Inspection path determination method and device for electric power inspection and electronic equipment

Abstract

The invention discloses an inspection path determination method and device for electric power inspection and electronic equipment. The method comprises the steps that inspection task information of a plurality of inspection target objects to be subjected to electric power inspection is acquired, the inspection task information at least comprises position information and inspection duration information, and the position information is used for representing the positions of the inspection target objects; an inspection path is determined based on the inspection task information by using a pre-trained deep learning model, an electric power inspection path is obtained, the pre-trained deep learning model at least comprises a first network model and a second network model, the input end of the second network model is connected with the output end of the first network model, and the input end of the second network model is connected with the output end of the second network model; the parameter number of the first network model is greater than that of the second network model, and the inspection starting point and the inspection ending point in the power inspection path are at the same position. According to the invention, the technical problem of low efficiency of determining the electric power inspection path in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the field of power inspection, and in particular to a method, device and electronic equipment for determining an inspection path for power inspection. Background Art

[0002] In the power sector, determining inspection routes for power equipment, lines, and various facilities—that is, determining power inspection paths—is crucial. By establishing reasonable inspection paths, we can effectively monitor the operating status of the power system, promptly identify and resolve potential faults and safety hazards, and ensure the safe and stable operation of the power system. Determining inspection paths typically requires considering factors such as the importance of the equipment, vulnerable parts, frequency of use, and the surrounding environment to ensure comprehensive and efficient inspections.

[0003] At present, related technologies usually use traditional optimization algorithms such as dynamic programming and greedy algorithms to determine power inspection paths. These algorithms are suitable for scenarios with a small number of target points, such as inspections of a small number of substations. Greedy algorithms can be used to quickly find suboptimal paths, but when the number of target points increases, the complexity of dynamic programming and exhaustive methods increases exponentially, which is not suitable for solving large-scale problems. Or traditional deep reinforcement learning may be required. The training process of traditional deep learning models is complex and time-consuming, and requires a large amount of historical data for model training. This is suitable for dynamically changing inspection scenarios, but model training requires a large amount of data and time, and the cost is high in the early stages of system development, which results in low efficiency in determining power inspection paths in related technologies.

[0004] To address the above-mentioned problems, no effective solutions have been proposed so far. Summary of the Invention

[0005] Embodiments of the present invention provide a method, device, and electronic device for determining an inspection path for power inspection, so as to at least solve the technical problem of low efficiency in determining the power inspection path in related technologies.

[0006] According to one aspect of an embodiment of the present invention, a method for determining an inspection path for power inspection is provided, comprising: obtaining inspection task information of multiple inspection target objects to be inspected for power, wherein the inspection task information includes at least location information and inspection duration information, the location information is used to characterize the locations of the multiple inspection target objects, and the inspection duration information is used to determine the inspection duration for inspecting the multiple inspection target objects; utilizing a pre-trained deep learning model to determine the inspection path based on the inspection task information to obtain a power inspection path, wherein the pre-trained deep learning model includes at least a first network model and a second network model, the input end of the second network model is connected to the output end of the first network model, the number of parameters of the first network model is greater than the number of parameters of the second network model, and the inspection starting point and the inspection end point in the power inspection path are at the same position.

[0007] Optionally, the model parameters of the second network model are determined based on the target loss function and a preset amplitude calculation function, and the target loss function is determined based on the first output information and the second output information, wherein the first output information is used to characterize the output information of the first network model, and the second output information is used to characterize the output information of the second network model.

[0008] Optionally, the method also includes: obtaining a target divergence between the first output information and the second output information, wherein the target divergence is used to characterize the similarity between the first output information and the second output information; constructing a target loss function based on the target divergence and preset adjustment parameters, wherein the preset adjustment parameters are used to control the optimization direction of the pre-trained deep learning model.

[0009] Optionally, based on the target divergence and preset adjustment parameters, a target loss function is constructed, including: obtaining a target cross entropy loss, wherein the target cross entropy loss is used to represent the cross entropy loss between the second network model and a preset true label, and the preset true label is used to represent the reward signal provided by the environment during reinforcement learning of the pre-trained deep learning model; based on the target cross entropy loss, the target divergence and the preset adjustment parameters, a target loss function is constructed.

[0010] Optionally, based on the target cross entropy loss, target divergence and preset adjustment parameters, a target loss function is constructed, including: obtaining the complement of the preset adjustment parameter; determining the product of the preset adjustment parameter and the target cross entropy loss to obtain a first product; determining the product of the complement and the target divergence to obtain a second product; and determining the target loss function based on the sum of the first product and the second product.

[0011] Optionally, the method also includes: sending the power inspection path to a mobile terminal; using the power inspection path received by the mobile terminal to guide at least one inspection personnel to perform power inspection and obtain inspection results, wherein the inspection results are used to characterize the operating status of multiple inspection target objects.

[0012] According to another aspect of an embodiment of the present invention, a patrol path determination device for power inspection is also provided, including: an acquisition module for acquiring patrol task information of multiple patrol target objects to be inspected for power, wherein the patrol task information includes at least location information and patrol duration information, the location information is used to characterize the locations of the multiple patrol target objects, and the patrol duration information is used to determine the patrol duration for inspecting the multiple patrol target objects; a determination module for determining the patrol path based on the patrol task information using a pre-trained deep learning model to obtain a power patrol path, wherein the pre-trained deep learning model includes at least a first network model and a second network model, the input end of the second network model is connected to the output end of the first network model, the number of parameters of the first network model is greater than the number of parameters of the second network model, and the patrol start point and the patrol end point in the power patrol path are at the same position.

[0013] According to another aspect of an embodiment of the present invention, an electronic device is provided, including: a memory storing an executable program; and a processor for running the program, wherein the program executes the methods of various embodiments of the present invention when running.

[0014] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is provided. The computer-readable storage medium includes a stored executable program, wherein when the executable program is running, the device where the computer-readable storage medium is located is controlled to execute the methods in various embodiments of the present invention.

[0015] According to another aspect of an embodiment of the present invention, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the method in each embodiment of the present invention is implemented.

[0016] According to another aspect of an embodiment of the present invention, a computer program product is provided, including a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method in each embodiment of the present invention is implemented.

[0017] According to another aspect of the embodiments of the present invention, a computer program is provided. When the computer program is executed by a processor, the methods in various embodiments of the present invention are implemented.

[0018] In an embodiment of the present invention, a method for determining an inspection path for power inspection is provided, including: obtaining inspection task information of multiple inspection target objects to be inspected for power, wherein the inspection task information includes at least location information and inspection duration information, the location information is used to characterize the locations of the multiple inspection target objects, and the inspection duration information is used to determine the inspection duration for inspecting the multiple inspection target objects; using a pre-trained deep learning model to determine the inspection path based on the inspection task information to obtain a power inspection path, wherein the pre-trained deep learning model includes at least a first network model and a second network model, the input end of the second network model is connected to the output end of the first network model, the number of parameters of the first network model is greater than the number of parameters of the second network model, and the inspection starting point and the inspection end point in the power inspection path are at the same position. It is easy to note that the present application uses a pre-trained deep learning model to determine the inspection path based on the inspection task information. By setting the pre-trained deep learning model to include at least a first network model and a second network model, the input end of the second network model is connected to the output end of the first network model, and the number of parameters of the first network model is greater than the number of parameters of the second network model. The first network model has a larger number of parameters and can learn more complex and deep feature representations, thereby providing more accurate prediction results. The second network model has a simpler network structure and fewer parameters, and has a lower computational cost. The second network model learns the solution to the complex task by receiving the output of the first network model, realizing knowledge transfer and extraction. In this way, the second network model can converge to the optimal solution faster when determining the inspection path, and can better utilize existing knowledge and experience. The collaboration between the first network model and the second network model can realize knowledge transfer and sharing, improve the generalization ability and efficiency of the model, and thus improve the efficiency and accuracy of path determination, which can effectively improve the efficiency and accuracy of power inspections, provide better protection for the safe operation of power equipment, and thus solve the technical problem of low efficiency in power inspection path determination in related technologies. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0020] Figure 1 This is a flow chart of a method for determining an inspection path for power inspection according to an embodiment of the present invention;

[0021] Figure 2 is a flow chart of an optional method for determining an inspection path for power inspection according to an embodiment of the present invention;

[0022] Figure 32 is a schematic diagram of a device for determining an inspection path for power inspection according to an embodiment of the present invention. DETAILED DESCRIPTION

[0023] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0024] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0025] According to one aspect of an embodiment of the present invention, a method for predicting over-limit of substation voltage is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0026] Figure 1 FIG. 1 is a flow chart of a method for determining an inspection path for power inspection according to an embodiment of the present invention. Figure 1 As shown, the method includes the following steps:

[0027] Step S102, obtain inspection task information of multiple inspection target objects to be inspected for power, wherein the inspection task information includes at least location information and inspection duration information, the location information is used to characterize the locations of the multiple inspection target objects, and the inspection duration information is used to determine the inspection duration for the multiple inspection target objects.

[0028] The above-mentioned power inspection may refer to a work of inspecting, checking and evaluating power equipment, lines and power systems, which can timely discover and solve faults, hidden dangers and problems of power equipment, and ensure the safe, stable and efficient operation of the power system. Specifically, power inspection may include but is not limited to visual inspection of power equipment, checking whether the surface of the equipment is damaged, rusted, loose, etc., to ensure that the appearance of the equipment is intact; operating status inspection of power equipment, checking the operating status, temperature, sound and other parameters of the equipment, to ensure the normal operation of the equipment; inspecting power lines, checking whether the insulation of the power lines is intact, whether there is any damage and leakage, to ensure the safety and reliability of the power lines; inspecting the power system, checking the grounding condition of the power system, whether the protection device is operating normally, etc., to ensure the stable operation of the power system. Power inspection can be carried out through manual inspection, drone inspection, infrared thermal imager inspection, etc. The specific method of power inspection can be determined according to actual needs and is not limited here.

[0029] The aforementioned inspection targets may refer to power equipment, power lines, substations, and other facilities and equipment related to power generation, transmission, and distribution. Specifically, these may include, but are not limited to, power equipment, including generators, transformers, switchgear, distribution cabinets, and other types of electrical equipment; power lines, including transmission lines, distribution lines, grounding wires, and other types of power line facilities; and substations, including substation equipment, control equipment, and monitoring equipment. Inspections may primarily include visual inspections of equipment, performance testing, temperature measurement, electrical parameter measurement, and other aspects of testing. Specific inspection targets can be determined based on actual needs and are not limited here.

[0030] The above-mentioned inspection task information may refer to the inspection demand or inspection requirement information for power inspection, which may include but is not limited to the location information, inspection duration information, inspection content, etc. of multiple inspection target objects. Among them, the location information may include the specific location of the inspection target object, which may be latitude and longitude coordinates, address information or a specific area range. The location information can be used to guide inspection personnel to reach the location of the inspection target object. The inspection duration information may indicate the time required to inspect each inspection target object, which may be an estimated time range or a specific time period. The inspection duration information is used to reasonably arrange the time of the inspection work and ensure that all inspection target objects can be inspected in a timely manner. The inspection content may describe the specific inspection content and requirements that need to be carried out on each inspection target object, including inspection items, inspection methods, inspection standards, etc. The inspection task information can be determined based on actual needs and is not limited here.

[0031] In an optional embodiment, the inspection target objects that need to be inspected can first be determined. The inspection target objects can be various power equipment, such as transformers, cables, switchgear, etc. Based on the actual situation and inspection requirements, multiple inspection target objects that need to be inspected can be determined. For each inspection target object, its location information can be obtained. The precise location information of the inspection target object can be obtained through positioning systems, map information, etc. This location information is the basis for determining the inspection path and is used to determine the distance and relative position relationship between inspection target objects. For each inspection target object, the inspection duration can be determined. The inspection duration can be determined based on factors such as the type of equipment, the complexity of the inspection content, and the inspection frequency. The inspection duration can be estimated based on historical data or professional experience. Next, the inspection task information can be integrated. The collected location information and inspection duration information can be integrated into the inspection task information to ensure that each inspection target object has corresponding location information and inspection duration information, which facilitates subsequent path planning and inspection plan formulation. Through the specific implementation of the above steps, the inspection task information of multiple inspection target objects to be inspected can be obtained, which provides effective support and basis for subsequent inspection path determination and inspection plan formulation, can improve the efficiency and accuracy of power inspection, and ensure the safe operation and reliability of power equipment.

[0032] Step S104: Use a pre-trained deep learning model to determine the inspection path based on the inspection task information to obtain a power inspection path, wherein the pre-trained deep learning model includes at least a first network model and a second network model, the input end of the second network model is connected to the output end of the first network model, the number of parameters of the first network model is greater than the number of parameters of the second network model, and the inspection starting point and inspection end point in the power inspection path are at the same position.

[0033] The above-mentioned pre-trained deep learning model is pre-trained and can be used to process inspection task information to perform path planning. The specific type of the pre-trained deep learning model can be determined according to actual needs and is not limited here.

[0034] In an optional embodiment, a pre-trained deep learning model can be used to determine the inspection route for power inspections. First, power inspection task information, including data on inspection equipment, equipment location, inspection frequency, and so on, can be collected and annotated to form a training dataset. A suitable deep learning model, such as a convolutional neural network or a recurrent neural network, can be selected for pre-training. Pre-training can train the model using a large amount of data, enabling it to learn the characteristics and patterns of the data. Next, a first network model and a second network model can be constructed, with the input of the second network model connected to the output of the first network model, and the number of parameters of the first network model greater than that of the second network model. Finally, the pre-trained deep learning model can be applied to the inspection task information, and the power inspection route can be determined based on the patterns and characteristics learned by the model. When determining the route, the inspection start and end points must be aligned to ensure the integrity and accuracy of the inspection task. The deep learning model can also be optimized and adjusted based on actual application scenarios to improve the accuracy and efficiency of inspection route determination.

[0035] The first network model and the second network model in this application may refer to two neural network models of different scales. The first network model may adopt a complex neural network model with a large number of parameters and a deeper number of layers, and the second network model may adopt a simple neural network model with a smaller number of parameters and a shallower number of layers. In the pre-trained deep learning model, the first network model can provide more accurate output labels or probability distributions to guide the learning of the second network model. By training on a large-scale data set, the first network model can learn more complex and deep feature representations, thereby providing more accurate prediction results. The second network model has a simpler network structure and fewer parameters, and has lower computational cost. The second network model learns solutions to complex tasks by receiving the output of the first network model as a supervisory signal, thereby realizing the transfer and extraction of knowledge. The collaboration between the first network model and the second network model can realize the transfer and sharing of knowledge, thereby improving the generalization ability and efficiency of the model.

[0036] This application utilizes a pre-trained deep learning model to determine inspection paths based on inspection task information. By configuring the pre-trained deep learning model to include at least a first network model and a second network model, with the input of the second network model connected to the output of the first network model, and the number of parameters in the first network model being greater than that of the second network model, the knowledge of the first network model can be transferred to the second network model, thereby accelerating the training process of the second network model and improving the model's generalization and performance. Specifically, the first network model can be a deep learning model pre-trained on a large-scale dataset. The first network model has learned general feature representations and is well-suited for power inspection tasks. The second network model can be a deep learning model fine-tuned for a specific inspection task to better adapt to specific inspection scenarios and requirements. Connecting the input of the second network model to the output of the first network model allows the second network model to directly benefit from the knowledge learned by the first network model. This allows the second network model to converge to the optimal solution more quickly when determining the inspection path, and can better leverage existing knowledge and experience to improve the efficiency and accuracy of path determination. This can effectively improve the efficiency and accuracy of power inspections and provide better protection for the safe operation of power equipment.

[0037] Optionally, the model parameters of the second network model are determined based on the target loss function and a preset amplitude calculation function, and the target loss function is determined based on the first output information and the second output information, wherein the first output information is used to characterize the output information of the first network model, and the second output information is used to characterize the output information of the second network model.

[0038] The above-mentioned preset argument calculation function may refer to a pre-set function for calculating the argument of a complex number, preferably, it may be an arg() function, etc. The preset argument calculation function may be determined according to actual needs and is not limited here.

[0039] In an optional embodiment, a second network model can be trained, with its input connected to the output of the first network model to utilize the feature information learned by the first network model. The model parameters of the second network model can be determined based on a target loss function and a preset amplitude calculation function. The target loss function is determined based on first output information and second output information, where the first output information is used to characterize the output information of the first network model, and the second output information is used to characterize the output information of the second network model. The parameters of the second network model can be updated using a backpropagation algorithm to minimize the loss function. After training is complete, this pre-trained deep learning model can be used to determine the power inspection path, obtain a suitable inspection plan, and improve inspection efficiency and accuracy.

[0040] Optionally, the method also includes: obtaining a target divergence between the first output information and the second output information, wherein the target divergence is used to characterize the similarity between the first output information and the second output information; constructing a target loss function based on the target divergence and preset adjustment parameters, wherein the preset adjustment parameters are used to control the optimization direction of the pre-trained deep learning model.

[0041] In an optional embodiment, a target divergence between the first output information and the second output information can be obtained, and a calculation method for the target divergence can be defined. The target divergence can use a variety of distance measurement methods to calculate the similarity between the first output information and the second output information, such as Euclidean distance, cosine similarity, etc., and a suitable distance measurement method can be selected to determine the similarity between the first output information and the second output information. Then, a target loss function can be constructed based on the target divergence and preset adjustment parameters. The target loss function can control the direction of optimization of the model during the training process by adjusting the parameters, so that the similarity between the first output information and the second output information is effectively enhanced. During the training process, the deep learning model can be optimized by minimizing the target loss function, so that the model can better learn the association between the first output information and the second output information, thereby improving the performance and generalization ability of the model. By defining the target divergence calculation method, constructing the target loss function, and adjusting the parameters to control the optimization direction of the model, it is possible to determine the inspection path of the power inspection and obtain the similarity between the first output information and the second output information, thereby improving the training effect and performance of the model.

[0042] Optionally, based on the target divergence and preset adjustment parameters, a target loss function is constructed, including: obtaining a target cross entropy loss, wherein the target cross entropy loss is used to represent the cross entropy loss between the second network model and a preset true label, and the preset true label is used to represent the reward signal provided by the environment during reinforcement learning of the pre-trained deep learning model; based on the target cross entropy loss, the target divergence and the preset adjustment parameters, a target loss function is constructed.

[0043] In an optional embodiment, a target cross-entropy loss can be obtained, and the cross-entropy loss between the second network model and the preset true label can be defined. This can be achieved by calculating the cross-entropy loss between the output of the second network model and the preset true label. The preset true label refers to the reward signal provided by the environment during the reinforcement learning process. The preset true label can be defined based on the specific requirements of the power inspection task and environmental feedback. Next, a target loss function can be constructed based on the target cross-entropy loss, target divergence, and preset adjustment parameters. The target divergence refers to the difference measure between the output distribution of the target network and the output distribution of the second network. By minimizing the target divergence, the target network can better fit the second network model. Finally, the target cross-entropy loss, target divergence, and preset adjustment parameters can be combined to construct a target loss function. The design of the target loss function can effectively guide the deep reinforcement learning model to learn better strategies and path planning in the power inspection task. Through the above steps, the target cross-entropy loss can be obtained and the target loss function can be constructed, thereby realizing the training and optimization of the pre-trained deep reinforcement learning model in the power inspection task.

[0044] Optionally, based on the target cross entropy loss, target divergence and preset adjustment parameters, a target loss function is constructed, including: obtaining the complement of the preset adjustment parameter; determining the product of the preset adjustment parameter and the target cross entropy loss to obtain a first product; determining the product of the complement and the target divergence to obtain a second product; and determining the target loss function based on the sum of the first product and the second product.

[0045] The aforementioned complement may refer to a complement operation in mathematical operations.

[0046] In an optional embodiment, a target loss function can be constructed. Specifically, a target cross-entropy loss function and a target divergence function can be defined. The target cross-entropy loss function measures the difference between the model's prediction and the target, while the target divergence function measures the difference between the probability distribution output by the model and the target probability distribution. The complement of a preset adjustment parameter can be obtained by operating on the preset adjustment parameter, such as by taking its complement. Next, the product of the preset adjustment parameter and the target cross-entropy loss can be determined to obtain a first product. This step can be achieved by simply multiplying the preset adjustment parameter by the target cross-entropy loss. Next, the product of the complement and the target divergence can be determined to obtain a second product. This step can be achieved by multiplying the complement by the target divergence. Finally, the target loss function can be determined based on the sum of the first and second products. The value of the target loss function can be obtained by adding the first and second products. Through the above steps, a target loss function can be constructed that includes the target cross-entropy loss, target divergence, and preset adjustment parameters to optimize the performance of deep reinforcement learning models in patrol path determination for power inspections.

[0047] Optionally, the method also includes: sending the power inspection path to a mobile terminal; using the power inspection path received by the mobile terminal to guide at least one inspection personnel to perform power inspection and obtain inspection results, wherein the inspection results are used to characterize the operating status of multiple inspection target objects.

[0048] In an optional embodiment, during the actual operation of a power inspection, the power inspection route can be sent to a mobile terminal by uploading the power inspection plan and route data to a cloud server. The mobile terminal can then access the cloud server via a network connection to obtain the latest power inspection route information. After receiving the power inspection route, the mobile terminal can display the entire inspection route on a map and, in conjunction with the navigation system, provide real-time guidance to inspectors along the route. While using the inspection route received by the mobile terminal to guide inspectors during power inspections, inspectors can use the mobile terminal application to record inspection results for each inspection point, including equipment operating status, abnormalities, and issues requiring repair. These inspection results are uploaded to the cloud server in real time, allowing monitoring personnel to view and analyze them in real time. After obtaining the inspection results, they can be analyzed and processed using a deep reinforcement learning model, which can characterize the operating status of multiple inspection targets. By learning from a large amount of inspection data and results, the deep reinforcement learning model can automatically identify equipment health, potential problems, and early warning information, helping managers make timely decisions and take measures to ensure the safe operation of power equipment. The sending, receiving and guidance of power inspection routes can be achieved through cloud servers and mobile terminals. The deep reinforcement learning model can analyze and characterize the inspection results, providing a scientific basis and decision-making support for the management and maintenance of power equipment.

[0049] The following is an explanation of the technical solution proposed in this application in combination with an optional application scenario. In the power industry, when it comes to the positioning and inspection of power rooms, meters, and substations, challenges similar to the "travelling salesman problem" are often encountered. The traveling salesman has to visit n cities, and the distance between each city is known. The traveling salesman has to find the shortest path so that each city is visited only once and finally returns to the starting city. Specifically, staff need to inspect each target point along the optimal path. This problem can be solved by using a path optimization algorithm to improve inspection efficiency and reduce time and cost. Based on this, the present application proposes a deep reinforcement learning model based on knowledge distillation. By extracting knowledge from large-scale pre-trained models and using this knowledge for smaller and more efficient models, it can quickly adjust the path when a new inspection target is added without completely retraining the model. This can significantly improve the real-time and flexibility of the model. Compared with the original deep learning model, the pre-trained deep learning model proposed in this application is smaller in size, consumes less computing resources, and can run on resource-constrained edge computing devices. Combining the exploration capabilities of reinforcement learning and the advantages of model simplification, it is possible to find a near-optimal inspection path in a shorter time. At the same time, it retains the experience of the large model, allowing the model to quickly make better decisions when planning the path. After the model is trained, it can be directly used for inspection tasks of various sizes, reducing the need for manual adjustments. When power inspection tasks change frequently, it is easier to update and deploy the model. In summary, the method proposed in this application has obvious advantages in the optimization of inspection paths in the power industry, especially in scenarios with a large number of target points and rapid dynamic changes. It not only optimizes the model performance, but also reduces the computational cost. It is a more effective solution to the "traveling salesman problem".

[0050] This application proposes a deep reinforcement learning model that can be used to solve the "traveling salesman problem" in power inspection tasks. This approach extracts knowledge from a large model and transfers it to a smaller model, achieving accuracy close to that of the large model while requiring fewer computational resources. The large model, a complex deep reinforcement learning model, undergoes large-scale pre-training to generate a stable path planning strategy. This experience is then transferred to a lightweight model, enabling it to quickly arrive at a near-optimal solution for large-scale inspection tasks.

[0051] Assume that the output of the large model is q T (x), which is the first output information; the output of the small model is q S (x), that is, the second output information, can achieve knowledge transfer by minimizing the following loss function, that is, the target loss function:

[0052] L KD =α·L CE (q S (x),y)+(1-α)·LKL (q S (x),q T (x));

[0053] Among them, L CE is the cross entropy loss between the small model and the true label y, that is, the target cross entropy loss; L KL It is the Kullback-Leibler (KL) divergence between the outputs of the small model and the large model, that is, the target divergence, which is used to measure the similarity between the two. α is the adjustment parameter, that is, the preset adjustment parameter; thus, the computational complexity of the small model can be significantly optimized.

[0054] This application can enhance the real-time response capability through adaptive reinforcement learning. The pre-trained deep learning model uses deep reinforcement learning (DRL), which is the reinforcement learning mentioned above, in path planning. The inspection path is adjusted in real time during the inspection process, and the optimal action strategy under different inspection point states is learned to adapt to the dynamic changes of the inspection target.

[0055] Compared to traditional algorithms that require recalculating routes when target points change, pre-trained deep learning models can learn multiple inspection route strategies during the training phase. This allows them to quickly determine the optimal route after adjustments when inspection points are added or removed. This allows the model to quickly respond to sudden equipment failures and changes in inspection tasks during power facility inspections.

[0056] The reinforcement learning model optimizes the policy π by maximizing the reward function R:

[0057]

[0058] Where γ is the discount factor, r t is the reward at time t. Through training, the pre-trained deep learning model learns the optimal inspection path under different states, thereby improving real-time performance.

[0059] This application can prevent path planning from falling into local optima. In reinforcement learning, the pre-trained deep learning model uses a greedy (ε-greedy) strategy to balance exploration and exploitation. That is, during training, a trade-off is made between randomly exploring new paths and selecting the current optimal path. This allows the small model to further optimize path planning capabilities while retaining the global perspective of the larger model.

[0060] The pre-trained deep learning model uses the exploration capability of reinforcement learning to more effectively find the global optimal path in the solution space, and is suitable for power inspection tasks with complex and uneven distribution of inspection points.

[0061] The action selection process in the greedy (ε-greedy) strategy can be expressed as:

[0062]

[0063] Among them, a t For state s t The action selected when Q(s t ,a) is the state-action value function. Through this strategy, the model can avoid converging to a suboptimal solution too early during training.

[0064] This application can reduce structural complexity and maintenance costs. The smaller model size and fewer parameters enable it to run efficiently on edge computing devices. Simultaneously, the simplified model structure makes deployment and maintenance easier. Pretrained deep learning models can be directly applied to different inspection scales and scenarios without repeated adjustments. This is especially true when power inspection points are complex and unstable, as pretrained deep learning models offer lower maintenance costs.

[0065] The simplified small model parameter θ S This can be optimized by:

[0066]

[0067] In this way, it means that the model parameters of the second network model are determined based on the target loss function and the preset angle calculation function. The small model inherits the knowledge of the large model and significantly reduces the model complexity.

[0068] This application improves the efficiency of route optimization. The pre-trained deep learning model retains the experience of the larger model through reinforcement learning during route planning, generating high-quality inspection routes in a short period of time. This not only speeds up route planning but also brings route quality closer to the global optimal solution.

[0069] The use of pre-trained deep learning models significantly reduces computing time and computing resource requirements while ensuring path quality.

[0070] During the path optimization process, the pre-trained deep learning model updates the path selection strategy through the following value function:

[0071]

[0072] Among them, α is the learning rate. By continuously updating the value function, the model can find the optimal path during the inspection process.

[0073] Through the above technical solution, a pre-trained deep learning model can provide efficient and flexible path planning solutions for power industry inspection tasks. By simplifying the model, it improves computational efficiency and real-time responsiveness, effectively avoiding local optimal solutions during path planning, thus overcoming many limitations of existing technologies. By transferring experience from a large model to a smaller model, the smaller model can efficiently solve the traveling salesman problem even with limited computing resources, adapting to power inspection scenarios. The reinforcement learning-based path optimization model can rapidly adjust the optimized path when the inspection target changes, meeting the real-time requirements of power facility inspection tasks. The model is lightweight, enabling it to run on edge computing devices, making it suitable for the deployment of mobile devices during power inspections. By leveraging the experience transfer mechanism of reinforcement learning, the model's global search capabilities are enhanced, optimizing inspection paths and avoiding local optimal solutions. The lightweight, small model structure reduces deployment and maintenance complexity, making it suitable for flexible application in large-scale power equipment inspection tasks. This not only improves inspection efficiency but also reduces the system's dependence on hardware resources, better meeting the actual needs of the power industry.

[0074] Figure 2 FIG. 1 is a flow chart of an optional method for determining an inspection path for power inspection according to an embodiment of the present invention. Figure 2 As shown, the system includes: S202, obtaining inspection task information for multiple inspection targets to be inspected for power lines. S204, using a pre-trained deep learning model to determine an inspection path based on the inspection task information, thereby obtaining a power line inspection path. S206, transmitting the power line inspection path to a mobile terminal. S208, using the power line inspection path received by the mobile terminal, guiding at least one inspector to perform a power line inspection and obtaining an inspection result.

[0075] According to another aspect of an embodiment of the present invention, a device for determining an inspection path for power inspection is also provided. The device can execute the method for determining an inspection path for power inspection of the above embodiment. The specific implementation method and preferred application scenario are the same as those of the above embodiment and will not be repeated here.

[0076] Figure 3 is a schematic diagram of a device for determining an inspection path for power inspection according to an embodiment of the present application, such as Figure 3 As shown, the apparatus includes the following: an acquisition module 302 and a determination module 304 .

[0077] Among them, the acquisition module is used to obtain inspection task information of multiple inspection target objects to be inspected for power, wherein the inspection task information includes at least location information and inspection duration information, the location information is used to characterize the locations of multiple inspection target objects, and the inspection duration information is used to determine the inspection duration of multiple inspection target objects; the determination module is used to use a pre-trained deep learning model to determine the inspection path based on the inspection task information to obtain a power inspection path, wherein the pre-trained deep learning model includes at least a first network model and a second network model, the input end of the second network model is connected to the output end of the first network model, the number of parameters of the first network model is greater than the number of parameters of the second network model, and the inspection starting point and inspection end point in the power inspection path are at the same position.

[0078] In which, the model parameters of the second network model are determined based on the target loss function and the preset amplitude calculation function, and the target loss function is determined based on the first output information and the second output information, wherein the first output information is used to characterize the output information of the first network model, and the second output information is used to characterize the output information of the second network model.

[0079] Among them, the determination module is also used to obtain the target divergence between the first output information and the second output information, wherein the target divergence is used to characterize the similarity between the first output information and the second output information; based on the target divergence and the preset adjustment parameters, a target loss function is constructed, wherein the preset adjustment parameters are used to control the optimization direction of the pre-trained deep learning model.

[0080] Among them, the determination module is also used to obtain the target cross entropy loss, wherein the target cross entropy loss is used to represent the cross entropy loss between the second network model and the preset true label, and the preset true label is used to represent the reward signal provided by the environment during the reinforcement learning process of the pre-trained deep learning model; based on the target cross entropy loss, the target divergence and the preset adjustment parameters, a target loss function is constructed.

[0081] Among them, the determination module is also used to obtain the complement of the preset adjustment parameter; determine the product of the preset adjustment parameter and the target cross entropy loss to obtain a first product; determine the product of the complement and the target divergence to obtain a second product; and determine the target loss function based on the sum of the first product and the second product.

[0082] Among them, the determination module is also used to send the power inspection path to the mobile terminal; use the power inspection path received by the mobile terminal to guide at least one inspection personnel to perform power inspection and obtain inspection results, wherein the inspection results are used to characterize the operating status of multiple inspection target objects.

[0083] An embodiment of the present application further provides an electronic device, comprising: a memory storing an executable program; and a processor for running the program, wherein the program executes the methods of various embodiments of the present invention when running.

[0084] The above-mentioned memory may refer to a device inside a computer for storing data and programs, and may include memory, hard disk, etc., wherein the memory may be used to temporarily store running programs and data, the hard disk may be used to store programs and data for a long time, and the memory may be used to enable the computer to read and write data, as well as execute programs; the above-mentioned processor may be responsible for executing instructions in computer programs and performing data processing, and may be responsible for controlling and executing various operations, including arithmetic operations, logical operations, data transmission, etc.

[0085] An embodiment of the present application further provides a computer-readable storage medium, which includes a stored executable program, wherein when the executable program is running, the device where the computer-readable storage medium is located is controlled to execute the methods in various embodiments of the present invention.

[0086] The above-mentioned computer storage medium may refer to a medium in a computer memory used to store certain discontinuous physical quantities. Computer storage media mainly include semiconductors, magnetic cores, magnetic drums, magnetic tapes, laser disks, etc. The stored program included in the computer-readable storage medium may be a set of instructions that can be recognized and executed by a computer, running on an electronic computer, and serving as an information tool to meet certain needs of people.

[0087] An embodiment of the present application further provides a computer program product, including a computer program, which implements the methods in various embodiments of the present invention when executed by a processor.

[0088] The above-mentioned computer program product may refer to a software program that has been written, tested and released, which can be run on a computer or other device. The computer program product may include an application, an operating system, tool software, etc., which is used to implement specific functions or solve specific problems.

[0089] An embodiment of the present application further provides a computer program product, including a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium is used to store a computer program, and when the computer program is executed by a processor, the method in each embodiment of the present invention is implemented.

[0090] The above-mentioned non-volatile computer-readable storage medium may refer to a medium for storing data. The non-volatile computer-readable storage medium can keep the data from being lost when the power is off, and can be used to store long-term data, such as operating systems, applications and user files. The non-volatile storage medium may include hard disk drives, solid-state drives, optical disks and flash memory storage devices, etc.

[0091] The embodiments of the present application further provide a computer program, which implements the methods in the above-mentioned embodiments of the present invention when executed by a processor.

[0092] The above-mentioned computer program may refer to a collection of instructions used to tell a computer to perform a specific task or operation. A computer program may be written by a programmer using a specific programming language and may include algorithms, data structures, logic, and control flows. Computer programs may be used for a variety of purposes, including application software, operating systems, and the like.

[0093] In the above embodiments of the present invention, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0094] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only exemplary. For example, the division of the units can be a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0095] The units described as separate components may or may not be physically separate, and 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 units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0096] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0097] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it 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 all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk, etc. Various media that can store program codes.

[0098] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A method for determining an inspection path for power inspection, characterized in that: include: Obtaining inspection task information of multiple inspection target objects to be inspected for power, wherein the inspection task information includes at least location information and inspection duration information, the location information is used to characterize the locations of the multiple inspection target objects, and the inspection duration information is used to determine the inspection duration for inspecting the multiple inspection target objects; A pre-trained deep learning model is used to determine the inspection path based on the inspection task information to obtain a power inspection path, wherein the pre-trained deep learning model includes at least a first network model and a second network model, the input end of the second network model is connected to the output end of the first network model, the number of parameters of the first network model is greater than the number of parameters of the second network model, and the inspection starting point and the inspection end point in the power inspection path are at the same position.

2. The method for determining the inspection path of power inspection according to claim 1, characterized in that: The model parameters of the second network model are determined based on a target loss function and a preset angle calculation function, and the target loss function is determined based on first output information and second output information, wherein the first output information is used to characterize the output information of the first network model, and the second output information is used to characterize the output information of the second network model.

3. The method for determining the inspection path of power inspection according to claim 2, characterized in that: The method further comprises: Acquire a target divergence between the first output information and the second output information, wherein the target divergence is used to characterize the similarity between the first output information and the second output information; Based on the target divergence and preset adjustment parameters, the target loss function is constructed, wherein the preset adjustment parameters are used to control the optimization direction of the pre-trained deep learning model.

4. The method for determining the inspection path of power inspection according to claim 3, characterized in that: Based on the target divergence and the preset adjustment parameters, constructing the target loss function includes: Obtaining a target cross entropy loss, wherein the target cross entropy loss is used to represent a cross entropy loss between the second network model and a preset true label, and the preset true label is used to represent a reward signal provided by the environment during reinforcement learning of the pre-trained deep learning model; The target loss function is constructed based on the target cross entropy loss, the target divergence and the preset adjustment parameters.

5. The method for determining the inspection path of power inspection according to claim 4, characterized in that: Based on the target cross entropy loss, the target divergence and the preset adjustment parameter, constructing the target loss function, including: Obtaining the complement of the preset adjustment parameter; Determine the product of the preset adjustment parameter and the target cross entropy loss to obtain a first product; determining a product of the complement and the target divergence to obtain a second product; The target loss function is determined based on the sum of the first product and the second product.

6. The method for determining an inspection path for power inspection according to claim 1, characterized in that: The method further comprises: Sending the power inspection path to a mobile terminal; The power inspection path received by the mobile terminal is used to guide at least one inspector to perform a power inspection to obtain an inspection result, wherein the inspection result is used to characterize the operating status of the multiple inspection target objects.

7. A device for determining an inspection path for power inspection, characterized in that: include: An acquisition module is used to acquire inspection task information of multiple inspection target objects to be inspected for power, wherein the inspection task information includes at least location information and inspection duration information, the location information is used to characterize the locations of the multiple inspection target objects, and the inspection duration information is used to determine the inspection duration for inspecting the multiple inspection target objects; A determination module is used to use a pre-trained deep learning model to determine the inspection path based on the inspection task information to obtain a power inspection path, wherein the pre-trained deep learning model includes at least a first network model and a second network model, the input end of the second network model is connected to the output end of the first network model, the number of parameters of the first network model is greater than the number of parameters of the second network model, and the inspection starting point and the inspection end point in the power inspection path are at the same position.

8. An electronic device, characterized in that: include: A memory storing an executable program; A processor is used to run the program, wherein the program, when running, executes the method for determining the inspection path for power inspection as described in any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored executable program, wherein when the executable program is executed, the device where the storage medium is located is controlled to execute the method for determining the inspection path for power inspection according to any one of claims 1 to 7.

10. A computer program product, characterized in that The invention comprises a computer program, which, when executed by a processor, implements the method for determining a patrol path for power patrol according to any one of claims 1 to 7.

Images