Transformer substation loop phasor detection method based on lightweight artificial intelligence and related equipment

Abstract

The invention provides a transformer substation loop phasor detection method based on lightweight artificial intelligence and related equipment. The method comprises the following steps: acquiring a loop phasor detection task of an intelligent transformer substation; constructing a multi-target phasor detection model; acquiring historical loop detection data of the intelligent substation, and extracting static topological characteristics of a substation loop; responding to the loop phasor detection task and controlling the current output device and the voltage output device to carry out pressurization and through-flow in the transformer substation loop so as to simulate the on-load operation process of the transformer substation loop; in the pressurization through-flow process, collecting equipment data of all loop equipment in a transformer substation loop through an intelligent sensor network; and inputting the equipment data into the multi-target phasor detection model, and outputting a loop phasor detection result of the intelligent substation through the multi-target phasor detection model. The method has the effect of intelligently and accurately completing the phasor detection process of the transformer substation loop.

Description

Technical Field

[0001] The present invention belongs to the technical field of loop phasor detection, and specifically relates to a substation loop phasor detection method and related equipment based on lightweight artificial intelligence. Background Art

[0002] Due to the increasing power demand and the diversification of the energy structure, the power grid needs to support various power supply and demand changes, which poses great challenges to the management and monitoring of substations. In this context, traditional substation detection systems usually rely on centralized management of a central control station, which was more applicable in the past single power supply mode. However, with the increasing complexity of the power generation and consumption distribution, this mode is obviously difficult to meet the flexibility requirements of today's power system. For example, the instability of new energy such as wind energy and solar energy and the load fluctuations brought by distributed generation have greatly increased the complexity of substation operation and maintenance.

[0003] Existing substation loop detection systems usually have a complex architecture, and data processing depends on a remote large-scale data center. Such an architecture design leads to a large amount of data transmission between multiple links, and the reliability and efficiency of data transmission are affected by many factors such as network conditions, information transmission protocols, and device performance. Therefore, in traditional methods, during the process from data acquisition to analysis and then to result output, the detection accuracy may be reduced due to the accumulation of delays and errors. In addition, due to the diversity of devices in the substation and the intricate state of each device and their mutual influence, it is difficult for the static and centralized processing mode of traditional methods to capture the true state and complex state changes of the system in real time. Summary of the Invention

[0004] The present invention provides a substation loop phasor detection method and related equipment based on lightweight artificial intelligence to solve the problem of poor accuracy of substation loop detection results.

[0005] In a first aspect, the present invention provides a substation loop phasor detection method based on lightweight artificial intelligence, which is applied to an intelligent substation. The intelligent substation is deployed with edge computing devices, and an intelligent sensor network is deployed in the substation loop of the intelligent substation. The intelligent sensor network is communicatively connected to the edge computing devices. A current output device and a voltage output device are also deployed on the primary circuit side of the substation loop;

[0006] The method includes the following steps:

[0007] Obtain the loop phasor detection task of the intelligent substation;

[0008] Based on a convolutional neural network model, a multi-object phasor detection model is constructed according to the task type in the loop phasor detection task, and the multi-object phasor detection model is deployed in the edge computing device. The multi-object phasor detection model includes a feature sharing layer and a multi-object detection task layer;

[0009] Obtain the historical loop detection data of the intelligent substation, and extract the static topological features of the substation loop;

[0010] Use the static topological features to complete the training of the feature sharing layer, and use the historical loop detection data to complete the training of the multi-object detection task layer;

[0011] In response to the loop phasor detection task, control the current output device and the voltage output device to apply voltage and current in the substation loop to simulate the load operation process of the substation loop;

[0012] During the process of applying voltage and current, collect the device data of all loop devices in the substation loop through the intelligent sensor network;

[0013] Input the device data into the multi-object phasor detection model, and output the loop phasor detection result of the intelligent substation through the multi-object phasor detection model.

[0014] Optionally, the loop phasor detection task is any one or more of the loop device phase sequence correctness detection, loop device polarity correctness detection, loop device turn ratio correctness detection, and cable connection performance detection.

[0015] Optionally, the constructing a multi-object phasor detection model based on a convolutional neural network model according to the task type in the loop phasor detection task includes the following steps:

[0016] Construct a feature sharing layer and multiple initial detection sub-networks based on the convolutional neural network model. The number of the initial detection sub-networks is the same as the number of task types in the loop phasor detection task, and all the initial detection sub-networks have independent output layers;

[0017] Adjust the layer structure and model parameters of each of the initial detection sub-networks according to the task type in the loop phasor detection task to obtain multiple basic detection sub-networks;

[0018] Connect the input layers of multiple basic detection sub-networks to the output layer in the feature sharing layer respectively to obtain a multi-object phasor detection model. All the basic detection sub-networks are arranged in parallel to form a multi-object detection task layer.

[0019] Optionally, the steps of obtaining the historical loop detection data of the intelligent substation and extracting the static topological features of the substation loop are as follows:

[0020] Obtain the historical loop detection data of the intelligent substation and the device parameters of all loop devices in the substation loop;

[0021] Use the loop devices as device nodes, and use the electrical connection relationships between all the loop devices in the substation loop as device node edges to construct a substation electrical connection graph, and assign node attributes to all device nodes based on the device parameters;

[0022] Use a graph traversal algorithm based on recursion to identify all electrical loops in the substation electrical connection graph;

[0023] Through the node clustering analysis of all the electrical loops, mark the bus structure features in all the electrical loops;

[0024] Integrate all the device parameters and all the electrical loops into the static topological features of the substation loop.

[0025] Optionally, the steps of using the static topological features to complete the training of the feature sharing layer and using the historical loop detection data to complete the training of the multi-objective detection task layer are as follows:

[0026] Use the feature sharing layer as an encoder, and map the static topological features to a low-dimensional space through the encoder;

[0027] Reconstruct the static topological features in the low-dimensional space through a preset decoder, and complete the training of the feature sharing layer when the reconstruction error reaches the minimum value;

[0028] Define a sub-network loss function for each of the basic monitoring sub-networks, and use the weighted sum of all the sub-network loss functions as the total loss function of the multi-objective detection task layer;

[0029] Classify and input the historical loop detection data into all the basic monitoring sub-networks according to all the task types of the loop phasor detection task, train all the basic monitoring sub-networks using the mini-batch gradient descent method, and use a gradient descent type algorithm to minimize the total loss function to complete the training of the multi-objective detection task layer.

[0030] Optionally, the steps of inputting the device data into the multi-objective phasor detection model and outputting the loop phasor detection result of the intelligent substation through the multi-objective phasor detection model are as follows:

[0031] Obtain the device resource status of the edge computing device;

[0032] Determine the current task set and the task priorities of all tasks in the current task set by parsing the loop phasor detection task;

[0033] Construct a computing resource allocation model for the multi-object phasor detection model by combining the device resource status, the current task set, and the task priorities;

[0034] Generate a reward function for the computing resource allocation model with the shortest time to complete all tasks in the current task set and the optimal resource utilization rate of the edge computing device;

[0035] Solve the computing resource allocation model based on the reward function and using the deep Q-network algorithm. When the reward function reaches the maximum value, obtain the optimal computing resource allocation strategy for the edge computing device to execute all tasks in the current task set;

[0036] Allocate the remaining device resources of the edge computing device to the feature sharing layer and multiple target basic detection sub-networks corresponding to the current task set according to the optimal computing resource allocation strategy;

[0037] Output the loop phasor detection results of the intelligent substation through all the target basic detection sub-networks.

[0038] Optionally, constructing the computing resource allocation model of the multi-object phasor detection model by combining the device resource status, the current task set, and the task priorities includes the following steps:

[0039] Construct a target state space based on the Markov decision model and by combining the device resource status, the current task set, and the task priorities;

[0040] Construct a target action space with the selection step of selecting the next task in the current task set and the preset initial resource allocation strategy of the edge computing device;

[0041] Construct the computing resource allocation model of the multi-object phasor detection model by combining the target state space, the target action space, and the preset state transition probability function.

[0042] Optionally, the expression formula of the optimal computing resource allocation strategy is as follows:

[0043]

[0044] In the formula: π * represents the optimal computing resource allocation strategy, argmax πdenotes selecting a computing resource allocation strategy that maximizes the reward function, E denotes the cumulative reward expected value based on the reward function, t denotes the time step, γ t denotes the discount factor, and both α and β denote weight coefficients, T(s t , a t ) denotes the reward value in the task completion time dimension when taking the action a in the target action space under the state s in the target state space t ; R(s t , a t ) denotes the penalty value in the resource utilization dimension when taking the action a in the target action space under the state s in the target state space t . t t

[0045] In a second aspect, the present invention further provides a loop phasor detection device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the substation loop phasor detection method based on lightweight artificial intelligence as described in the first aspect.

[0046] In a third aspect, the present invention further provides a substation loop phasor detection system based on lightweight artificial intelligence, including:

[0047] the loop phasor detection device described in the second aspect;

[0048] an edge computing device, deployed in an intelligent substation, for hosting a multi-target phasor detection model constructed based on a convolutional neural network model;

[0049] an intelligent sensor network, deployed in the substation loop of the intelligent substation and communicatively connected to the edge computing device, for collecting device data of all loop devices in the substation loop;

[0050] a current output device, deployed on the primary loop side of the substation loop, for injecting current into the substation loop when performing a loop phasor detection task;

[0051] a voltage output device, deployed on the primary loop side of the substation loop, for injecting voltage into the substation loop when performing a loop phasor detection task.

[0052] The beneficial effects of the present invention are:

[0053] ​​The present invention has significant beneficial effects compared with the prior art. Especially in improving the real-time performance, accuracy of phasor detection in substation circuits and the flexibility of the system, it shows unique technical advantages. First of all, by deploying edge computing devices in intelligent substations and introducing lightweight artificial intelligence technologies such as convolutional neural networks into the phasor detection system, the present invention can effectively reduce the latency in the data processing and transmission processes. This adjustment of the architecture enables the detection tasks to be carried out locally in the substation, greatly improving the real-time performance of the detection and being able to more quickly reflect the actual operating state of the substation circuit. Secondly, by constructing a multi-object phasor detection model and training it using an intelligent sensor network and topological features, the present invention has higher detection accuracy and response capabilities. The introduction of the multi-object detection model can not only identify and process various types of detection tasks in a complex power system, but also perform adaptive learning and adjustment according to the dynamically changing operating environment, thereby improving the accuracy of the detection results. In addition, by simulating the load conditions of the substation circuit, the present invention can more realistically reflect the state and behavior of the equipment under various operating conditions, providing more reliable data support and decision-making basis for the operation and maintenance of intelligent substations. In summary, by leveraging advanced artificial intelligence and edge computing technologies, the present invention transforms the centralized mode of traditional power system detection into a distributed stream processing mode, not only improving the system's response speed and detection accuracy, but also enhancing the adaptability to complex and changing power environments, laying a solid foundation for the management and monitoring of future intelligent substations. Description of the Drawings

[0054] Figure 1 It is a schematic flowchart of a method for detecting phasors in a substation circuit based on lightweight artificial intelligence in one embodiment of the present application.

[0055] Figure 2 It is a schematic diagram of the model structure of a multi-object phasor detection model in one embodiment of the present application.

[0056] Figure 3 It is a system structure diagram of a system for detecting phasors in a substation circuit based on lightweight artificial intelligence in one embodiment of the present application. Detailed Embodiments

[0057] Next, the technical solutions in the embodiments of the present application will be clearly described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application belong to the scope of protection of the present application.

[0058] The terms "first", "second", etc. in the description and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances so that the embodiments of this application can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first", "second", etc. are generally of the same type and do not limit the number of objects. For example, the first object can be one or multiple. In addition, "and / or" in the description and claims means at least one of the connected objects, and the character " / ", generally indicates an "or" relationship between the associated objects before and after.

[0059] The present invention discloses a substation loop phasor detection method based on lightweight artificial intelligence technology, which is applied to intelligent substations. The present invention provides support by deploying edge computing devices in intelligent substations. Advanced intelligent sensor networks are installed in the substation loops of intelligent substations, and these sensors are distributed at various important nodes of the substation and can collect the operation data of various power equipment in real time. The sensor network and the edge computing device are connected through high-speed communication to achieve fast data transmission and processing. This design breaks the excessive dependence of traditional detection systems on central data processing. By locally processing data in the substation, the delay in the data transmission process is significantly reduced, and at the same time, the efficiency of data processing is improved.

[0060] A current output device and a voltage output device are specifically deployed on the main circuit side of the substation loop. The introduction of these devices enables pressurization and current conduction operations in the loop to simulate the operating state of the substation under actual load conditions. This simulation function enables the detection system to better understand and evaluate the dynamic response of the substation loop under various operating conditions, providing more real and reliable data support for the monitoring and fault diagnosis of the substation. The current output device can be a three-phase digital intelligent current source, and its parameters are as follows:

[0061] (1) Three-phase output current: 0 - 300 A; (2) Output current accuracy: 1% (30 - 300 A); (3) Maximum output power of phase current: 2400 VA; (4) Phase output range: 0 - 360°; (5) Phase accuracy: 0.5 degrees; (6) Frequency output range: 50 Hz; (7) Frequency accuracy: 0.01 Hz.

[0062] The voltage output device can be a three-phase digital intelligent high-voltage source, and its parameters are as follows:

[0063] (1) Three-phase output voltage: 0 - 6000V (phase voltage), 0 - 10400V (line voltage); (2) Output voltage accuracy: 1% (500 - 6000V); (3) Phase voltage output power: 300VA; (4) Phase output range: 0 - 360°; (5) Phase accuracy: 0.5 degrees; (6) Frequency output range: 50Hz; (7) Frequency accuracy: 0.01Hz.

[0064] Figure 1 It is a schematic flowchart of a method for detecting phasors in a substation circuit based on lightweight artificial intelligence in an embodiment. It should be understood that although Figure 1 each step in the flowchart is shown in sequence according to the indication of the arrow, these steps do not necessarily execute in the order indicated by the arrow. Unless otherwise clearly stated in this article, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover, Figure 1 at least a part of the steps in Figure 1 may include multiple sub-steps or multiple stages. These sub-steps or stages do not necessarily execute at the same moment, but can execute at different moments. The execution order of these sub-steps or stages is not necessarily sequential either, but can be executed alternately or in turn with at least a part of other steps or sub-steps or stages of other steps. As

[0065] S101. Obtain the phasor detection task of the intelligent substation circuit.

[0066] Among them, the loop phasor detection task is any one or more of the detection of the phase sequence correctness of loop devices, the detection of the polarity correctness of loop devices, the detection of the turns ratio correctness of loop devices, and the detection of cable connection performance. Such a wide range of detection is aimed at ensuring that each loop component of the substation is in the best working state, thus ensuring the normal operation of the power grid. The phase sequence correctness detection is used to confirm whether the order of each phase signal conforms to the predetermined standard, because incorrect phase sequence may cause the equipment to fail to start properly or operate unstably. The polarity correctness detection ensures that parameters such as current and voltage are measured and transmitted in the correct direction, and deviations may lead to system failures. The turns ratio correctness detection evaluates whether the turns ratio of the transformer operates according to the design parameters, which is particularly crucial for the effective operation of the transformer. The cable connection performance detection checks the connection quality in the line, thereby ensuring electrical transmission safety. In implementation, the intelligent substation automatically captures any one or more of the above detection tasks through the deployed sensors and edge computing devices. This is achieved by configuring the intelligent control strategy of the system to allocate and prioritize multiple tasks. The sensors continuously monitor various parameters in the loop and send them to the edge devices for processing. Without manual intervention, the intelligent system automatically defines the most urgent detection tasks in combination with historical data and actual situations, ensuring the continuous and stable power transmission of the substation.

[0067] S102. Based on the convolutional neural network model, construct a multi-objective phasor detection model according to the task type in the loop phasor detection task, and deploy the multi-objective phasor detection model in the edge computing device.

[0068] Among them, the multi-object phasor detection model includes a feature sharing layer and a multi-object detection task layer. To build an efficient multi-object phasor detection model, it is necessary to introduce convolutional neural network (CNN) technology. CNN is widely used in the fields of computer vision and signal processing due to its powerful feature extraction ability. By adopting CNN to build the multi-object phasor detection model, the detection accuracy can be significantly improved. First, according to different loop phasor detection tasks, the model needs to have the ability to handle multiple detection tasks, such as the phase sequence, polarity, turns ratio, and cable connection functions mentioned above. The core of the multi-task model lies in its feature sharing layer, which makes full use of the common features required in the phasor detection tasks and extracts multi-level features from the input data through deep learning. On this basis, the feature sharing layer can reduce feature redundancy and improve the detection speed and efficiency. Next, the multi-object detection task layer optimizes the output of its respective detection tasks with different task loss functions according to the features extracted by the feature sharing layer. This multi-task collaborative optimization method takes into account the possible mutual dependencies between various detection tasks. Through training with various historical task data and specific types of current and voltage data obtained in advance, the multi-object phasor detection model is deployed in real time on edge computing devices. During the practical process, the model continuously processes real-time data and adapts to the actual detection conditions and environmental changes through self-learning and adjustment. Therefore, efficient and accurate multi-task phasor detection of substation loops can be achieved.

[0069] S103. Obtain the historical loop detection data of the intelligent substation and extract the static topological features of the substation loop.

[0070] Among them, before carrying out intelligent detection, it is necessary to obtain the historical loop detection data of the substation. These data provide the operating conditions of the substation loop at different time nodes, including electrical parameters under various load conditions. These data are accumulated over a long time through the sensor network and edge computing devices, truly reflecting the operation mode and potential problems of the substation. After obtaining these data, the next step is to extract the static topological features of the substation loop. These features include the connection mode of the loop, the rated parameters of the transformer, the layout of the switches, and the load distribution status, etc. These topological information runs through the entire physical connection structure and is crucial for system analysis. Through this technical solution, the design parameters are directly extracted from the data using an algorithm, enabling the model to more accurately depict the corresponding positions of each part of the loop and the relationships between them. Combining historical data with topological features, the system sorts and encodes the obtained data, laying a foundation for feature sharing and optimization of the task detection layer. Analyzing and extracting these features not only supports improving the efficiency of model training but also plays a clear guiding role in subsequent intelligent detection.

[0071] S104. Complete the training of the feature sharing layer using static topological features, and complete the training of the multi-objective detection task layer using historical loop detection data.

[0072] Among them, after extracting the topological features and historical data, the next step is how to effectively apply them to the training of the model. This process includes two main parts: the training of the feature sharing layer and the training of the multi-objective detection task layer. During the training of the feature sharing layer, with the help of the previously extracted static topological features, an adaptive learning algorithm in deep learning is used to iteratively optimize the network parameters. Since these features provide static information about the substation loop and are closely related to the actual operation state of the power system, during training, the feature sharing layer uses this information to capture the basic features shared in each task, which not only improves the performance of the model but also avoids repeated calculations. After completing the training of the feature sharing layer, the training of the multi-objective detection task layer in the model becomes particularly important. Its training is mainly based on historical loop detection data, and using labeled tag data such as phase sequence, polarity, turns ratio, etc., to specifically optimize the loss function of the task layer to enhance its recognition ability and accuracy for handling specific tasks. The entire training process is iteratively carried out. During the process, the task layer continuously updates its strategy to adapt to the needs of multi-task processing and maximize the collaborative efficiency between tasks. This training method makes the model more sensitive to the discovery of potential problems in the substation loop, realizing rapid response and positioning of anomalies.

[0073] S105. In response to the loop phasor detection task, control the current output device and the voltage output device to apply voltage and current in the substation loop to simulate the load operation process of the substation loop.

[0074] Among them, the voltage and current are applied in the loop by controlling the current output device and the voltage output device. This process simulates the operation state of the substation under real load conditions. First, the system configures the output parameters of the current and voltage devices according to the preset load curve or the power demand at different times. Through precise control, the power load is gradually increased to the target point. This not only allows the detection system to evaluate the behavior parameters of the loop under overload or underload conditions but also provides a risk-free environment to identify potential hidden danger information. Through load tests under stable and dynamic conditions, the system can verify the correlation between its performance in actual working conditions and the model prediction, thereby evaluating or adjusting the parameter settings of the detection model.

[0075] S106. During the voltage and current application process, collect the device data of all loop devices in the substation loop through the intelligent sensor network.

[0076] Among them, the circuit equipment includes transformers, circuit breakers, disconnectors, current transformers, voltage transformers, lightning arresters, busbars, cables and overhead lines, protection devices (relay protection devices), etc. The sensors can capture various key parameters such as current, voltage, phase, frequency, and temperature in real time. With the help of highly sensitive and accurate measurement technologies, a large amount of real-time data collected is transmitted to the edge computing device. This connection provides an accurate and timely information source for subsequent intelligent analysis, ensuring the continuity and consistency of the data from collection to processing. To ensure the validity of the data, the sensor network converts the data stream into a structured set of power grid characteristics, and through a priority processing mechanism and a data integrity protocol, avoids information loss or error accumulation caused by the transmission process. After all the collected data is integrated, it is converted into an effective data signal input for model analysis to achieve comprehensive and accurate monitoring of the status of the substation circuit equipment.

[0077] S107. Input the equipment data into the multi-object phasor detection model, and output the phasor detection results of the circuits in the intelligent substation through the multi-object phasor detection model.

[0078] Among them, after the training of the multi-object phasor detection model and the data collection are completed, all the collected equipment data is input into the model for real-time calculation and analysis. The multi-object phasor detection model, based on the parameters obtained from its learning and training, outputs a series of phasor detection results for the circuits in the intelligent substation through its feature sharing and multi-task optimization mechanism. In specific implementation, the model comprehensively evaluates different types of data, such as determining whether the phase sequence is normal, whether the polarity is accurate, whether the transformation ratio is consistent with the expectation, and whether the connection performance of the cable meets the standard. After rigorous algorithm analysis, these results are presented in various forms such as tables or graphs, providing a user-friendly interface interaction, so that technicians can intuitively understand and make decisions. The output detection results are usually accompanied by problem diagnosis prompts to help technicians quickly identify and solve potential problems, ensure the normal operation of the substation, and continuously optimize the reliability and safety of the power system. At the same time, the historical archive of the detection results will become a powerful basis for subsequent improvement and analysis, supporting the optimization of further power grid management strategies.

[0079] In one implementation, based on the convolutional neural network model, constructing a multi-object phasor detection model according to the task type in the circuit phasor detection task includes the following steps:

[0080] Construct a feature sharing layer and multiple initial detection sub-networks based on the convolutional neural network model. The number of initial detection sub-networks is the same as the number of types of task types in the circuit phasor detection task, and all initial detection sub-networks have independent output layers;

[0081] Adjust the layer structure and model parameters of each initial detection subnet according to the task type in the loop phasor detection task to obtain multiple basic detection subnets;

[0082] Connect the input layers in multiple basic detection subnets to the output layer in the feature sharing layer to obtain a multi-object phasor detection model.

[0083] In this embodiment, referring to Figure 2 , in this process, it is first necessary to create a feature sharing layer, which is designed to extract shared underlying features from the input data. Convolutional neural networks are widely used due to their powerful feature extraction capabilities, especially in the fields of computer vision and signal processing. The feature sharing layer can be considered as a set of convolutional layers and pooling layers, and these layers combined can effectively capture local characteristics and cross-feature correlations in the input data. The convolutional layer scans the input data through a set of convolutional kernels (or filters) to extract various detailed features, while the pooling layer compresses the size by performing downsampling operations on the input feature map, thereby reducing the size of the feature map and alleviating the computational pressure. The feature sharing layer uses a combination of various convolutional and pooling operations, which can adapt to a wide range of phasor detection task scenarios and provide rich basic features for subsequent detection subnets. The key to this step lies in designing appropriate numbers, sizes, and layers of convolutional kernels in order to achieve maximized feature extraction generality in various detection tasks.

[0084] After constructing the feature sharing layer, the design of the initial detection subnet follows. The initial detection subnet is a set of independent basic networks, each corresponding to a specific detection task type. The main role of these subnets is to receive the basic features extracted by the feature sharing layer to further optimize specific types of detection tasks. The initial detection subnet includes a specific set of convolutional layers or fully connected layers, and the structure composed of these layers and the parameters therein need to be adjusted according to the characteristics of the target detection task. Each detection subnet contains its own output layer, whose role is to transfer the features processed by its respective specific layers to the final output of the model. This ensures that even under the same input, each subnet can independently generate output results according to its dedicated training objective to adapt to different types of detection tasks, such as phase sequence detection, polarity check, turns ratio verification, and cable connection performance testing, etc. The implementation effect of this step is to lay the basic framework for subsequent network adjustment and optimization through the diversification of the initial network structure and ensure that each subnet has an independent learning channel to self-adjust.

[0085] Based on the construction of the initial detection sub-networks, the next step is to perform personalized adjustments to these initial sub-networks for different types of detection tasks. This process involves optimizing the layer structure and model parameters of the detection sub-networks one by one to meet the specific requirements of different tasks. The objects of adjustment include the depth of the layers, the size of the convolutional kernels, the selection of pooling strategies, and the configuration of activation functions and loss functions, etc. Different task types may require different depths of feature extraction and pattern recognition capabilities, so meticulous parameter adjustment needs to be performed for each sub-network. For example, for phase sequence detection, the focus is on the relative position relationship of three-phase signals, so the network may emphasize capturing the characteristics of periodic signals and amplifying the phase differences. For polarity detection, the network needs to have stronger direction sensitivity to ensure the correctness of the current detection signal direction. After these personalized adjustments, each initial detection sub-network is transformed into a basic detection sub-network highly optimized for a specific task. Through detailed optimization and parameter debugging, the construction of the basic detection sub-networks ensures the efficient recognition and accuracy improvement of multiple task types, and can accurately respond to the complex signal processing requirements in actual application scenarios.

[0086] After the construction of the basic detection sub-networks is completed, the input layers of these networks need to be connected to the output layer of the feature sharing layer respectively to form an overall multi-object phasor detection model. The feature sharing layer has extracted rich shared features, while the basic detection sub-networks have the processing capabilities for specific detection tasks. By combining the basic detection sub-networks with the feature sharing layer, the goal of simultaneous task processing can be achieved. The structure of the multi-object phasor detection model is equivalent to stacking each optimized basic detection sub-network in parallel above the feature sharing layer, and these sub-networks together constitute the multi-object detection task layer. This structure design aims to achieve the simultaneous detection and evaluation of multiple operating parameters in the power grid through the unified feature extraction of the sharing layer and the dedicated task recognition of the basic sub-networks. The connected multi-object phasor detection model can not only significantly improve the recognition efficiency by using shared features, but also form a collaborative processing relationship among multiple tasks, accelerate the processing speed of detection data through parallel computing, and provide a comprehensive phasor detection report at the task output layer. The implementation effect of the overall model is to greatly improve the reaction speed and accuracy of the detection tasks, making real-time monitoring and fault diagnosis in a complex power grid environment more efficient and reliable.

[0087] In one implementation, the steps for obtaining the historical loop detection data of an intelligent substation and extracting the static topological features of the substation loop are as follows:

[0088] Obtain the historical loop detection data of the intelligent substation and the equipment parameters of all loop devices in the substation loop;

[0089] Using circuit equipment as equipment nodes and taking the electrical connection relationships among all circuit equipment in the substation circuit as the edges of the equipment nodes to construct a substation electrical connection diagram, and assigning node attributes to all equipment nodes based on equipment parameters;

[0090] Using a graph traversal algorithm based on recursion to identify all electrical circuits in the substation electrical connection diagram;

[0091] Through the node aggregation analysis of all electrical circuits, marking the bus structure characteristics in all electrical circuits;

[0092] Integrating all equipment parameters and all electrical circuits into the static topological characteristics of the substation circuit.

[0093] In this embodiment, first of all, it is crucial to collect the historical circuit detection data of the intelligent substation. These data can provide information about the electrical system performance and operation history, helping to analyze the health status and fault patterns of the power grid. The historical circuit detection data includes data records of electrical parameters such as voltage, current, power, frequency, etc. These records are accumulated during the long-term operation process and can help analysts identify potential anomalies or instability factors in the electrical system. At the same time, it is also necessary to obtain the equipment parameters of all circuit equipment in the substation circuit. These parameters usually include equipment type, rated capacity, operating voltage, rated current, short-circuit current capacity, manufacturer information, installation date, operating status, etc. The equipment parameters provide detailed static information, making the roles and characteristics of the equipment in the electrical connection diagram more clear. The collection of these data can be achieved through various channels such as the substation monitoring system, maintenance logs, operation records, etc. By comprehensively obtaining the static parameters of the equipment and the dynamic detection data of the circuit, it provides detailed basic information for subsequent analysis and model construction. Finally, the data collected and sorted in this part will provide effective support for establishing an accurate electrical connection diagram and analyzing the behavior of electrical circuits.

[0094] Taking circuit equipment as equipment nodes and using the electrical connection relationships among all circuit equipment in the substation circuit as the edges of the equipment nodes to construct a substation electrical connection diagram, the electrical equipment and its interconnection relationships are transformed into a graph theory model during this process. First, each circuit equipment is regarded as a node. These equipment include transformers, circuit breakers, instrument transformers, etc., which play different roles in the electrical system. The definition of node attributes is based on equipment parameters and may include important information such as the technical specifications and locations of the equipment. The connection relationships of the equipment nodes are represented as edges in the graph, and the construction of the edges reflects the actual electrical connections among the equipment in the circuit, such as physical connections in the form of cable lines, conductors, etc. To construct this electrical connection diagram, it is necessary to make full use of the equipment parameters and historical detection data collected in the previous step to ensure the integrity and accuracy of the graph. To achieve this process, nodes and edges in graph theory are usually used to build an abstract model of the entire electrical system. In addition, this connection diagram is not only an intuitive structural representation but also needs to be continuously updated and maintained through a digital information system. Along with the addition, deletion, or modification of equipment, the structure of the graph also needs to be dynamically adjusted.

[0095] Using a recursive graph traversal algorithm to identify all electrical circuits in the substation electrical connection diagram is an important step in analyzing the complex relationships of the electrical system. The recursive algorithm is a computational method suitable for solving problems with self-similar structures. By dividing and solving sub-problems, the ultimate solution to the entire problem is achieved. For the electrical connection diagram, recursive graph traversal mainly means starting from a node in the graph, sequentially visiting each node along the connection path of the edges, and returning to the starting node to form a complete electrical circuit. Graph traversal algorithms mainly include depth-first search (DFS) and breadth-first search (BFS), etc. The application of each method depends on the focus of the requirements, such as finding the shortest path or a complete circuit. During the traversal process, the algorithm recursively visits each node and tracks the path of the nodes that have been visited to avoid repeated visits and getting into infinite loops. To identify an electrical circuit, it is necessary to check each possible path to confirm whether they form a closed circuit. The efficiency of the recursive traversal algorithm lies in its ability to adaptively handle a large number of nodes and complex connection structures, which is particularly applicable in the case of a substation electrical diagram with dense connections and numerous nodes.

[0096] The nodal aggregation analysis of all electrical circuits and the marking of the bus structure features in all electrical circuits are important steps for the refined identification of the topological characteristics of the electrical network. The nodal aggregation analysis involves in-depth parsing of the traversed and identified electrical circuits, with a focus on identifying the aggregation (i.e., the bus) between key nodes and its importance in the entire cycle. The bus is a key location where multiple devices in the electrical circuit converge, usually used for aggregating and distributing power, so its role in the electrical system is crucial. In the aggregation analysis, by examining the connectivity of each node in the circuit, the nodes with high connectivity are identified as potential bus structures. Connectivity refers to the number of connections of a node to other nodes, and a higher connectivity means that the node plays a converging role in multiple circuits. To mark these bus structures, it is necessary to analyze the connection density and distribution characteristics between nodes and identify them in combination with the physical attributes and functions of the devices. Once the key bus structures are marked, they can be visualized in the form of a topological view and a data table, presenting the important connected positions in the electrical circuit. Finally, all device parameters and all electrical circuits are integrated into the static topological characteristics of the substation circuit. The determination of the static topological characteristics is based on the results of identification and analysis in the previous steps, and the device parameters, circuit structure, and bus characteristics are systematically integrated. This integration process is not just a mechanical superposition of each element, but takes into account the mutual functional coupling and influence between devices to form an overall model with logical connections and practical value.

[0097] In one implementation, using the static topological characteristics to complete the training of the feature sharing layer and using the historical loop detection data to complete the training of the multi-objective detection task layer includes the following steps:

[0098] Taking the feature sharing layer as an encoder, mapping the static topological characteristics to a low-dimensional space through the encoder;

[0099] Reconstructing the static topological characteristics in the low-dimensional space through a preset decoder, and completing the training of the feature sharing layer when the reconstruction error reaches the minimum value;

[0100] Defining a sub-network loss function for each basic monitoring sub-network, and taking the weighted sum of all sub-network loss functions as the total loss function of the multi-objective detection task layer;

[0101] Classifying and inputting the historical loop detection data into all basic monitoring sub-networks according to all task types of the loop phasor detection task, training all basic monitoring sub-networks using the mini-batch gradient descent method, and minimizing the total loss function using a gradient descent type algorithm to complete the training of the multi-objective detection task layer.

[0102] In this embodiment, the encoder is a neural network model whose task is to compress the input data into a latent space with a lower dimension through a series of non-linear transformations. Static topological features refer to the structural and connectivity information in the power system. These information are usually high-dimensional and complex. Through the processing of the encoder, these high-dimensional data are transformed into a low-dimensional form that can effectively express their essential information. To achieve this process, first, an appropriate encoder network structure needs to be constructed, specifically including an input layer, several hidden layers, and an output layer. Among them, the input layer is responsible for receiving the data of static topological features, the hidden layer realizes feature extraction through superimposed convolution, activation, and pooling operations, and finally the output layer maps the data to a lower-dimensional space. At this time, by appropriately selecting the depth and width of the feature-sharing layer, while compressing the data, the original information of the input can be maximally retained. In addition, the coding effect can be further improved through parameter optimization during the training process. When the features are mapped to the low-dimensional space, the computational efficiency of the data is significantly improved, laying a foundation for subsequent analysis and processing. In addition, the low-dimensional feature representation also reduces the burden of data storage and transmission and improves the response speed of the model.

[0103] Reconstructing the static topological features in the low-dimensional space through a preset decoder, and completing the training of the feature-sharing layer when the reconstruction error reaches the minimum value, describes the reverse operation process of the autoencoder. The decoder is usually a neural network structure symmetric to the encoder, and its role is to map the features that have been compressed into the low-dimensional space back to the high-dimensional space, that is, the space of the original data. The design of the decoder generally follows the same principle: its input is the output of the encoder. Through training processes such as backpropagation and gradient descent, the weights in the network are gradually adjusted to make the output of the decoder as close as possible to the original input data. The evaluation criterion for training is the reconstruction error, which is usually measured by a certain loss function, such as the mean square error (MSE), to calculate the difference between the input data and the reconstructed output of the decoder. When the value of the loss function decreases to a certain set threshold or reaches a local minimum, the training process is considered to converge. At this time, the feature-sharing layer has been fully trained, which can ensure that the low-dimensional features output by the encoder are accurately reconstructed into the original form in the decoder. This adaptive training mechanism enables the feature-sharing layer to retain key information in a limited data representation space and can also accurately reconstruct the original data when necessary.

[0104] Define a sub-network loss function for each basic monitoring sub-network, and use the weighted sum of all sub-network loss functions as the total loss function of the multi-objective detection task layer, that is, to achieve objective optimization in the multi-task deep learning framework. This step emphasizes the importance of multi-objective optimization. Each basic monitoring sub-network represents different monitoring tasks, such as voltage monitoring, frequency monitoring, etc. They each have specific learning objectives and scopes, corresponding to independent task functions. Each task function is embodied as a loss function, which measures the difference between the predicted output and the true value of its corresponding task. The loss functions of these sub-networks can be mean square error, cross entropy, etc., and are selected according to specific task requirements. Then, the loss functions of all sub-networks are weighted to reflect the importance and priority of different tasks, forming a comprehensive total loss function. The way of weighted sum is usually to preset weight coefficients in advance, or dynamically adjust according to the real-time performance of the task to better coordinate the learning among tasks. Through the optimization of the total loss function, it is ensured that each basic monitoring sub-network can work together in the overall network architecture to achieve the unified detection and optimization of multiple objectives. This multi-objective optimization strategy helps to improve the generalization ability of the entire model, ensure that the performance can be effectively balanced when dealing with different types of tasks, and thus obtain more robust prediction results.

[0105] Classify and input the historical loop detection data into all basic monitoring sub-networks according to all task types of the loop phasor detection task. Use the mini-batch gradient descent method to train all basic monitoring sub-networks, and use gradient descent algorithms to minimize the total loss function to complete the training of the multi-objective detection task layer. This step details the training mechanism and optimization strategy of deep learning used in power system detection. First, the historical loop detection data involves a large number of records, including parameters such as current, voltage, phase angle, etc. It is classified and sorted according to the different requirements of the phasor detection task and assigned to the corresponding sub-network modules. Through this targeted input classification, it is ensured that each sub-network can focus on its specific task data, thereby improving the learning efficiency. Then, during the model training process, the mini-batch gradient descent method is used for parameter update. Mini-batch gradient descent is an algorithm that balances computational efficiency and model stability. Compared with the gradient descent of the entire data, it can reduce the computational overhead and prevent falling into local minima. For all basic monitoring sub-networks, the goal is to reduce the prediction error of each sub-network by adjusting the weights, thereby gradually reducing the value of the total loss function. To achieve this, advanced gradient descent algorithms such as Adam and RMSProp are used to further improve the optimization efficiency and convergence speed. Finally, after multiple rounds of iteration, when the total loss function reaches the preset minimization goal, it indicates that the sub-networks of the model have been successfully trained, providing a reliable basis for subsequent practical applications.

[0106] In one implementation, inputting device data into a multi-object phasor detection model, and outputting the loop phasor detection result of an intelligent substation through the multi-object phasor detection model includes the following steps:

[0107] Obtain the device resource status of the edge computing device;

[0108] Determine the current task set and the task priorities of all tasks in the current task set by parsing the loop phasor detection task;

[0109] Construct a computing resource allocation model for the multi-object phasor detection model by combining the device resource status, the current task set, and the task priorities;

[0110] Generate a reward function for the computing resource allocation model with the shortest time to complete all tasks in the current task set and the optimal resource utilization rate of the edge computing device;

[0111] Solve the computing resource allocation model based on the reward function and using the deep Q-network algorithm. When the reward function reaches the maximum value, obtain the optimal computing resource allocation strategy for the edge computing device to execute all tasks in the current task set;

[0112] Allocate the remaining device resources of the edge computing device to the feature sharing layer and multiple target basic detection sub-networks corresponding to the current task set according to the optimal computing resource allocation strategy;

[0113] Output the loop phasor detection result of the intelligent substation through all target basic detection sub-networks.

[0114] In this implementation, the status of resources usually includes computing capabilities (such as the usage of CPUs and GPUs), remaining storage space, network bandwidth, and memory usage, etc. This information can be obtained by querying the device's operating system or through specific monitoring tools. Real-time monitoring and collection of these data can adopt an agent-based monitoring scheme, which can deploy a lightweight agent process on the device, and this process is responsible for collecting and reporting the status of resource usage. These status information usually needs to be recorded and analyzed to provide data support for subsequent resource scheduling decisions. The collection of resource status information needs to be formatted, such as through data serialization technologies (such as JSON or XML) to ensure the readability of the data and the simplicity of transmission. When storing this information, a time series database can also be selected, which can efficiently process real-time data streams.

[0115] The task scheduling system is used to classify and sort tasks according to their nature. During this process, the priority needs to be set based on the urgency of the tasks, the scope of influence, and the significance for system stability and security. This can be achieved through preset task attribute tags and weight parameters, where task importance and the amount of resources required for execution are common evaluation criteria. Through the division of priorities, it can be ensured that the most important and urgent tasks are processed first, avoiding arbitrary waste of system resources and task delays. The reasonable allocation of priorities can improve the overall task processing ability, optimize the system response time, and ensure the real-time and effectiveness of power system monitoring.

[0116] To build a computing resource allocation model, it is necessary to comprehensively consider the resource status of the devices and the requirements of the current tasks, including the computing power of edge computing devices, the complexity of the tasks, and the data processing requirements. First, resource allocation equations can be formulated through methods such as linear programming or integer programming. These equations are used to define the resource requirements of the tasks and the resource limitations that the devices can provide. In a multi-objective scenario, it is often necessary to construct a balance equation to ensure that while meeting the basic resource requirements of each task, the resource utilization rate of the entire computing system is improved. Specifically, by setting an objective function, it can be made to complete the task set in the shortest time and at the same time consider the efficiency of resource use, and solve it under a multi-variable optimization framework. By determining the unit resource benefit of different tasks (for example, the amount of tasks that can be completed per unit of computing resource), the model can be dynamically adjusted to adapt to different tasks and resource changes.

[0117] Generating a reward function for the computing resource allocation model with the shortest time to complete all tasks in the current task set and the optimal resource utilization rate of edge computing devices is part of establishing a feedback mechanism. The reward function is used in the computing field to guide and optimize the decision-making process. Its core idea is to seek a solution that minimizes the completion time of all tasks and maximizes the resource utilization rate in a specific computing resource and task allocation strategy. When constructing the reward function, it is necessary to first define evaluation indicators, such as the completion time of tasks, the usage ratio and utility of resources, etc. These indicators need to be combined through reasonable mathematical formula relationships to form a comprehensive reward value. For the measurement of time and resource rate, the commonly used method is the weighted average of the unit task completion time and the device utilization rate. The completion time can be calculated by counting the start and end times of each task, and the resource rate is measured by the ratio of the available resources of the device to the allocated resources. To balance the calculation results, a penalty parameter can be introduced to give negative rewards for the situation where the resource utilization rate is lower than a certain threshold or the task completion exceeds the time limit. Through such a reward function design, it is possible to guide the resource allocation algorithm to converge to the optimal direction without sacrificing performance requirements, thus achieving the high efficiency of resource allocation and the rationality of strategies in a complex computing environment.

[0118] Based on the reward function and using the Deep Q-Network (DQN) algorithm to solve the computing resource allocation model, when the reward function reaches its maximum value, the optimal computing resource allocation strategy for the edge computing device to execute all tasks in the current task set is obtained. This process involves the application of deep reinforcement learning. The Deep Q-Network (DQN) algorithm is a reinforcement learning method that combines deep learning and Q-learning. It uses a neural network to estimate the Q-values of each state-action pair, that is, to evaluate the long-term benefits of taking a certain action in a specific state. To optimize the computing resource allocation model, DQN uses the device resource status and the task set as inputs to predict the reward values of each possible resource allocation strategy. By continuously updating the weights of the neural network, DQN can gradually approach the optimal strategy, that is, take the most correct resource allocation behavior in each task state. The reward function plays a role in evaluating the decision output in this process, guiding the model to continuously adjust itself to achieve the optimization of the resource allocation strategy. Through iterative experiments and the experience replay mechanism, the DQN algorithm can converge quickly and effectively in a dynamic and complex computing environment and find a strategy with high rewards. Under the guidance of this strategy, the edge computing device can efficiently complete the allocation of task loads, maximizing the utilization of computing resources.

[0119] Allocating the remaining device resources of the edge computing device to the feature sharing layer and multiple target basic detection subnets corresponding to the current task set according to the optimal computing resource allocation strategy is a key measure to effectively achieve multi-task processing. In this implementation process, the feature sharing layer is used to extract and learn the common features in the task data, so it requires a certain amount of computing resource support. In addition, multiple basic detection subnets perform in-depth analysis and processing for specific tasks, such as voltage detection, phasor calculation, etc. According to the obtained optimal resource allocation strategy, the remaining resources will be reasonably allocated to different subnets and the feature sharing layer to ensure that all tasks are processed most effectively with appropriate computing resource support. This allocation process can be achieved by adjusting the resource allocation ratio, ensuring that the resources are moderately tilted according to the priorities and the requirements of the task workloads, and improving the satisfaction of specific task requirements. At the same time, a dynamic resource scheduling mechanism can also be introduced to continuously adjust the device resource allocation strategy according to the real-time feedback of task execution to adapt to the changes in task loads. Under this optimized resource allocation, the device can operate more flexibly and efficiently, greatly enhancing its ability to handle large-scale multi-tasks and complex computations.

[0120] Each target basic detection sub-network is dedicated to solving specific power system detection problems and achieving accurate detection of phasor data through an efficient computational resource allocation strategy. The final detection results include the measurement of various power parameters and their deterministic analysis, and these data are published through the output interface and used for further operation and maintenance and technical decision-making. Through the verification and feedback of the detection results, deficiencies in the previous resource allocation and computational models can be discovered and optimized, forming a complete closed-loop control and continuously improving in future task processing. In addition, the detection results can be updated to the power grid monitoring platform through the information sharing interface to achieve more intelligent analysis and management.

[0121] In this embodiment, the expression formula of the optimal computational resource allocation strategy is as follows:

[0122]

[0123] Where: π * represents the optimal computational resource allocation strategy, argmax π represents selecting the computational resource allocation strategy that can maximize the reward function, E represents the cumulative reward expectation value based on the reward function, t represents the time step, γ t represents the discount factor, α and β both represent weight coefficients, T(s t , a t ) represents the reward value in the task completion time dimension when taking the action a t in the target action space under the state s t in the target state space, and R(s t , a t ) represents the penalty value in the resource utilization rate dimension when taking the action a t in the target action space under the state s t in the target state space.

[0124] In one of the embodiments, the computational resource allocation model for constructing a multi-target phasor detection model by combining the device resource status, the current task set, and the task priority includes the following steps:

[0125] Construct a target state space based on the Markov decision model and by combining the device resource status, the current task set, and the task priority;

[0126] Construct a target action space with the selection step of choosing the next task in the current task set and the preset initial resource allocation strategy of the edge computing device;

[0127] Construct a computational resource allocation model for the multi-target phasor detection model by combining the target state space, the target action space, and the preset state transition probability function.

[0128] In this embodiment, as a dynamic decision-making tool, the Markov decision model can effectively describe decision-making problems with randomness and temporal correlation. When defining the state space, it is necessary to combine the resource status of the device, including information on hardware resources such as currently available CPU and storage, and also consider the characteristics of the device tasks, such as the scale and complexity of the current task set. Each state can be regarded as the specific configuration and load situation of the device at a certain moment. The priority of the task is an important reference factor for constructing the state space, and the value and transition path of the state can be affected by setting the priority weight. Combining this information, the state space will cover all potential configuration states and have the ability to abstract complex system conditions. The state transitions under different behavioral strategies need to be analyzed in detail to form a state transition matrix, which depicts how the system transitions between various states under different action selections.

[0129] Construct the target action space based on the selection step of the next task in the current task set and the preset initial resource allocation strategy of the edge computing device. In this process, the action space contains all possible behaviors that can be taken in a given state, determining how the current selection affects future states. First, considering the selection of each task in the current task set as a scheduling behavior, each selection step corresponds to a decision-making action. This step is not just about picking a task from the task set, but rather involves evaluating the importance, urgency of the task and its impact on the overall system, and then selecting the task worthy of being executed first. In addition, the initial resource allocation strategy defines how to initially allocate computing resources under the current device resource conditions to achieve the best performance. Combining these elements, the construction of the action space needs to consider the differential impacts of different task combination selections and their resource allocation strategies. In this process, detailed action descriptions need to be formulated, and possible combination schemes are presented in the form of charts or matrices. The reasonable construction of the action space helps to include potential strategies and ensure that an optimized solution that meets the conditions can always be found under different strategic paths.

[0130] Construct a computational resource allocation model for a multi-object phasor detection model by combining the target state space, the target action space, and a preset state transition probability function. The computational resource allocation model forms an integrated decision-making model by integrating the previously constructed state space and action space and combining the transition probabilities between states. The role of the state transition probability function in this process is to describe the likelihood of a state transitioning to another state after taking a certain action. These probabilities can be estimated based on historical data statistics or designed according to system simulations or expert knowledge. During the model construction process, the probability distribution of each state transition process needs to be defined and calibrated to ensure that the model can accurately define the behavioral responses in various possible scenarios. By constructing a multi-object phasor detection model, the model aims to simultaneously meet multiple resource optimization goals, such as maximizing resource utilization, minimizing task response time, and satisfying task priorities. In the resource allocation model, by analyzing each strategy, the optimal order of task execution and resource allocation can be identified, enabling all goals to receive effective decision support under each strategy. Finally, through the iterative optimization of the model, a dynamic resource allocation strategy framework that conforms to the current device resource status and task requirements is formed, providing an intelligent decision support system for edge computing devices when executing complex tasks.

[0131] The present invention also discloses a loop phasor detection device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the substation loop phasor detection method based on lightweight artificial intelligence in any of the above-mentioned embodiments.

[0132] Among them, the processor can adopt a central processing unit (CPU). Of course, according to the actual usage situation, other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. can also be used. The general-purpose processor can adopt a microprocessor or any conventional processor, etc. This application does not make any restrictions in this regard.

[0133] Among them, the memory can be an internal storage unit of the computer device, for example, the hard disk or memory of the computer device, or it can also be an external storage device of the computer device, for example, a plug-in hard disk, a smart media card (SMC), a secure digital card (SD), or a flash card (FC) equipped on the computer device. Moreover, the memory can also be a combination of the internal storage unit and the external storage device of the computer device. The memory is used to store the computer program and other programs and data required by the computer device. The memory can also be used to temporarily store the data that has been output or will be output. This application does not make any restrictions in this regard.

[0134] The present invention also discloses a substation loop phasor detection system based on lightweight artificial intelligence. Referring to Figure 3 , the system includes:

[0135] A loop phasor detection device;

[0136] An edge computing device, deployed in an intelligent substation, for carrying a multi-target phasor detection model constructed based on a convolutional neural network model;

[0137] An intelligent sensor network, deployed in the substation loop of the intelligent substation and communicatively connected to the edge computing device, for collecting device data of all loop devices in the substation loop;

[0138] A current output device, deployed on the primary circuit side of the substation loop, for injecting current into the substation loop when performing a loop phasor detection task;

[0139] A voltage output device, deployed on the primary circuit side of the substation loop, for injecting voltage into the substation loop when performing a loop phasor detection task.

[0140] Those of ordinary skill in the art should understand that the discussion of any of the above embodiments is only exemplary and is not intended to imply that the scope of protection of this application is limited to these examples; under the concept of this application, the technical features in the above embodiments or different embodiments can also be combined, and the steps can be implemented in any order, and there are many other variations in different aspects of one or more of the above embodiments of this application, which are not provided in detail for the sake of brevity.

[0141] One or more embodiments of this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of this application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments of this application shall be included within the scope of protection of this application.

Claims

1. A substation circuit phasor detection method based on lightweight artificial intelligence, characterized in that: Applied to a smart substation, the smart substation is deployed with an edge computing device, the substation loop in the smart substation is deployed with an intelligent sensor network, the intelligent sensor network is communicatively connected with the edge computing device, and a current output device and a voltage output device are also deployed on the primary loop side of the substation loop; The method comprises the following steps: Obtaining a loop phasor detection task of the smart substation; Based on the convolutional neural network model, a multi-target phasor detection model is constructed according to the task type in the loop phasor detection task, and the multi-target phasor detection model is deployed in the edge computing device, wherein the multi-target phasor detection model includes a feature sharing layer and a multi-target detection task layer; Acquire historical loop detection data of the smart substation and extract static topological features of the substation loop; The static topological features are used to complete the training of the feature sharing layer, and the historical loop detection data are used to complete the training of the multi-target detection task layer; In response to the loop phasor detection task, the current output device and the voltage output device are controlled to pressurize and flow in the substation loop to simulate the load operation process of the substation loop; During the pressurized flow process, the equipment data of all circuit equipment in the substation circuit are collected through the intelligent sensor network; The device data is input into the multi-objective phasor detection model, and the loop phasor detection result of the smart substation is output through the multi-objective phasor detection model.

2. The substation circuit phasor detection method based on lightweight artificial intelligence according to claim 1 is characterized in that: The loop phase quantity detection task is any one or more of loop device phase sequence correctness detection, loop device polarity correctness detection, loop device transformation ratio correctness detection and cable connection performance detection.

3. The substation circuit phasor detection method based on lightweight artificial intelligence according to claim 1 is characterized in that: The multi-target phasor detection model is constructed based on the convolutional neural network model according to the task type in the loop phasor detection task, and includes the following steps: Building a feature sharing layer and a plurality of initial detection subnetworks based on a convolutional neural network model, wherein the number of the initial detection subnetworks is the same as the number of task types in the loop phasor detection task, and all the initial detection subnetworks have independent output layers; According to the task type in the loop phasor detection task, the layer structure and model parameters of each of the initial detection sub-networks are adjusted respectively to obtain a plurality of basic detection sub-networks; The input layers in the plurality of basic detection subnetworks are respectively connected to the output layers in the feature sharing layer to obtain a multi-target phasor detection model, and all the basic detection subnetworks are arranged in parallel to form a multi-target detection task layer.

4. The substation circuit phasor detection method based on lightweight artificial intelligence according to claim 3 is characterized in that: The step of obtaining the historical loop detection data of the smart substation and extracting the static topological features of the substation loop comprises the following steps: Acquire historical loop detection data of the smart substation and equipment parameters of all loop equipment in the substation loop; The circuit devices are used as device nodes, and the electrical connection relationships between all the circuit devices in the substation circuit are used as device node edges to construct a substation electrical connection diagram, and node attributes are assigned to all device nodes based on the device parameters; Identifying all electrical circuits in the electrical connection diagram of the substation using a recursive graph traversal algorithm; By performing node aggregation analysis on all the electrical circuits, busbar structural features in all the electrical circuits are marked; All the equipment parameters and all the electrical circuits are integrated into static topological characteristics of the substation circuit.

5. The substation circuit phasor detection method based on lightweight artificial intelligence according to claim 4 is characterized in that: The method of completing the training of the feature sharing layer by using the static topological features and completing the training of the multi-target detection task layer by using the historical loop detection data comprises the following steps: Using the feature sharing layer as an encoder, mapping the static topological features to a low-dimensional space through the encoder; Reconstructing the static topological features in the low-dimensional space by a preset decoder, and completing the training of the feature sharing layer when the reconstruction error reaches a minimum value; Defining a subnetwork loss function for each of the basic monitoring subnetworks, and taking the weighted sum of all the subnetwork loss functions as the total loss function of the multi-target detection task layer; According to all task types of the loop phasor detection task, the historical loop detection data are classified and input into all the basic monitoring sub-networks, all the basic monitoring sub-networks are trained using the small batch gradient descent method, and the total loss function is minimized using a gradient descent algorithm to complete the training of the multi-target detection task layer.

6. The substation circuit phasor detection method based on lightweight artificial intelligence according to claim 3 is characterized in that: The step of inputting the device data into the multi-objective phasor detection model and outputting the loop phasor detection result of the smart substation through the multi-objective phasor detection model comprises the following steps: Obtaining a device resource status of the edge computing device; Determine the task priorities of a current task set and all tasks in the current task set by parsing the loop phasor detection task; Constructing a computing resource allocation model of the multi-objective phasor detection model in combination with the device resource status, the current task set and the task priority; Generate a reward function for the computing resource allocation model based on the shortest time to complete all tasks in the current task set and the optimal resource utilization of the edge computing device; Based on the reward function and using the deep Q network algorithm to solve the computing resource allocation model, when the reward function When the maximum value is reached, the optimal computing resource allocation strategy for the edge computing device to execute all tasks in the current task set is obtained; Allocating the remaining device resources of the edge computing device to the feature sharing layer and a plurality of target basic detection subnetworks corresponding to the current task set according to the optimal computing resource allocation strategy; The loop phasor detection result of the smart substation is output through all the target basic detection subnetworks.

7. The substation circuit phasor detection method based on lightweight artificial intelligence according to claim 6 is characterized in that: The computing resource allocation model for constructing the multi-objective phasor detection model in combination with the device resource status, the current task set and the task priority comprises the following steps: Constructing a target state space based on a Markov decision model and in combination with the device resource state, the current task set and the task priority; Constructing a target action space by selecting a selection step of a next task in the current task set and an initial resource allocation strategy preset by the edge computing device; A computing resource allocation model of the multi-target phasor detection model is constructed by combining the target state space, the target action space and a preset state transition probability function.

8. The substation circuit phasor detection method based on lightweight artificial intelligence according to claim 7 is characterized in that: The expression formula of the optimal computing resource allocation strategy is as follows: Where: π * represents the optimal computing resource allocation strategy, argmax π represents the computing resource allocation strategy that can maximize the reward function, E represents the expected value of the cumulative reward based on the reward function, t represents the time step, γ t represents the discount factor, α and β represent the weight coefficients, T(s t , a t ) represents the state s in the target state space t Take action a in the target action space t The reward value in the task completion time dimension, R(s t , a t ) represents the state s in the target state space t Take action a in the target action space t The penalty value in the resource utilization dimension.

9. A loop phase detection device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the substation circuit phasor detection method based on lightweight artificial intelligence as described in any one of claims 1 to 8 is implemented.

10. A substation loop phasor detection system based on lightweight artificial intelligence, characterized in that: include: The loop phasor detection device according to claim 9; Edge computing devices are deployed in smart substations to carry multi-target phasor detection models built based on convolutional neural network models; An intelligent sensor network, deployed in a substation loop of the intelligent substation and in communication connection with the edge computing device, for collecting device data of all loop devices in the substation loop; A current output device, deployed on the primary circuit side of the substation circuit, for injecting current into the substation circuit when performing a circuit phasor detection task; The voltage output device is deployed on the primary circuit side of the substation circuit and is used to inject voltage into the substation circuit when performing the circuit phasor detection task.

Images