Substation equipment fault early warning method and system

Abstract

The invention relates to a substation equipment fault early warning method and system, and belongs to the technical field of intelligent substations, and the method comprises the steps: obtaining the target operation state data of substation equipment, the target operation state data comprising target sensor data and target image data obtained by an unmanned plane which carries out the field inspection of a substation; inputting the target operation state data into a completely trained fault diagnosis model, extracting spatial features of target image data by the completely trained fault diagnosis model through a long-short-term memory network, extracting time features of target sensor data through a convolutional neural network, and splicing and fusing the spatial features and the time features to obtain a fault diagnosis result; outputting a fault prediction result; and performing early warning based on the fault prediction result. According to the invention, the fault diagnosis model is combined with the spatial features of the image data and the time sequence features of the sensor data to carry out early warning on the substation equipment fault, and multi-modal data fusion and fault prediction of the substation equipment are realized.

Description

Technical Field

[0001] This application relates to the technical field of intelligent substations, and particularly to a method and system for early warning of substation equipment failures. Background Art

[0002] A substation is an important part of the power system, used to convert high-voltage electrical energy into low-voltage electrical energy, or vice versa, for various equipment and users in the power system. It is one of the nodes for power transmission, distribution, conversion, and control in the power system.

[0003] In the existing substation equipment management and operation and maintenance mode, a monitoring system based on simple sensors often can only provide limited data, with single data, making it difficult to achieve accurate prediction and diagnosis of equipment failures. Summary of the Invention

[0004] In view of this, it is necessary to provide a method and system for early warning of substation equipment failures to solve the problem in the prior art that the data of the monitoring system based on simple sensors is single and it is difficult to achieve accurate prediction and diagnosis of equipment failures.

[0005] To solve the above problems, this application provides a method for early warning of substation equipment failures, including: Obtaining target operation status data of substation equipment, where the target operation status data includes target sensor data and target image data obtained by an unmanned aerial vehicle (UAV) patrolling at the substation site; Inputting the target operation status data into a trained fault diagnosis model, where the trained fault diagnosis model extracts spatial features of the target image data through a long short-term memory network, extracts temporal features of the target sensor data through a convolutional neural network, and splices and fuses the spatial features and the temporal features to output a fault prediction result; Performing early warning based on the fault prediction result.

[0006] In some possible implementation manners, obtaining target operation status data of substation equipment, where the target operation status data includes target sensor data and target image data obtained by an unmanned aerial vehicle (UAV) patrolling at the substation site, includes: Obtaining original operation status data of substation equipment, where the original operation status data includes original sensor data and original image data obtained by an unmanned aerial vehicle (UAV) patrolling at the substation site; Preprocessing the original sensor data to obtain intermediate sensor data, and preprocessing the original image data to obtain intermediate image data; Performing time synchronization calibration on the intermediate sensor data and the intermediate image data to obtain target operation status data including target sensor data and target image data.

[0007] In some possible implementation manners, the original image data includes original visible light image data and original infrared thermal image data. The original visible light image data is acquired by a high-definition camera carried on the unmanned aerial vehicle, and the original infrared thermal image data is acquired by an infrared thermal imager carried on the unmanned aerial vehicle.

[0008] In some possible implementation manners, preprocessing the original sensor data to obtain intermediate sensor data, including: Removing high-frequency noise in the original sensor data to obtain first timing data; Performing outlier processing on the first timing data to obtain second timing data; Performing normalization processing on the second timing data to obtain third timing data; Segmenting the third timing data into sequence segments of a preset length to obtain intermediate timing data.

[0009] In some possible implementation manners, preprocessing the original image data to obtain intermediate image data, including: Performing image stitching, image enhancement, and target detection on the original image data to obtain intermediate image data.

[0010] In some possible implementation manners, the fault diagnosis model is trained using a cross-entropy loss function.

[0011] In some possible implementation manners, the fault prediction result includes a fault type and a fault probability. Based on the fault prediction result, early warning is performed, including: Obtaining substation mapping data; Constructing a digital twin model of the substation according to the substation mapping data and the target sensor data; When the fault probability is greater than a preset threshold, or the fault type is a preset type, synchronizing the fault prediction result to the digital twin model for visual display.

[0012] In some possible implementation manners, constructing a digital twin model of the substation according to the substation mapping data and the target sensor data, including: Constructing a three-dimensional digital model of the substation based on the Unreal Engine platform and the substation mapping data; Importing the target sensor data into the three-dimensional digital model and mapping it to the corresponding devices in the three-dimensional digital model to obtain a digital twin model of the substation.

[0013] This application also provides a substation equipment fault early warning system, including: The terminal inspection module includes a drone body for on-site inspection in the substation and an image data acquisition unit carried on the drone, and the image data acquisition unit is used to acquire original image data; The edge service module is deployed near the substation and is used to receive the original sensor data transmitted from the substation and the original image data transmitted from the terminal inspection module, preprocess the original sensor data and the original image data to obtain target sensor data and target image data, and transmit the target sensor data and the target image data to the cloud service module; The cloud service module is used to extract the spatial features of the target image data through a long short-term memory network based on a trained and complete fault diagnosis model, extract the temporal features of the target sensor data through a convolutional neural network, splice and fuse the spatial features and the temporal features, and output a fault prediction result; The warning module is used to give a warning based on the fault prediction result.

[0014] In some possible implementation manners, the warning module is a digital twin model, and the digital twin model is constructed according to the substation mapping data and the original operation state data and is used to visually display the fault prediction result.

[0015] The beneficial effect of this application is that: the substation equipment fault warning method provided by this application uses a long short-term memory network to extract the spatial features of image data, uses a convolutional neural network to extract the temporal features of the time series data, and splices and fuses the spatial features and the temporal features and outputs, realizing multi-modal data fusion and fault prediction of substation equipment, making full use of the complementary information in the drone image data and the sensor time series data, and improving the accuracy of fault prediction. Description of the Drawings

[0016] Figure 1 It is a schematic flowchart of an embodiment of the substation equipment fault warning method provided by this application; Figure 2 For this application Figure 1 It is a schematic flowchart of an embodiment of step S101 in this application; Figure 3 For this application Figure 2 It is a schematic flowchart of an embodiment of step S202 in this application; Figure 4 For this application Figure 1 It is a schematic flowchart of an embodiment of step S102 in this application; Figure 5 For this application Figure 1 It is a schematic flowchart of an embodiment of step S103 in this application; Figure 6 For this applicationFigure 5 Schematic flowchart of an embodiment of step S502; Figure 7 Schematic structural diagram of an embodiment of the substation equipment fault warning system provided by the present application. Detailed implementation manners

[0017] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts shall fall within the protection scope of the present application.

[0018] It should be understood that the schematic drawings are not drawn to actual scale. The flowcharts used in the present application illustrate the operations implemented according to some embodiments of the present application. It should be understood that the operations in the flowchart may not be implemented in sequence, and steps without logical context relationships may be reversed or implemented simultaneously. In addition, those skilled in the art can add one or more other operations to the flowchart or remove one or more operations from the flowchart under the guidance of the content of the present application. Some of the block diagrams shown in the drawings are functional entities, which do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor systems and / or microcontroller systems.

[0019] The descriptions such as "first" and "second" involved in the embodiments of the present application are only for the purpose of implicit description, and cannot be understood as indicating or implying their relative importance or implicitly indicating the quantity of the indicated technical features. Therefore, the technical features defined with "first" and "second" may explicitly or implicitly include at least one of such features. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist, for example: A and / or B may represent: A exists alone, A and B exist simultaneously, and B exists alone.

[0020] Referring to "embodiment" in this text means that the specific features, structures or characteristics described in combination with the embodiment may be included in at least one embodiment of the present application. The phrase appears in various positions in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0021] The present application provides a substation equipment fault warning method, which will be described separately below.

[0022] Figure 1 A schematic flowchart of an embodiment of the substation equipment fault warning method provided by this application is shown as Figure 1 shown. The substation equipment fault warning method includes: S101. Obtain the target operation status data of the substation equipment, where the target operation status data includes target sensor data and target image data obtained by a drone performing on-site inspections at the substation; S102. Input the target operation status data into a trained fault diagnosis model. The trained fault diagnosis model extracts the spatial features of the target image data through a long short-term memory network, extracts the temporal features of the target sensor data through a convolutional neural network, splices and fuses the spatial features and temporal features, and outputs a fault prediction result; S103. Perform a warning based on the fault prediction result.

[0023] Compared with the prior art, this application uses a long short-term memory network to extract the spatial features of image data, uses a convolutional neural network to extract the temporal features of time series data, and splices and fuses the spatial features and temporal features, realizing multi-modal data fusion and fault prediction of substation equipment, making full use of the complementary information in the drone image data and sensor time series data, and improving the accuracy of fault prediction.

[0024] In order to improve the data quality of the target operation status data, in some embodiments, it is necessary to preprocess the original operation status data. Specifically, as Figure 2 shown, step S101 includes: S201. Obtain the original operation status data of the substation equipment, where the original operation status data includes original sensor data and original image data obtained by a drone performing on-site inspections at the substation; It should be noted that the original sensor data includes the internal status, operation status, and environmental parameters during the operation of the equipment collected in real time by various types of sensors deployed at the substation site. For example, temperature sensors monitor the temperature changes of key parts such as the oil temperature and bushing temperature of the transformer, pressure sensors monitor the oil pressure and gas pressure of the transformer, current and voltage transformers monitor the current and voltage changes of the equipment, and vibration sensors monitor the vibration frequency and amplitude of the equipment.

[0025] It should also be noted that the original image data is specifically collected by sensors such as high-definition cameras and infrared thermal imagers carried on the drone body performing on-site inspections at the substation, including the original visible light image data collected by the high-definition camera and the original infrared thermal image data collected by the infrared thermal imager, which can provide visual information and temperature information of the equipment and provide a data basis for subsequent fault diagnosis.

[0026] S202. Preprocess the original sensor data to obtain intermediate sensor data, and preprocess the original image data to obtain intermediate image data; S203. Perform time synchronization calibration on the intermediate sensor data and the intermediate image data to obtain target operating state data including target sensor data and target image data.

[0027] It should be noted that to ensure data consistency and timeliness, both the original sensor data and the original image data carry timestamps and are time-synchronized and calibrated to ensure that different types of data can be aligned in time.

[0028] Furthermore, to better preprocess the original sensor data, in some embodiments, the preprocessing includes data cleaning and data segmentation. Among them, data cleaning can remove errors and interferences in the data and improve data accuracy and reliability. Specifically, as Figure 3 shown, the preprocessing of the original sensor data in step S202 to obtain intermediate sensor data includes: S301. Remove high-frequency noise from the original sensor data to obtain first-time series data; It should be noted that denoising specifically uses methods such as moving average filters and Kalman filters to smooth the original sensor data, remove high-frequency noise in it, and reduce the interference of noise on the model.

[0029] S302. Perform outlier processing on the first-time series data to obtain second-time series data; It should be noted that outlier processing specifically uses statistical methods, such as box plots, 3 sigma principle, etc., to identify and process outliers in the data, such as missing values and outliers. For missing values, methods such as deletion, filling (such as mean, median, mode imputation, linear interpolation, K-nearest neighbor interpolation) can be used according to specific situations. For obvious outliers, methods such as truncation and replacement can be used for processing.

[0030] S303. Perform normalization processing on the second-time series data to obtain third-time series data.

[0031] It should be noted that since the data collected by different sensors have different dimensions and value ranges, in order to eliminate this difference, it is necessary to perform normalization processing on the data. Usually, the Min-Max normalization method is used to scale all data into the [0,1] interval; for example, the temperature data may vary between 20 and 80 degrees Celsius, while the current data may vary between 0 and 100 amperes. After normalization processing, all data are within the same numerical range, eliminating the influence brought by dimensional differences, which helps to improve the training efficiency and prediction accuracy of the model.

[0032] S304. Split the third time series data into sequence segments of a preset length to obtain intermediate time series data.

[0033] It should be noted that data splitting is to split data into sequence segments of a fixed length according to a time window for input into a model for training and diagnosis. For example, data continuously collected for 30 seconds or 1 minute can be used as a data segment, and each segment is used as an independent sample.

[0034] Furthermore, in order to better preprocess the original image data, in some embodiments, preprocessing the original image data in step S202 to obtain intermediate image data includes: performing grayscale conversion, size adjustment, image stitching, image enhancement, and target detection on the original image data to obtain intermediate image data.

[0035] It should be noted that data augmentation specifically uses methods such as rotation, cropping, scaling, brightness adjustment, contrast adjustment, and adding noise to expand the scale of the image dataset and increase data diversity. These operations can simulate images under different perspectives and different lighting conditions, so that the trained model is more robust and can adapt to various complex actual scenarios.

[0036] In some embodiments, the fault diagnosis model includes a feature extraction layer, a fully connected layer, and an output layer connected in sequence. Among them, the feature extraction layer includes a parallel long short-term memory network module and a convolutional neural network module. Specifically, as Figure 4 shown, step S102 specifically includes: S401. Extract features from the target sensor data based on the long short-term memory network module to obtain a time series feature vector, extract features from the target image data based on the convolutional neural network module to obtain a spatial feature vector, and splice the time series feature vector and the spatial feature vector to obtain a spliced feature vector; It should be noted that the long short-term memory network (LSTM) is a special recurrent neural network (RNN) that can capture long-term dependencies in time series data. In a specific embodiment, a two-layer LSTM network is used, with each layer containing 128 LSTM units. The long-term dependencies in the time series data are captured through the input gate, forget gate, and output gate inside the LSTM, and the hidden state of the last time step of the LSTM network is output as the time series feature vector.

[0037] It should also be noted that the Convolutional Neural Network (CNN) uses ResNet-50 pre-trained on the ImageNet dataset as the backbone network and is fine-tuned according to the characteristics of substation equipment images. The input image data is first processed through a series of convolutional layers, pooling layers, and activation functions. In the convolutional layer, the convolutional kernel slides on the image to extract local features of different scales, such as edges, corners, textures, etc. After the convolutional operation, a pooling layer (such as max pooling) is used to reduce the dimension of the feature map, retain the most important features while reducing the computational amount, and enhance the translational invariance of the model. The activation function (such as ReLU) introduces non-linearity to improve the expressive ability of the model. After iterative processing through multiple convolutional layers, pooling layers, and activation functions, the feature map output by the last pooling layer is flattened to obtain a one-dimensional spatial feature vector, which highly summarizes the spatial information related to equipment defects in the image.

[0038] It should also be noted that the concatenation specifically concatenates the spatial feature vector of the CNN (e.g., 25088 dimensions) and the temporal feature vector of the LSTM (e.g., 128 dimensions) directly together to form a longer concatenated feature vector (e.g., 25216 dimensions). This fused feature vector contains both the spatial information of the image data and the temporal information of the sensor data, providing more comprehensive information for subsequent fault prediction.

[0039] S402. Extract the fused feature vector based on the fully connected layer. It should be noted that a series of fully connected layers further extract and transform the features of the concatenated feature vector. For example, the first fully connected layer contains 512 neurons, the second fully connected layer contains 128 neurons, and each fully connected layer is followed by a ReLU activation function to introduce non-linear transformation. The fully connected layer maps the fused feature vector to a lower-dimensional space by weighted summation.

[0040] S403. Perform fault classification on the fused feature vector based on the output layer to obtain the fault prediction result.

[0041] It should be noted that for the binary classification task of judging whether the equipment is faulty, the output layer is a single neuron, using the Sigmoid activation function to output a probability value between 0 and 1, representing the probability of the equipment having a fault. For the multi-class classification task of judging the type of equipment fault (e.g., 5 types of faults), the output layer has the same number of neurons as the number of fault types (e.g., 5), using the Softmax activation function to output a probability distribution, representing the probability of the equipment belonging to each fault type.

[0042] The training process of the above model is as follows: First, divide the target sensor data and target image data into a training set, a validation set, and a test set according to a certain ratio (e.g., 8:2); among them, the training set is used to train the parameters of the model, the validation set is used to monitor the performance of the model during training and perform hyperparameter tuning, and the test set is used to evaluate the final performance of the model.

[0043] Then, input the training set data into the fault diagnosis model for training; when training the model, the Adam optimizer is used for parameter update, and the cross-entropy loss function is used to measure the difference between the prediction result and the true label. For binary classification problems, the binary cross-entropy loss function is used, and for multi-class classification problems, the categorical cross-entropy loss function is used; during the training process, the Mini-batch Training method is adopted, the training set is divided into multiple batches, and each time a batch of data is input into the model for training, and backpropagation is performed according to the loss value of this batch to update the model parameters; during each iterative training process, in addition to the training set data, the validation set data is also used to monitor the performance of the model, and the model parameters (such as learning rate, batch size, number of LSTM units, number of neurons in the fully connected layer, etc.) are adjusted according to the performance metrics (such as accuracy, recall, F1-score, etc.) on the validation set; in order to avoid overfitting of the model on the training set, the early stopping method is adopted, that is, when the performance of the model on the validation set does not improve within a certain number of rounds, the training is stopped in advance; in addition, techniques such as regularization and batch normalization are also adopted to improve the generalization ability of the model. Among them, regularization is to add an L1 or L2 regularization term to the loss function to constrain the model parameters, and batch normalization is to add a batch normalization layer after each convolutional layer and fully connected layer to accelerate the training process and improve the stability of the model.

[0044] After the training is completed, use the test set to conduct the final performance evaluation of the model. The test set data does not participate in the training process of the model and is used to simulate the performance of the model in actual applications. The evaluation metrics include Accuracy, Recall, F1-score, Precision, and AUC value, etc. These metrics reflect the performance of the model from different perspectives; for example, accuracy measures the proportion of samples predicted correctly by the model, recall measures the proportion of positive example samples identified by the model among all true positive example samples, and the F1-score is the harmonic mean of precision and recall, comprehensively reflecting the performance of the model; by evaluating the performance of the model on the test set, the generalization ability and practicality of the model can be objectively understood. According to the evaluation results, the model can be further adjusted and optimized until the model performance meets the expected requirements.

[0045] Through the above steps, a fault diagnosis model with good performance can be trained for substation equipment fault early warning. This model can make full use of the information in the UAV image data and sensor time-series data to improve the accuracy and reliability of fault prediction, providing strong support for the intelligent operation and maintenance of substations.

[0046] Finally, in order to better conduct early warning, in some embodiments, the fault prediction results include the fault type and the fault probability. For example, Figure 5 as shown, step S103 includes: S501. Obtain substation surveying and mapping data; It should be noted that the substation surveying and mapping data is obtained by using means such as Internet open-source substation three-dimensional digital model datasets, on-site investigations, and geographic information systems (GIS) to conduct detailed surveying and recording of the physical layout of the substation, equipment locations, connection relationships, etc.

[0047] S502. Based on the Unreal Engine platform and the substation surveying and mapping data, construct a digital twin model of the substation; It should be noted that the digital twin model of the substation is constructed by using digital twin technology, the Unreal Engine platform, and virtual simulation interaction technology, including the topological structure of the substation, equipment characteristics, and their mutual relationships. Combining simulation technology, the model can dynamically reflect the operating state of the substation and be updated in real time. Therefore, it is not only a three-dimensional mapping of the substation physical entity in the virtual space but also integrates multi-dimensional information such as the operating state of the equipment and environmental parameters to achieve real-time, dynamic, and visual display of the operating state of the substation.

[0048] S503. When the fault probability is greater than the preset threshold or the fault type is a preset type, synchronize the fault prediction results to the digital twin model for visual display.

[0049] To better construct the digital twin model, in some embodiments, for example, Figure 6 as shown, step S502 includes: S601. Based on the Unreal Engine platform and the substation surveying and mapping data, construct a three-dimensional digital model of the substation; It should be noted that due to the powerful rendering and interaction capabilities of the Unreal Engine platform, in a specific embodiment, the Unreal Engine platform is used as the development environment for the digital twin model, and a three-dimensional scene of the substation is built in the Unreal Engine platform according to the substation surveying and mapping data, including three-dimensional digital models of primary equipment such as transformers, circuit breakers, disconnectors, busbars, and arresters, as well as secondary equipment such as control rooms and relay rooms, and ensure that the geometric dimensions, relative positions, etc. of the models are consistent with the actual equipment.

[0050] S602. Import the target sensor data into the 3D digital model and map it to the corresponding devices in the 3D digital model to obtain the digital twin model of the substation.

[0051] It should be noted that specifically, real-time data streams such as the target sensor data and the geographical information data obtained through the GIS system are imported into the 3D digital model through methods such as API interfaces or message queues to achieve data synchronization between the digital twin model and the actual substation. Through the interactive settings inside the Unreal Engine, such as setting the display method of the data, the data is mapped to specific devices. For example, the value of the temperature sensor is displayed in real time on the transformer model, and the color of the model is changed according to the value size (such as showing red when the temperature is too high); through the above steps, a digital twin model that can reflect the operating state of the substation in real time is constructed.

[0052] Furthermore, in order to ensure the accuracy and feasibility of the digital twin model, in some embodiments, the digital twin model is compared and verified with the operating state of the actual substation through a method combining quantitative analysis and qualitative evaluation.

[0053] Specifically, in the model verification stage, first, the outputs of the digital twin model, such as the operating state of the device, environmental parameters, etc., are compared with the operating data of the actual substation to calculate the error between the two; then, according to the size and distribution of the error, the accuracy of the model is evaluated; in addition, on-site operation and maintenance personnel are invited to evaluate the digital twin model to ensure the feasibility and practicality of the model; for the errors or deficiencies in the model, optimization algorithms are used for correction. For example, the accuracy of the model can be improved by adjusting model parameters, optimizing model structures, etc.; at the same time, using the internal interactive function of the Unreal Engine, the actual device state, such as the opening and closing state of the switch, is imported into the digital twin model, and according to the change of the actual state, the parameters or state of the model are adjusted to make it consistent with the actual state; in addition, according to the requirements of multi-modal state superposition display such as "sound, light, electricity, magnetism" and equipment risk assessment, the model is optimized to make it more realistically reflect the actual operating state of the substation; through continuous verification and correction, a digital twin model that can accurately and reliably reflect the actual operation of the substation is finally obtained, providing support for subsequent fault prediction and emergency response.

[0054] To better implement a substation equipment fault warning method in the embodiments of the present application, correspondingly, on the basis of a substation equipment fault warning method, as Figure 7 shown, the embodiments of the present application further provide a substation equipment fault warning system 700, including: The terminal inspection module 701 is responsible for executing the intelligent inspection tasks of substation equipment, including the UAV body for on-site inspection of substations and the image data acquisition unit carried on the UAV body. The image data acquisition unit is used to collect the original image data, specifically sensors such as high-definition cameras and infrared thermal imagers; It should be noted that the UAV body is also equipped with a 5G communication unit, a high-precision positioning unit (such as GPS, RTK), an on-board edge computing unit, and an obstacle detection unit. The 5G communication unit is used for real-time data transmission with the edge service module 702, ensuring the timeliness and reliability of the data. The high-precision positioning unit is used for accurate positioning and autonomous flight according to the navigation instructions provided by the preset route digital twin model, realizing all-round and non-blind-spot inspection of substation equipment. During the inspection process, the on-board edge computing unit performs preliminary processing on the collected original image data, such as image compression and quality assessment, and then transmits the processed data and the flight status data of the UAV itself (such as position, attitude, power, etc.) to the edge service module 702 through the 5G network for further processing and analysis. The obstacle detection unit is based on on-board lidar or vision sensors and is used to detect obstacles during flight and avoid them, ensuring flight safety.

[0055] The edge service module 702 consists of intelligent inspection workstations deployed near the substation, and is used to receive the original sensor data transmitted from the substation and the original image data transmitted from the terminal inspection module 701, and preprocess the original sensor data and the original image data to obtain the target sensor data and the target image data, and transmit the target sensor data and the target image data to the cloud service module 703 and the digital twin model 704; It should be noted that each intelligent inspection workstation is equipped with a high-performance computing unit (such as a GPU) and a large-capacity storage unit to support the rapid processing and storage of data. On the one hand, the edge service module 702 receives the inspection data transmitted in real time by the drone through the 5G network, including visible light images, infrared thermal imaging data, and the flight status data of the drone, and stores them; on the other hand, the edge service module 702 is also combined with the digital twin model to map the data collected by the drone into the digital twin model in real time, realizing the linkage between the virtual and the real. For example, the images taken by the drone are superimposed on the corresponding devices in the digital twin model to intuitively display the operating status of the devices; with its powerful computing power, the edge service module 702 performs rapid preprocessing and analysis on the received data, including image stitching, image enhancement, target detection, data cleaning, format conversion, etc. These preprocessing steps provide high-quality input data for the subsequent deep learning model; in addition, the edge service module 702 is also responsible for managing and scheduling the drones. According to the inspection tasks issued by the cloud service module 703 and the real-time status of the drones, it assigns tasks, plans paths, and conducts cooperative control for the drones, realizing multi-drone cooperative inspection and improving the inspection efficiency.

[0056] The cloud service module 703, which consists of multiple high-performance servers deployed in a remote data center, is used to extract the spatial features of the target image data through a long short-term memory network and the temporal features of the target sensor data through a convolutional neural network based on a well-trained fault diagnosis model, and splice and fuse the spatial features and temporal features to output a fault prediction result.

[0057] It should be noted that the cloud data module has powerful computing and storage capabilities, and can support the training of complex deep learning models and the storage of massive data. On the one hand, the cloud service module 703 receives the inspection data uploaded by the edge service module 702, including preprocessed image data, sensor data, etc. On the other hand, the cloud service module 703 also receives data from the digital twin model 704, such as the operating status of the device, environmental parameters, etc. These data will be stored in the database of the cloud service module 703 and used for subsequent deep learning model training and fault diagnosis. The cloud service module 703 utilizes its powerful computing power to deeply analyze and mine the received data. For example, it uses deep learning algorithms to perform refined defect identification on image data and extract temporal features and fault prediction on sensor data. In addition, the cloud service module 703 is also responsible for formulating inspection plans. According to factors such as the operating status of substation equipment, fault risks, and historical inspection data, it generates optimized inspection tasks through intelligent optimization algorithms (such as genetic algorithms, particle swarm algorithms) and issues them to the drones through the edge service module 702 to guide the inspection work of the drones. At the same time, the cloud service module 703 is also responsible for managing and updating the fault diagnosis model, continuously optimizing the model parameters according to the actual operation data, such as adjusting hyperparameters such as the network structure and learning rate of the model to improve the accuracy of fault prediction. Finally, the cloud service module 703 provides the results of in-depth analysis, including fault diagnosis results, inspection plans, etc., to the operation and maintenance personnel through methods such as API interfaces, providing decision-making support for the operation and maintenance management of the substation.

[0058] The warning module is the digital twin model 704, which is constructed based on the substation mapping data and target sensor data and is used to visually display the fault prediction results.

[0059] It should be noted that the cloud service module 703 returns the prediction results to the edge service module 702. If the predicted fault probability exceeds the preset threshold, or the predicted fault type belongs to the predefined serious fault type, the edge service module 702 will trigger the warning mechanism and notify the operation and maintenance personnel by popping up an alarm notification in the digital twin system. At the same time, the warning information is synchronized to the digital twin model 704 for visual display, such as highlighting and marking on the corresponding device model and displaying information such as the fault type, occurrence time, and prediction probability.

[0060] The substation equipment fault warning system 700 provided by the above embodiments can implement the technical solutions described in the embodiments of the substation equipment fault warning method. The specific implementation principles of the above units can be referred to the corresponding content in the embodiments of the substation equipment fault warning method, which will not be elaborated here.

[0061] The above has introduced in detail a substation equipment fault warning method provided by the present application. Specific examples are used in this article to elaborate on the principle and implementation manner of the present application. The description of the above embodiments is only used to help understand the method and its core idea of the present application; at the same time, for those skilled in the art, according to the idea of the present application, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present application.

[0062] As mentioned above, the above is only a preferred specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present application should be covered within the protection scope of the present application.

Claims

1. A method for early warning of substation equipment failure, characterized in that: include: Acquire target operating status data of the substation equipment, wherein the target operating status data includes target sensor data and target image data acquired by a drone inspecting the substation on-site; The target operating state data is input into a well-trained fault diagnosis model, the well-trained fault diagnosis model extracts the spatial features of the target image data through a long short-term memory network, extracts the temporal features of the target sensor data through a convolutional neural network, and splices and fuses the spatial features and the temporal features to output a fault prediction result; An early warning is issued based on the fault prediction result.

2. The transformer substation equipment failure early warning method according to claim 1, characterized in that: Obtaining target operating status data of substation equipment, the target operating status data including target sensor data and target image data acquired by a drone inspecting the substation on-site, including: Acquire original operation status data of the substation equipment, wherein the original operation status data includes original sensor data and original image data acquired by a drone inspecting the substation on site; Preprocessing the raw sensor data to obtain intermediate sensor data, and preprocessing the raw image data to obtain intermediate image data; The intermediate sensor data and the intermediate image data are time-synchronized and calibrated to obtain target operating state data including target sensor data and target image data.

3. The transformer substation equipment failure early warning method according to claim 2, characterized in that: The original image data includes original visible light image data and original infrared thermal image data. The original visible light image data is collected by a high-definition camera carried by the drone, and the original infrared thermal image data is collected by an infrared thermal imager carried by the drone.

4. The transformer substation equipment failure early warning method according to claim 2, characterized in that: Preprocessing the raw sensor data to obtain intermediate sensor data includes: Removing high-frequency noise from the original sensor data to obtain first time series data; Performing outlier processing on the first time series data to obtain second time series data; Normalizing the second time series data to obtain third time series data; The third time series data is divided into sequence segments of a preset length to obtain intermediate time series data.

5. The transformer substation equipment failure early warning method according to claim 2, characterized in that: Preprocessing the original image data to obtain intermediate image data includes: The original image data is subjected to image stitching, image enhancement and target detection to obtain intermediate image data.

6. The transformer substation equipment failure early warning method according to claim 1, characterized in that: The fault diagnosis model is trained using a cross entropy loss function.

7. The method for early warning of substation equipment failure according to claim 2, characterized in that: The fault prediction result includes the fault type and the fault probability, and the early warning is performed based on the fault prediction result, including: Obtain substation mapping data; Constructing a digital twin model of the substation based on the substation mapping data and the target sensor data; When the fault probability is greater than a preset threshold, or the fault type is a preset type, the fault prediction result is synchronized to the digital twin model for visual display.

8. The transformer substation equipment failure early warning method according to claim 7, characterized in that: According to the substation mapping data and the target sensor data, a digital twin model of the substation is constructed, including: Based on the Unreal Engine platform and the substation surveying and mapping data, construct a three-dimensional digital model of the substation; The target sensor data is imported into the three-dimensional digital model and mapped to corresponding devices in the three-dimensional digital model to obtain a digital twin model of the substation.

9. A substation equipment failure early warning system, characterized in that: include: The terminal inspection module includes a drone body for on-site inspection of the substation and an image data acquisition unit carried on the drone, wherein the image data acquisition unit is used to acquire original image data; The edge service module is deployed near the substation, and is used to receive the original sensor data transmitted from the substation and the original image data transmitted from the terminal inspection module, and pre-process the original sensor data and the original image data to obtain target sensor data and target image data, and transmit the target sensor data and target image data to the cloud service module; A cloud service module is used to extract the spatial features of the target image data through a long short-term memory network based on a well-trained fault diagnosis model, extract the temporal features of the target sensor data through a convolutional neural network, and splice and fuse the spatial features and the temporal features to output a fault prediction result; The early warning module is used to issue an early warning based on the fault prediction result.

10. The transformer substation equipment failure early warning system according to claim 9, characterized in that: The early warning module is a digital twin model, which is constructed based on the substation mapping data and the original operating status data, and is used to visualize the fault prediction results.

Images