Power network monitoring method and system

Abstract

The invention relates to the technical field of power dispatching data networks, in particular to a power network monitoring method and system. Operation interface image data of all machine room equipment can be acquired through image acquisition equipment such as a monitoring system in a machine room, and image processing is performed on the operation interface image data of all the machine room equipment by utilizing a classification device based on a depth algorithm technology. Machine room equipment operation state diagnosis based on video data can be realized; therefore, real-time diagnosis and visual display of the operation state of each machine room device can be realized based on the existing monitoring system in the machine room.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power dispatching data network, and in particular to a method and system for monitoring an electric power network. Background Art

[0002] The power dispatching data network refers to a dedicated wide area data network serving the power dispatching production. It is the infrastructure for real-time and quasi-real-time data communication between power dispatching production departments and between power dispatching production departments and power plants and substations, such as computer monitoring systems. The power dispatching data network is the basis for the automation and modernization of power grid dispatching and management. It is an important means to ensure the safe, stable and economical operation of the power grid. It is an important infrastructure of the power system. It plays an important role in coordinating the joint operation of the power system's generation, transmission, transformation, distribution, and power consumption components and ensuring the safe, economical, stable and reliable operation of the power grid. It also effectively guarantees the communication needs of power production, power dispatching, reservoir dispatching, fuel dispatching, relay protection, safety automatic devices, telecontrol, power grid dispatching automation, etc., and plays an irreplaceable role in power production and management.

[0003] The data network for power dispatching mainly consists of the following structures: the access layer structure of the power dispatching data network, the backbone layer structure of the power dispatching data network, and the core layer structure of the power dispatching data network. For the access layer structure of the power dispatching data network, it is actually the access and collection of information, mainly collecting and transmitting power information and data in various links of the power system; for the backbone layer structure of the power dispatching data network, it is mainly composed of power transmission methods and power operation protocols in different links; the core layer of the power dispatching data network structure is the most critical structural component of the entire power data network structure. Generally, the core layer structure of the power dispatching data network is located in the center of each region, so as to realize the information and data exchange of the entire power system. Through the construction of the power dispatching data network, the control and management level of power management enterprises over power dispatching can be effectively improved.

[0004] The specific work content of the existing power dispatching is to judge the safety, economic operation and production status of the power grid based on the data flow information fed back by various information collection devices, or the information provided by monitoring personnel and monitoring equipment through recording, combined with the actual parameters generated in the power grid, such as voltage, current, frequency, load and resistance data, and comprehensively consider the operation of various production work, to ensure the continuous safe and stable operation of the power grid.

[0005] In recent years, with the continuous development of technology, the continuous improvement of modern supervision and control devices, the technical support and effective management of power dispatching have become increasingly powerful. The most important equipment in video surveillance equipment is the camera, which can be used as the front-end video image signal collection. The video surveillance system is a comprehensive system with strong prevention capabilities. Video surveillance is widely used in many occasions for its intuitive, accurate, timely and rich information content. With the rapid development of computers, networks, image processing and transmission technologies, video surveillance technology has also made great progress.

[0006] At present, based on video data, the real-time operating status of various front-end devices in the power dispatching data network, such as computer room equipment, can be obtained; in theory, based on image processing of video data, it is possible to extract the operating interface image of each computer room equipment from the monitoring video data, and then realize the operating status judgment of the computer room equipment based on image classification processing. Although the image processing technology based on deep algorithms is relatively complete, when it is applied to the image processing of computer room equipment, the following issues need to be considered: 1. In the same computer room, there are generally multiple computer room equipment with different models and specifications, and different computer room equipment has different (size, color, pattern, etc.) operating interfaces, that is, it is basically impossible to complete the judgment of the operating status of different equipment using the same image classification algorithm model; 2. Generally speaking, the number of image acquisition devices in a computer room is far less than the number of computer room equipment, that is, in the monitoring video data, there must be a large number of computer room equipment operating interface images that are not collected in the positive projection posture, which brings huge challenges to the training process and recognition accuracy of the image classification algorithm model. Summary of the invention

[0007] The present invention provides a power network monitoring method and system, which can overcome certain defects of the prior art.

[0008] A power network monitoring method according to the present invention comprises:

[0009] Based on the image acquisition equipment, the monitoring video data during the operation of the computer room is collected; wherein the monitoring video data is composed of multiple frames of continuous image data, and any image data contains the operating interface image data of the covered computer room equipment;

[0010] Based on the processing of image data in the monitoring video data, the operating status of each equipment room is obtained.

[0011] Based on the above, it is possible to better realize real-time diagnosis and intuitive display of the operating status of each equipment in the computer room based on the existing monitoring system in the computer room.

[0012] Preferably, the monitoring video data collected during the operation of the computer room based on the image acquisition device includes:

[0013] The computer room coordinate system is constructed by taking the installation point of the image acquisition device as the origin, the length direction of the computer room in the actual scene as the X-axis, the width direction of the computer room in the actual scene as the Y-axis, and the height direction of the computer room in the actual scene as the Z-axis;

[0014] Based on the coordinate points of the display screen of the equipment in the computer room, obtain the spatial normal vector of the plane where the display screen of each equipment in the computer room is located;

[0015] Based on the spatial normal vectors of the planes where the display screens of all the equipment in the computer room are located, a constraint is established that the angles between the imaging plane of the image acquisition device and the plane where the display screens of any equipment in the computer room are located reach a set threshold and the average of the angles is the smallest, and the optimal solution is used as the optical axis vector of the image acquisition device;

[0016] The azimuth angle required by the image acquisition device is acquired based on the optical axis vector, and the posture of the image acquisition device is adjusted based on the azimuth angle.

[0017] It can be understood that the more parallel the display screen of the computer room equipment is to the imaging plane of the image acquisition device, the more comprehensively its operation interface will be captured in the monitoring video data. In the present disclosure, by constructing a computer room coordinate system, the imaging plane of the image acquisition device and the plane where the display screen of the computer room equipment is located can be characterized in the form of a spatial vector; by constructing constraints and based on the operation of spatial vectors, the acquisition of the optimal optical axis direction of the image acquisition device can be achieved, wherein the optical axis direction can be characterized as an optical axis vector; based on the acquisition of the posture information of the optical axis vector in space, the acquisition of the optimal posture of the image acquisition device can be achieved; wherein, the optimal posture of the image acquisition device can be characterized as an azimuth in space; and after the azimuth is obtained, the posture adjustment of the image acquisition device can be achieved, and then any image acquisition device can better achieve the acquisition of the operation interface of the computer room equipment it covers on a global basis.

[0018] As a preferred embodiment, for any equipment i in the computer room, its space normal vector n i The methods for obtaining include:

[0019] Get the three non-collinear coordinate points A of the display screen of equipment i in the computer room i , B i and C i The coordinates of and

[0020] Get the spatial normal vector n of equipment i in the computer room i , n i =(a i , b i , c i );in,

[0021]

[0022]

[0023]

[0024] Based on the above, it is possible to better obtain the spatial normal vector of the plane where the display screen of any equipment in the computer room is located.

[0025] Preferably, the optical axis vector n 0 With all space normal vectors n i The constraints between them are as follows:

[0026]

[0027] Among them, n 0 =(j, k, l), N is the total number of equipment in the computer room, θ i is the angle between the imaging plane of the image acquisition device and the plane where the display screen of the equipment i in the computer room is located;

[0028]

[0029] Through the above, it is possible to constrain the angle between the imaging plane of the image acquisition device and the plane where the display screen of any computer room equipment is located to be smaller than the preset angle threshold θ. At the same time, based on the minimum average angle between the imaging plane and the plane where the display screens of all computer room equipment are located, it can be achieved globally that the operating interface of each computer room equipment can project a projection image that meets the preset projection ratio on the imaging plane of the image acquisition device, so that the operating interface image data projected by each operating interface can be achieved globally with the greatest comprehensiveness.

[0030] Preferably, the azimuth angle required by the image acquisition device includes an angle α with the X-axis, an angle β with the Y-axis, and an angle γ with the Z-axis, wherein:

[0031]

[0032]

[0033]

[0034] Based on the above, the azimuth angle in space can be preferably acquired based on the optical axis vector, which is beneficial to the posture adjustment of the image acquisition device.

[0035] Preferably, the operation status of each equipment room device is obtained based on the processing of the image data in the monitoring video data, including:

[0036] Segment the image data to obtain the operating interface image data corresponding to each equipment in the computer room;

[0037] The operating interface image data of each equipment in the computer room is sent to the corresponding image feature extraction model to extract image features; wherein the image feature extraction model adopts a convolutional neural network;

[0038] The extracted image features are sent to the state classification model to obtain the operating status of the corresponding equipment in the computer room; wherein the state classification model adopts a fully connected neural network.

[0039] Based on the above, it is possible to first obtain single-frame image data from the monitoring video data, and then based on the segmentation processing of the image data, the operation interface image data of each computer room equipment in the image data can be obtained, and by performing feature extraction on the corresponding operation interface image data with a matching image feature extraction model, it is possible to achieve better extraction of image features of the operation interface image data of different computer room equipment; at the same time, the operation interface image data of each computer room equipment can be classified at the same state classification model after feature extraction, thereby realizing the judgment of the operation state, so it is possible to take into account the differences between different computer room equipment while also being able to better reduce the training workload of related models.

[0040] Preferably, the image feature extraction model corresponding to any equipment in the computer room is trained and obtained in the following manner:

[0041] Construct an initial image feature extraction model;

[0042] Acquire multiple groups of sample pairs and construct a sample pair library; wherein any sample pair includes orthographic projection image data and non-orthographic projection image data of the corresponding equipment in the computer room under the same operation interface;

[0043] The initial image feature extraction model is trained based on the sample pair library to obtain the corresponding image feature extraction model; wherein a loss function is constructed based on the output value of the sample pair.

[0044] Based on the above, the final image feature extraction model obtained after training can extract image features similar to the image features extracted from the orthographic projection image data from the non-orthographic projection image data; this method makes it necessary for the image feature extraction model to be trained only for the types of computer room equipment. After the training, the image feature extraction model obtained can be applied to the processing of operating interface image data at the corresponding computer room equipment in multiple different computer room environments; therefore, it has good versatility.

[0045] Preferably, the loss function is constructed based on the similarity of the output values ​​of the sample pairs, so that the construction of the loss function can be better realized.

[0046] As a preferred method, the state classification model is trained and obtained in the following manner:

[0047] An initial operation state classification model is constructed, wherein the initial operation state classification model has an initial feature extraction layer and an initial classification output layer connected in sequence; wherein the initial feature extraction layer has the same structure as the image feature extraction model, and the initial classification output layer adopts a fully connected neural network;

[0048] Establish a sample library, the samples in the sample library include the operation interface images of each equipment in the computer room under different operation states, the operation interface images selected in each sample are all orthographic projection image data under the operation interface, and the sample label of each sample is the operation state of the corresponding operation interface image;

[0049] The initial operation status classification model is trained based on the sample library to obtain the operation status classification model; the network parameters of the classification output layer in the operation status classification model are extracted;

[0050] Based on the extracted network parameters of the classification output layer, a state classification model is constructed.

[0051] Based on the above, the acquisition of a state classification model with wide versatility is achieved.

[0052] A power network monitoring system according to the present invention comprises:

[0053] An image acquisition device used to collect monitoring video data during the operation of the computer room; wherein the monitoring video data is composed of multiple frames of continuous image data, and any image data contains the operating interface image data of the covered computer room equipment;

[0054] A classification device for obtaining the operating status of each computer room device based on monitoring video data. Based on the above, the operating interface image data of all computer room devices can be collected through image acquisition devices such as the monitoring system in the computer room, and the operating interface image data of all computer room devices can be processed using a classification device based on, for example, deep algorithm technology, so that the operating status diagnosis of computer room devices based on video data can be realized; therefore, based on the existing monitoring system in the computer room, the operating status of each computer room device can be diagnosed in real time and displayed intuitively. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a schematic diagram of the operation of a power network monitoring system disclosed in the present invention;

[0056] Figure 2 A schematic diagram of a method for determining the operating status of equipment in a computer room disclosed in the present invention;

[0057] Figure 3 A schematic diagram of the training process of an image feature extraction model disclosed in the present invention. DETAILED DESCRIPTION

[0058] In order to further understand the content of the present invention, the present invention is described in detail in conjunction with the embodiments. It should be understood that the embodiments are only for explaining the present invention and are not intended to limit it.

[0059] In order to monitor and manage the operating status of each device in a computer room such as a data center, the present invention provides a power network monitoring system in one embodiment.

[0060] See Figure 1 , a power network monitoring system disclosed in the above embodiment may include:

[0061] An image acquisition device used to collect monitoring video data during the operation of the computer room; wherein the monitoring video data is composed of multiple frames of continuous image data, and any image data contains the operating interface image data of the covered computer room equipment;

[0062] A classification device for obtaining the operating status of each equipment room device based on monitoring video data.

[0063] Based on the above, the operating interface image data of all computer room equipment can be collected through image acquisition devices such as the monitoring system in the computer room, and the operating interface image data of all computer room equipment can be processed using, for example, a classification device based on deep algorithm technology, thereby realizing the diagnosis of the operating status of computer room equipment based on video data; therefore, based on the existing monitoring system in the computer room, real-time diagnosis and intuitive display of the operating status of each computer room equipment can be realized.

[0064] Based on the power network monitoring system in the above embodiment, the present invention also provides a power network monitoring method, which includes:

[0065] Based on the image acquisition equipment, the monitoring video data during the operation of the computer room is collected; wherein the monitoring video data is composed of multiple frames of continuous image data, and any image data contains the operating interface image data of the covered computer room equipment;

[0066] Based on the processing of image data in the monitoring video data, the operating status of each equipment room is obtained.

[0067] Based on the above, it is possible to better realize real-time diagnosis and intuitive display of the operating status of each equipment in the computer room based on the existing monitoring system in the computer room.

[0068] In the above-mentioned embodiment of the power network monitoring method, it is necessary to consider that one image acquisition device in the computer room will cover multiple computer room devices, and the function of the existing video monitoring system in the computer room is only to monitor the real-time environment in the computer room. Based on this monitoring purpose, it is difficult to take into account the good acquisition of the operating interface of each computer room device. In particular, it is impossible to ensure that the operating interface of each computer room device can be captured more comprehensively.

[0069] Therefore, in a preferred implementation, the monitoring video data collected during the operation of the computer room based on the image acquisition device can include:

[0070] The computer room coordinate system can be constructed by taking the installation point of the image acquisition device as the origin, the length direction of the computer room in the actual scene as the X-axis, the width direction of the computer room in the actual scene as the Y-axis, and the height direction of the computer room in the actual scene as the Z-axis;

[0071] Based on the coordinate points of the display screen of the equipment in the computer room, obtain the spatial normal vector of the plane where the display screen of each equipment in the computer room is located;

[0072] Based on the spatial normal vectors of the planes where the display screens of all the equipment in the computer room are located, a constraint is established that the angles between the imaging plane of the image acquisition device and the plane where the display screens of any equipment in the computer room are located reach a set threshold and the average of the angles is the smallest, and the optimal solution is used as the optical axis vector of the image acquisition device;

[0073] The azimuth angle required by the image acquisition device is acquired based on the optical axis vector, and the posture of the image acquisition device is adjusted based on the azimuth angle.

[0074] It can be understood that the more parallel the display screen of the computer room equipment is to the imaging plane of the image acquisition device, the more comprehensively its operation interface will be captured in the monitoring video data. In this embodiment, by constructing a computer room coordinate system, the imaging plane of the image acquisition device and the plane where the display screen of the computer room equipment is located can be characterized in the form of a spatial vector; by constructing constraints and based on spatial vector operations, the acquisition of the optimal optical axis direction of the image acquisition device can be achieved, wherein the optical axis direction can be characterized as an optical axis vector; based on the acquisition of the posture information of the optical axis vector in space, the acquisition of the optimal posture of the image acquisition device can be achieved; wherein, the optimal posture of the image acquisition device can be characterized as an azimuth in space; and after obtaining the azimuth, the posture adjustment of the image acquisition device can be achieved, and then any image acquisition device can better achieve the acquisition of the operation interface of the computer room equipment it covers on a global basis.

[0075] In fact, when acquiring the azimuth angle in space, the spatial posture of the image acquisition device can be automatically adjusted based on existing equipment such as a pan / tilt.

[0076] Specifically, for any equipment i in the computer room, its space normal vector n i The methods for obtaining include:

[0077] Get the three non-collinear coordinate points A of the display screen of equipment i in the computer room i , B i and C i The coordinates of and

[0078] Get the spatial normal vector n of equipment i in the computer room i , n i =(a i , b i , c i );in,

[0079]

[0080]

[0081]

[0082] Based on the above, it is possible to better obtain the spatial normal vector of the plane where the display screen of any equipment in the computer room is located.

[0083] In fact, the display screens of current equipment in computer rooms are all rectangular, so it is only necessary to select the positions of any three corner points of the display screen as three non-collinear coordinate points.

[0084] In a preferred embodiment, the optical axis vector n 0 With all space normal vectors n i The following constraints can be established:

[0085]

[0086] Among them, n 0 =(j, k, l), N is the total number of equipment in the computer room, θ i is the angle between the imaging plane of the image acquisition device and the plane where the display screen of the equipment i in the computer room is located;

[0087]

[0088] Through the above, it is possible to constrain the angle between the imaging plane of the image acquisition device and the plane where the display screen of any computer room equipment is located to be smaller than the preset angle threshold θ. At the same time, based on the minimum average angle between the imaging plane and the plane where the display screens of all computer room equipment are located, it can be achieved globally that the operating interface of each computer room equipment can project a projection image that meets the preset projection ratio on the imaging plane of the image acquisition device, so that the operating interface image data projected by each operating interface can be achieved globally with the greatest comprehensiveness.

[0089] The azimuth angles required by the image acquisition device include an angle α with the X-axis, an angle β with the Y-axis, and an angle γ with the Z-axis, wherein:

[0090]

[0091]

[0092]

[0093] Based on the above, the azimuth angle in space can be preferably acquired based on the optical axis vector, which is beneficial to the posture adjustment of the image acquisition device.

[0094] In the above method embodiment, the obtaining of the operating status of each equipment room device based on the processing of the image data in the monitoring video data may include:

[0095] Segment the image data to obtain the operating interface image data corresponding to each equipment in the computer room;

[0096] The operating interface image data of each equipment in the computer room is sent to the corresponding image feature extraction model to extract image features; wherein the image feature extraction model adopts a convolutional neural network;

[0097] The extracted image features are sent to the state classification model to obtain the operating status of the corresponding equipment in the computer room; wherein the state classification model adopts a fully connected neural network.

[0098] Based on the above, it is possible to first obtain single-frame image data from the monitoring video data, and then based on the segmentation processing of the image data, the operation interface image data of each computer room equipment in the image data can be obtained, and by performing feature extraction on the corresponding operation interface image data with a matching image feature extraction model, it is possible to achieve better extraction of image features of the operation interface image data of different computer room equipment; at the same time, the operation interface image data of each computer room equipment can be classified at the same state classification model after feature extraction, thereby realizing the judgment of the operation state, so it is possible to take into account the differences between different computer room equipment while also being able to better reduce the training workload of related models.

[0099] Specifically, in a specific application scenario such as a data center, the number of types of computer room equipment involved is actually limited, so the state classification model can be trained based on all the types of computer room equipment. In this way, the state classification model obtained through training can be applied to classify and judge the operating status of almost all computer room equipment; that is, after the state classification model is trained in a targeted manner, it will have strong versatility, so there is no need to conduct separate training for each different computer room equipment, thereby reducing the training workload.

[0100] In addition, it can be known that when the same image acquisition device acquires the operating interface of the computer room equipment in the area it covers, due to the existence of the shooting angle, the acquired operating interface image data will hardly be shot under the orthographic projection perspective, which will inevitably result in the loss of some image information; at the same time, it can be understood that even the operating interface image data acquired from the same computer room equipment at different shooting angles will be different; therefore, in this embodiment, the operating interface of the corresponding computer room equipment is captured in a targeted manner based on different image feature extraction models trained separately, which can better overcome this difference and improve the accuracy of operating status judgment.

[0101] by Figure 2 Take this as an example to explain this in detail; Figure 2 The single frame image data in the image contains four kinds of operation interface image data of different equipment in the computer room, which are 01-04 respectively; after segmentation processing, the operation interface image data 01-04 can be segmented and extracted separately; then, the operation interface image data 01-04 can be sent to the corresponding image feature extraction model 01-04 respectively for image feature extraction processing; then, the extracted four image features can all be sent to the same state classification model for classification, so as to realize the judgment of the operation state.

[0102] The segmentation process of the image data can be realized based on current means such as segmentation algorithms, and the improvement points that need to be emphasized in the present invention are not here, so they will not be elaborated.

[0103] In the above, the image feature extraction model can be implemented based on a conventional convolutional neural network, such as a 3D-CNN model, a 2D-CNN, etc.; it can be understood that the convolutional neural network can be composed of one or more convolutional units connected in sequence, and the convolutional unit can have a convolutional layer and a pooling layer connected in sequence, etc.; it will not be repeated in this embodiment.

[0104] In the above, the state classification model can be implemented based on a conventional fully connected neural network, which can have a fully connected layer and an output layer, and the output layer can be implemented based on, for example, a Softmax function, which will not be described in detail in this embodiment.

[0105] Based on the above, it can be known that the image feature extraction model for each computer room device needs to be trained separately; wherein, the image feature extraction model corresponding to any computer room device can be trained and obtained in the following manner:

[0106] Construct an initial image feature extraction model;

[0107] Acquire multiple groups of sample pairs and construct a sample pair library; wherein any sample pair includes orthographic projection image data and non-orthographic projection image data of the corresponding equipment in the computer room under the same operation interface;

[0108] The initial image feature extraction model is trained based on the sample pair library to obtain the corresponding image feature extraction model; wherein a loss function is constructed based on the output value of the sample pair.

[0109] Based on the above, the final image feature extraction model obtained after training can extract image features similar to the image features extracted from the orthographic projection image data from the non-orthographic projection image data; this method makes it necessary for the image feature extraction model to be trained only for the types of computer room equipment. After the training, the image feature extraction model obtained can be applied to the processing of operating interface image data at the corresponding computer room equipment in multiple different computer room environments; therefore, it has good versatility.

[0110] by Figure 3 As an illustration, when the image feature extraction model is trained, the initial image feature extraction model can be trained based on sample pairs, and a loss function can be constructed based on two output values ​​of the same sample pair, thereby updating the model parameters of the initial image feature extraction model.

[0111] Specifically, the loss function can be constructed based on the similarity of the output values ​​of the sample pairs, so that the construction of the loss function can be better achieved.

[0112] It is understandable that the similarity can be, for example, Euclidean distance, cosine similarity, etc. After the loss function is constructed, the goal of model optimization can be set to similarity reaching a set threshold, etc., thereby achieving the final image feature extraction model in an iterative manner.

[0113] In addition, the state classification model can also be trained and obtained in the following way:

[0114] An initial operation state classification model is constructed, wherein the initial operation state classification model has an initial feature extraction layer and an initial classification output layer connected in sequence; wherein the initial feature extraction layer has the same structure as the image feature extraction model, and the initial classification output layer adopts a fully connected neural network;

[0115] Establish a sample library, the samples in the sample library include the operation interface images of each equipment in the computer room under different operation states, the operation interface images selected in each sample are all orthographic projection image data under the operation interface, and the sample label of each sample is the operation state of the corresponding operation interface image;

[0116] The initial operation status classification model is trained based on the sample library to obtain the operation status classification model; the network parameters of the classification output layer in the operation status classification model are extracted;

[0117] Based on the extracted network parameters of the classification output layer, a state classification model is constructed.

[0118] Based on the above, the acquisition of a state classification model with wide versatility is achieved.

[0119] Of course, those skilled in the art can understand that, in the process of training the initial operating status classification model, iterative optimization can be performed based on the label value and the predicted value, which can be achieved by conventional means and will not be elaborated in this embodiment.

[0120] It is easy to understand that those skilled in the art can combine, split, reorganize, etc. the embodiments of the present application to obtain other embodiments based on one or several embodiments provided in the present application, and these embodiments do not exceed the protection scope of the present application.

[0121] The present invention and its implementation methods are described schematically above, and the description is not restrictive. The embodiments shown in the embodiments are only part of the implementation methods of the present invention, and the actual structure is not limited thereto. Therefore, if a person skilled in the art is inspired by the embodiments and designs a structure and an implementation method similar to the technical solution without creativity without departing from the purpose of the invention, they should all fall within the protection scope of the present invention.

Claims

1. A power network monitoring method, comprising: Based on the image acquisition equipment, the monitoring video data during the operation of the computer room is collected; wherein the monitoring video data is composed of multiple frames of continuous image data, and any image data contains the operating interface image data of the covered computer room equipment; Based on the processing of image data in the monitoring video data, the operating status of each equipment room is obtained.

2. A power network monitoring method according to claim 1, characterized in that: The monitoring video data collected during the operation of the computer room based on the image acquisition device includes: The computer room coordinate system is constructed by taking the installation point of the image acquisition device as the origin, the length direction of the computer room in the actual scene as the X-axis, the width direction of the computer room in the actual scene as the Y-axis, and the height direction of the computer room in the actual scene as the Z-axis; Based on the coordinate points of the display screen of the equipment in the computer room, obtain the spatial normal vector of the plane where the display screen of each equipment in the computer room is located; Based on the spatial normal vectors of the planes where the display screens of all the equipment in the computer room are located, a constraint is established that the angles between the imaging plane of the image acquisition device and the plane where the display screens of any equipment in the computer room are located reach a set threshold and the average of the angles is the smallest, and the optimal solution is used as the optical axis vector of the image acquisition device; The azimuth angle required by the image acquisition device is acquired based on the optical axis vector, and the posture of the image acquisition device is adjusted based on the azimuth angle.

3. A power network monitoring method according to claim 2, characterized in that: For any equipment i in the computer room, its space normal vector n i The methods for obtaining include: Get the three non-collinear coordinate points A of the display screen of equipment i in the computer room i , B i and C i The coordinates of and Get the spatial normal vector n of equipment i in the computer room i , n i =(a i , b i , c i );in, 4. A power network monitoring method according to claim 3, characterized in that: The optical axis vector n0 and the total space normal vector n i The constraints between them are as follows: Where n0 = (j, k, l), N is the total number of equipment in the computer room, θ i is the angle between the imaging plane of the image acquisition device and the plane where the display screen of the equipment i in the computer room is located; 5. A power network monitoring method according to claim 4, characterized in that: The azimuth angles required by the image acquisition device include the angle α with the X-axis, the angle β with the Y-axis, and the angle γ with the Z-axis, where:

6. A power network monitoring method according to claim 1, characterized in that: The operation status of each equipment room device is obtained based on the processing of the image data in the monitoring video data, including: Segment the image data to obtain the operating interface image data corresponding to each equipment in the computer room; The operating interface image data of each equipment in the computer room is sent to the corresponding image feature extraction model to extract image features; wherein the image feature extraction model adopts a convolutional neural network; The extracted image features are sent to the state classification model to obtain the operating status of the corresponding equipment in the computer room; wherein the state classification model adopts a fully connected neural network.

7. A power network monitoring method according to claim 6, characterized in that: The image feature extraction model corresponding to any equipment in the computer room is trained and obtained in the following way: Construct an initial image feature extraction model; Acquire multiple groups of sample pairs and construct a sample pair library; wherein any sample pair includes orthographic projection image data and non-orthographic projection image data of the corresponding equipment in the computer room under the same operation interface; The initial image feature extraction model is trained based on the sample pair library to obtain the corresponding image feature extraction model; wherein a loss function is constructed based on the output value of the sample pair.

8. A power network monitoring method according to claim 7, characterized in that: The loss function is constructed based on the similarity of the output values ​​of the sample pairs.

9. A power network monitoring method according to claim 6, characterized in that: The state classification model is trained and obtained in the following way: An initial running state classification model is constructed, wherein the initial running state classification model has an initial feature extraction layer and an initial classification output layer connected in sequence; wherein the initial feature extraction layer has the same structure as the image feature extraction model, and the initial classification output layer adopts a fully connected neural network; Establish a sample library, the samples in the sample library include the operation interface images of each equipment in the computer room under different operation states, the operation interface images selected in each sample are all orthographic projection image data under the operation interface, and the sample label of each sample is the operation state of the corresponding operation interface image; The initial operation status classification model is trained based on the sample library to obtain the operation status classification model; the network parameters of the classification output layer in the operation status classification model are extracted; Based on the extracted network parameters of the classification output layer, a state classification model is constructed.

10. A power network monitoring system, comprising: Image acquisition equipment used to collect monitoring video data during the operation of the computer room; wherein, The monitoring video data is composed of multiple frames of continuous image data, and any image data contains the operating interface image data of the covered equipment room equipment; A classification device for obtaining the operating status of each equipment room device based on monitoring video data.

Images