtain the status monitoring results of the transmission line.

[0097] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0098] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the transmission line condition monitoring method based on sensing fusion provided by the above methods, the method comprising: Based on the current status of the transmission line, multiple different monitoring tasks are generated; the monitoring tasks include data filtering sub-tasks, risk prediction sub-tasks, and comprehensive analysis sub-tasks. Based on the computational requirements of each monitoring task, the target model partitioning strategy, target resource allocation strategy, and target data acquisition signals are determined. Based on the target model partitioning strategy, the target resource allocation strategy, and the target data acquisition signal, each monitoring task is executed in parallel to obtain the status monitoring results of the transmission line.

[0099] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the transmission line condition monitoring method based on sensor fusion provided by the above methods, the method comprising: Based on the current status of the transmission line, multiple different monitoring tasks are generated; the monitoring tasks include data filtering sub-tasks, risk prediction sub-tasks, and comprehensive analysis sub-tasks. Based on the computational requirements of each monitoring task, the target model partitioning strategy, target resource allocation strategy, and target data acquisition signals are determined. Based on the target model partitioning strategy, the target resource allocation strategy, and the target data acquisition signal, each monitoring task is executed in parallel to obtain the status monitoring results of the transmission line.

[0100] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0101] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0102] It should be noted that, in this invention, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this invention is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0103] It should also be noted that the terms "target," "first," and "second" in this invention are used to distinguish similar objects, not to describe a specific order or sequence. It should be understood that such terms can be used interchangeably where appropriate so that embodiments of the invention can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first" and "second" are generally of the same class, without limiting the number of objects; for example, the first object can be one or more.

[0104] In this embodiment of the invention, "determine B based on A" means that factor A must be considered when determining B. It is not limited to "B can be determined based solely on A," but should also include: "determine B based on A and C," "determine B based on A, C, and E," "determine C based on A, and further determine B based on C," etc. Additionally, it can include using A as a condition for determining B, for example, "when A meets the first condition, determine B using the first method"; or "when A meets the second condition, determine B," etc.; or "when A meets the third condition, determine B based on the first parameter," etc. Of course, it can also be a condition where A is a factor in determining B, for example, "when A meets the first condition, determine C using the first method, and further determine B based on C," etc.

[0105] In this invention, the term "multiple" refers to two or more, and other quantifiers are similar.

[0106] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for monitoring the condition of transmission lines based on sensor fusion, characterized in that, include: Based on the current state of the transmission line, multiple different monitoring tasks are generated; The monitoring task includes data filtering sub-tasks, risk prediction sub-tasks, and comprehensive analysis sub-tasks. Based on the computational requirements of each monitoring task, the target model partitioning strategy, target resource allocation strategy, and target data acquisition signals are determined. Based on the target model partitioning strategy, the target resource allocation strategy, and the target data acquisition signal, each monitoring task is executed in parallel to obtain the status monitoring results of the transmission line.

2. The method for monitoring the condition of transmission lines based on sensor fusion according to claim 1, characterized in that, Based on the current state of the transmission line, several different monitoring tasks are generated, including: Based on the current state of the transmission line, determine the potential risk category of the transmission line; Based on the potential risk categories, the multiple different monitoring tasks are generated.

3. The method for monitoring the condition of transmission lines based on sensor fusion according to claim 1, characterized in that, The determination of the target model partitioning strategy, target resource allocation strategy, and target data acquisition signals based on the computational requirements of each monitoring task includes: Based on the computational requirements of each monitoring task, the preset iterative steps are repeatedly executed to update the intermediate probability parameters until the change range of the intermediate probability parameters is within the preset range. The intermediate probability parameters obtained in the last iteration are then used as the target probability parameters. Based on the target probability parameters, a target model partitioning strategy is generated for each monitoring task; Based on the target model partitioning strategy, determine the target resource allocation strategy for each monitoring task; Based on the target model partitioning strategy and the target resource allocation strategy, the target data acquisition signal corresponding to each monitoring task is determined.

4. The method for monitoring the condition of transmission lines based on sensor fusion according to claim 3, characterized in that, The iterative steps include: Based on the intermediate probability parameters obtained in the previous iteration, the intermediate model partitioning strategy obtained in the previous iteration is updated to obtain the intermediate model partitioning strategy obtained in this iteration; wherein, each intermediate probability parameter corresponds to each monitoring task; each monitoring task corresponds to multiple intermediate model partitioning strategies; Based on the intermediate model partitioning strategy obtained in this iteration, an intermediate resource allocation strategy obtained in this iteration is generated. Based on the intermediate model partitioning strategy and intermediate resource allocation strategy obtained in this iteration, the intermediate data acquisition signal obtained in the previous iteration is updated to obtain the intermediate data acquisition signal obtained in this iteration. Based on the intermediate model partitioning strategy, the intermediate resource allocation strategy, and the intermediate data acquisition signals obtained in this iteration, each monitoring task is executed in parallel to obtain the corresponding task execution time. Based on the task execution time, a preset number of intermediate model partitioning strategies are selected from the intermediate model partitioning strategies obtained in this round of iteration as the first sample set; For each intermediate model partitioning strategy in the strategy sample set, a neighborhood search optimization is performed to obtain a second sample set; Based on the second sample set, the intermediate probability parameters obtained in the previous iteration are updated to obtain the intermediate probability parameters obtained in the current iteration. The intermediate probability parameters used in the first round of iterations are randomly generated based on the Bernoulli distribution; the intermediate data acquisition signals used in the first round of iterations are determined based on the data acquisition requirements and priorities of each monitoring task; and the priorities are determined based on preset rules.

5. The transmission line condition monitoring method based on sensor fusion as described in claim 4, characterized in that, The intermediate model partitioning strategy obtained from the current iteration generates an intermediate resource allocation strategy for the current iteration, including: Based on the intermediate model partitioning strategy obtained in this iteration, the deep learning model corresponding to each monitoring task is divided into a first intermediate sub-model, a second intermediate sub-model, and a third intermediate sub-model; the first intermediate sub-model is used to perform the data filtering sub-task; the second intermediate sub-model is used to perform the risk prediction sub-task; and the third intermediate sub-model is used to perform the comprehensive analysis sub-task. Based on the computational volume corresponding to the first intermediate sub-model, the second intermediate sub-model, and the third intermediate sub-model, the server computing resource requirements corresponding to the first intermediate sub-model, the second intermediate sub-model, and the third intermediate sub-model are determined. Based on the server's computing resource requirements, the intermediate resource allocation strategy obtained in this iteration is generated.

6. The method for monitoring the condition of transmission lines based on sensor fusion as described in claim 5, characterized in that, The method of executing each monitoring task in parallel based on the target model partitioning strategy, the target resource allocation strategy, and the target data acquisition signal to obtain the status monitoring results of the transmission line includes: Based on the target model partitioning strategy, the deep learning model corresponding to each monitoring task is divided into a first target sub-model, a second target sub-model, and a third target sub-model; the first target sub-model is used to perform the data filtering sub-task; the second target sub-model is used to perform the risk prediction sub-task; and the third target sub-model is used to perform the comprehensive analysis sub-task. Based on the target resource allocation strategy, corresponding first server computing resources, second server computing resources and third server computing resources are allocated to the first target sub-model, the second target sub-model and the third target sub-model, respectively. Based on the target data acquisition signal, the sensing data required for each monitoring task are collected respectively; The first target sub-model and the first server computing resources are used to filter the perceived data to obtain key features; Risk prediction is performed based on the key features using the second target sub-model and the second server computing resources to obtain prediction results; The prediction results are comprehensively analyzed using the third target sub-model and the third server computing resources to obtain the monitoring results corresponding to each monitoring task. The monitoring results corresponding to each monitoring task are used as the status monitoring results of the transmission line.

7. A transmission line condition monitoring device based on sensor fusion, characterized in that, include: The generation module is used to generate multiple different monitoring tasks based on the current state of the transmission line; The monitoring task includes data filtering sub-tasks, risk prediction sub-tasks, and comprehensive analysis sub-tasks. The allocation module is used to determine the target model partitioning strategy, target resource allocation strategy, and target data acquisition signals based on the computational requirements of each monitoring task. The monitoring module is used to execute each monitoring task in parallel based on the target model partitioning strategy, the target resource allocation strategy, and the target data acquisition signal to obtain the status monitoring results of the transmission line.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the transmission line condition monitoring method based on sensor fusion as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the transmission line condition monitoring method based on sensor fusion as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the transmission line condition monitoring method based on sensor fusion as described in any one of claims 1 to 6.

Images