Ground automatic meteorological station operation state prediction method based on multivariate data

Abstract

The invention discloses a ground automatic meteorological station operation state prediction method based on multivariate data, and relates to the technical field of meteorological monitoring, and the method comprises the steps: obtaining an operation environment image and an equipment state data sequence of a ground automatic meteorological station; performing image quality detection on the operation environment image to obtain a prediction index result of a camera state index; and performing target identification on the operation environment image by using the observation environment identification model to obtain a prediction index result of the environment target state index. And obtaining a prediction index result of the working parameters based on the equipment state data sequence by using the equipment state time sequence prediction model. And determining a prediction index result of the equipment support index. And fusing prediction index results of the camera state index, the environment target state index, the working parameters and the equipment support index at the future moment to obtain an operation state prediction result. The running state of the ground automatic weather station at the future moment is predicted, and the maintenance efficiency and the maintenance quality are improved.

Description

Technical Field

[0001] The present application relates to the field of meteorological monitoring technology, and in particular to a method for predicting the operating status of a ground automatic weather station based on multivariate data. Background Art

[0002] Automatic ground-based weather stations are a vital component of the ground-based meteorological observation network and play a key role in weather forecasting, warning, and disaster prevention and mitigation services. They provide fundamental data for weather, temperature, wind speed, and precipitation forecasts, and are crucial for preventing extreme weather events such as strong winds and heavy rain.

[0003] However, ground-based automatic weather stations are affected by various factors during operation, leading to performance degradation and failure losses. Currently, the main method for ensuring the operation of ground-based automatic weather stations is passive maintenance after a failure occurs, which affects the continuity and accuracy of meteorological monitoring. Summary of the Invention

[0004] In response to the above-mentioned problems and technical needs, this application proposes a method for predicting the operating status of a ground-based automatic weather station based on multivariate data. The technical solution of this application is as follows:

[0005] A method for predicting the operating status of a ground automatic weather station based on multivariate data comprises the following steps:

[0006] An environmental monitoring camera is used to obtain an operating environment image of the ground automatic weather station's working environment at any time t, and a device status data sequence of the ground automatic weather station at time t is obtained; wherein the device status data sequence at time t includes the device status data of the ground automatic weather station at time t and several consecutive historical times before that, and the device status data at each time includes the operating parameters of the ground automatic weather station and the observation data obtained by the ground automatic weather station;

[0007] Perform image quality testing on the operating environment image at time t, and obtain prediction results for camera status indicators at several future times after time t based on the image quality testing results. Perform target recognition on the operating environment image at time t using the observation environment recognition model, and obtain prediction results for environmental target status indicators at several future times after time t based on the target recognition results. Environmental target status indicators are used to characterize the quality of the working environment of the ground automatic weather station, and camera status indicators are used to characterize the working performance of the environmental monitoring camera.

[0008] The equipment status time series prediction model is used to obtain the prediction index results of the working parameters at several future moments after time t based on the equipment status data sequence at time t;

[0009] Determine the predicted index results of the equipment support index of the ground automatic weather station at several future moments after time t. The equipment support index is used to characterize the inherent station construction characteristics and historical operation and maintenance records of the ground automatic weather station;

[0010] The prediction index results of the camera status indicators, the environmental target status indicators, the working parameters, and the equipment support indicators at each future moment after time t are integrated to obtain the operation status prediction results of the ground automatic weather station at the future moment; the better the working performance of the environmental monitoring camera indicated by the prediction index results of the camera status indicators, the better the working environment of the ground automatic weather station indicated by the prediction index results of the environmental target status indicators, the better the working performance of the meteorological observation equipment of the ground automatic weather station indicated by the prediction index results of the working parameters, and the better the operation and maintenance conditions of the ground automatic weather station indicated by the prediction index results of the equipment support indicators, the healthier the operation status of the ground automatic weather station indicated by the obtained operation status prediction results.

[0011] A further technical solution is that the observation environment recognition model includes a target detection model and a deep learning classification model; the observation environment recognition model is used to perform target recognition on the operating environment image at time t and obtain the prediction index results of the environmental target state index, including:

[0012] The target detection model is used to identify meteorological observation equipment and abnormal targets in the operating environment image, and the change of the device position of the meteorological observation equipment is obtained based on the device position of each meteorological observation equipment;

[0013] Use deep learning classification models to detect the status of meteorological observation equipment and spider web occlusion areas in operating environment images;

[0014] The smaller the device position change of each meteorological observation device in the ground automatic weather station, the fewer abnormal targets in the working environment of the ground automatic weather station, the more normal the status of each meteorological observation device, and the smaller the spider web occlusion area in the operating environment image, the better the prediction index results of the environmental target status index indicate that the working environment of the ground automatic weather station is.

[0015] Its further technical solution is that the target detection model is constructed by introducing a visual perception module and a feature selection module between the backbone network and the head network of the YOLOv6 model. The visual perception module uses the attention mechanism to enhance the features of the input feature map and then outputs it; the feature selection module is used to weight feature maps of different scales according to different weighted weights and then output multi-scale features.

[0016] Its further technical solution is that the visual perception module includes a feature extraction layer and a channel attention module that are cascaded in sequence. The feature extraction layer is used to extract spatial features from the input feature map. The channel attention module obtains attention weights based on the spatial features. The attention weights are weighted with the input feature map to obtain a feature map with enhanced features.

[0017] Its further technical solution is that the feature selection module includes a feature processing layer and a weighted processing layer. The feature processing layer is used to perform feature processing on the feature map of each scale to obtain a feature response. The weighted processing layer is used to weight the feature response of the feature map of each scale with the weight of the feature map of the corresponding scale and then output multi-scale features. The larger the scale of the feature map, the higher the corresponding weight.

[0018] A further technical solution is to perform image quality detection on the operating environment image at time t and obtain the predicted index results of the camera status index, including:

[0019] The operating environment image at time t is filtered using multiple edge detection filters to obtain image features; all the obtained image features are input into an edge detection classification model trained based on a decision tree classification model to obtain image quality detection results. The lower the clarity of the operating environment image represented by the image quality detection result, the worse the working performance of the environmental monitoring camera is indicated by the obtained camera status index prediction index result; the higher the clarity of the operating environment image represented by the image quality detection result, the better the working performance of the environmental monitoring camera is indicated by the obtained camera status index prediction index result.

[0020] A further technical solution is to use multiple edge detection filters to filter the operating environment image at time t to obtain image features including:

[0021] Each edge detection filter is used to filter the global image and the central local area image of the operating environment image at time t to obtain image features. The edge detection filters used include Sobel filter, Canny filter and Laplacian filter.

[0022] Its further technical solution is that the observation data include temperature, humidity, air pressure, wind speed, and precipitation; the working parameters include mainboard temperature, mainboard voltage, signal strength, signal stability, and signal transmission rate. The more normal the mainboard temperature, the more normal the mainboard voltage, the higher the signal strength, the higher the signal stability, and the faster the signal transmission rate in the prediction index results of the working parameters, the better the working performance of the meteorological observation equipment of the ground automatic weather station indicated.

[0023] Its further technical solution is that the inherent site construction characteristics include site construction time, the distance between the site construction location and the support center, the difficulty of traveling from the site construction location to the support center, personnel on-duty status, spare parts support status, site safety conditions, and equipment installation conditions; historical operation and maintenance records include the number of failures that have occurred, the longest single failure time, and the time since the last inspection; among them, the shorter the site construction time, the shorter the distance between the site construction location and the support center, the easier the travel from the site construction location to the support center, the more on-duty personnel, the more spare parts support, the safer the site, the more favorable the equipment installation conditions, the fewer failures that have occurred, the shorter the longest single failure time, and the shorter the time since the last inspection in the prediction indicator results of the equipment support index, the better the operation and maintenance conditions of the ground automatic weather station indicated.

[0024] A further technical solution is to obtain the predicted operating status of the ground automatic weather station at each future moment, including:

[0025] Obtain the prediction score corresponding to the prediction index result of the camera status index at the future moment, obtain the prediction score corresponding to the prediction index result of the environmental target status index, obtain the prediction score corresponding to the prediction index result of the working parameter, and obtain the prediction score corresponding to the prediction index result of the equipment support index;

[0026] The health value of the ground automatic weather station at a future time is calculated by weighting the corresponding prediction scores of the camera status indicator, environmental target status indicator, working parameter, and equipment support indicator, and an early warning message is output when the health value is lower than the health value threshold;

[0027] Among them, the prediction index results of the camera status index indicate that the better the working performance of the environmental monitoring camera, the higher the prediction score corresponding to the camera status index; the prediction index results of the environmental target status index indicate that the better the working environment of the ground automatic weather station, the higher the prediction score corresponding to the environmental target status index; the prediction index results of the working parameters indicate that the better the working performance of the meteorological observation equipment of the ground automatic weather station, the higher the prediction score corresponding to the working parameters; the prediction index results of the equipment support index indicate that the better the operation and maintenance conditions of the ground automatic weather station, the higher the prediction score corresponding to the equipment support index.

[0028] The beneficial technical effects of this application are:

[0029] This application proposes a method for predicting the operating status of ground-based automatic weather stations based on multivariate data. By designing multiple prediction models based on the characteristics of multivariate data, such as operating environment images and equipment status data, this method enables the prediction of the operating status of ground-based automatic weather stations at future moments, effectively resolving the current technical challenge of being unable to predict the operating status of ground-based automatic weather stations. By fully considering the impact of multivariate data on prediction performance, the prediction indicators derived through fusion prediction further improve the accuracy and reliability of the prediction results, providing important guidance for predictive maintenance of ground-based automatic weather stations.

[0030] Through deep learning-based image detection and recognition and time series prediction methods, real-time automatic predictions are achieved in complex environments, while ensuring the accuracy of prediction results and improving prediction efficiency. By combining edge detection methods with machine learning methods, the difficulty of detecting camera lens clarity under complex weather factors is effectively solved, ensuring the image quality of the operating environment images collected by the ground-based automatic weather station, and providing a guarantee for detecting and identifying various abnormal conditions in the ground-based automatic weather station working environment.

[0031] By quantifying the predicted indicators and using the predicted scores to quantify the operating status of the ground-based automatic weather station, the health of the ground-based automatic weather station can be clearly and intuitively determined. Combined with the early warning system, timely warning information is provided, enabling predictive maintenance of the ground-based automatic weather station and effectively reducing the occurrence of failures. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a flow chart of the method for predicting the operating status of ground automatic weather stations.

[0033] Figure 2 It is the visualization result of the trained edge detection classification model.

[0034] Figure 3 This is an image of the operating environment with spider web occlusion.

[0035] Figure 4 It is the structural diagram of the target detection model.

[0036] Figure 5 This is the structural diagram of the visual perception module.

[0037] Figure 6 It is the structural diagram of the feature selection module.

[0038] Figure 7 It is the device location identification result of the meteorological observation equipment.

[0039] Figure 8 It is the recognition result of abnormal targets in the operating environment image. DETAILED DESCRIPTION

[0040] The specific implementation of this application will be further described below with reference to the accompanying drawings.

[0041] This application proposes a method for predicting the operating status of ground automatic weather stations based on multivariate data. Please refer to Figure 1 The specific steps in the flowchart are as follows:

[0042] Step 1: Collect multivariate data generated by the operation of the ground automatic weather station and perform data preprocessing such as data cleaning, outlier and missing value removal, and data standardization to obtain valid operation data. The operation data of the ground automatic weather station includes: operating environment images, equipment status data, and equipment support data. The specific method for obtaining each type of operation data is as follows:

[0043] An environmental monitoring camera is used to obtain the operating environment image of the ground automatic weather station at any time t. The environmental monitoring camera is a device installed at a specified location to capture the working environment of the ground automatic weather station. Its position is fixed, and the field of view covers the working environment of the ground automatic weather station. The shooting angle remains basically unchanged, so the operating environment image obtained in a short period of time at adjacent times remains basically unchanged.

[0044] A device status data sequence of the ground automatic weather station at any time t is obtained from the data system of the ground automatic weather station, wherein the device status data sequence at time t includes the device status data of the ground automatic weather station at time t and several consecutive historical times before it, and the device status data at each time includes the working parameters of the ground automatic weather station and the observation data obtained by the ground automatic weather station. The working parameters of the ground automatic weather station are various parameters representing the working performance of the meteorological observation equipment. In one embodiment, the working parameters of the ground automatic weather station include mainboard temperature, mainboard voltage, signal strength, signal stability, and signal transmission rate. The observation data obtained by the ground automatic weather station are various meteorological data observed by the ground automatic weather station. In one embodiment, the observation data obtained by the ground automatic weather station include temperature, humidity, air pressure, wind speed, and precipitation.

[0045] Equipment support data for a ground-based automatic weather station is obtained based on various relevant information about the station's inherent construction characteristics and historical operation and maintenance records. Some inherent construction characteristics are determined during initial station construction. In one embodiment, these include construction duration, distance from the station location to the support center, travel difficulty from the station location to the support center, personnel availability, spare parts availability, site safety conditions, and equipment installation conditions. Historical operation and maintenance records include the number of faults that have occurred, the longest single fault duration, and the time since the last inspection.

[0046] Step 2: Process the operational data at time t and predict the forecast indicators for several future time periods after time t. The various indicators used to evaluate the operational status of ground-based automatic weather stations are obtained by analyzing the influencing factors of actual operations and meteorological observation equipment using statistical analysis and hierarchical analysis methods based on various operational data.

[0047] (1) Since the state of the environmental monitoring camera directly determines the quality of the captured operating environment image, the operating environment image at time t is subjected to image quality detection, and the prediction index results of the camera state index at several future moments after time t are obtained based on the image quality detection results. The camera state index is used to characterize the working performance of the environmental monitoring camera.

[0048] The lens of an environmental monitoring camera is contaminated by rain, fog, and other weather conditions, resulting in blurry and unclear images of the operating environment. Therefore, image quality testing is required to determine the camera's status. In one embodiment, considering that lens contaminants are mostly patches with edges, the operating environment image at time t is filtered using multiple edge detection filters to obtain image features. All obtained image features are then input into an edge detection classification model trained based on a decision tree classification model to obtain image quality detection results. The lower the clarity of the operating environment image represented by the image quality detection results, the worse the performance of the environmental monitoring camera is. The higher the clarity of the operating environment image represented by the image quality detection results, the better the performance of the environmental monitoring camera is.

[0049] Since the lens contaminants of the environmental monitoring camera in the external environment are mainly concentrated in the central area of ​​the lens, when extracting the image features of the operating environment image, each edge detection filter is used to filter the global image and the central local area image of the operating environment image at time t to obtain image features. The edge detection filters used include sobel filters, canny filters and laplacian filters. Specifically, several different combinations of edge detection filters can be used according to needs. For example, 6 image features can be obtained by using sobel filters, canny filters and laplacian filters to filter the global image and the central local area image of the operating environment image at time t respectively. The size of the central local area can be customized. Among them, the specific processing process of each edge detection filter is the content of the existing technology and will not be repeated in this application.

[0050] In one embodiment, a total of 3744 operating environment images and their central local area images are filtered using Sobel filter, Canny filter and Laplacian filter to extract image features. All the image features obtained by filtering are input into a decision tree classification model with a depth of 3. The visualization results of the trained edge detection classification model are shown as follows: Figure 2 The edge detection classification model trained based on the decision tree classification model has strong interpretability, making the entire classification process clearly visible, as shown in Figure 2. Figure 2 As shown in , the classification process of the edge detection classification model trained in this example is to first determine the image features obtained by canny filtering of the global image. When the image features obtained by canny filtering are greater than 449.86, the image features obtained by sobel filtering of the global image are further determined. When the image features obtained by sobel filtering are less than or equal to 4086.11, the image features obtained by laplacian filtering of the central local area image are further determined. Finally, it is determined whether the image features obtained by laplacian filtering are greater than 334.3. Finally, the classification result of this branch is obtained. Among the results greater than 334.3, n=115 operating environment image samples are clear images.

[0051] (2) Abnormalities may occur due to the location and status of the meteorological observation equipment in the ground automatic weather station, such as the rain gauge being overturned, lost, covered by foreign objects, the rain gauge having a lid, and abnormal targets appearing in the ground automatic weather station. In addition, since the working environment of the ground automatic weather station is in the outdoor natural environment, spiders often weave webs in front of the lens of the environmental monitoring camera, causing the spider web to cover the camera lens and thus affect the quality of the captured image, especially for the image quality of night shots. The image quality is more serious. The operating environment image captured when the spider web covers the camera lens is as follows: Figure 3 As shown in the figure, spider web obstruction detection is necessary to ensure timely maintenance measures such as pest control and removal of surrounding vegetation. The operating environment image can be used to determine the location and status of meteorological observation equipment, as well as abnormal targets in the ground-based automatic weather station. The observation environment recognition model is used to identify targets in the operating environment image at time t. Based on the target recognition results, predictions for environmental target status indicators for several future moments after time t are obtained. These environmental target status indicators are used to characterize the quality of the working environment of the ground-based automatic weather station.

[0052] In one embodiment, the observation environment recognition model includes a target detection model and a deep learning classification model. The target detection model is used to identify meteorological observation equipment and abnormal targets in the operating environment image, and the change in the device position of each meteorological observation device is obtained based on the device position of each meteorological observation device. The deep learning classification model is used to detect the status of the meteorological observation equipment in the operating environment image and the spider web occlusion area in the operating environment image. The smaller the device position change of each meteorological observation device in the ground automatic weather station, the fewer abnormal targets in the working environment of the ground automatic weather station, the more normal the status of each meteorological observation device, and the smaller the spider web occlusion area in the operating environment image, the better the predicted index result of the environmental target status index indicates that the working environment of the ground automatic weather station is.

[0053] Since the installation position of the environmental observation camera is fixed, the angle of the operating environment image taken is relatively fixed, and the object size of the meteorological observation equipment that needs to be detected is also relatively fixed. At the same time, the types and numbers of meteorological observation equipment and abnormal targets that need to be detected in the ground automatic weather station are limited, and the working environment is not overly complicated. In addition, the sizes of the meteorological observation equipment and abnormal targets to be detected are generally large, and the smallest target has at least dozens to hundreds of pixels in the operating environment image. Based on the above practical application characteristics, in one embodiment, the model structure of the constructed target detection model is as follows: Figure 4 As shown in the figure, considering the fixed shooting angle and limited target types, a lightweight YOLOv6 model is selected as the basic model to ensure the model operation efficiency, and an improved YOLOv6 model is obtained by introducing a visual perception module and a feature selection module between the backbone network Backbone and the head network Head of the YOLOv6 model. The visual perception module uses the attention mechanism to enhance the features of the input feature map and then output it; the feature selection module is used to weight feature maps of different scales according to different weights and then output multi-scale features.

[0054] Based on the relatively fixed position of each meteorological observation device in the operating status image, the introduced visual perception module includes a feature extraction layer and a channel attention module. For the specific structure, please refer to Figure 5. The feature extraction layer is used to extract spatial features from the input feature map. The feature extraction layer includes a convolutional layer, a BatchNorm layer, a Relu activation function layer, and a convolutional layer that are cascaded in sequence. The first convolutional layer uses a 3×3 convolution kernel to extract local features of the input feature map and capture the spatial layout pattern of each meteorological observation device; the BatchNorm layer is used to normalize the extracted local features; the Relu activation function layer is used to increase the nonlinearity of local features and improve the generalization performance of the neural network; the second convolutional layer uses a 1×1 convolution kernel to map the extracted local features back to the original channel to obtain the spatial features of the input feature map.

[0055] The channel attention module derives attention weights based on spatial features. It consists of a cascade of pooling layers, downsampling layers, Relu activation layers, upsampling layers, and Sigmoid activation layers. The pooling layer uses an average pooling layer to compress the dimensionality of the input spatial features. The downsampling layer performs dimensionality reduction based on the compression achieved by the average pooling layer. The Relu activation layer adds nonlinearity to the input spatial features. The upsampling layer increases the dimensionality of the features processed by the Relu activation layer to maintain the original data format. Both the downsampling and upsampling layers use convolution layers with a 1×1 kernel. The Sigmoid activation layer generates attention weights between 0 and 1.

[0056] Finally, the attention weights are weighted and added to the input feature map to enhance the feature responses at important locations, resulting in an enhanced feature map. For example, a rain gauge on a ground-based automatic weather station always appears in the lower right corner of the operating environment image, so the visual perception module learns to pay special attention to the features in this area.

[0057] Based on the characteristics that the size of the target to be detected in the operating environment image is relatively fixed and large, the introduced feature selection module includes a feature processing layer and a weighted processing layer. For the specific structure, please refer to Figure 6 The feature processing layer is used to perform feature processing on the feature map of each scale to obtain feature responses. The weighted processing layer is used to weight the feature responses of the feature map of each scale with the weight of the feature map of the corresponding scale and output multi-scale features. The larger the scale of the feature map, the higher the corresponding weight.

[0058] The deep learning classification model in the observation environment recognition model uses existing methods. First, a meteorological observation device location recognition algorithm is used to determine the location of the meteorological observation device. Then, an image containing the meteorological observation device is cut and input into a trained meteorological observation device classification model to output the detected meteorological observation device status. The location recognition algorithm can use any conventional location recognition method, and the specific details are not detailed in this application.

[0059] The results of using the target detection model of this application to identify the location of meteorological observation equipment in the operating environment image are as follows: Figure 7 As shown in the figure, the results of using the target detection model to identify abnormal targets in the operating environment image are as follows: Figure 8 As shown in the figure, two abnormal human targets appearing in the working environment of various meteorological observation equipment and ground automatic weather stations are identified in the boxes.

[0060] (3) Based on the time series characteristics of the working parameters and their correlation with time, the device status time series prediction model is used to obtain the prediction index results of the working parameters at several future moments after time t based on the device status data sequence at time t. The more normal the motherboard temperature, the more normal the motherboard voltage, the higher the signal strength, the higher the signal stability, and the faster the signal transmission rate in the prediction index results of the working parameters, the better the working performance of the meteorological observation equipment of the ground automatic weather station indicated. Among them, the more the motherboard temperature / voltage conforms to the normal motherboard temperature range, the more normal the motherboard temperature / voltage is; the smaller the change in signal strength, the higher the signal stability.

[0061] Since the observation data obtained by the ground automatic weather station will affect the prediction index results of the working parameters of the ground automatic weather station, for example, the motherboard temperature will be affected by the temperature, and the motherboard temperature will also increase under high temperature conditions. Therefore, an array consisting of the observation data obtained by the ground automatic weather station and the working parameters of the ground automatic weather station is used as input, and the working parameters of the ground automatic weather station are used as output to train a deep learning timing prediction model to obtain a device status timing prediction model. The deep learning timing prediction model adopts a conventional timing prediction model such as a long short-term memory network LSTM or a gated recurrent unit GRU. The deep learning timing prediction model is a prior art and will not be described in detail in this application.

[0062] (4) Some information about the inherent station construction characteristics can be determined at the initial station construction moment, and the historical operation and maintenance records can also be directly calculated based on time. Therefore, the prediction index results of the equipment support index of the ground automatic weather station at several future moments after time t can be determined. The equipment support index is used to characterize the inherent station construction characteristics and historical operation and maintenance records of the ground automatic weather station. Among them, the shorter the station construction time, the shorter the distance between the station construction location and the support center, the easier the travel distance between the station construction location and the support center, the more on-duty personnel, the more spare parts support, the safer the site, the more favorable the equipment installation conditions, the fewer the number of faults that have occurred, the shorter the longest single fault time, and the shorter the time since the last inspection, the better the operation and maintenance conditions of the ground automatic weather station indicated.

[0063] Step 3: Fuse the predicted index results of the camera status index, the environmental target status index, the working parameters, and the equipment support index for each future moment after time t to obtain the predicted operation status result of the ground automatic weather station at the future moment. The better the predicted index result of the camera status index indicates the working performance of the environmental monitoring camera, the better the predicted index result of the environmental target status index indicates the working environment of the ground automatic weather station, the better the predicted index result of the working parameters indicates the working performance of the meteorological observation equipment of the ground automatic weather station, and the better the predicted index result of the equipment support index indicates the operation and maintenance conditions of the ground automatic weather station. The obtained predicted operation status result indicates that the operation status of the ground automatic weather station is healthier.

[0064] In one embodiment, score each predicted index result according to a pre-established index scoring table to obtain the predicted score corresponding to the predicted index result of the camera status index at the future moment, the predicted score corresponding to the predicted index result of the environmental target status index, the predicted score corresponding to the predicted index result of the working parameters, and the predicted score corresponding to the predicted index result of the equipment support index. Among them, the index scoring table is constructed based on the results of expert rating and scoring according to the influence degree of each index on the actual business and meteorological observation equipment. For example, for the single longest fault time of the equipment support index in the index scoring table, if the single longest fault time is 0 hours, the score is 1 point; if 0 hours < h ≤ 12 hours, the score is 0.75 points; if 12 hours < h ≤ 24 hours, the score is 0.5 points; if 24 hours < h ≤ 36 hours, the score is 0.25 points; if h ≥ 36 hours, the score is 0 points.

[0065] Calculate the health value of the ground automatic weather station at the future moment by weighted calculation of the predicted scores corresponding to the camera status index, the environmental target status index, the working parameters, and the equipment support index respectively, and output a warning message when the health value is lower than the health value threshold. Among them, the better the predicted index result of the camera status index indicates the working performance of the environmental monitoring camera, the higher the predicted score corresponding to the camera status index; the better the predicted index result of the environmental target status index indicates the working environment of the ground automatic weather station, the higher the predicted score corresponding to the environmental target status index; the better the predicted index result of the working parameters indicates the working performance of the meteorological observation equipment of the ground automatic weather station, the higher the predicted score corresponding to the working parameters; the better the predicted index result of the equipment support index indicates the operation and maintenance conditions of the ground automatic weather station, the higher the predicted score corresponding to the equipment support index.

[0066] Warning information can take a variety of different forms of warning methods. At the same time, you can customize different levels of health value thresholds and set different degrees of warning reminders based on each health value threshold. For example, use light prompts. The lower the health value, the faster the light flashes.

[0067] The above description is only a preferred embodiment of the present application, and the present application is not limited to the above embodiments. It is understood that other improvements and variations directly derived or imagined by those skilled in the art without departing from the spirit and concept of the present application should be considered to be included in the scope of protection of the present application.

Claims

1. A method for predicting the operating status of a ground automatic weather station based on multivariate data, characterized in that: The method for predicting the operating status of the ground automatic weather station comprises: Using an environmental monitoring camera to obtain an operating environment image of the working environment of the ground automatic weather station at any time t, and obtain a device status data sequence of the ground automatic weather station at time t; wherein the device status data sequence at time t includes device status data of the ground automatic weather station at time t and several consecutive historical times before and before, and the device status data at each time includes the working parameters of the ground automatic weather station and the observation data obtained by the ground automatic weather station; Perform image quality detection on the operating environment image at time t, and obtain the prediction index results of the camera state index at several future moments after time t based on the image quality detection results; perform target recognition on the operating environment image at time t using the observation environment recognition model, and obtain the prediction index results of the environmental target state index at several future moments after time t based on the target recognition results; the environmental target state index is used to characterize the quality of the working environment of the ground automatic weather station, and the camera state index is used to characterize the working performance of the environmental monitoring camera; Using the equipment status time series prediction model to obtain the prediction index results of the working parameters at several future moments after time t based on the equipment status data sequence at time t; Determine the prediction index results of the equipment support index of the ground automatic weather station at several future moments after time t, where the equipment support index is used to characterize the inherent station construction characteristics and historical operation and maintenance records of the ground automatic weather station; The prediction index results of the camera state index, the prediction index results of the environmental target state index, the prediction index results of the working parameters, and the prediction index results of the equipment support index at each future moment after time t are integrated to obtain the prediction result of the operation state of the ground automatic weather station at the future moment; the prediction index results of the camera state index indicate that the better the working performance of the environmental monitoring camera, the better the working environment of the ground automatic weather station indicated by the prediction index results of the environmental target state index, the better the working performance of the meteorological observation equipment of the ground automatic weather station indicated by the prediction index results of the working parameters, and the better the operation and maintenance conditions of the ground automatic weather station indicated by the prediction index results of the equipment support index, the obtained operation state prediction result indicates that the operation state of the ground automatic weather station is healthier.

2. The method for predicting the operating status of a ground automatic weather station according to claim 1, characterized in that: The observation environment recognition model includes a target detection model and a deep learning classification model; The observation environment recognition model is used to identify the target of the operating environment image at time t and obtain the prediction index results of the environmental target state index, including: Using the target detection model to identify meteorological observation equipment and abnormal targets in the operating environment image, and obtaining the change of the device position of each meteorological observation device according to the device position of the meteorological observation device; Use deep learning classification models to detect the status of meteorological observation equipment in operating environment images and spider web occlusion areas in operating environment images; The smaller the equipment position change of each meteorological observation device in the ground automatic weather station, the fewer abnormal targets in the working environment of the ground automatic weather station, the more normal the status of each meteorological observation device, and the smaller the spider web occlusion area in the operating environment image, the predicted index results of the environmental target state index indicate that the working environment of the ground automatic weather station is better.

3. The method for predicting the operating status of a ground automatic weather station according to claim 2, characterized in that: The target detection model is constructed by introducing a visual perception module and a feature selection module between the backbone network and the head network of the YOLOv6 model. The visual perception module uses an attention mechanism to enhance the features of the input feature map and then outputs it; the feature selection module is used to weight feature maps of different scales according to different weighted weights and then output multi-scale features.

4. The method for predicting the operating status of a ground automatic weather station according to claim 3, characterized in that: The visual perception module includes a feature extraction layer and a channel attention module cascaded in sequence. The feature extraction layer is used to extract spatial features from an input feature map. The channel attention module obtains attention weights based on the spatial features. The attention weights are weighted with the input feature map to obtain a feature map with enhanced features.

5. The method for predicting the operating status of a ground automatic weather station according to claim 3, characterized in that: The feature selection module includes a feature processing layer and a weighted processing layer. The feature processing layer is used to perform feature processing on the feature map of each scale to obtain a feature response. The weighted processing layer is used to weight the feature response of the feature map of each scale with the weight of the feature map of the corresponding scale and then output a multi-scale feature. The larger the scale of the feature map, the higher the corresponding weight.

6. The method for predicting the operating status of a ground automatic weather station according to claim 1, characterized in that: The image quality of the operating environment image at time t is detected and the prediction index results of the camera status index are obtained, including: The operating environment image at time t is filtered using a variety of edge detection filters to obtain image features; all the obtained image features are input into an edge detection classification model trained based on a decision tree classification model to obtain image quality detection results. The lower the clarity of the operating environment image represented by the image quality detection result, the worse the working performance of the environment monitoring camera is indicated by the prediction index result of the camera status index; the higher the clarity of the operating environment image represented by the image quality detection result, the better the working performance of the environment monitoring camera is indicated by the prediction index result of the camera status index.

7. The method for predicting the operating status of a ground automatic weather station according to claim 6, characterized in that: The filtering process of the operating environment image at time t using a plurality of edge detection filters to obtain image features includes: Each edge detection filter is used to filter the global image and the central local area image of the operating environment image at time t to obtain image features. The edge detection filters used include sobel filter, canny filter and laplacian filter.

8. The method for predicting the operating status of a ground automatic weather station according to claim 1, characterized in that: The observation data include temperature, humidity, air pressure, wind speed, and precipitation; the working parameters include mainboard temperature, mainboard voltage, signal strength, signal stability, and signal transmission rate. The more normal the mainboard temperature, the more normal the mainboard voltage, the higher the signal strength, the higher the signal stability, and the faster the signal transmission rate in the prediction index results of the working parameters, the better the working performance of the meteorological observation equipment of the ground automatic weather station indicated.

9. The method for predicting the operating status of a ground automatic weather station according to claim 1, characterized in that: The inherent site construction characteristics include site construction time, the distance between the site construction location and the support center, the difficulty of traveling from the site construction location to the support center, personnel on duty, spare parts support, site safety conditions, and equipment installation conditions; historical operation and maintenance records include the number of failures that have occurred, the longest single failure time, and the time since the last inspection; among them, the shorter the site construction time, the shorter the distance between the site construction location and the support center, the easier the travel from the site construction location to the support center, the more on-duty personnel, the more spare parts support, the safer the site, the more favorable the equipment installation conditions, the fewer failures that have occurred, the shorter the longest single failure time, and the shorter the time since the last inspection in the prediction indicator results of the equipment support index, the better the operation and maintenance conditions of the ground automatic weather station indicated.

10. The method for predicting the operating status of a ground automatic weather station according to claim 1, characterized in that: The obtaining of the prediction result of the operation status of the ground automatic weather station at each future moment includes: Obtain the prediction score corresponding to the prediction index result of the camera state index at the future moment, obtain the prediction score corresponding to the prediction index result of the environmental target state index, obtain the prediction score corresponding to the prediction index result of the working parameter, and obtain the prediction score corresponding to the prediction index result of the equipment support index; The predicted scores corresponding to the camera status index, the environmental target status index, the working parameter, and the equipment support index are weighted and calculated to obtain the health value of the ground automatic weather station at the future time, and output warning information when the health value is lower than the health value threshold; Among them, the prediction index results of the camera status indicators indicate that the better the working performance of the environmental monitoring camera, the higher the prediction score corresponding to the camera status indicators; the prediction index results of the environmental target status indicators indicate that the better the working environment of the ground automatic weather station, the higher the prediction score corresponding to the environmental target status indicators; the prediction index results of the working parameters indicate that the better the working performance of the meteorological observation equipment of the ground automatic weather station, the higher the prediction score corresponding to the working parameters; the prediction index results of the equipment support indicators indicate that the better the operation and maintenance conditions of the ground automatic weather station, the higher the prediction score corresponding to the equipment support indicators.

Images