Intelligent power grid operation and maintenance system based on cloud service

Abstract

The invention relates to the technical field of intelligent management, and particularly discloses a power grid intelligent operation and maintenance system based on cloud service, which adopts a cloud computing technology based on deep learning to monitor and analyze operation parameters of power grid equipment, and extracts dynamic change mode characteristics of each operation parameter of the power grid equipment in a time domain. And cross-modal interaction fusion is carried out on the multi-modal operation parameter time domain correlation mode characteristics to mine abnormal operation state information of the power grid equipment, so that intelligent identification of the abnormal operation state of the power grid equipment is realized. Therefore, power grid equipment can be monitored in real time, and abnormal conditions of the equipment can be found in time, so that operation and maintenance personnel are helped to formulate a reasonable maintenance plan, the operation and maintenance efficiency is improved, and the service life of the equipment is prolonged.

Description

Technical Field

[0001] The present application relates to the field of intelligent management technology, and more specifically, to a cloud service-based intelligent power grid operation and maintenance system. Background Art

[0002] With the rapid development of society, the stability and reliability of the power grid, as the infrastructure for the operation of modern society, are of great significance to the sustainable development of the economy and society. However, due to the complexity of power grid equipment and the variability of the operating environment, the operating status monitoring and maintenance of power grid equipment has always been a major challenge in power system operation and maintenance.

[0003] Traditional power grid equipment operation and maintenance methods mostly rely on manual inspections and regular maintenance. However, due to the diverse and widespread distribution of power grid equipment, manual inspections consume significant manpower and time, and comprehensive coverage is difficult. Furthermore, collecting and organizing inspection data also consumes significant time and manpower, potentially leading to delayed data updates and difficulties in supporting accurate and timely operation and maintenance decisions, posing a challenge to the stable operation of power grid equipment. Therefore, a cloud-based intelligent power grid operation and maintenance system is highly sought after. Summary of the Invention

[0004] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a cloud service-based smart power grid operation and maintenance system, which uses cloud computing technology based on deep learning to monitor and analyze the operating parameters of power grid equipment, extract the dynamic change pattern characteristics of each operating parameter of the power grid equipment in the time domain, and perform cross-modal interactive fusion of the multi-modal operating parameter time domain correlation pattern characteristics to mine the abnormal operating status information of the power grid equipment, thereby realizing intelligent identification of the abnormal operating status of the power grid equipment. In this way, the power grid equipment can be monitored in real time and the abnormal conditions of the equipment can be discovered in time, thereby helping the operation and maintenance personnel to formulate reasonable maintenance plans, improve operation and maintenance efficiency, and extend the life of the equipment.

[0005] Accordingly, according to one aspect of the present application, a cloud service-based smart power grid operation and maintenance system is provided, which includes:

[0006] an operating parameter acquisition module, configured to acquire a time series of operating parameters of a monitored power grid device collected by a sensor group deployed on the monitored power grid device, wherein the monitored power grid device is a transformer, and the operating parameters of the monitored power grid device include temperature, current, and oil level;

[0007] A data transmission module, configured to transmit the time series of the operating parameters of the monitored power grid equipment to a cloud server;

[0008] a time series correlation analysis module, configured to extract time series features from the time series of the operating parameters of the monitored power grid equipment on the cloud server to obtain a temperature time series correlation feature vector, a current time series correlation feature vector, and an oil level time series correlation feature vector;

[0009] a cross-modal interaction module, configured to perform cross-modal interaction on the temperature time series correlation feature vector, the current time series correlation feature vector, and the oil level time series correlation feature vector on the cloud server to obtain a transformer multi-modal time series characterization feature vector;

[0010] An operation and maintenance voice prompt generation module is used to determine whether to generate an operation and maintenance voice prompt based on the transformer multimodal time series characterization feature vector on the cloud server.

[0011] In the above-mentioned cloud service-based smart power grid operation and maintenance system, the timing correlation analysis module includes: a data sorting unit, which is used to sort the time series of the operating parameters of the monitored power grid equipment according to the parameter sample dimension to obtain a temperature timing input vector, a current timing input vector and an oil level timing input vector; a data correction unit, which is used to perform gamma correction on the temperature timing input vector, the current timing input vector and the oil level timing input vector to obtain a gamma-corrected temperature timing input vector, a gamma-corrected current timing input vector and a gamma-corrected oil level timing input vector; a timing pattern feature extraction unit, which is used to pass the gamma-corrected temperature timing input vector, the gamma-corrected current timing input vector and the gamma-corrected oil level timing input vector through a timing pattern feature extractor based on a one-dimensional extended convolutional neural network model to obtain the temperature timing correlation feature vector, the current timing correlation feature vector and the oil level timing correlation feature vector.

[0012] In the above-mentioned cloud service-based power grid intelligent operation and maintenance system, the time series pattern feature extraction unit is used to: process the gamma-corrected temperature time series input vector using the following one-dimensional extended convolution formula to obtain the temperature time series correlation feature vector; wherein the one-dimensional extended convolution formula is:

[0013] l'=l+(l-1)×(d-1)

[0014] C i =f(ω·X i:i+j-l‘ +b)

[0015] C=[C1,C2,...,C n-l‘+1 ]

[0016] Among them, the size of the one-dimensional expanded convolution kernel ω is 1×l', l is the length of the original convolution kernel, d is the expansion rate, X i:i+j-l‘represents the time window of length l' headed by the eigenvalue of the ith position in the gamma-corrected temperature time series input vector, b is the bias term, and b∈R, f(·) is the nonlinear activation function, C i is the i-th local convolutional coding feature vector, n is the dimension of the temperature time series input vector after gamma correction, C is the temperature time series correlation feature vector, and [·,·] represents cascade.

[0017] In the above-mentioned cloud service-based smart power grid operation and maintenance system, the cross-modal interaction module includes: a common space mapping unit, which is used to pass the temperature time series association feature vector, the current time series association feature vector and the oil level time series association feature vector through a common space mapper based on a fully connected layer to obtain a mapped temperature time series association feature vector, a mapped current time series association feature vector and a mapped oil level time series association feature vector; a cross-modal fusion unit, which is used to pass the mapped temperature time series association feature vector, the mapped current time series association feature vector and the mapped oil level time series association feature vector through a cross-modal interaction architecture to obtain the transformer multi-modal time series representation feature vector.

[0018] In the above-mentioned cloud service-based power grid intelligent operation and maintenance system, the cross-modal fusion unit is used to: cascade the mapped temperature time series correlation feature vector, the mapped current time series correlation feature vector and the mapped oil level time series correlation feature vector to obtain a multi-modal cascade feature vector; perform pooling, dimensionality reduction and activation processing on the multi-modal cascade feature vector to obtain a concentrated channel information representation feature vector; pass the concentrated channel information representation feature vector through the first fully connected layer, the second fully connected layer and the third fully connected layer respectively to obtain the first channel information representation encoding feature vector, the second channel information A characterization coding feature vector and a third channel information characterization coding feature vector, wherein the first fully connected layer, the second fully connected layer, and the third fully connected layer have the same number of nodes; the first channel information characterization coding feature vector and the mapped temperature time series association feature vector, the second channel information characterization coding feature vector and the mapped current time series association feature vector, and the third channel information characterization coding feature vector and the mapped oil level time series association feature vector are multiplied by broadcast multiplication, and the multiplication results are added together to obtain the transformer multimodal time series characterization feature vector.

[0019] In the above-mentioned cloud service-based smart power grid operation and maintenance system, the operation and maintenance voice prompt generation module includes: an operation status monitoring unit, which is used to pass the transformer multimodal time series characterization feature vector through a classifier-based operation status monitor to obtain a monitoring result, and the monitoring result is used to indicate whether the operation status of the monitored power grid equipment is abnormal; a prompt generation judgment unit, which is used to generate the operation and maintenance voice prompt in response to the monitoring result indicating that the operation status of the monitored power grid equipment is abnormal.

[0020] In the above-mentioned cloud service-based power grid intelligent operation and maintenance system, the operating status monitoring unit is used to: use the classifier-based operating status monitor to process the transformer multimodal time series characterization feature vector using the following classification formula to obtain the monitoring result; wherein the classification formula is:

[0021] O=softmax{(W n ,B n ):…:(W1,B1)│X}

[0022] Among them, W1 to W n is the weight matrix, B1 to B n is the bias vector, X is the transformer multimodal time series characterization feature vector, softmax represents the normalized exponential function, and O represents the monitoring result.

[0023] The above-mentioned cloud service-based smart power grid operation and maintenance system also includes a training module for training the temporal pattern feature extractor based on the one-dimensional extended convolutional neural network model, the co-space mapper based on the fully connected layer, the cross-modal interaction architecture and the classifier-based operation status monitor.

[0024] In the above-mentioned cloud service-based smart grid operation and maintenance system, the training module includes: a training data acquisition unit for acquiring training data, wherein the training data includes a time series of training operating parameters of the monitored grid equipment and a real value of whether the operating state of the monitored grid equipment is abnormal; a training data transmission unit for transmitting the time series of training operating parameters of the monitored grid equipment to a cloud server; a training data sorting unit for sorting the time series of training operating parameters of the monitored grid equipment according to parameter sample dimensions on the cloud server to obtain a training temperature time series input vector, a training current time series input vector and a training oil level time series input vector. a training data gamma correction unit for performing gamma correction on the training temperature time series input vector, the training current time series input vector, and the training oil level time series input vector on the cloud server to obtain a training gamma-corrected temperature time series input vector, a training gamma-corrected current time series input vector, and a training gamma-corrected oil level time series input vector; a training data time series feature extraction unit for performing gamma correction on the training temperature time series input vector, the training gamma-corrected current time series input vector, and the training gamma-corrected oil level time series input vector on the cloud server through the time series pattern feature extractor based on the one-dimensional extended convolutional neural network model to obtain a training temperature time series correlation feature. a training data co-space mapping unit for mapping the training temperature time series association feature vector, the training current time series association feature vector and the training oil level time series association feature vector through the co-space mapper based on the fully connected layer on the cloud server to obtain the training mapped temperature time series association feature vector, the training mapped current time series association feature vector and the training mapped oil level time series association feature vector; a training data cross-modal interaction unit for mapping the training mapped temperature time series association feature vector, the training mapped current time series association feature vector and the training mapped oil level time series association feature vector on the cloud server; The cross-modal interaction architecture passes the feature vector of the multimodal temporal representation of the training transformer through the cross-modal interaction architecture to obtain a classification loss function value for the multimodal temporal representation feature vector of the training transformer on the cloud server; the model training unit is used to train the temporal pattern feature extractor based on the one-dimensional extended convolutional neural network model, the co-space mapper based on the fully connected layer, the cross-modal interaction architecture and the classifier-based running status monitor with the classification loss function value, wherein in each round of iteration of the training, the multimodal temporal representation feature vector of the training transformer is optimized.

[0025] Compared to existing technologies, the cloud-based smart grid operation and maintenance system provided by this application uses deep learning-based cloud computing technology to monitor and analyze the operating parameters of grid equipment, extracting the dynamic change pattern characteristics of each operating parameter of grid equipment in the time domain, and performing cross-modal interactive fusion of multi-modal operating parameter time-domain correlation pattern characteristics to mine abnormal operating status information of grid equipment, thereby realizing intelligent identification of abnormal operating status of grid equipment. In this way, grid equipment can be monitored in real time and abnormal conditions of equipment can be discovered in a timely manner, thereby helping operation and maintenance personnel to formulate reasonable maintenance plans, improve operation and maintenance efficiency, and extend equipment life. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0027] Figure 1 2 is a block diagram of a cloud service-based smart grid operation and maintenance system according to an embodiment of the present application.

[0028] Figure 2 Schematic diagram of the architecture of a cloud service-based smart grid operation and maintenance system according to an embodiment of the present application.

[0029] Figure 3 This is a block diagram of a timing correlation analysis module in a cloud service-based smart grid operation and maintenance system according to an embodiment of the present application.

[0030] Figure 4 This is a block diagram of a cross-modal interaction module in a cloud service-based smart grid operation and maintenance system according to an embodiment of the present application.

[0031] Figure 5 This is a block diagram of an operation and maintenance voice prompt generation module in a cloud service-based power grid intelligent operation and maintenance system according to an embodiment of the present application.

[0032] Figure 6 This is a block diagram of a training module in a cloud service-based smart grid operation and maintenance system according to an embodiment of the present application. DETAILED DESCRIPTION

[0033] Below, the embodiments of the present application will be described in more detail with reference to the accompanying drawings, and the above-mentioned and other purposes, features and advantages of the present application will become more apparent. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments of the present application, and it should be understood that the present application is not limited to the example embodiments described herein. At the same time, the accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0034] Figure 1 FIG is a block diagram of a cloud service-based smart grid operation and maintenance system according to an embodiment of the present application. Figure 1 As shown, according to an embodiment of the present application, the cloud service-based smart power grid operation and maintenance system 100 includes: an operating parameter acquisition module 110, which is used to obtain a time series of operating parameters of the monitored power grid device collected by a sensor group deployed on the monitored power grid device, wherein the monitored power grid device is a transformer, and the operating parameters of the monitored power grid device include temperature value, current value and oil level value; a data transmission module 120, which is used to transmit the time series of operating parameters of the monitored power grid device to a cloud server; a time series correlation analysis module 130, which is used to perform time series feature extraction on the time series of operating parameters of the monitored power grid device on the cloud server to obtain a temperature time series correlation feature vector, a current time series correlation feature vector and an oil level time series correlation feature vector; a cross-modal interaction module 140, which is used to perform cross-modal interaction on the temperature time series correlation feature vector, the current time series correlation feature vector and the oil level time series correlation feature vector on the cloud server to obtain a transformer multi-modal time series representation feature vector; an operation and maintenance voice prompt generation module 150, which is used to determine whether to generate an operation and maintenance voice prompt based on the transformer multi-modal time series representation feature vector on the cloud server.

[0035] As mentioned in the background technology above, the traditional operation and maintenance method of power grid equipment is mainly based on manual inspection and regular maintenance. Although this method can ensure the normal operation of power grid equipment to a certain extent, its low efficiency, long inspection cycle, high inspection labor cost, and difficulty in updating inspection data in real time are becoming increasingly prominent. In response to the above technical problems, the technical concept of this application is to use cloud computing technology based on deep learning to monitor and analyze the operating parameters of power grid equipment, extract the dynamic change pattern characteristics of each operating parameter of power grid equipment in the time domain, and perform cross-modal interactive fusion of the multi-modal operating parameter time domain correlation pattern characteristics to mine abnormal operating status information of power grid equipment, thereby realizing intelligent identification of abnormal operating status of power grid equipment. In this way, power grid equipment can be monitored in real time and abnormal conditions of equipment can be discovered in a timely manner, thereby helping operation and maintenance personnel to formulate reasonable maintenance plans, improve operation and maintenance efficiency, and extend equipment life.

[0036] In the aforementioned cloud-based smart grid operation and maintenance system 100, the operating parameter acquisition module 110 is configured to acquire a time series of operating parameters of the monitored grid device collected by a sensor group deployed on the monitored grid device. The monitored grid device is a transformer, and the operating parameters include temperature, current, and oil level. It should be understood that as a critical component of the power grid, the operating status of the transformer has a significant impact on the stability and reliability of the grid. During transformer operation, the transformer's temperature, current, and oil level are important indicators of its operating status. Specifically, the temperature reflects the transformer's thermal state; abnormal temperatures may indicate internal overload or poor heat dissipation; the current reflects the transformer's load; abnormal currents may indicate load imbalance, a short circuit, or overload; and the oil level reflects the transformer's insulation condition; abnormal oil level fluctuations may indicate leakage or other problems. Based on this, by collecting the transformer's operating parameters such as temperature, current, and oil level through a sensor group, we can understand the operating status of the transformer equipment in real time, and analyze the abnormal conditions of the transformer by integrating multi-dimensional information to improve the management efficiency and operation quality of the transformer.

[0037] In the above-mentioned cloud service-based smart grid operation and maintenance system 100, the data transmission module 120 is used to transmit the time series of the operating parameters of the monitored grid equipment to the cloud server. It should be understood that the cloud server has a powerful data storage capacity and can realize cross-regional data sharing and collaboration. In the technical solution of the present application, by transmitting the time series of the operating parameters of the monitored grid equipment to the cloud server, a large amount of operating parameter data of the grid equipment can be centrally stored and uniformly managed. In this way, no matter where the monitored grid equipment is located, transmitting the data to the cloud server allows operation and maintenance personnel in multiple locations to share data and work together across regions. In addition, the cloud server has high-performance computing resources. By transmitting the time series of the operating parameters of the monitored grid equipment to the cloud server, its powerful computing resources can be used to perform complex data analysis and mining, realize remote monitoring and intelligent analysis of grid equipment, and improve operation and maintenance efficiency and accuracy.

[0038] In the above-mentioned cloud service-based power grid intelligent operation and maintenance system 100, the time series correlation analysis module 130 is used to extract time series features of the time series of the operating parameters of the monitored power grid equipment on the cloud server to obtain temperature time series correlation feature vectors, current time series correlation feature vectors, and oil level time series correlation feature vectors. Specifically, Figure 3 FIG is a block diagram of a time series correlation analysis module in a cloud service-based smart grid operation and maintenance system according to an embodiment of the present application. Figure 3 As shown, the timing correlation analysis module 130 includes: a data sorting unit 131, which is used to sort the time series of the operating parameters of the monitored power grid equipment according to the parameter sample dimension to obtain a temperature timing input vector, a current timing input vector and an oil level timing input vector; a data correction unit 132, which is used to perform gamma correction on the temperature timing input vector, the current timing input vector and the oil level timing input vector to obtain a gamma-corrected temperature timing input vector, a gamma-corrected current timing input vector and a gamma-corrected oil level timing input vector; a timing pattern feature extraction unit 133, which is used to pass the gamma-corrected temperature timing input vector, the gamma-corrected current timing input vector and the gamma-corrected oil level timing input vector through a timing pattern feature extractor based on a one-dimensional extended convolutional neural network model to obtain the temperature timing correlation feature vector, the current timing correlation feature vector and the oil level timing correlation feature vector.

[0039] Specifically, the data sorting unit 131 is used to sort the time series of the operating parameters of the monitored power grid equipment according to the parameter sample dimension to obtain a temperature time series input vector, a current time series input vector, and an oil level time series input vector. It should be understood that considering that the operating parameters of the transformer equipment have a time series nature, that is, the parameter values ​​of each operating parameter sample will change over time. Therefore, in order to better capture the time series change pattern of the operating parameters of the transformer equipment, the time series of the operating parameters of the monitored power grid equipment are further sorted according to the parameter sample dimension. That is, the operating parameters such as the temperature value, current value, and oil level value of the transformer equipment are arranged in time sequence to form their respective corresponding time series input vectors, so as to preserve the time sequence relationship of the operating parameter data, thereby more clearly understanding the change trend of each parameter over time and providing basic data for subsequent data analysis and anomaly detection.

[0040] Specifically, the data correction unit 132 is configured to perform gamma correction on the temperature time series input vector, the current time series input vector, and the oil level time series input vector to obtain gamma-corrected temperature time series input vector, gamma-corrected current time series input vector, and gamma-corrected oil level time series input vector. It should be understood that during the acquisition of transformer equipment operating parameters, environmental factors, sensor performance, and other factors may affect the collected data, resulting in bias or noise. Therefore, to improve data quality and accuracy, the temperature time series input vector, the current time series input vector, and the oil level time series input vector are further gamma-corrected to reduce data bias and noise. Specifically, gamma correction is a nonlinear operation that compensates for the nonlinear response of the time series input vectors of each operating parameter to remove nonlinear distortion caused by sensor noise, thereby improving data quality and reliability. This ensures that the corrected time series input vectors more accurately reflect the time series characteristics of the transformer equipment operating parameters, providing more accurate data support for subsequent data analysis and anomaly detection.

[0041] Specifically, the time series pattern feature extraction unit 133 is configured to pass the gamma-corrected temperature time series input vector, the gamma-corrected current time series input vector, and the gamma-corrected oil level time series input vector through a time series pattern feature extractor based on a one-dimensional dilated convolutional neural network model to obtain the temperature time series correlation feature vector, the current time series correlation feature vector, and the oil level time series correlation feature vector. It should be understood that a one-dimensional dilated convolutional neural network (1D Dilated Convolutional Neural Network) is a special convolutional neural network with powerful feature extraction capabilities. In the technical solution of the present application, a time series pattern feature extractor based on a one-dimensional dilated convolutional neural network model is used to perform one-dimensional convolution operations on the gamma-corrected temperature time series input vector, the gamma-corrected current time series input vector, and the gamma-corrected oil level time series input vector, respectively. This can fully capture the dynamic temporal change patterns of various operating parameters of power grid equipment, including periodic changes, long-term change trends, short-term mutations, etc. Furthermore, compared to traditional convolutional neural networks, the one-dimensional dilated convolutional neural network model introduces an expansion rate during the convolution process to control the spacing between elements in the convolution kernel. By adjusting the expansion rate, the convolution kernel's perceptual range can be expanded without increasing network parameters, thereby capturing temporal correlation features over a longer time range and reducing model complexity.

[0042] In a specific example of the present application, the time series pattern feature extraction unit 133 is configured to process the gamma-corrected temperature time series input vector using the following one-dimensional extended convolution formula to obtain the temperature time series correlation feature vector; wherein the one-dimensional extended convolution formula is:

[0043] l'=l+(l-1)×(d-1)

[0044] C i =f(ω·X i:i+j-l‘ +b)

[0045] C=[C1,C2,...,C n-l‘+1 ]

[0046] Among them, the size of the one-dimensional expanded convolution kernel ω is 1×l', l is the length of the original convolution kernel, d is the expansion rate, X i:i+j-l‘ represents the time window of length l' headed by the eigenvalue of the ith position in the gamma-corrected temperature time series input vector, b is the bias term, and b∈R, f(·) is the nonlinear activation function, C i is the i-th local convolutional coding feature vector, n is the dimension of the temperature time series input vector after gamma correction, C is the temperature time series correlation feature vector, and [·,·] represents cascade.

[0047] In the above-mentioned cloud service-based smart grid operation and maintenance system 100, the cross-modal interaction module 140 is used to perform cross-modal interaction on the temperature time series correlation feature vector, the current time series correlation feature vector and the oil level time series correlation feature vector on the cloud server to obtain a transformer multi-modal time series characterization feature vector. It should be understood that, considering that there is usually a correlation between the operating parameters of the transformer equipment, such as the temperature value, current value and oil level value, that is, the changes in these parameters may affect each other, for example, the magnitude of the current may affect the change in temperature. Therefore, in order to make full use of the correlation between these parameters and further improve the accuracy and efficiency of transformer equipment anomaly detection, the temperature time series correlation feature vector, the current time series correlation feature vector and the oil level time series correlation feature vector can be interactively fused.

[0048] Figure 4 FIG is a block diagram of a cross-modal interaction module in a cloud service-based smart grid operation and maintenance system according to an embodiment of the present application. Figure 4 As shown, the cross-modal interaction module 140 includes: a common space mapping unit 141, which is used to pass the temperature time series association feature vector, the current time series association feature vector and the oil level time series association feature vector through a common space mapper based on a fully connected layer to obtain a mapped temperature time series association feature vector, a mapped current time series association feature vector and a mapped oil level time series association feature vector; a cross-modal fusion unit 142, which is used to pass the mapped temperature time series association feature vector, the mapped current time series association feature vector and the mapped oil level time series association feature vector through a cross-modal interaction architecture to obtain the transformer multi-modal time series representation feature vector.

[0049] Specifically, the co-space mapping unit 141 is used to pass the temperature time-series correlation feature vector, the current time-series correlation feature vector, and the oil level time-series correlation feature vector through a co-space mapper based on a fully connected layer to obtain a mapped temperature time-series correlation feature vector, a mapped current time-series correlation feature vector, and a mapped oil level time-series correlation feature vector. It should be understood that, considering that the various operating parameters of the transformer equipment come from different data sources and may have different feature spaces and dimensions, it may be difficult to directly perform feature fusion analysis on them. Therefore, in order to eliminate the feature space differences between different operating parameters, the temperature time-series correlation feature vector, the current time-series correlation feature vector, and the oil level time-series correlation feature vector are further subjected to feature space mapping conversion through a co-space mapper based on a fully connected layer. Specifically, the fully connected layer can learn and fit the nonlinear mapping relationship between the time-series correlation feature vectors of various operating parameters, map the temperature time-series correlation feature vector, the current time-series correlation feature vector, and the oil level time-series correlation feature vector into a unified feature space, and adjust them to the same feature dimension to eliminate the feature space differences between the various operating parameters, so that they can be interactively fused and analyzed in the same feature space.

[0050] Specifically, the cross-modal fusion unit 142 is used to obtain the transformer multi-modal time series characterization feature vector by passing the mapped temperature time series correlation feature vector, the mapped current time series correlation feature vector and the mapped oil level time series correlation feature vector through a cross-modal interaction architecture. It should be understood that the cross-modal interaction architecture can utilize the information interaction between different modal data to obtain a more comprehensive and accurate data representation. In the technical solution of the present application, the mapped temperature time series correlation feature vector, the mapped current time series correlation feature vector and the mapped oil level time series correlation feature vector of different modalities are subjected to information interaction fusion through a cross-modal interaction architecture, which can mine the potential relationship between different sensor data and capture the dependency between the operating parameters of different modalities, thereby better understanding the overall operation of the transformer and providing a more comprehensive feature representation for subsequent abnormality detection tasks.

[0051] In a specific example of the present application, the cross-modal fusion unit 142 is used to: cascade the mapped temperature time series association feature vector, the mapped current time series association feature vector and the mapped oil level time series association feature vector to obtain a multimodal cascade feature vector; perform pooling, dimensionality reduction and activation processing on the multimodal cascade feature vector to obtain a concentrated channel information representation feature vector; pass the concentrated channel information representation feature vector through a first fully connected layer, a second fully connected layer and a third fully connected layer respectively to obtain a first channel information representation coding feature vector, a second channel information representation coding feature vector and a third channel information representation coding feature vector, wherein the first fully connected layer, the second fully connected layer and the third fully connected layer have the same number of nodes; use broadcast multiplication to multiply the first channel information representation coding feature vector and the mapped temperature time series association feature vector, the second channel information representation coding feature vector and the mapped current time series association feature vector, the third channel information representation coding feature vector and the mapped oil level time series association feature vector, and then add the multiplication results to obtain the transformer multimodal time series representation feature vector.

[0052] In the above-mentioned cloud service-based power grid intelligent operation and maintenance system 100, the operation and maintenance voice prompt generation module 150 is used to determine whether to generate an operation and maintenance voice prompt based on the transformer multimodal time series characterization feature vector on the cloud server. Specifically, Figure 5 This is a block diagram of an operation and maintenance voice prompt generation module in a cloud service-based power grid smart operation and maintenance system according to an embodiment of the present application. Figure 5 As shown, the operation and maintenance voice prompt generation module 150 includes: an operation status monitoring unit 151, which is used to pass the transformer multimodal time series characterization feature vector through a classifier-based operation status monitor to obtain a monitoring result, and the monitoring result is used to indicate whether the operation status of the monitored power grid equipment is abnormal; a prompt generation judgment unit 152, which is used to generate the operation and maintenance voice prompt in response to the monitoring result indicating that the operation status of the monitored power grid equipment is abnormal.

[0053] Specifically, the operating status monitoring unit 151 is used to pass the transformer multimodal time series characterization feature vector through a classifier-based operating status monitor to obtain a monitoring result, and the monitoring result is used to indicate whether there is an abnormality in the operating status of the monitored power grid equipment. It should be understood that the classifier can divide the data set into two or more categories based on the input data features. In the technical solution of the present application, the classifier is trained with a large amount of labeled power grid equipment operating data (various operating parameters of the transformer equipment in normal and abnormal states) so that it can accurately distinguish between normal and abnormal operating states. Then, the transformer multimodal time series characterization feature vector is passed through the trained classifier-based operating status monitor, and the classifier can map the transformer multimodal time series characterization feature vector to the corresponding category according to the mapping rule learned during the training process to indicate whether the operating status of the equipment is normal.

[0054] In a specific example of the present application, the operating status monitoring unit 151 is configured to use the classifier-based operating status monitor to process the transformer multimodal time series characterization feature vector using the following classification formula to obtain the monitoring result; wherein the classification formula is:

[0055] O=softmax{(W n ,B n ):…:(W1,B1)│X}

[0056] Among them, W1 to W n is the weight matrix, B1 to B n is the bias vector, X is the transformer multimodal time series characterization feature vector, softmax represents the normalized exponential function, and O represents the monitoring result.

[0057] Specifically, the prompt generation and determination unit 152 is configured to generate the operation and maintenance voice prompt in response to the monitoring result indicating an abnormality in the operating status of the monitored power grid equipment. It should be understood that the operation and maintenance voice prompt can directly convey information and promptly notify relevant personnel of abnormal equipment status without requiring personnel to review monitoring results or alarm information. This saves processing time, enables them to quickly take necessary measures to address the problem, and improves response speed and efficiency. This real-time, intuitive prompt method helps improve the level and efficiency of power grid operation and maintenance management, strengthens the timely response of operation and maintenance personnel to abnormal situations, and reduces the impact of failures on the power grid system.

[0058] It should be understood that before using the above-mentioned neural network model, it is necessary to train the temporal pattern feature extractor based on the one-dimensional extended convolutional neural network model, the co-space mapper based on the fully connected layer, the cross-modal interaction architecture, and the classifier-based operating status monitor. In other words, the cloud-based power grid smart operation and maintenance system of the present application also includes a training module for training the temporal pattern feature extractor based on the one-dimensional extended convolutional neural network model, the co-space mapper based on the fully connected layer, the cross-modal interaction architecture, and the classifier-based operating status monitor.

[0059] Figure 6 FIG is a block diagram of a training module in a cloud service-based smart grid operation and maintenance system according to an embodiment of the present application. Figure 6As shown, the training module 200 includes: a training data acquisition unit 210 for acquiring training data, wherein the training data includes a time series of training operating parameters of the monitored power grid device and a real value of whether the operating state of the monitored power grid device is abnormal; a training data transmission unit 220 for transmitting the time series of training operating parameters of the monitored power grid device to a cloud server; a training data sorting unit 230 for sorting the time series of training operating parameters of the monitored power grid device according to the parameter sample dimension on the cloud server to obtain a training temperature time series input vector, a training current time series input vector and a training oil level time series input vector; a training data gamma The correction unit 240 is used to perform gamma correction on the training temperature timing input vector, the training current timing input vector and the training oil level timing input vector on the cloud server to obtain the training gamma-corrected temperature timing input vector, the training gamma-corrected current timing input vector and the training gamma-corrected oil level timing input vector; the training data timing feature extraction unit 250 is used to pass the training gamma-corrected temperature timing input vector, the training gamma-corrected current timing input vector and the training gamma-corrected oil level timing input vector on the cloud server through the timing pattern feature extractor based on the one-dimensional extended convolutional neural network model to obtain the training temperature timing association feature vector, the training data timing feature extraction unit 250. The training current time series association feature vector and the training oil level time series association feature vector are respectively provided; a training data co-space mapping unit 260 is used for, on the cloud server, passing the training temperature time series association feature vector, the training current time series association feature vector and the training oil level time series association feature vector through the co-space mapper based on the fully connected layer to obtain the training mapped temperature time series association feature vector, the training mapped current time series association feature vector and the training mapped oil level time series association feature vector; a training data cross-modal interaction unit 270 is used for, on the cloud server, passing the training mapped temperature time series association feature vector, the training mapped current time series association feature vector and the training mapped oil level time series association feature vector The eigenvector is passed through the cross-modal interaction architecture to obtain a training transformer multimodal temporal representation feature vector; a classification loss calculation unit 280 is used to pass the training transformer multimodal temporal representation feature vector through the classifier-based operating status monitor on the cloud server to obtain a classification loss function value; a model training unit 290 is used to train the temporal pattern feature extractor based on the one-dimensional extended convolutional neural network model, the fully connected layer-based co-space mapper, the cross-modal interaction architecture and the classifier-based operating status monitor with the classification loss function value, wherein in each round of training iteration, the training transformer multimodal temporal representation feature vector is optimized.

[0060] In the technical solution described above, the training temperature time series association feature vector, the training current time series association feature vector and the training oil level time series association feature vector respectively express the local time series association pattern characteristics of the training temperature value, the training current value and the training oil level value after nonlinear response correction. However, considering the source time series distribution differences of the training current value, the training voltage value and the training temperature value and the enhancement of the main time series distribution trend by the nonlinear response correction, the training temperature time series association feature vector, the training current time series association feature vector and the training oil level time series association feature vector will also have feature distribution misalignment, so that the training mapped temperature time series association feature vector, the training mapped current time series association feature vector and the training mapped oil level time series association feature vector obtained by the co-space mapper based on the fully connected layer will have significant inconsistency between the local distributions of the feature distribution.

[0061] In this way, when the training mapped temperature time series associated feature vector, the training mapped current time series associated feature vector and the training mapped oil level time series associated feature vector are passed through the cross-modal interaction architecture, the local distribution of the feature distributions therebetween will be inconsistent, resulting in the discretized local feature distribution of the training transformer multimodal time series representation feature vector obtained by cross-modal feature interaction fusion, thereby affecting the convergence effect of the training transformer multimodal time series representation feature vector to the class probability density space when classified by the classifier.

[0062] Therefore, in the technical solution of the present application, each time the multimodal temporal characterization feature vector of the training transformer is classified and iterated by the classifier, the multimodal temporal characterization feature vector of the training transformer is optimized using the following optimization formula, wherein the optimization formula is:

[0063]

[0064]

[0065] Where V is the multimodal temporal representation feature vector of the trained transformer, v i and v j are the eigenvalues ​​of the i-th and j-th positions of the multimodal temporal representation feature vector V of the training transformer, M μ (i, j) is the mean between every two eigenvalues ​​of the multimodal time series representation feature vector of the training transformer, M μ is the mean matrix of the multimodal temporal representation feature vector of the training transformer, M σ (i, j) is the average difference between every two eigenvalues ​​of the multimodal temporal representation feature vector of the trained transformer, M μis the mean difference matrix of the multimodal temporal representation feature vector of the training transformer, and V′ is the optimized multimodal temporal representation feature vector of the training transformer.

[0066] That is, by introducing the local statistical information distribution of the training transformer multimodal time series representation feature vector V as an external information source to perform feature vector retrieval enhancement, so as to avoid the distribution illusion of the training transformer multimodal time series representation feature vector V caused by local overflow information distribution based on local statistical dense information structuring, thereby obtaining the information trustworthy response reasoning of the training transformer multimodal time series representation feature vector V based on the retention of local distribution group dimension, so as to obtain the trustworthy distribution response of the training transformer multimodal time series representation feature vector V in the probability density space based on the discretized local feature distribution, thereby improving the convergence effect of the probability density space, so as to improve the training speed and the accuracy of the training results.

[0067] In summary, the cloud-based smart grid operation and maintenance system according to the embodiment of the present application is described. It uses deep learning-based cloud computing technology to monitor and analyze the operating parameters of grid equipment, extract the dynamic change pattern characteristics of each operating parameter of the grid equipment in the time domain, and perform cross-modal interactive fusion of the multi-modal operating parameter time domain correlation pattern characteristics to mine abnormal operating status information of the grid equipment, thereby realizing intelligent identification of abnormal operating status of the grid equipment. In this way, the grid equipment can be monitored in real time and abnormal conditions of the equipment can be discovered in a timely manner, thereby helping operation and maintenance personnel to formulate reasonable maintenance plans, improve operation and maintenance efficiency, and extend equipment life.

[0068] The basic principles of the present invention have been described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, and effects mentioned in the present invention are merely illustrative and non-limiting, and should not be construed as necessarily possessed by each embodiment of the present invention. Furthermore, the specific details of the above embodiments are provided for illustrative purposes and to facilitate understanding, and are not intended to be limiting. These details do not necessarily limit the present invention to being implemented using these specific details.

[0069] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, please refer to the relevant description of other embodiments. In the several embodiments provided by the present invention, it should be understood that the disclosed system and method can be implemented in other ways. For example, the system embodiment described above is only schematic. For example, the module division is only a logical function division, and there may be other division methods in actual implementation. The modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0070] In addition, the functional modules in various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or hardware plus software functional modules.

[0071] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be encompassed therein. Any reference to a figure in a claim should not be construed as limiting the claim to which it relates.

[0072] In addition, it is obvious that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units stated in the system claims can also be implemented by one unit through software or hardware.

[0073] Finally, it should be noted that the above description has been provided for purposes of illustration and description. Furthermore, the above embodiments are intended only to illustrate the technical solutions of the present invention and are not intended to be limiting. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art will appreciate that the technical solutions of the present invention may be modified or replaced with equivalents without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A cloud service-based power grid intelligent operation and maintenance system, characterized by: include: an operating parameter acquisition module, configured to acquire a time series of operating parameters of a monitored power grid device collected by a sensor group deployed on the monitored power grid device, wherein the monitored power grid device is a transformer, and the operating parameters of the monitored power grid device include temperature, current, and oil level; A data transmission module, configured to transmit the time series of the operating parameters of the monitored power grid equipment to a cloud server; a time series correlation analysis module, configured to extract time series features from the time series of the operating parameters of the monitored power grid equipment on the cloud server to obtain a temperature time series correlation feature vector, a current time series correlation feature vector, and an oil level time series correlation feature vector; a cross-modal interaction module, configured to perform cross-modal interaction on the temperature time series correlation feature vector, the current time series correlation feature vector, and the oil level time series correlation feature vector on the cloud server to obtain a transformer multi-modal time series characterization feature vector; An operation and maintenance voice prompt generation module is used to determine whether to generate an operation and maintenance voice prompt based on the transformer multimodal time series characterization feature vector on the cloud server.

2. The cloud service-based smart grid operation and maintenance system according to claim 1, characterized in that: The time series correlation analysis module includes: a data arranging unit, configured to arrange the time series of the operating parameters of the monitored power grid equipment according to parameter sample dimensions to obtain a temperature time series input vector, a current time series input vector, and an oil level time series input vector; a data correction unit, configured to perform gamma correction on the temperature time series input vector, the current time series input vector, and the oil level time series input vector to obtain a gamma-corrected temperature time series input vector, a gamma-corrected current time series input vector, and a gamma-corrected oil level time series input vector; A timing pattern feature extraction unit is used to pass the gamma-corrected temperature timing input vector, the gamma-corrected current timing input vector and the gamma-corrected oil level timing input vector through a timing pattern feature extractor based on a one-dimensional extended convolutional neural network model to obtain the temperature timing association feature vector, the current timing association feature vector and the oil level timing association feature vector.

3. The cloud service-based smart power grid operation and maintenance system according to claim 2, characterized in that: The temporal pattern feature extraction unit is configured to: The gamma-corrected temperature time series input vector is processed using the following one-dimensional extended convolution formula to obtain the temperature time series correlation feature vector; wherein the one-dimensional extended convolution formula is: l‘＝l+(l-1)×(d-1) C i =f(ω·X i:i+j-l‘ +b) C=[C1,C2,...,C n-l‘+1 ] Among them, the size of the one-dimensional expanded convolution kernel ω is 1×l', l is the length of the original convolution kernel, d is the expansion rate, X i:i+j-l‘ represents the time window of length l' headed by the eigenvalue of the ith position in the gamma-corrected temperature time series input vector, b is the bias term, and b∈R, f(·) is the nonlinear activation function, C i is the i-th local convolutional coding feature vector, n is the dimension of the temperature time series input vector after gamma correction, C is the temperature time series correlation feature vector, and [·,·] represents cascade.

4. The cloud service-based smart power grid operation and maintenance system according to claim 3, characterized in that: The cross-modal interaction module includes: a common space mapping unit, configured to pass the temperature time series correlation feature vector, the current time series correlation feature vector, and the oil level time series correlation feature vector through a common space mapper based on a fully connected layer to obtain a mapped temperature time series correlation feature vector, a mapped current time series correlation feature vector, and a mapped oil level time series correlation feature vector; A cross-modal fusion unit is used to obtain the transformer multi-modal time series representation feature vector by using the mapped temperature time series correlation feature vector, the mapped current time series correlation feature vector and the mapped oil level time series correlation feature vector through a cross-modal interaction architecture.

5. The cloud service-based smart power grid operation and maintenance system according to claim 4, characterized in that: The cross-modal blending unit is used to: cascading the mapped temperature time series correlation feature vector, the mapped current time series correlation feature vector, and the mapped oil level time series correlation feature vector to obtain a multi-modal cascade feature vector; Performing pooling, dimensionality reduction, and activation processing on the multimodal cascade feature vector to obtain a concentrated channel information representation feature vector; Passing the concentrated channel information representation feature vector through a first fully connected layer, a second fully connected layer, and a third fully connected layer, respectively, to obtain a first channel information representation encoding feature vector, a second channel information representation encoding feature vector, and a third channel information representation encoding feature vector, wherein the first fully connected layer, the second fully connected layer, and the third fully connected layer have the same number of nodes; Use broadcast multiplication to multiply the first channel information representation coding feature vector and the mapped temperature time series association feature vector, the second channel information representation coding feature vector and the mapped current time series association feature vector, and the third channel information representation coding feature vector and the mapped oil level time series association feature vector, and then add the multiplication results to obtain the transformer multimodal time series representation feature vector.

6. The cloud service-based smart power grid operation and maintenance system according to claim 5, characterized in that: The operation and maintenance voice prompt generation module includes: An operating status monitoring unit, configured to pass the transformer multimodal time series characterization feature vector through a classifier-based operating status monitor to obtain a monitoring result, wherein the monitoring result is used to indicate whether the operating status of the monitored power grid equipment is abnormal; The prompt generation judgment unit is used to generate the operation and maintenance voice prompt in response to the monitoring result that the operating status of the monitored power grid equipment is abnormal.

7. The cloud service-based smart power grid operation and maintenance system according to claim 6, characterized in that: The operating status monitoring unit is used to: The classifier-based operating status monitor is used to process the transformer multimodal time series characterization feature vector using the following classification formula to obtain the monitoring result; wherein the classification formula is: O=softmax{(W n ,B n ):…:(W1,B1)│X} Among them, W1 to W n is the weight matrix, B1 to B n is the bias vector, X is the transformer multimodal time series characterization feature vector, softmax represents the normalized exponential function, and O represents the monitoring result.

8. The cloud service-based smart power grid operation and maintenance system according to claim 7, characterized in that: It also includes a training module for training the temporal pattern feature extractor based on the one-dimensional extended convolutional neural network model, the co-space mapper based on the fully connected layer, the cross-modal interaction architecture and the classifier-based operation status monitor.

9. The cloud service-based smart power grid operation and maintenance system according to claim 8, characterized in that: The training module includes: a training data acquisition unit, configured to acquire training data, the training data including a time series of training operating parameters of the monitored power grid device and a true value of whether an abnormality exists in the operating state of the monitored power grid device; A training data transmission unit, configured to transmit a time series of training operating parameters of the monitored power grid equipment to a cloud server; a training data arranging unit, configured to, on the cloud server, arrange the time series of the training operating parameters of the monitored power grid equipment according to parameter sample dimensions to obtain a training temperature time series input vector, a training current time series input vector, and a training oil level time series input vector; a training data gamma correction unit, configured to perform gamma correction on the training temperature time series input vector, the training current time series input vector, and the training oil level time series input vector on the cloud server to obtain a training gamma-corrected temperature time series input vector, a training gamma-corrected current time series input vector, and a training gamma-corrected oil level time series input vector; a training data time series feature extraction unit, configured to, on the cloud server, pass the training gamma-corrected temperature time series input vector, the training gamma-corrected current time series input vector, and the training gamma-corrected oil level time series input vector through the time series pattern feature extractor based on the one-dimensional extended convolutional neural network model to obtain a training temperature time series association feature vector, a training current time series association feature vector, and a training oil level time series association feature vector; a training data co-space mapping unit, configured to, on the cloud server, pass the training temperature time series association feature vector, the training current time series association feature vector, and the training oil level time series association feature vector through the fully connected layer-based co-space mapper to obtain a trained mapped temperature time series association feature vector, a trained mapped current time series association feature vector, and a trained mapped oil level time series association feature vector; A training data cross-modal interaction unit is configured to, on the cloud server, pass the training mapped temperature time series correlation feature vector, the training mapped current time series correlation feature vector, and the training mapped oil level time series correlation feature vector through the cross-modal interaction architecture to obtain a training transformer multi-modal time series representation feature vector; A classification loss calculation unit is used to pass the multimodal time series representation feature vector of the training transformer through the classifier-based operation status monitor on the cloud server to obtain a classification loss function value; A model training unit is used to train the temporal pattern feature extractor based on the one-dimensional extended convolutional neural network model, the co-space mapper based on the fully connected layer, the cross-modal interaction architecture, and the classifier-based operating status monitor using the classification loss function value, wherein in each round of iteration of the training, the multimodal temporal representation feature vector of the training transformer is optimized.

Images