System intelligent image recognition technical method for low-temperature congelation weather of power grid

Abstract

The invention relates to the technical field of power grid low-temperature congelation weather system intelligent graph identification, in particular to a power grid low-temperature congelation weather system intelligent graph identification technical method, which comprises the steps of collecting data; carrying out data annotation processing; analyzing weather influence characteristics of low-temperature congelation events; according to the date of the event, the sea level air pressure type and the high altitude potential height type are matched with the event, and a high altitude and low altitude configuration weather combination of each low-temperature congelation event is obtained; drawing a gray contour map by using a Python language; forming a sea level air pressure field sample data set by the sea level air pressure field pictures; on the basis of a Pytorch framework, lightweight convolutional neural network GhostNet image recognition models are respectively constructed for two sample data sets, so that the technology can be applied to a mobile terminal and embedded equipment, and functionality and multi-scene expansion application are considered; the capsule network is adopted to improve the defects of the lightweight convolutional neural network, so that the model obtains richer and more comprehensive spatial information, and the accuracy of the model is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent map recognition for low-temperature freezing weather systems in power grids, and in particular to a technical method for intelligent map recognition for low-temperature freezing weather systems in power grids. Background Art

[0002] The power grid consists of four main links: power generation, transmission, distribution and consumption. It includes transmission and distribution networks. The function of the transmission network is to transmit the electricity generated by power plants to areas where electricity is consumed, or to transmit electricity between adjacent power grids to form an interconnected power grid. The function of the distribution network is to receive the electricity transmitted by the transmission network, and then redistribute it and transmit it to cities and rural areas, and further distribute and supply it to industry, agriculture, commerce, residents and electricity users with special needs.

[0003] Among the extreme weather conditions that affect the power grid, low temperature condensation and heavy rainfall account for 80% of the impact. Ice on wires in low temperature condensation weather can cause major electrical accidents such as disconnection, flashover, tripping and even tower collapse. Summary of the invention

[0004] In view of the above problems or problems existing in the prior art, the present invention is proposed.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0006] In a first aspect, the present invention provides the following technical solutions: a technical method for intelligent image recognition of a low-temperature freezing weather system of a power grid, comprising the following steps;

[0007] Including the following steps;

[0008] Collect daily ERA5 meteorological data for the same period;

[0009] Conduct data annotation and processing on the daily ERA5 meteorological data during the same period;

[0010] The weather characteristics of ERA5 meteorology and low temperature freezing events during the same period were classified;

[0011] According to the date of the event, the classified weather characteristics are matched with the event;

[0012] Draw a grayscale contour map through the above steps;

[0013] The sea level pressure field images are combined into a sea level pressure field sample data set;

[0014] Construct lightweight convolutional neural network and GhostNet image recognition model for the dataset respectively;

[0015] Use capsule network to improve GhostNet image recognition model;

[0016] The contour image of the new date is input into the corresponding image recognition model, and the corresponding weather type is output through the GhostNet image recognition model.

[0017] As a preferred solution of the intelligent map recognition technology method for the low-temperature freezing weather system of the power grid of the present invention, the ERA5 meteorological data includes setting the proposed study area, collecting the cases of low-temperature freezing events of the power grid in the proposed study area in the past 10 years through tension sensor observation or artificial ice observation; downloading the daily ERA5 meteorological reanalysis data of the same period, and the elements of the analysis data include sea level pressure and high-altitude potential height.

[0018] As a preferred solution of the technical method for intelligent mapping of low-temperature freezing weather system of power grid of the present invention, data annotation processing includes annotating ERA5 meteorological reanalysis data based on the date of low-temperature freezing event case of power grid, wherein ERA5 meteorology includes "weather situation when low-temperature freezing event occurs" and "weather situation when low-temperature freezing event does not occur".

[0019] As a preferred solution of the present invention for the intelligent map recognition technology method of the low-temperature freezing weather system of the power grid, wherein: weather feature classification, that is, using objective weather classification software to perform PCT objective classification, including analyzing the sea level pressure type and the high altitude potential height type.

[0020] As a preferred solution of the intelligent map recognition technology method for the low-temperature freezing weather system of the power grid of the present invention, the weather characteristics and event matching content includes matching the sea level pressure type, the high-altitude potential height type and the event to obtain the high-altitude and low-altitude configuration weather combination for each low-temperature freezing event.

[0021] As a preferred solution of the intelligent map recognition technology method for the low-temperature freezing weather system of the power grid of the present invention, the grayscale contour map includes drawing grayscale contour maps according to various types of sea level pressure and various types of high-altitude potential heights using Python language and low-temperature freezing event meteorological data sets; selecting meteorological data sets in which no low-temperature freezing events occurred, and drawing grayscale contour maps of sea level pressure and high-altitude potential heights.

[0022] As a preferred solution of the present invention for the intelligent image recognition technology method of the low-temperature freezing weather system of the power grid, wherein: the sea level pressure field sample data set includes, the sea level pressure field sample data set is composed of various types of sea level pressure field pictures where low-temperature freezing events occur and sea level pressure field pictures where low-temperature freezing events do not occur, and the number of each type of pictures in the data set remains consistent.

[0023] As a preferred solution of the intelligent image recognition technology method for the low-temperature freezing weather system of the power grid of the present invention, wherein: the lightweight convolutional neural network and the GhostNet image recognition model both include an input layer, a convolutional layer, an average pooling layer, and a fully connected layer;

[0024] The GhostNet image recognition model also includes the GhostNet network structure.

[0025] As a preferred solution of the present invention for the intelligent image recognition technology method of the low-temperature freezing weather system of the power grid, the capsule network improvement method includes replacing the average pooling layer with the capsule network, and training modeling and parameter tuning and finalization of the capsule network to form a GhostNet image recognition model.

[0026] As a preferred solution of the intelligent map recognition technology method for the low-temperature freezing weather system of the power grid of the present invention, the weather type includes identifying the sea level pressure field image through the GhostNet image recognition model to obtain the "low pressure control type", and identifying the high-altitude potential height contour line image through the GhostNet image recognition model to obtain the "south branch trough type".

[0027] The beneficial effects of the present invention are as follows: the whole process from weather classification of low-temperature freezing events in power grids to image recognition is objective, efficient and accurate, which is helpful to realize the popularization and application of high-threshold meteorological professional knowledge in the power grid security industry, and can assist personnel without professional meteorological knowledge to complete the rapid identification and timely response to the weather affected by low-temperature freezing events in power grids; general weather classification research only considers a single layer on the ground or at high altitude, and this patent adopts a new double-layer weather classification method, taking into account both high-altitude and ground configurations, which is more in line with current actual meteorological business operations; image recognition adopts a lightweight convolutional neural network, so that the technology can be applied to mobile terminals and embedded devices, taking into account functionality and multi-scenario expansion applications; the capsule network is used to improve the defects of the lightweight convolutional neural network, so that the model can obtain richer and more comprehensive spatial information, thereby improving the accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0029] Figure 1 The present invention is an overall flow chart of the intelligent image recognition technology method for the low temperature and freezing weather system of the power grid.

[0030] Figure 2This is a diagram of the GhostNet structure of the intelligent image recognition technology method for the low-temperature freezing weather system of the power grid.

[0031] Figure 3 This is the structure diagram of GhostModule.

[0032] Figure 4 To improve the GhostNet model structure diagram. DETAILED DESCRIPTION

[0033] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0034] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0035] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0036] Example 1

[0037] Reference Figures 1 to 4 , which is the first embodiment of the present invention, and which provides a technical method for intelligent image recognition of a low-temperature freezing weather system of a power grid, comprising the following steps;

[0038] Collect daily ERA5 meteorological data for the same period;

[0039] Conduct data annotation and processing on the daily ERA5 meteorological data during the same period;

[0040] The weather characteristics of ERA5 meteorology and low temperature freezing events during the same period were classified;

[0041] According to the date of the event, the classified weather characteristics are matched with the event;

[0042] Draw a grayscale contour map through the above steps;

[0043] The sea level pressure field images are combined into a sea level pressure field sample data set;

[0044] Construct lightweight convolutional neural network and GhostNet image recognition model for the dataset respectively;

[0045] Use capsule network to improve GhostNet image recognition model;

[0046] The contour image of the new date is input into the corresponding image recognition model, and the corresponding weather type is output through the GhostNet image recognition model.

[0047] Furthermore, the ERA5 meteorological data includes setting the proposed study area, collecting cases of power grid low-temperature freezing events in the proposed study area over the past 10 years through tension sensor observation or artificial ice observation; downloading the daily ERA5 meteorological reanalysis data for the same period, and analyzing the data elements including sea level pressure and high-altitude potential height.

[0048] Furthermore, the data annotation processing includes annotating the ERA5 meteorological reanalysis data based on the date of the power grid low-temperature freezing event case, wherein the ERA5 meteorological includes “the weather situation when the low-temperature freezing event occurred” and “the weather situation when the low-temperature freezing event did not occur”.

[0049] It should be noted that, taking the sea level pressure classification as an example, the analysis results in the ground circulation situation: low pressure front type, uniform pressure field type, low pressure control type, high pressure rear type... Similarly, the high-altitude potential height can be divided into several types.

[0050] Furthermore, weather characteristic classification is carried out by using objective weather classification software to conduct PCT objective classification, including analysis of sea level pressure type and high altitude potential height type.

[0051] Furthermore, the matching of weather characteristics and events includes matching the sea level pressure type, the high altitude geopotential height type and the event, and obtaining the high altitude and low altitude configuration weather combination of each low temperature freezing event.

[0052] It should be noted that, combined with the nature of low-temperature freezing events, further analysis was conducted to obtain the nature and proportion of low-temperature freezing events caused by each high- and low-altitude weather combination, and using the ice thickness as the event nature, we obtained conclusions such as "80% of the ice events under the control of the southern trough + low pressure have deeper ice thickness".

[0053] Furthermore, the grayscale contour map includes drawing grayscale contour maps according to various types of sea level pressure and various types of upper air potential height using Python language and the meteorological data set of low temperature condensation events; and drawing grayscale contour maps of sea level pressure and upper air potential height by selecting meteorological data sets where no low temperature condensation events occurred.

[0054] Furthermore, the sea level pressure field sample data set includes a sea level pressure field sample data set composed of various types of sea level pressure field images where low temperature freezing events occurred and sea level pressure field images where low temperature freezing events did not occur, and the number of each type of images in the data set remains consistent.

[0055] It should be noted that the sample data set is divided into a training set and a test set in a ratio of 8:2. Similarly, the sample data set of the upper-air potential height field is composed and divided.

[0056] Furthermore, both the lightweight convolutional neural network and the GhostNet image recognition model include an input layer, a convolutional layer, an average pooling layer, and a fully connected layer;

[0057] The GhostNet image recognition model also includes the GhostNet network structure.

[0058] It should be noted that if Figure 2 As shown in the figure, the GhostModule in the Ghost bottleneck structure is divided into two operations. First, a convolution operation is performed through the input feature layer to obtain m original feature maps T. The second part maps and linearly transforms the original feature maps T respectively, stacks the mapping object m and the linear transformation object, and finally obtains the output feature, where the number of channels of the input feature is consistent with that of the output feature.

[0059] Furthermore, the capsule network is improved by replacing the average pooling layer with a capsule network, and training and modeling the capsule network and performing parameter tuning to form a GhostNet image recognition model.

[0060] Among them, the steps to improve GhostNet are as follows:

[0061] The size of the grayscale image to be inputted is uniformly processed into a matrix of 224×224×1.

[0062] Preferably, through a convolution operation of a convolution layer, a convolution kernel of size 3×3 is used, the sliding step is set to 2, the number of convolution kernels is 16, the convolution kernel slides on the image, the image features are extracted, and a 112×112×16 feature map is obtained.

[0063] Through 16 times of GhostBottleneck, this process uses bottlenecks with stacked stride=1 and stride=2, inputs 112×112×16 feature maps, and outputs 7×7×160 feature maps.

[0064] Through a 1×1 ordinary convolution operation, a 7×7×960 feature map is output.

[0065] Preferably, the 7×7×960 feature map is capsule-encoded and input into the capsule network. Through 8 sets of parallel convolution operations, it is converted into 8 sets of 32×3×3 capsules, and 288 capsules can be obtained.

[0066] A dynamic routing algorithm is used to perform dynamic routing between layers through 288 capsules, and the digital capsule layer is mapped, which is composed of a 14×16 vector.

[0067] Preferably, the maximum norm of the 14×16 vector is calculated to output a unique prediction category, for example, the south branch trough type.

[0068] The improved GhostNet was used for training and modeling based on the training set, and the parameters were optimized based on the test set before finalization, resulting in the sea level pressure field image recognition model and the high altitude potential height field image recognition model for low temperature freezing events.

[0069] Furthermore, the weather type includes identifying the sea level pressure field image through the GhostNet image recognition model to obtain the "low pressure controlled type", and identifying the high-altitude potential height contour image through the GhostNet image recognition model to obtain the "south branch trough type".

[0070] It should be noted that by matching and comparing this combination of high and low altitude weather types with the analysis results of matching the sea level pressure type, high altitude geopotential height type and the event according to the date of the event, it is possible to determine whether a low temperature freezing event will occur in the high and low altitude situation on that day, as well as the nature of the low temperature freezing event. It is judged that a low temperature freezing event will occur in the "Southern Branch Trough + Low Pressure Control Type", and 90% of it will be rime.

[0071] In summary, the beneficial effects of the present invention are as follows: the whole process from weather classification of low-temperature freezing events in power grids to image recognition is objective, efficient and accurate, which is helpful to realize the popularization and application of high-threshold meteorological professional knowledge in the power grid security industry, and can assist personnel without professional meteorological knowledge to complete the rapid identification and timely response to the weather affected by low-temperature freezing events in power grids; general weather classification research only considers a single layer on the ground or at high altitude, and this patent adopts a new double-layer weather classification method, taking into account both high-altitude and ground configurations, which is more in line with current actual meteorological business operations; image recognition adopts a lightweight convolutional neural network, so that the technology can be applied to mobile terminals and embedded devices, taking into account functionality and multi-scenario expansion applications; the capsule network is used to improve the defects of the lightweight convolutional neural network, so that the model can obtain richer and more comprehensive spatial information, and improve the accuracy of the model.

[0072] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for intelligent image recognition in a low-temperature freezing weather system of a power grid, characterized in that: Including the following steps; Collect daily ERA5 meteorological data for the same period; Data annotation and processing of ERA5 meteorological data for the same period; The weather characteristics of ERA5 meteorology and low temperature freezing events during the same period were classified; According to the date of the event, the classified weather characteristics are matched with the event; Draw a grayscale contour map through the above steps; The sea level pressure field images are combined into a sea level pressure field sample data set; Construct lightweight convolutional neural network and GhostNet image recognition model for the dataset respectively; Use capsule network to improve GhostNet image recognition model; The contour image of the new date is input into the corresponding image recognition model, and the corresponding weather type is output through the GhostNet image recognition model.

2. The intelligent image recognition technology method for low temperature freezing weather system of power grid according to claim 1, characterized in that: The ERA5 meteorological data includes setting the proposed study area, collecting cases of power grid low-temperature freezing events in the proposed study area over the past 10 years through tension sensor observation or artificial ice observation; downloading daily ERA5 meteorological reanalysis data for the same period, and analyzing data elements including sea level pressure and high-altitude potential height.

3. The intelligent image recognition technology method for low temperature freezing weather system of power grid as claimed in claim 2 is characterized by: The data labeling process includes labeling the ERA5 meteorological reanalysis data based on the date of the power grid low-temperature freezing event case, wherein the ERA5 meteorology includes "weather conditions when low-temperature freezing events occur" and "weather conditions when low-temperature freezing events do not occur".

4. The intelligent image recognition technology method for low temperature freezing weather system of power grid as claimed in claim 3 is characterized by: The weather characteristic classification is to use objective weather classification software to perform PCT objective classification, including analysis of sea level pressure type and high altitude potential height type.

5. The intelligent image recognition technology method for low temperature freezing weather system of power grid as claimed in claim 4, characterized in that: The weather feature and event matching content includes matching the sea level pressure type, the high altitude potential height type and the event to obtain the high altitude and low altitude configuration weather combination for each low temperature freezing event.

6. The intelligent image recognition technology method for low temperature and freezing weather system of power grid according to claim 5, characterized in that: The grayscale contour map includes drawing a grayscale contour map according to various types of sea level pressure and various types of high-altitude potential height using Python language and a low-temperature freezing event meteorological data set; Meteorological data sets without low-temperature freezing events were selected to draw grayscale contour maps of sea level pressure and high-altitude geopotential height.

7. The intelligent image recognition technology method for low temperature freezing weather system of power grid according to claim 6, characterized in that: The sea level pressure field sample data set includes sea level pressure field sample data sets composed of various types of sea level pressure field images where low temperature freezing events occurred and sea level pressure field images where low temperature freezing events did not occur, and the number of each type of images in the data set remains consistent.

8. The intelligent image recognition technology method for low temperature freezing weather system of power grid as claimed in claim 7, characterized in that: The lightweight convolutional neural network and GhostNet image recognition model both include an input layer, a convolutional layer, an average pooling layer, and a fully connected layer; The GhostNet image recognition model also includes the GhostNet network structure.

9. The intelligent image recognition technology method for low temperature freezing weather system of power grid according to claim 8, characterized in that: The capsule network improvement method includes replacing the average pooling layer with a capsule network, and forming a GhostNet image recognition model through training modeling and parameter tuning of the capsule network.

10. A method for using the intelligent image recognition technology method for a power grid low temperature freezing weather system as claimed in any one of claims 7 to 9, characterized in that: The weather type includes identifying the sea level pressure field image through the GhostNet image recognition model to obtain the "low pressure control type", and identifying the high-altitude potential height contour line image through the GhostNet image recognition model to obtain the "south branch trough type".

Images