Cloud collaborative comprehensive energy supply equipment fault identification method and system

Abstract

The invention relates to the technical field of fault identification, and provides a cloud collaborative comprehensive energy supply equipment fault identification method and system. Through a cloud collaborative mode, pre-judgment is carried out based on operation data of equipment; when a fault is identified, collecting operation data and corresponding image data of the current equipment and the associated equipment, and transmitting the operation data and the corresponding image data to a cloud end, and performing fusion judgment on the image data and the operation data through the cloud end; and through pre-judgment, only the equipment data which is judged to be possibly faulty is uploaded, so that the data communication pressure and the data processing amount of the cloud are reduced, and the accuracy and the efficiency of fault prediction are improved. Meanwhile, the data information is identified at the cloud through combination of causal expansion convolution of an attention mechanism and an equipment fault identification network model of a residual convolutional network, and through collaborative analysis of the data information and image information of the integrated energy supply equipment, the defect of a single identification method is overcome, and rapid and accurate identification of faults is realized.

Description

Technical Field

[0001] The present disclosure relates to the technical field related to fault identification, and more specifically, to a cloud-coordinated integrated energy supply equipment fault identification method and system. Background Art

[0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art.

[0003] In traditional energy systems, a large amount of energy loss often occurs during the transmission, distribution and conversion of energy, resulting in low energy utilization. In contrast, the Integrated Energy System (IES) achieves cascade utilization of energy and optimization of multi-energy transfer through the coordinated work of multiple energy devices, thereby greatly improving energy utilization efficiency. The essence of the integrated energy system lies in the coordinated complementarity of its multiple energy devices, making it possible to jointly supply multiple energy sources such as heat, electricity, cooling and gas.

[0004] However, a key challenge of the integrated energy system lies in the high coupling of the subsystems within the system. Since there are multiple connections between energy supply devices, such as energy flow, information flow, and control flow, when a certain energy supply device in the system fails, its failure impact may quickly spread to other devices in the system, resulting in a chain reaction of failures. This coupling relationship makes the fault identification and fault management of the system particularly complex. In particular, when the failure occurs in key equipment (such as heat pumps, gas turbines, or energy storage systems), it may directly threaten the stability of the entire system and the continuity of energy supply.

[0005] The inventors found in their research that existing fault identification methods mainly rely on rule-based diagnostic models, model-driven physical modeling methods, or data-based statistical learning methods. These methods perform well in single-device fault diagnosis, but when faced with complex multi-energy equipment coordinated integrated energy systems, traditional methods have significant shortcomings: (1) Limitations of single-device diagnosis: Traditional methods usually focus on single-device fault diagnosis and are difficult to achieve joint diagnosis of multiple devices. When a device fails, the state changes of other devices cannot be effectively perceived, and potential faults of associated devices cannot be identified in a timely manner. (2) Insufficient real-time performance and response speed: Traditional fault detection and diagnosis methods usually rely on complete data collection and deep data analysis in the cloud, but this model requires the transmission of a large amount of device data to the cloud for processing, resulting in data transmission delays and diagnostic delays. Especially when the number of edge devices is large, the data communication pressure is too high, making it difficult to meet the real-time response requirements for faults. Summary of the invention

[0006] In order to solve the above problems, the present disclosure proposes a cloud-coordinated integrated energy supply equipment fault identification method and system. The integrated energy fault identification system uses the cloud-coordinated mode to pre-judge and screen the collected information on the terminal side, and concentrates the model-based information processing process on the cloud, thereby improving the fault identification efficiency. Through the collaborative analysis of data information and image information, the shortcomings of a single identification method are overcome, and rapid and accurate fault identification is achieved.

[0007] In order to achieve the above objectives, the present disclosure adopts the following technical solutions: One or more embodiments provide a cloud-coordinated integrated energy supply equipment fault identification method, including the following steps: The client side performs fault pre-judgment based on the device operation data information to obtain the device suspected of having a fault, and the cloud side obtains the device operation data information and image information of the device suspected of having a fault and related devices; The cloud side re-arranges the acquired equipment operation data information based on relevance, intercepts the sorted data information through a sliding window, and transmits it to the trained equipment fault identification network model that incorporates the causal expansion convolution and residual convolution network with an attention mechanism to obtain data recognition results; The cloud side processes the acquired image information to obtain image recognition results; The data recognition results and image recognition results are fused to obtain the fault recognition results.

[0008] A cloud-coordinated integrated energy supply equipment fault identification system includes a terminal-side system and a cloud-side system. The cloud-side system includes: The data analysis module is configured to rearrange the acquired equipment operation data information based on correlation, intercept the sorted data information by sliding window, and transmit it to the trained equipment fault identification network model of causal expansion convolution combined with residual convolution network with attention mechanism to obtain data identification results; An image analysis module is configured to perform image processing on the acquired image information to obtain an image recognition result; The fusion identification module is configured to fuse the data identification result and the image identification result to obtain the fault identification result.

[0009] Compared with the prior art, the present invention has the following beneficial effects: The disclosed end side makes a pre-judgment based on the operation data of the device. When a fault is identified, the operation data of the current device and the associated devices and the corresponding image data are collected and transmitted to the cloud, and the image data and the operation data are fused and judged through the cloud; By prejudging and uploading the data of devices that may have faults, the pressure on data communication and the amount of data processing in the cloud are reduced, and the accuracy and efficiency of fault prediction are improved.

[0010] At the same time, in the cloud, data information is identified through the equipment fault identification network model that combines causal dilation convolution with attention mechanism and residual convolution network. Through the collaborative analysis of data information and image information of comprehensive energy supply equipment, the shortcomings of a single identification method are overcome, and rapid and accurate fault identification is achieved.

[0011] The advantages of the present disclosure and the advantages of additional aspects will be described in detail in the following specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The accompanying drawings constituting a part of the present disclosure are used to provide a further understanding of the present disclosure. The exemplary embodiments of the present disclosure and the description thereof are used to explain the present disclosure but do not constitute a limitation of the present disclosure.

[0013] Figure 1 is a block diagram of a fault identification system according to Embodiment 2 of the present disclosure; Figure 2 This is a device fault identification network model diagram of the causal expansion convolution combined with the residual convolution network in the embodiment 1 of the present disclosure that incorporates the attention mechanism; Figure 3 is a schematic diagram of the processing flow of the spatial attention module of Embodiment 1 of the present disclosure; Figure 4 It is a schematic diagram of the processing flow of the channel attention module of Example 1 of the present disclosure. DETAILED DESCRIPTION

[0014] The present disclosure is further described below in conjunction with the accompanying drawings and embodiments.

[0015] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present disclosure belongs.

[0016] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof. It should be noted that, in the absence of conflict, the various embodiments in the present disclosure and the features in the embodiments can be combined with each other. The embodiments will be described in detail below in conjunction with the accompanying drawings.

[0017] Example 1 In the technical solutions disclosed in one or more embodiments, Figures 1 to 4 As shown, a cloud-coordinated integrated energy supply equipment fault identification method includes the following steps: Step 1: The client side performs fault pre-judgment based on the device operation data information to obtain the device suspected of having a fault, and the cloud side obtains the device operation data information and image information of the device suspected of having a fault and related devices; Step 2: The cloud side re-arranges the acquired equipment operation data information based on the correlation, and intercepts the sorted data information through a sliding window to obtain fused data containing the variable dimension information and time series dimension information of each equipment collection point, and transmits it to the trained equipment fault identification network model of causal expansion convolution combined with residual convolution network with attention mechanism to obtain data recognition results; Step 3: The cloud side processes the acquired image information to obtain an image recognition result; Step 4: Fuse the data recognition result and the image recognition result to obtain the fault recognition result.

[0018] In this embodiment, the terminal side makes a preliminary judgment based on the operation data of the equipment. When a fault is identified, the operation data of the current equipment and related equipment and the corresponding image data are collected and transmitted to the cloud, and the image data and operation data are fused and judged through the cloud. Through the preliminary judgment, only the data of the equipment that may be faulty is uploaded, which reduces the pressure of data communication and the amount of data processing in the cloud, and improves the accuracy and efficiency of fault prediction. At the same time, in the cloud, the equipment fault identification network model that combines causal expansion convolution with attention mechanism and residual convolution network is used to identify data information. Through the collaborative analysis of data information and image information of comprehensive energy supply equipment, the shortcomings of a single identification method are overcome, and rapid and accurate identification of faults is achieved.

[0019] In some embodiments, the end-side system of the end-side includes a data perception module and an image acquisition module for acquiring real-time equipment operation data information of energy supply equipment of the integrated energy system; The data perception module is used to use sensors to capture the operating data and environmental data of each energy supply equipment in the integrated energy system in real time, check the collected data before uploading it to the cloud side, and upload it to the cloud side system as equipment operation data information after it is correct; The data perception module of the end-side system takes into account the fault characteristics of different energy supply equipment and collects sensor data information in the corresponding area.

[0020] Specifically, the data information collected from the energy supply equipment of the integrated energy system includes photovoltaic sensor information, wind turbine sensor information, gas turbine sensor information, and gas boiler sensor information; Optionally, photovoltaic sensor information may include but is not limited to data such as voltage, current, power, frequency, temperature, statistical time, humidity around photovoltaic units, and real-time light intensity at each location of the photovoltaic unit; wind turbine sensor information may include but is not limited to data such as voltage, current, power, temperature, speed, vibration amplitude, frequency, phase, wind direction, and wind speed at each location of the wind turbine unit; gas turbine sensor information may include but is not limited to data such as temperature, speed, oil quality, and gas composition of each component of the gas turbine; gas boiler sensor information may include but is not limited to data such as temperature of each component, pressure at each location, flow rate, and water level. The information required for the energy supply equipment to be acquired can be determined based on the specific equipment model, application scenario, etc.

[0021] The data perception module checks the collected data. If the collected data is missing or has obvious abnormal values, the sensor will re-collect the data. If the inspection is correct, it will be uploaded to the cloud system. The image acquisition module collects visible light, infrared and other images of components of large energy supply equipment such as photovoltaic and wind turbines, and checks the collected image information and uploads it to the cloud system after it is correct.

[0022] Specifically, the image acquisition module of the terminal system can use drone intelligent inspections to effectively capture infrared and visible light components of photovoltaic, wind power and other energy supply units, generate image data of energy supply equipment, and indicate location information.

[0023] Optionally, photovoltaic image information includes but is not limited to visible light and infrared images of photovoltaic panels, visible light images of various brackets and structural components, etc.; wind turbine image information includes but is not limited to infrared and visible light images of various blades, towers, etc., infrared images of motors, etc.; gas turbine image information includes but is not limited to infrared images of combustion chambers, images of turbine blades, etc.; gas boiler image information includes but is not limited to visible light and infrared images of combustion areas, infrared images of boiler shells, etc. The information required for each of the above energy supply equipment can be specifically determined based on the specific equipment model, application scenario, etc.

[0024] The image acquisition module checks the collected image data. If the image is obviously blurred or shaken, the drone is arranged to re-sample and shoot. If the inspection is correct, the collected image data is uploaded to the cloud-side system.

[0025] In step 1, a device suspected of having a fault and associated devices are determined on the terminal side, and the method includes: Step 11: The terminal side constructs a device fault association topology map based on the coupling relationship for the integrated energy system device to be identified; Specifically, each device is used as a node in the topology graph, and nodes with coupling relationships are connected through edges, with the probability of failure of the connected node when the current device fails as the weight of the edge; Step 12: Based on the equipment operation data, a threshold judgment method is used to determine the probability of failure of each equipment, and the equipment suspected of failure is obtained; Step 13: Identify associated devices in the associated topology map based on the associated topology map.

[0026] A further technical solution is to establish a specific implementation method of the association topology map, including the following steps: Step 1-1, establish a directed graph, and use each device in the integrated energy system as a node in the topology graph; Specifically, the types of equipment that a node can represent include generators, energy storage systems, combined cooling, heating and power equipment, heat exchange equipment, gas turbines, heat pumps, air conditioning equipment, etc. Each node represents a specific device, and the attributes of the node include device ID, device type, device status (such as whether it is faulty), device location, and device operating status variables (current, temperature, speed, etc.); Furthermore, nodes are divided based on the type of device, including: Core nodes, core equipment in the trigeneration of cooling, heating and power, such as gas turbines, gas boilers, etc.; Edge nodes are devices responsible for a single type of energy conversion, such as energy storage devices, refrigerators, etc. Auxiliary nodes are devices used to monitor specific states or signals, such as environmental sensors, flow monitors, etc. Step 1-2, edge establishment and weight calculation: Establish edges between nodes. The direction of the edge indicates the direction of fault propagation, and the weight of the edge indicates the probability of a fault propagating from one node to its adjacent nodes. A directed edge E={(vi,vj)} is established on the coupling path between devices, indicating the fault propagation path from device vi to device vj; The direction of the edge is from the upstream device to the downstream device.

[0027] Edge weight : The weight of the edge represents the probability of a fault propagating from node vi to node vj, and the calculation formula is: in, represents the number of times device j fails when device i fails. Indicates the total number of failures of device i.

[0028] Steps 1-3: Obtain real-time operating data of all equipment in the integrated energy system; Specifically, the operating data includes, but is not limited to, the operating status of the equipment, input / output energy flow, temperature, pressure, voltage, current and other key characteristic parameters; Step 1-4, device coupling relationship identification: Integrate physical models and data-driven methods to identify the coupling relationship between devices and determine the strength and propagation direction of the coupling relationship; The association relationship between devices is represented by the adjacency matrix A. Assuming there are n devices in the system, the size of matrix A is n×n, where Aij represents the coupling relationship between device i and device j. If devices i and j have a coupling relationship, Aij=1, otherwise Aij=0.

[0029] Step 1-41, using a physical model-based approach, using the system physical model and the energy exchange model between devices, analyze the energy flow path between the devices and identify the direct relationship between the devices.

[0030] For example, in a combined heat and power system, the output of the gas turbine (heat and electricity) will directly affect the input of the heat exchange equipment; Step 1-42, using a data-driven approach, through data correlation analysis, using a sliding window method to process time series data, calculate the dynamic correlation between equipment operation data, and identify the strong temporal correlation of equipment state variables.

[0031] Specifically, the correlation can be expressed using the Pearson correlation coefficient or the mutual information. When the operating states of two devices (such as power output, temperature, flow rate, etc.) have a high correlation, it is considered that there is a coupling relationship between them.

[0032] Step 1-43, weighted fusion of the physical model and data-driven analysis results, updating the adjacency matrix of the association relationship between devices based on the fused association relationship, and forming a device coupling relationship matrix; Steps 1-5: Dynamic update of the associated topology map: dynamically adjust the edge weights according to the changes in real-time data to ensure the timeliness and accuracy of the topology map; Specifically, the weights of the edges of the associated topology graph are updated as follows: During system operation, the coupling weight of the device is updated in real time , the update formula is:

[0033] in, is the smoothing coefficient; When a new device is connected to the system or an old device is offline, the node set V of the topology graph is dynamically adjusted.

[0034] For newly added devices, the coupling relationship with other devices is re-identified through the physical model, and the weights of the edges with other devices are recalculated.

[0035] In the above solution, the accuracy of the topology map is ensured by dynamically adjusting the coupling relationship of the devices and the weight of the edges according to real-time data; when the device is connected or exited, the topology map can be dynamically updated and has good scalability.

[0036] In this embodiment, in order to realize fault monitoring and rapid identification of the integrated energy system, a method for constructing an association topology map based on the coupling relationship between devices is proposed. By constructing the association topology map, the mutual relationship between the various devices in the system and their fault propagation paths can be clearly represented, so that when a device fails, the associated devices that may be affected can be quickly identified, thereby achieving accurate fault location and efficient collaborative control.

[0037] In some embodiments, the above steps 2 to 4 are implemented on the cloud side by setting up a cloud side system, and the cloud side system includes a data analysis module, an image analysis module, and a fusion recognition module; The data analysis module of the cloud-side system is configured to implement step 2, such as Figure 2 The equipment operation data information shown implements the energy supply equipment fault identification process, and the data analysis module is used to receive and save the equipment operation data information sample data transmitted by the data acquisition module on the terminal side, and perform fault identification to obtain data identification results; In step 2, the identification method of the data identification result is configured to be implemented in the data analysis module, including the following steps: Step 21: Calculate the correlation distance between variables to obtain the coupling relationship between the variables of the sample data, and rearrange the order of the variables according to the calculated correlation distance so that the variables with stronger correlation are closer; Specifically, the correlation distance calculation formula is as follows:

[0038] In the formula, is the correlation coefficient between variables x and y, The larger the absolute value of indicates, the stronger the correlation between variables x and y; k is the number of samples of each variable in the first running data; and is the mean of variables x and y.

[0039] Step 22: Standardize the rearranged data; Considering the autocorrelation between variable data, we first perform min-max standardization on the data. The calculation formula is as follows:

[0040] Step 23: intercept the standardized data using a sliding window to obtain fused data containing variable dimension information and time series dimension information of each device collection point; In some embodiments, in step 2, the device fault identification network model of the causal expansion convolution with attention mechanism combined with the residual convolution network has a structure as follows: Figure 2 As shown, specifically, it includes a feature extraction network and a discriminator; the feature extraction network includes sequentially cascaded convolutional layers, four residual blocks and an average pooling layer, each residual block contains four convolutional layers incorporating causal expansion convolution operations; a channel attention module is connected after the third residual block; a spatial attention module is connected after the fourth residual block, and the spatial attention module is connected to the discriminator through an average pooling layer; the discriminator includes two fully connected layers and an activation layer; In this embodiment, the residual block uses multiple causal expansion convolution operations to capture the long-term and short-term time series dependencies, improving the feature extraction capability of dynamic data; the channel and spatial attention modules effectively amplify the key features, improving the accuracy of fault identification; Further technical solutions, the channel attention module structure diagram is as follows Figure 3 As shown in the figure, the channel attention module performs global maximum pooling and global average pooling on the input features in the spatial dimension, and then performs one-dimensional convolution and ReLU activation on the two results to obtain two channel attention weights. Then, the original input features of the channel attention module are multiplied by the two channel attention weights, and then element-by-element addition is performed to obtain the channel attention features, which are used as the input of the fourth residual block.

[0041] Further technical solutions, the structure of the spatial attention module is as follows Figure 4 As shown in the figure, the spatial attention module processes the input features through global maximum pooling and global average pooling in the channel dimension, and then processes the two results through one-dimensional convolution and ReLU activation to obtain two spatial attention weights. Then, the original input features of the spatial attention module are multiplied by the two spatial attention weights and then element-by-element added to obtain the spatial attention features, which are used as the input of the final average pooling layer of the network.

[0042] In this embodiment, by setting the channel attention module and the spatial attention module between the residual blocks, the key feature information in the device data is amplified, and the model's ability to capture multi-dimensional data features is significantly improved. Combined with the causal expansion convolution operation, the model effectively extracts multi-scale time series features, enhances the recognition accuracy of complex faults, reduces the false alarm rate and missed alarm rate, and improves the real-time and reliability of fault diagnosis.

[0043] In step 2, the fused data obtained by interception is transmitted to the trained equipment fault identification network model of causal expansion convolution combined with residual convolution network with attention mechanism. The fault identification process includes the following steps: Step 2-1: Extract initial features through a convolutional layer; Step 2-2: The extracted initial features are sequentially passed through four residual blocks for multi-scale feature extraction. Each residual block extracts multi-scale time series features through multiple causal dilation convolutions to capture the long-term and short-term dependencies of time series data. Step 2-3: After the third residual block, the global features of each channel are extracted through the channel attention mechanism, and the key channel features are amplified. Specifically, the following steps are included: Step 2-31: The output features of the third residual block are processed through global maximum pooling and global average pooling, and then through one-dimensional convolution and ReLU activation to generate two channel attention weights; Step 2-32: Multiply the output features of the third residual block by the two channel attention weights and add them element by element to obtain the channel attention features; Step 2-4: After the fourth residual block, the feature weights of the spatial dimension are extracted through the spatial attention mechanism to further improve the expression ability of key features. The specific steps include the following: Step 2-41, the output features of the fourth residual block are subjected to global maximum pooling and global average pooling operations in the channel dimension, and then to one-dimensional convolution and ReLU activation respectively to generate two spatial attention weights; Step 2-42, multiply the output features of the fourth residual block by the two spatial attention weights respectively and then add them element by element to obtain the spatial attention features.

[0044] Step 2-5: Perform average pooling on the output of spatial attention and aggregate the extracted high-dimensional features to generate fault feature encoding; Step 2-6: Encode the obtained fault features, perform nonlinear mapping and further feature compression through two fully connected layers; then use the softmax activation function to normalize the output to obtain the probability distribution of each category of faults, and select the category with the largest probability as the fault category identification result, that is, to obtain the data recognition result.

[0045] In this embodiment, a comprehensive energy supply equipment fault identification method combining causal expansion convolution with an attention mechanism and a residual convolution network is used to rearrange and intercept the data using the correlation between data, so as to facilitate the accurate extraction of fault features of the feature extraction network. The causal expansion convolution combined with the residual convolution network realizes the coupled extraction of variable dimension and time series dimension features, and the useful information in the equipment data is amplified through the channel and spatial attention mechanism, which helps the network better capture features and effectively improves the accuracy of fault identification.

[0046] The cloud-side system data analysis module reports the data information and data recognition results to the fusion recognition module.

[0047] The image analysis module of the cloud-side system is configured to implement step 3. The image analysis module can use a neural network model such as a convolutional neural network to perform fault identification based on image information to obtain an image recognition result, specifically: The image analysis module on the cloud side receives and saves the image data and location information transmitted by the image acquisition module on the terminal side; specifically, the image analysis module on the cloud side stores the received image data to form structured data storage, including image files, acquisition timestamps, device IDs, and device location information.

[0048] Use image technology to analyze image data, input the image data into the trained fault identification network model based on convolutional neural network, analyze the abnormal information of each device, and finally report the analysis results and location information to the fusion identification module; Step 31: Perform image preprocessing: Resolution adjustment: Adjust the input image to a fixed resolution required by the training model to ensure consistency in subsequent analysis.

[0049] Denoising: Remove noise from the image through filtering (such as Gaussian filtering) to improve image quality.

[0050] Image normalization: Normalize the image pixel values, for example, normalize the pixel values ​​to the range [0, 1]) to meet the input requirements of the convolutional neural network; Step 32: input the preprocessed image into a fault identification network model based on a convolutional neural network (CNN) to extract features; Convolutional layer processing: The convolution layer of CNN extracts local features of the image, such as edges and textures, through convolution kernels. Each layer extracts features at different levels. The convolution operation extracts key area information in the image through a sliding window, and introduces nonlinearity in combination with activation functions (such as ReLU) to enhance feature expression capabilities.

[0051] Pooling layer processing: Through maximum pooling or average pooling operations, the extracted features are reduced in dimension and compressed to retain key feature information while reducing the amount of model calculation.

[0052] Multi-layer processing: convolution and pooling operations are performed alternately to gradually extract high-level features of the image.

[0053] Step 33: Identify abnormal areas in the image through feature mapping, use target detection algorithm to mark the abnormal areas, and output the corresponding coordinate frame; Step 34: Map the extracted high-dimensional features to specific fault categories through the fully connected layer and the softmax activation function, output the probability value of each category, and select the category with the highest probability as the fault category; Step 35: Combine the fault type identified by the image analysis module with the location information transmitted by the terminal side to indicate the fault type and location of the specific device.

[0054] Specifically, the abnormal area can be marked in the image, and a visual fault diagnosis result map can be generated in combination with the location information, including the device name, fault type, location coordinates, etc.

[0055] In this embodiment, image analysis can intuitively reflect the surface status and fault characteristics of the equipment, such as cracks, deformation, contamination, etc.; with the help of the feature extraction ability of deep learning, it can accurately identify tiny fault characteristics. The entire process does not require human intervention and has a high degree of automation. The image processing module can quickly identify anomalies and shorten the fault diagnosis time. In addition, the combination of image information and location information makes fault location more accurate, providing a basis for rapid maintenance.

[0056] The fusion identification module on the cloud side is configured to implement step 4, fusing the data identification result and the image identification result to obtain the fault identification result, including the following steps: Step 41: Obtain data recognition results and image recognition results; The cloud-side fusion recognition module integrates the data information and preliminary recognition results from the data analysis module and the image analysis results from the image analysis module to perform decision-level fusion.

[0057] Step 42: If both the data analysis and image analysis results identify a fault in a certain energy supply device, and the fault type is consistent, it is determined that the corresponding device has a fault; 42.1) Immediately issue a fault alarm for the device.

[0058] 42.2) Locate the specific faulty component based on the image analysis results, and eliminate misjudgments caused by dust, dirt, etc.; 42.3) Further diagnose subtle defects of components based on data analysis results to achieve refined fault diagnosis.

[0059] Step 43: If a certain energy supply equipment is identified as faulty in the data analysis result but is judged to be normal in the image recognition result, the image data of the corresponding equipment is re-acquired to execute step 3 for image recognition; Specifically, the fusion recognition module replans the drone's cruising path and capture position and angle based on the fault analyzed by the data analysis module, obtains new image information through the image acquisition module on the end side, and re-performs image recognition.

[0060] Step 44: If the data analysis result is normal but the image analysis result identifies a fault in a certain energy supply equipment, new operation data of the corresponding equipment is obtained again, and step 2 is executed to re-identify the fault of the data information to verify whether there is a potential fault; Furthermore, the fusion recognition module analyzes the fault information output by the image analysis module, determines the changes in the operating physical quantities of electrical, thermal and other equipment that may be caused by the fault, generates data collection instructions, and the terminal-side data collection module re-collects new operating data of related equipment.

[0061] Step 43: If both the data analysis result and the image analysis result are judged to be normal, the fusion recognition module sends a normal signal to the end side, and each acquisition module on the end side maintains the daily monitoring state.

[0062] The above-mentioned integrated energy fault identification system uses the cloud collaboration mode to pre-judge and screen information on the terminal side, and concentrates the model-based information processing process on the cloud, thereby improving the efficiency of fault identification. At the same time, through the collaborative analysis of data information and image information, the diagnostic results from the data analysis module are used to identify small defects that are difficult to identify in the image analysis module. At the same time, the image analysis module analysis results and location information are used to verify the fault identification results in the data analysis module and realize component-level fault location. The final fault identification results are obtained and an alarm is issued, overcoming the shortcomings of a single identification method, realizing rapid and accurate fault identification, and providing a basis for the formulation of subsequent operation and maintenance strategies.

[0063] Example 2 Based on Example 1, this embodiment provides a cloud-coordinated integrated energy supply equipment fault identification system, including a terminal-side system and a cloud-side system, and the cloud-side system includes: The data analysis module is configured to rearrange the acquired equipment operation data information based on correlation, intercept the sorted data information by sliding window, and transmit it to the trained equipment fault identification network model of causal expansion convolution combined with residual convolution network with attention mechanism to obtain data identification results; An image analysis module is configured to perform image processing on the acquired image information to obtain an image recognition result; The fusion identification module is configured to fuse the data identification result and the image identification result to obtain the fault identification result.

[0064] Furthermore, a device fault identification network model that incorporates causal expansion convolution with attention mechanism and residual convolution network includes a feature extraction network and an identifier; The feature extraction network includes a cascaded convolutional layer, four residual blocks, and an average pooling layer. Each residual block contains four convolutional layers that incorporate causal dilation convolution operations. A channel attention module is connected after the third residual block. A spatial attention module is connected after the fourth residual block, and the spatial attention module is connected to the identifier through an average pooling layer. The discriminator consists of two fully connected layers and an activation layer.

[0065] Furthermore, the end-side system includes a data perception module, an image acquisition module, and an association relationship recognition module; Data perception module, used to use sensors to capture the operating data and environmental data of each energy supply equipment in the integrated energy system in real time; The image acquisition module is used to acquire image information of the device.

[0066] The association relationship identification module is configured to construct and dynamically update the association topology map and identify the association relationship of device failures.

[0067] In the association relationship identification module, the specific implementation method of establishing the association topology map is specifically described in Example 1 and will not be repeated here.

[0068] The above description is only a preferred embodiment of the present disclosure and is not intended to limit the present disclosure. For those skilled in the art, the present disclosure may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

[0069] Although the above describes the specific implementation methods of the present disclosure in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present disclosure. Technical personnel in the relevant field should understand that on the basis of the technical solution of the present disclosure, various modifications or variations that can be made by those skilled in the art without creative work are still within the scope of protection of the present disclosure.

Claims

1. A cloud-coordinated integrated energy supply equipment fault identification method, characterized in that: The steps include: The client side performs fault pre-judgment based on the device operation data information to obtain the device suspected of having a fault, and the cloud side obtains the device operation data information and image information of the device suspected of having a fault and related devices; The cloud side re-arranges the acquired equipment operation data information based on relevance, intercepts the sorted data information through a sliding window, and transmits it to the trained equipment fault identification network model that incorporates the causal expansion convolution and residual convolution network with an attention mechanism to obtain data recognition results; The cloud side processes the acquired image information to obtain image recognition results; The data recognition results and image recognition results are fused to obtain the fault recognition results.

2. A cloud-coordinated integrated energy supply equipment fault identification method as claimed in claim 1, characterized in that: The method for determining a suspected faulty device and associated devices on a terminal side includes the following steps: On the client side, for the integrated energy system equipment to be identified, a device fault correlation topology map is constructed based on the coupling relationship. Based on the equipment operation data, the threshold judgment method is used to determine the probability of failure of each device and obtain the equipment suspected of failure; Identify the associated devices in the associated topology map based on the associated topology map.

3. A cloud-coordinated integrated energy supply equipment fault identification method as claimed in claim 2, characterized in that: The specific implementation method of establishing the association topology graph includes the following steps: Establish a directed graph and use each device in the integrated energy system as a node in the topology graph; Establish edges between nodes. The direction of the edge indicates the direction of fault propagation, and the weight of the edge indicates the probability of a fault propagating from one node to its adjacent nodes. Obtain real-time operating data of all equipment in the integrated energy system; By integrating physical models and data-driven methods, the coupling relationship between devices is identified based on real-time operation data, the strength and propagation direction of the coupling relationship are determined, and the weight of the edge is dynamically adjusted. The dynamic adjustment formula is: in, is the smoothing coefficient; represents the number of times device j fails when device i fails. Indicates the total number of failures of device i.

4. The cloud-coordinated integrated energy supply equipment fault identification method according to claim 1, characterized in that: The equipment fault identification network model that integrates causal expansion convolution with attention mechanism and residual convolution network, including feature extraction network and identifier; The feature extraction network includes a cascaded convolutional layer, four residual blocks, and an average pooling layer. Each residual block contains four convolutional layers that incorporate causal dilation convolution operations. A channel attention module is connected after the third residual block. A spatial attention module is connected after the fourth residual block, and the spatial attention module is connected to the identifier through an average pooling layer. The discriminator consists of two fully connected layers and an activation layer.

5. A cloud-coordinated integrated energy supply equipment fault identification method as claimed in claim 4, characterized in that: The channel attention module processes the input features through global maximum pooling and global average pooling in the spatial dimension, and then processes the two results through one-dimensional convolution and ReLU activation to obtain two channel attention weights. Then, the original input features of the channel attention module are multiplied by the two channel attention weights and then element-wise added to obtain the channel attention features, which are used as the input of the fourth residual block. Alternatively, the spatial attention module processes the input features through global maximum pooling and global average pooling in the channel dimension, and then processes the two results through one-dimensional convolution and ReLU activation to obtain two spatial attention weights; then, the original input features of the spatial attention module are multiplied by the two spatial attention weights, and then element-by-element addition is performed to obtain the spatial attention features, which are used as the input of the final average pooling layer of the network.

6. The cloud-coordinated integrated energy supply equipment fault identification method according to claim 1, characterized in that: The equipment fault identification network model using causal expansion convolution with attention mechanism combined with residual convolution network is used. The fault identification process includes the following steps: Through a convolution layer, the initial features are extracted; The extracted initial features are sequentially passed through four residual blocks for multi-scale feature extraction. Each residual block extracts multi-scale time series features through multiple causal dilation convolutions to capture the long-term and short-term dependencies of time series data. After the third residual block, the global features of each channel are extracted through the channel attention mechanism: After the fourth residual block, the feature weights of the spatial dimension are extracted through the spatial attention mechanism; The output of spatial attention is average-pooled, and the extracted high-dimensional features are aggregated to generate fault feature encoding; The obtained fault features are encoded, and the features are nonlinearly mapped and further compressed through two fully connected layers. Then the output is normalized to obtain the probability distribution of each type of fault, that is, the data recognition result. The image recognition result is obtained by fault identification based on image information, using convolutional neural network.

7. The cloud-coordinated integrated energy supply equipment fault identification method according to claim 1, characterized in that: The data recognition result and the image recognition result are integrated to obtain the fault recognition result, including the following steps: Obtain data recognition results and image recognition results; If both the data analysis and image analysis results identify a fault in the energy supply device, and the fault type is consistent, then it is determined that the corresponding device has a fault; If the energy supply equipment is identified as faulty in the data analysis results but is judged to be normal in the image recognition results, the image data of the corresponding equipment is re-acquired for image recognition; If the data analysis result is normal but the image analysis result identifies a fault in the energy supply equipment, new operating data of the corresponding equipment is obtained again and the fault identification of the data information is re-performed; If both the data analysis results and the image analysis results are judged to be normal, it is judged to be normal.

8. A cloud-coordinated integrated energy supply equipment fault identification system, characterized in that: It includes the terminal-side system and the cloud-side system. The cloud-side system includes: The data analysis module is configured to rearrange the acquired equipment operation data information based on correlation, intercept the sorted data information by sliding window, and transmit it to the trained equipment fault identification network model of causal expansion convolution combined with residual convolution network with attention mechanism to obtain data identification results; An image analysis module is configured to perform image processing on the acquired image information to obtain an image recognition result; The fusion identification module is configured to fuse the data identification result and the image identification result to obtain the fault identification result.

9. A cloud-coordinated integrated energy supply equipment fault identification system as claimed in claim 8, characterized in that: The equipment fault identification network model that integrates causal expansion convolution with attention mechanism and residual convolution network, including feature extraction network and identifier; The feature extraction network includes a cascaded convolutional layer, four residual blocks, and an average pooling layer. Each residual block contains four convolutional layers that incorporate causal dilation convolution operations. A channel attention module is connected after the third residual block. A spatial attention module is connected after the fourth residual block, and the spatial attention module is connected to the identifier through an average pooling layer. The discriminator consists of two fully connected layers and an activation layer.

10. A cloud-coordinated integrated energy supply equipment fault identification system as claimed in claim 8, characterized in that: The end-side system includes a data perception module, an image acquisition module, and an association relationship recognition module; Data perception module, used to use sensors to capture the operating data and environmental data of each energy supply equipment in the integrated energy system in real time; An image acquisition module, used for acquiring image information of a device; The association relationship identification module is configured to construct and dynamically update the association topology map and identify the association relationship of device failures.

Images