Icing prediction model training method and power transmission line icing prediction method and device

Abstract

The invention provides an icing prediction model training method and a power transmission line icing prediction method and device, and relates to the technical field of power grid safety. The icing prediction model training method comprises the following steps: acquiring icing data including micrometeorological data, topographic data and icing monitoring data of a power transmission line at continuous moments; performing feature extraction on the icing data to obtain time sequence feature data; inputting the time sequence characteristic data into a residual network to obtain a spatial implicit function relationship among the micrometeorological characteristic, the topographic characteristic and the icing monitoring characteristic; on the basis of the time sequence feature data and a time sequence implicit function relation obtained according to the space implicit function relation and the time sequence feature data, feature fusion processing is conducted through an EMA attention mechanism, and target time sequence feature data is obtained; and adjusting parameters of the icing prediction model based on the reference icing thickness corresponding to the to-be-predicted moment and the predicted icing thickness obtained based on the target time sequence characteristic data to obtain the icing prediction model with better universality.

Description

Technical Field

[0001] This application relates to the technical field of power grid security, and particularly to a method for training an icing prediction model, a method and device for predicting icing on transmission lines. Background Art

[0002] Icing refers to the ice layer formed by the freezing of moisture on transmission lines due to low temperature and humid weather conditions. During power transmission, icing on the surface of transmission lines is likely to cause various safety accidents. By predicting the growth trend of icing on transmission lines, measures can be taken to reduce the risks brought by icing.

[0003] Currently, the icing thickness of transmission lines is usually predicted by establishing a mathematical and physical model of the relationship between micro-meteorological factors and icing growth, and combining the autocorrelation of icing growth. However, in some scenarios, predicting the icing thickness of transmission lines in the above way has the problem of poor generality. Summary of the Invention

[0004] This application provides a method for training an icing prediction model, a method and device for predicting icing on transmission lines, so as to solve the problem of poor generality in predicting the icing thickness of transmission lines by the current method.

[0005] In a first aspect, this application provides a method for training an icing prediction model. The icing prediction model includes a residual network. The method for training the icing prediction model includes:

[0006] Obtain the icing data of the transmission line at consecutive moments. The icing data includes micro-meteorological data, terrain data, and icing monitoring data;

[0007] Extract features from the icing data to obtain time-series feature data. The time-series feature data includes micro-meteorological features, terrain features, and icing monitoring features at consecutive moments;

[0008] Input the time-series feature data into the residual network to obtain the spatial implicit function relationship between the micro-meteorological features, terrain features, and icing monitoring features output by the residual network;

[0009] Based on the time-series feature data and the time-series implicit function relationship, perform feature fusion processing using the exponential moving average (EMA) attention mechanism to obtain target time-series feature data. The time-series implicit function relationship is obtained based on the spatial implicit function relationship and the time-series feature data;

[0010] Based on the predicted icing thickness and the reference icing thickness corresponding to the moment to be predicted, adjust the parameters of the icing prediction model to obtain the trained icing prediction model. The predicted icing thickness corresponding to the moment to be predicted is obtained based on the target time-series feature data.

[0011] Optionally, the residual network includes a first residual block and a second residual block. Inputting the time-series feature data into the residual network to obtain the spatial implicit function relationship among the micro-meteorological features, terrain features, and ice-coverage monitoring features output by the residual network, including: inputting the time-series feature data into the first residual block to obtain the initial spatial implicit function relationship among the micro-meteorological features, terrain features, and ice-coverage monitoring features output by the first residual block; inputting the initial spatial implicit function relationship and the time-series feature data into the second residual block, and the second residual block performs feature fusion processing on the initial spatial implicit function relationship and the time-series feature data through residual connection to obtain the spatial implicit function relationship.

[0012] Optionally, based on the time-series feature data and the time-series implicit function relationship, an exponential moving average (EMA) attention mechanism is used for feature fusion processing to obtain the target time-series feature data, including: concatenating the time-series implicit function relationship with the time-series feature data to obtain the concatenated multi-modal feature data; applying an exponentially decaying weight to the concatenated multi-modal feature data using the EMA attention mechanism for feature weighted fusion processing to obtain the target time-series feature data.

[0013] Optionally, the ice-coverage prediction model further includes a long short-term memory (LSTM) network. The time-series implicit function relationship is obtained in the following manner: inputting the spatial implicit function relationship and the time-series feature data into the LSTM network to obtain the time-series implicit function relationship output by the LSTM network.

[0014] Optionally, the ice-coverage prediction model further includes a fully connected layer. The predicted ice-coverage thickness corresponding to the prediction time is obtained in the following manner: inputting the target time-series feature data into the fully connected layer to obtain the predicted ice-coverage thickness corresponding to the prediction time output by the fully connected layer.

[0015] Optionally, the ice-coverage prediction model further includes a feature extraction module to extract features from the ice-coverage data to obtain the time-series feature data, including: inputting the ice-coverage data into the feature extraction module to obtain the time-series feature data output by the feature extraction module.

[0016] In a second aspect, the present application provides a method for predicting ice-coverage on a transmission line, including:

[0017] In response to an ice-coverage prediction instruction for a transmission line, obtaining the ice-coverage data of the transmission line, where the ice-coverage data includes micro-meteorological data, terrain data, and ice-coverage monitoring data;

[0018] Inputting the ice-coverage data into the ice-coverage prediction model for ice-coverage prediction to obtain the ice-coverage thickness of the transmission line corresponding to the prediction time, where the ice-coverage prediction model is trained using the ice-coverage prediction model training method described in the first aspect of the present application.

[0019] In a third aspect, the present application provides an icing prediction model training device. The icing prediction model includes a residual network. The icing prediction model training device includes:

[0020] A first acquisition module, configured to acquire icing data of a transmission line at consecutive moments. The icing data includes micro-meteorological data, terrain data, and icing monitoring data;

[0021] A feature extraction module, configured to extract features from the icing data to obtain time-series feature data. The time-series feature data includes micro-meteorological features, terrain features, and icing monitoring features at consecutive moments;

[0022] A second acquisition module, configured to input the time-series feature data into the residual network to obtain the spatial implicit function relationship between the micro-meteorological features, terrain features, and icing monitoring features output by the residual network;

[0023] A fusion processing module, configured to perform feature fusion processing on the time-series feature data and the time-series implicit function relationship by using an exponential moving average (EMA) attention mechanism to obtain target time-series feature data. The time-series implicit function relationship is obtained based on the spatial implicit function relationship and the time-series feature data;

[0024] An adjustment module, configured to adjust the parameters of the icing prediction model based on the predicted icing thickness and the reference icing thickness corresponding to the moment to be predicted, and obtain the trained icing prediction model. The predicted icing thickness corresponding to the moment to be predicted is obtained based on the target time-series feature data.

[0025] Optionally, the residual network includes a first residual block and a second residual block. The second acquisition module is specifically configured to: input the time-series feature data into the first residual block to obtain the initial spatial implicit function relationship between the micro-meteorological features, terrain features, and icing monitoring features output by the first residual block; input the initial spatial implicit function relationship and the time-series feature data into the second residual block, and the second residual block performs feature fusion processing on the initial spatial implicit function relationship and the time-series feature data through residual connection to obtain the spatial implicit function relationship.

[0026] Optionally, the fusion processing module is specifically configured to: perform feature splicing on the time-series implicit function relationship and the time-series feature data to obtain the spliced multi-modal feature data; apply exponential decay weights to the spliced multi-modal feature data by using the EMA attention mechanism for feature weighted fusion processing to obtain the target time-series feature data.

[0027] Optionally, the icing prediction model further includes a long short-term memory (LSTM) network. The fusion processing module is further configured to obtain the time-series implicit function relationship in the following manner: input the spatial implicit function relationship and the time-series feature data into the LSTM network to obtain the time-series implicit function relationship output by the LSTM network.

[0028] Optionally, the icing prediction model further includes a fully connected layer, and the adjustment module is further configured to obtain the predicted icing thickness corresponding to the moment to be predicted in the following manner: input the target time series feature data into the fully connected layer, and obtain the predicted icing thickness corresponding to the moment to be predicted output by the fully connected layer.

[0029] Optionally, the icing prediction model further includes a feature extraction module, and the feature extraction module is specifically configured to: input the icing data into the feature extraction module, and obtain the time series feature data output by the feature extraction module.

[0030] In a fourth aspect, the present application provides a transmission line icing prediction device, including:

[0031] An acquisition module, configured to acquire icing data of a transmission line in response to an icing prediction instruction for the transmission line, where the icing data includes micro-meteorological data, terrain data, and icing monitoring data;

[0032] A prediction module, configured to input the icing data into the icing prediction model for icing prediction, and obtain the icing thickness of the transmission line corresponding to the moment to be predicted, where the icing prediction model is trained by using the icing prediction model training method described in the first aspect of the present application.

[0033] In a fifth aspect, the present application provides an electronic device, including: a processor, and a memory communicatively connected to the processor;

[0034] The memory stores computer-executable instructions;

[0035] The processor executes the computer-executable instructions stored in the memory to implement the icing prediction model training method described in the first aspect of the present application or the transmission line icing prediction method described in the second aspect.

[0036] In a sixth aspect, the present application provides a computer-readable storage medium, in which computer program instructions are stored, and when the computer program instructions are executed, the icing prediction model training method described in the first aspect of the present application or the transmission line icing prediction method described in the second aspect is implemented.

[0037] In a seventh aspect, the present application provides a computer program product, including a computer program, and when the computer program is executed, the icing prediction model training method described in the first aspect of the present application or the transmission line icing prediction method described in the second aspect is implemented.

[0038] A method for training an icing prediction model, a method and device for predicting transmission line icing provided by this application. By obtaining icing data of a transmission line at consecutive moments, the icing data includes micro-meteorological data, terrain data, and icing monitoring data; extracting features from the icing data to obtain time-series feature data, the time-series feature data includes micro-meteorological features, terrain features, and icing monitoring features at consecutive moments; inputting the time-series feature data into a residual network to obtain the spatial implicit function relationship between the micro-meteorological features, terrain features, and icing monitoring features output by the residual network; by introducing residual connections, the problem of gradient disappearance or explosion in deep networks can be effectively solved, thereby improving the learning ability and prediction accuracy of the icing prediction model; based on the time-series feature data and the time-series implicit function relationship, an EMA attention mechanism is used for feature fusion processing to obtain target time-series feature data, and the time-series implicit function relationship is obtained based on the spatial implicit function relationship and the time-series feature data; by using the EMA attention mechanism to comprehensively analyze the icing monitoring data, combined with the micro-meteorological data and terrain data, and deeply exploring the global context information, the dependence relationship between data at multiple scales can be captured, thereby better characterizing the dynamic change law of the icing phenomenon, effectively reducing the computational overhead and retaining the effective information of each channel, and by grouping the features, the spatial semantic distribution of each group of features can be balanced; furthermore, based on the predicted icing thickness and the reference icing thickness corresponding to the moment to be predicted, the parameters of the icing prediction model are adjusted to obtain the trained icing prediction model, and the predicted icing thickness corresponding to the moment to be predicted is obtained based on the target time-series feature data. By introducing the EMA attention mechanism and the residual network, this application can effectively cope with the complexity of multi-source icing data and improve the versatility of the icing prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The accompanying drawings here are incorporated into the specification and constitute a part of this specification, showing embodiments consistent with this application, and are used together with the specification to explain the principles of this application.

[0040] Figure 1 It is a flowchart of a method for training an icing prediction model provided by an embodiment of this application;

[0041] Figure 2 It is a flowchart of a method for training an icing prediction model provided by another embodiment of this application;

[0042] Figure 3 It is a flowchart of a method for predicting transmission line icing provided by an embodiment of this application;

[0043] Figure 4 It is a schematic structural diagram of an icing prediction model training device provided by an embodiment of this application;

[0044] Figure 5Schematic structural diagram of an ice accretion prediction device for a transmission line provided by an embodiment of the present application;

[0045] Figure 6 Schematic structural diagram of an electronic device provided by an embodiment of the present application.

[0046] Through the above-mentioned drawings, specific embodiments of the present application have been shown, and more detailed descriptions will be provided later. These drawings and textual descriptions are not intended to limit the scope of the concept of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. Detailed implementation manners

[0047] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Apparently, the described embodiments are some, but not all, of the embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0048] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or fully authorized by all parties. The collection, use, and processing of relevant data need to comply with relevant laws, regulations, and standards, and corresponding operation entrances are provided for the user to choose to authorize or refuse.

[0049] During the power transmission process, icing on the surface of the transmission line is likely to cause various safety accidents. By predicting the growth trend of ice accretion on the transmission line, measures can be taken to reduce the risks brought by icing.

[0050] Currently, the difficulty in predicting the growth trend of ice accretion on transmission lines lies in analyzing the influence of complex and variable environments on ice accretion growth. Usually, a mathematical and physical model of the relationship between micro-meteorological factors and ice accretion growth is established, and the autocorrelation of ice accretion growth is combined to predict the ice accretion thickness of the transmission line. However, the above method overly relies on the accuracy of the mathematical and physical model and has poor robustness. In some scenarios, for example, for transmission lines covering a large geographical area such as line-level or regional-level, predicting the ice accretion thickness of the transmission line by the above method has the problem of poor generality, and does not make full use of a large amount of historical data of ice accretion on the transmission line and does not extract the general laws that conform to the growth and change of ice accretion from it. Therefore, for transmission lines covering a large geographical area, how to construct an ice accretion prediction model with strong robustness and good generality according to the geographical environment, micro-meteorological conditions, and ice accretion situation of the transmission line is a technical problem urgently to be solved in the art.

[0051] Based on the above problems, the present application provides a method for training an icing prediction model, a method and device for predicting icing on a transmission line. Based on multi-source icing data (including micro-meteorological data, terrain data, and icing monitoring data), by introducing a residual network and combining it with an Exponential Moving Average (EMA) attention mechanism, an icing prediction model is trained. The obtained icing prediction model can effectively handle the complexity of multi-source icing data and improve the versatility of the icing prediction model. Thus, when predicting the icing thickness of a transmission line through the icing prediction model, the icing thickness can be predicted more accurately, providing strong technical support for the monitoring and warning system in actual engineering.

[0052] It should be noted that the method for training the icing prediction model and the method for predicting icing on a transmission line provided in the embodiments of the present application can be applied in a server, which can be an independent server or a service cluster, etc.

[0053] Figure 1 The figure is a flowchart of a method for training an icing prediction model provided in an embodiment of the present application, which is used to train the icing prediction model. The icing prediction model includes a residual network. As Figure 1 shown, the method of the embodiment of the present application includes:

[0054] S101. Obtain the icing data of the transmission line at consecutive moments. The icing data includes micro-meteorological data, terrain data, and icing monitoring data.

[0055] In the embodiments of the present application, exemplarily, based on the historical icing data of the transmission line obtained from actual icing detection points, the icing data of the transmission line at consecutive moments within a set time period can be obtained as training samples, so that multiple training samples can be obtained, and each training sample is used for training the icing prediction model. Among them, the icing data of the transmission line at consecutive moments within a set time period is, for example, the icing data sampled every half hour within the past 24 hours. Among them, the micro-meteorological data can include data such as temperature, humidity, wind speed, precipitation, and air pressure, and the micro-meteorological data is used to predict the environmental conditions for icing occurrence; the terrain data can include data such as slope, aspect, curvature, and elevation, as well as micro-topographic characteristic factors obtained based on these data (such as the length of the windward slope and the valley depth ratio); the icing monitoring data can include the equivalent icing thickness data obtained through the power grid icing monitoring terminal, the icing thickness data obtained through unmanned aerial vehicle line inspection, and the icing thickness data obtained through manual measurement at some detection points, and is used to provide real-time icing condition information.

[0056] S102. Extract the features of the icing data to obtain time-series feature data, which includes the micro-meteorological features, terrain features, and icing monitoring features at consecutive moments.

[0057] In this step, the features of the icing data can be extracted to obtain time-series feature data, which includes the micro-meteorological features, terrain features, and icing monitoring features at consecutive moments.

[0058] S103. Input the time-series feature data into the residual network to obtain the spatial implicit function relationship among the micro-meteorological features, terrain features, and icing monitoring features output by the residual network.

[0059] It can be understood that compared with traditional neural networks, the residual network (ResNet) can effectively solve the problem of gradient vanishing or explosion in deep networks by introducing residual connections, and at the same time fully consider the interaction characteristics of multi-dimensional data in complex scenarios, which helps to improve the learning ability and prediction accuracy of the model. In this step, after obtaining the time-series feature data, the time-series feature data can be input into the residual network to obtain the spatial implicit function relationship among the micro-meteorological features, terrain features, and icing monitoring features. For how to obtain the spatial implicit function relationship among the micro-meteorological features, terrain features, and icing monitoring features through the residual network specifically, reference can be made to the subsequent embodiments and will not be elaborated here.

[0060] S104. Based on the time-series feature data and the time-series implicit function relationship, perform feature fusion processing using the EMA attention mechanism to obtain the target time-series feature data, and the time-series implicit function relationship is obtained according to the spatial implicit function relationship and the time-series feature data.

[0061] It can be understood that for the multi-source data (i.e., micro-meteorological data, terrain data, and icing monitoring data) in the icing prediction task, by comprehensively analyzing the icing monitoring data using the EMA attention mechanism, combining the micro-meteorological data and terrain data, and deeply exploring the global context information, the dependence relationship between data at multiple scales can be captured, so as to better characterize the dynamic change law of the icing phenomenon, effectively reduce the computational overhead and retain the effective information of each channel. By grouping the features, the spatial semantic distribution of each group of features can be balanced. Exemplarily, after obtaining the spatial implicit function relationship among the micro-meteorological features, terrain features, and icing monitoring features, the time-series implicit function relationship can be used as a time-series feature to be incorporated into the overall feature splicing, that is, the time-series implicit function relationship is spliced with the time-series feature data to obtain the spliced multi-modal feature data, and then the EMA attention mechanism is used to perform feature weighted fusion processing on the spliced multi-modal feature data using exponential decay weights to obtain the target time-series feature data, which can strengthen the influence of recent time-series features and key features, smooth the historical data, and highlight the features at critical moments, thus helping to improve the accuracy and robustness of the icing prediction model.

[0062] S105. Adjust the parameters of the icing prediction model based on the predicted icing thickness and the reference icing thickness corresponding to the moment to be predicted, so as to obtain the trained icing prediction model. The predicted icing thickness corresponding to the moment to be predicted is obtained based on the target time-series feature data.

[0063] In this step, the moment to be predicted is, for example, a set moment after the current training moment. Specifically, for example, it is the moment corresponding to 6 hours after the current training moment. The predicted icing thickness corresponding to the moment to be predicted can be obtained based on the target time-series feature data, that is, the predicted icing thickness corresponding to the moment to be predicted is predicted based on the icing data of the transmission line at consecutive moments. For how to specifically obtain the predicted icing thickness corresponding to the moment to be predicted based on the target time-series feature data, reference can be made to the subsequent embodiments. The reference icing thickness corresponding to the moment to be predicted is the actual icing thickness corresponding to the moment to be predicted, which can be used as a label to be compared with the predicted icing thickness corresponding to the moment to be predicted, so as to obtain the value of the loss function, and then the parameters of the icing prediction model can be adjusted according to the value of the loss function to obtain the trained icing prediction model. Among them, for example, the mean squared error (MSE) can be used as the loss function. When the change rate of the value of the loss function is less than a threshold (such as 0.1%), it is determined that the icing prediction model converges. The optimizer can be, for example, the adaptive moment estimation (Adam) optimizer with an adaptive learning rate, which can automatically adjust the learning rate to improve the training effect.

[0064] Optionally, the icing prediction model can also be used to predict the icing occurrence probability corresponding to the moment to be predicted according to the icing thickness, so as to be compared with the label (whether icing occurs) corresponding to the moment to be predicted, and the parameters of the icing prediction model are adjusted.

[0065] The icing prediction model training method provided by the embodiments of the present application obtains the icing data of the transmission line at continuous time moments. The icing data includes micro-meteorological data, terrain data, and icing monitoring data. Feature extraction is performed on the icing data to obtain time-series feature data, which includes micro-meteorological features, terrain features, and icing monitoring features at continuous time moments. The time-series feature data is input into a residual network to obtain the spatial implicit function relationship between the micro-meteorological features, terrain features, and icing monitoring features output by the residual network. By introducing residual connections, the problem of gradient vanishing or explosion in deep networks can be effectively solved, thereby improving the learning ability and prediction accuracy of the icing prediction model. Based on the time-series feature data and the time-series implicit function relationship, the EMA attention mechanism is used for feature fusion processing to obtain target time-series feature data. The time-series implicit function relationship is obtained based on the spatial implicit function relationship and the time-series feature data. By comprehensively analyzing the icing monitoring data using the EMA attention mechanism, combining micro-meteorological data and terrain data, and deeply exploring the global context information, the dependence relationship between data at multiple scales can be captured, so as to better represent the dynamic change law of the icing phenomenon, effectively reduce the computational overhead, and retain the effective information of each channel. By grouping the features, the spatial semantic distribution of each group of features can be balanced. Furthermore, based on the predicted icing thickness and the reference icing thickness corresponding to the moment to be predicted, the parameters of the icing prediction model are adjusted to obtain the trained icing prediction model. The predicted icing thickness corresponding to the moment to be predicted is obtained based on the target time-series feature data. By introducing the EMA attention mechanism and the residual network, the embodiments of the present application can effectively cope with the complexity of multi-source icing data and improve the versatility of the icing prediction model.

[0066] Figure 2 It is a flowchart of the icing prediction model training method provided by another embodiment of the present application. On the basis of the above embodiments, the embodiments of the present application further illustrate the icing prediction model training method. As Figure 2 shown, the method of the embodiments of the present application may include:

[0067] S201. Obtain the icing data of the transmission line at continuous time moments. The icing data includes micro-meteorological data, terrain data, and icing monitoring data.

[0068] For the specific description of this step, reference can be made to the relevant description of S101 in the Figure 1 illustrated embodiment, which will not be elaborated here.

[0069] Considering that the icing prediction model further includes a feature extraction module, therefore, in the embodiments of the present application, Figure 1 step S102 in

[0070] S202. Input the icing data into the feature extraction module to obtain the time-series feature data output by the feature extraction module. The time-series feature data includes micro-meteorological features, terrain features, and icing monitoring features at consecutive moments.

[0071] Exemplarily, the feature extraction module is used to extract the data features of the icing data. The feature extraction module includes, for example, a first convolutional layer, a second convolutional layer, and a max pooling layer. Among them, the first convolutional layer is, for example, a one-dimensional convolutional layer, the number of filters it contains is, for example, 64, the convolutional kernel size is, for example, 3, and the activation function uses, for example, the Rectified Linear Unit (ReLU) function. The second convolutional layer is, for example, a one-dimensional convolutional layer, the number of filters it contains is, for example, 128, the convolutional kernel size is, for example, 3, and the activation function uses, for example, the ReLU function. The pooling size of the max pooling layer is, for example, 2, which is used to perform dimensionality reduction processing on the features. It can be understood that by inputting the icing data into the feature extraction module, the first convolutional layer of the feature extraction module first performs preliminary feature extraction on the icing data, then the second convolutional layer further extracts features, and finally the max pooling layer performs dimensionality reduction processing on the extracted features to obtain the time-series feature data corresponding to the icing data. The time-series feature data includes micro-meteorological features, terrain features, and icing monitoring features at consecutive moments.

[0072] Considering that the residual network includes a first residual block and a second residual block, therefore, in the embodiments of the present application, Figure 1 Step S103 in can be further included two steps of S203 and S204 as follows:

[0073] S203. Input the time-series feature data into the first residual block to obtain the initial spatial implicit function relationship among the micro-meteorological features, terrain features, and icing monitoring features output by the first residual block.

[0074] Exemplarily, after obtaining the time-series feature data, the time-series feature data can be input into the first residual block. The first residual block performs feature fusion processing on the micro-meteorological features, terrain features, and icing monitoring features through residual connections, so as to obtain the initial spatial implicit function relationship among the micro-meteorological features, terrain features, and icing monitoring features. It can be understood that through the residual connections in this step, it can help the icing prediction model capture deep features. Among them, the first residual block can include a third convolutional layer and a fourth convolutional layer. The number of filters contained in the third convolutional layer is, for example, 256, the convolutional kernel size is, for example, 3, and the activation function uses, for example, the ReLU function. The number of filters contained in the fourth convolutional layer is, for example, 256, the convolutional kernel size is, for example, 3, and the activation function uses, for example, the ReLU function.

[0075] S204. Input the initial spatial implicit function relationship and the temporal feature data into the second residual block. The second residual block performs feature fusion processing on the initial spatial implicit function relationship and the temporal feature data through residual connection, and obtains the spatial implicit function relationship among the micro-meteorological feature, the terrain feature, and the icing monitoring feature.

[0076] Exemplarily, after obtaining the initial spatial implicit function relationship among the micro-meteorological feature, the terrain feature, and the icing monitoring feature, the initial spatial implicit function relationship and the temporal feature data can be input into the second residual block. The second residual block performs feature fusion processing on the initial spatial implicit function relationship and the temporal feature data through residual connection, so as to obtain the spatial implicit function relationship among the micro-meteorological feature, the terrain feature, and the icing monitoring feature. It can be understood that through the residual connection in this step, the icing prediction model can be helped to further capture deep features. Among them, the second residual block may include a fifth convolutional layer and a sixth convolutional layer. The number of filters included in the fifth convolutional layer is, for example, 512, the convolutional kernel size is, for example, 3, and the activation function is, for example, the ReLU function; the number of filters included in the sixth convolutional layer is, for example, 512, the convolutional kernel size is, for example, 3, and the activation function is, for example, the ReLU function.

[0077] Considering that the icing prediction model further includes a long short-term memory network, therefore, in the embodiment of the present application, Figure 1 Step S104 may further include the following three steps of S205 to S207:

[0078] S205. Input the spatial implicit function relationship and the temporal feature data into a long short-term memory network (LSTM), and obtain the temporal implicit function relationship output by the long short-term memory network.

[0079] Exemplarily, after obtaining the spatial implicit function relationship among the micro-meteorological feature, the terrain feature, and the icing monitoring feature, the spatial implicit function relationship and the temporal feature data can be input into the long short-term memory network, and the temporal implicit function relationship output by the long short-term memory network is obtained. Among them, the long short-term memory network may include, for example, a first long short-term memory network and a second long short-term memory network. The output dimension of the first long short-term memory network is, for example, 64, and the activation function is, for example, the hyperbolic tangent function (tanh); the output dimension of the second long short-term memory network is, for example, 128, and the activation function is, for example, the tanh function.

[0080] S206. Perform feature splicing on the temporal implicit function relationship and the temporal feature data to obtain the spliced multi-modal feature data.

[0081] In this step, after obtaining the temporal implicit function relationship, the temporal implicit function relationship can be incorporated into the overall feature concatenation as a temporal feature, that is, the temporal implicit function relationship and the temporal feature data are concatenated to obtain the concatenated multi-modal feature data.

[0082] S207. Apply exponential decay weights to the concatenated multi-modal feature data using the EMA attention mechanism for feature weighted fusion processing to obtain the target temporal feature data.

[0083] It can be understood that the EMA attention mechanism enhances the weights of recent temporal features and key features by applying exponential decay weights to the concatenated multi-modal feature data, thereby strengthening the influence of recent temporal features and key features, smoothing historical data, highlighting the features at critical moments, and thus helping to improve the accuracy and robustness of the icing prediction model. In this step, the EMA attention mechanism can be used to apply exponential decay weights to the concatenated multi-modal feature data for feature weighted fusion processing, so as to obtain the target temporal feature data, which can improve the influence of important features.

[0084] Considering that the icing prediction model also includes a fully connected layer, therefore, in the embodiments of the present application, Figure 1 step S105 may further include the following two steps of S208 and S209:

[0085] S208. Input the target temporal feature data into the fully connected layer to obtain the predicted icing thickness corresponding to the moment to be predicted output by the fully connected layer.

[0086] Exemplarily, after obtaining the target temporal feature data, the target temporal feature data can be input into the fully connected layer to obtain the predicted icing thickness corresponding to the moment to be predicted output by the fully connected layer. Among them, the fully connected layer includes, for example, a first fully connected layer, a second fully connected layer, and a third fully connected layer. The number of neurons in the first fully connected layer is, for example, 256, and the activation function is, for example, the ReLU function; the number of neurons in the second fully connected layer is, for example, 128, and the activation function is, for example, the ReLU function; the third fully connected layer serves as the output layer, and the number of neurons is, for example, 1, and is used to predict the icing thickness corresponding to the moment to be predicted.

[0087] Optionally, the third fully connected layer can also be used to predict the icing occurrence probability corresponding to the moment to be predicted according to the icing thickness.

[0088] S209. Based on the predicted icing thickness corresponding to the moment to be predicted and the reference icing thickness, adjust the parameters of the icing prediction model to obtain the trained icing prediction model.

[0089] For the specific description of this step, reference can be made to Figure 1 the relevant description of S105 in the illustrated embodiment, which will not be elaborated here.

[0090] The ice-covering prediction model training method provided by the embodiment of the present application obtains ice-covering data of a transmission line at consecutive moments, where the ice-covering data includes micro-meteorological data, terrain data, and ice-covering monitoring data; inputs the ice-covering data into a feature extraction module to obtain time-series feature data output by the feature extraction module, and the time-series feature data includes micro-meteorological features, terrain features, and ice-covering monitoring features at consecutive moments; inputs the time-series feature data into a first residual block to obtain an initial spatial implicit function relationship among the micro-meteorological features, terrain features, and ice-covering monitoring features output by the first residual block, inputs the initial spatial implicit function relationship and the time-series feature data into a second residual block, and the second residual block performs feature fusion processing on the initial spatial implicit function relationship and the time-series feature data through residual connection to obtain a spatial implicit function relationship among the micro-meteorological features, terrain features, and ice-covering monitoring features; by introducing residual connection, the problem of gradient disappearance or explosion in deep networks can be effectively solved, thereby improving the learning ability and prediction accuracy of the ice-covering prediction model; inputs the spatial implicit function relationship and the time-series feature data into a long short-term memory network to obtain a time-series implicit function relationship output by the long short-term memory network, splices the time-series implicit function relationship with the time-series feature data to obtain spliced multi-modal feature data, and applies exponential decay weights to the spliced multi-modal feature data using the EMA attention mechanism for feature weighted fusion processing to obtain target time-series feature data; by comprehensively analyzing the ice-covering monitoring data using the EMA attention mechanism, combining the micro-meteorological data and the terrain data, and deeply exploring the global context information, the dependence relationship between data at multiple scales can be captured, thereby better characterizing the dynamic change law of the ice-covering phenomenon, effectively reducing the computational overhead and retaining the effective information of each channel, and by grouping the features, the spatial semantic distribution of each group of features can be balanced; then inputs the target time-series feature data into a fully connected layer to obtain a predicted ice-covering thickness corresponding to the moment to be predicted output by the fully connected layer, and based on the predicted ice-covering thickness corresponding to the moment to be predicted and the reference ice-covering thickness, adjusts the parameters of the ice-covering prediction model to obtain a trained ice-covering prediction model. By introducing the EMA attention mechanism and the residual network, the embodiment of the present application can effectively cope with the complexity of multi-source ice-covering data and improve the versatility and robustness of the ice-covering prediction model.

[0091] Based on the above embodiment, Figure 3 is a flowchart of a transmission line ice-covering prediction method provided by an embodiment of the present application. As Figure 3 shown, the method of the embodiment of the present application includes:

[0092] S301. In response to an ice-covering prediction instruction for a transmission line, obtain ice-covering data of the transmission line, where the ice-covering data includes micro-meteorological data, terrain data, and ice-covering monitoring data.

[0093] Exemplarily, the icing prediction instruction can be input by a user to the electronic device implementing the embodiments of this method, or can be automatically triggered by the electronic device implementing the embodiments of this method, or can be sent by other devices to the electronic device implementing the embodiments of this method. Specifically, for example, the user can trigger an icing prediction instruction for the transmission line based on the real-time icing monitoring data of the transmission line, or can automatically trigger an icing prediction instruction for the transmission line by setting an icing prediction period. Correspondingly, the electronic device implementing the embodiments of this method, in response to the icing prediction instruction for the transmission line, obtains the icing data of the transmission line, and the icing data includes micro-meteorological data, terrain data, and icing monitoring data. For example, the icing data of the transmission line at consecutive moments including the current moment within a set past duration can be obtained.

[0094] S302. Input the icing data into the icing prediction model for icing prediction to obtain the icing thickness of the transmission line corresponding to the moment to be predicted. The icing prediction model is trained by using the icing prediction model training method in any of the above method embodiments.

[0095] Exemplarily, the moment to be predicted is, for example, the next set moment. Specifically, the moment to be predicted is, for example, the moment corresponding to 6 hours after the current moment. In this step, inputting the icing data into the icing prediction model for icing prediction can obtain the icing thickness of the transmission line corresponding to the moment to be predicted. Among them, the icing prediction model is trained by using the icing prediction model training method in any of the above method embodiments. Optionally, if the icing prediction model can predict the icing occurrence probability corresponding to the moment to be predicted based on the icing thickness, then in the embodiments of this application, inputting the icing data into the icing prediction model for icing prediction can also obtain the icing occurrence probability corresponding to the moment to be predicted.

[0096] The icing prediction model training method provided by the embodiments of this application, by responding to the icing prediction instruction for the transmission line, obtains the icing data of the transmission line, and the icing data includes micro-meteorological data, terrain data, and icing monitoring data; inputs the icing data into the icing prediction model for icing prediction to obtain the icing thickness of the transmission line corresponding to the moment to be predicted. Since the icing prediction model of the embodiments of this application has good versatility, therefore, through the icing prediction model, the icing thickness of the transmission line can be predicted more accurately, with good stability, thus being able to provide strong technical support for the monitoring and early warning system in actual engineering.

[0097] The following are the device embodiments of this application, which can be used to execute the method embodiments of this application. For details not disclosed in the device embodiments of this application, please refer to the method embodiments of this application.

[0098] Figure 4The figure is a schematic structural diagram of an icing prediction model training device provided by an embodiment of the present application. The icing prediction model includes a residual network. As Figure 4 shown, the icing prediction model training device 400 of the embodiment of the present application includes: a first acquisition module 401, a feature extraction module 402, a second acquisition module 403, a fusion processing module 404, and an adjustment module 405. Among them:

[0099] The first acquisition module 401 is configured to acquire icing data of a transmission line at consecutive moments. The icing data includes micro-meteorological data, terrain data, and icing monitoring data.

[0100] The feature extraction module 402 is configured to extract features from the icing data to obtain time-series feature data. The time-series feature data includes micro-meteorological features, terrain features, and icing monitoring features at consecutive moments.

[0101] The second acquisition module 403 is configured to input the time-series feature data into the residual network to obtain the spatial implicit function relationship between the micro-meteorological features, terrain features, and icing monitoring features output by the residual network.

[0102] The fusion processing module 404 is configured to perform feature fusion processing on the time-series feature data and the time-series implicit function relationship by using the EMA attention mechanism to obtain target time-series feature data. The time-series implicit function relationship is obtained according to the spatial implicit function relationship and the time-series feature data.

[0103] The adjustment module 405 is configured to adjust the parameters of the icing prediction model based on the predicted icing thickness and the reference icing thickness corresponding to the moment to be predicted, and obtain the trained icing prediction model. The predicted icing thickness corresponding to the moment to be predicted is obtained based on the target time-series feature data.

[0104] In some embodiments, the residual network includes a first residual block and a second residual block. The second acquisition module 403 may be specifically configured to: input the time-series feature data into the first residual block to obtain the initial spatial implicit function relationship between the micro-meteorological features, terrain features, and icing monitoring features output by the first residual block; input the initial spatial implicit function relationship and the time-series feature data into the second residual block, and the second residual block performs feature fusion processing on the initial spatial implicit function relationship and the time-series feature data through residual connection to obtain the spatial implicit function relationship.

[0105] In some embodiments, the fusion processing module 404 may be specifically configured to: perform feature splicing on the time-series implicit function relationship and the time-series feature data to obtain the spliced multi-modal feature data; apply exponential decay weights to the spliced multi-modal feature data by using the EMA attention mechanism for feature weighted fusion processing to obtain the target time-series feature data.

[0106] Optionally, the icing prediction model further includes a long short-term memory network, and the fusion processing module 404 can also be used to obtain the temporal implicit function relationship in the following manner: input the spatial implicit function relationship and the temporal feature data into the long short-term memory network to obtain the temporal implicit function relationship output by the long short-term memory network.

[0107] Optionally, the icing prediction model further includes a fully connected layer, and the adjustment module 405 can also be used to obtain the predicted icing thickness corresponding to the moment to be predicted in the following manner: input the target temporal feature data into the fully connected layer to obtain the predicted icing thickness corresponding to the moment to be predicted output by the fully connected layer.

[0108] In some embodiments, the icing prediction model further includes a feature extraction module. The feature extraction module 402 can specifically be used to: input the icing data into the feature extraction module to obtain the temporal feature data output by the feature extraction module.

[0109] The device according to the embodiment of the present application can be used to execute the solution of the icing prediction model training method in any of the above method embodiments. The implementation principle and technical effect are similar and will not be elaborated here.

[0110] Figure 5 This is a schematic structural diagram of a transmission line icing prediction device provided by an embodiment of the present application. As Figure 5 shown, the transmission line icing prediction device 500 according to the embodiment of the present application includes: an acquisition module 501 and a prediction module 502. Among them:

[0111] The acquisition module 501 is configured to obtain the icing data of the transmission line in response to an icing prediction instruction for the transmission line. The icing data includes micro-meteorological data, terrain data, and icing monitoring data;

[0112] The prediction module 502 is configured to input the icing data into the icing prediction model for icing prediction to obtain the icing thickness of the transmission line corresponding to the moment to be predicted. The icing prediction model is trained by using the icing prediction model training method in any of the above method embodiments.

[0113] The device according to the embodiment of the present application can be used to execute the solution of the transmission line icing prediction method in any of the above method embodiments. The implementation principle and technical effect are similar and will not be elaborated here.

[0114] Figure 6 This is a schematic structural diagram of an electronic device provided by an embodiment of the present application. As Figure 6 shown, the electronic device 600 may include: at least one processor 601 and a memory 602.

[0115] The memory 602 is used to store a program. Specifically, the program may include program code, and the program code includes computer execution instructions.

[0116] The memory 602 may include a high-speed random access memory (RAM), and may also include non-volatile memory, such as at least one disk memory.

[0117] The processor 601 is configured to execute the computer-executable instructions stored in the memory 602 to implement the icing prediction model training method or the transmission line icing prediction method described in the foregoing method embodiments. Among them, the processor 601 may be a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application. Specifically, when implementing the icing prediction model training method or the transmission line icing prediction method described in the foregoing method embodiments, the electronic device may be an electronic device with processing functions such as a server.

[0118] Optionally, the electronic device 600 may further include a communication interface 603. In a specific implementation, if the communication interface 603, the memory 602, and the processor 601 are implemented independently, the communication interface 603, the memory 602, and the processor 601 may be interconnected through a bus and communicate with each other. The bus may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc., but it does not mean that there is only one bus or one type of bus.

[0119] Optionally, in a specific implementation, if the communication interface 603, the memory 602, and the processor 601 are integrated on a chip, the communication interface 603, the memory 602, and the processor 601 may communicate through an internal interface.

[0120] The present application also provides a computer-readable storage medium, in which computer program instructions are stored. When the processor executes the computer program instructions, the solutions of the icing prediction model training method or the transmission line icing prediction method as described above are implemented.

[0121] The present application also provides a computer program product, including a computer program which, when executed, implements the solution of the icing prediction model training method or the solution of the transmission line icing prediction method as described above.

[0122] The above computer-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read Only Memory (PROM), Read Only Memory (ROM), magnetic memory, flash memory, a magnetic disk or an optical disc. The readable storage medium can be any available medium accessible by a general or special-purpose computer.

[0123] An exemplary readable storage medium is coupled to the processor, enabling the processor to read information from the readable storage medium and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can be located in an application-specific integrated circuit. Of course, the processor and the readable storage medium can also exist as discrete components in the icing prediction model training device or the transmission line icing prediction device.

[0124] Those of ordinary skill in the art can understand that all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions. The foregoing program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps including the above method embodiments; and the foregoing storage medium includes various media that can store program codes, such as ROM, RAM, magnetic disks, or optical discs.

[0125] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not intended to limit them. Although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for training an ice cover prediction model, characterized in that: The icing prediction model includes a residual network, and the icing prediction model training method includes: Acquire ice coverage data of the power transmission line at consecutive moments, wherein the ice coverage data includes micro-meteorological data, terrain data and ice coverage monitoring data; Extracting features from the ice cover data to obtain time series feature data, wherein the time series feature data includes micro-meteorological features, terrain features, and ice cover monitoring features at consecutive moments; Inputting the time series feature data into the residual network to obtain the spatial implicit function relationship between the micro-meteorological features, the terrain features and the ice cover monitoring features output by the residual network; Based on the time series feature data and the time series implicit function relationship, an exponential moving average EMA attention mechanism is used to perform feature fusion processing to obtain target time series feature data, wherein the time series implicit function relationship is obtained based on the spatial implicit function relationship and the time series feature data; Based on the predicted ice thickness corresponding to the time to be predicted and the reference ice thickness, the parameters of the ice prediction model are adjusted to obtain a trained ice prediction model, wherein the predicted ice thickness corresponding to the time to be predicted is obtained based on the target time series feature data.

2. The ice cover prediction model training method according to claim 1, characterized in that: The residual network includes a first residual block and a second residual block, and the time series feature data is input into the residual network to obtain the spatial implicit function relationship between the micro-meteorological features, the terrain features and the ice cover monitoring features output by the residual network, including: Inputting the time series feature data into the first residual block to obtain an initial spatial implicit function relationship between the micrometeorological feature, the terrain feature and the ice cover monitoring feature output by the first residual block; The initial spatial implicit function relationship and the temporal feature data are input into the second residual block, and the second residual block performs feature fusion processing on the initial spatial implicit function relationship and the temporal feature data through residual connection to obtain the spatial implicit function relationship.

3. The ice cover prediction model training method according to claim 1, characterized in that: Based on the time series feature data and the time series implicit function relationship, the exponential moving average EMA attention mechanism is used to perform feature fusion processing to obtain the target time series feature data, including: Performing feature splicing on the time series implicit function relationship and the time series feature data to obtain spliced ​​multimodal feature data; The EMA attention mechanism is used to apply exponential decay weights to the spliced ​​multimodal feature data to perform feature weighted fusion processing to obtain the target time series feature data.

4. The ice cover prediction model training method according to claim 3, characterized in that: The ice cover prediction model also includes a long short-term memory network, and the temporal implicit function relationship is obtained in the following way: The spatial implicit function relationship and the temporal feature data are input into the long short-term memory network to obtain the temporal implicit function relationship output by the long short-term memory network.

5. The ice prediction model training method according to any one of claims 1 to 4, characterized in that: The ice coverage prediction model also includes a fully connected layer, and the predicted ice coverage thickness corresponding to the predicted time is obtained by: The target time series feature data is input into the fully connected layer to obtain the predicted ice thickness corresponding to the time to be predicted output by the fully connected layer.

6. The ice prediction model training method according to any one of claims 1 to 4, characterized in that: The ice cover prediction model further includes a feature extraction module, and the feature extraction of the ice cover data to obtain time series feature data includes: The ice coverage data is input into the feature extraction module to obtain the time series feature data output by the feature extraction module.

7. A method for predicting icing of a transmission line, characterized in that: include: In response to an ice-covering prediction instruction for a power transmission line, acquiring ice-covering data of the power transmission line, the ice-covering data including micro-meteorological data, terrain data and ice-covering monitoring data; The icing data is input into an icing prediction model for icing prediction to obtain the icing thickness of the transmission line corresponding to the time to be predicted, wherein the icing prediction model is trained using the icing prediction model training method according to any one of claims 1 to 6.

8. An icing prediction model training device, characterized in that: The icing prediction model includes a residual network, and the icing prediction model training device includes: A first acquisition module is used to acquire ice coverage data of the power transmission line at consecutive moments, wherein the ice coverage data includes micro-meteorological data, terrain data and ice coverage monitoring data; A feature extraction module is used to extract features from the ice cover data to obtain time series feature data, wherein the time series feature data includes micro-meteorological features, terrain features and ice cover monitoring features at consecutive moments; A second acquisition module is used to input the time series feature data into the residual network to obtain the spatial implicit function relationship between the micro-meteorological features, the terrain features and the ice cover monitoring features output by the residual network; A fusion processing module is used to perform feature fusion processing based on the time series feature data and the time series implicit function relationship by using an exponential moving average EMA attention mechanism to obtain target time series feature data, wherein the time series implicit function relationship is obtained based on the spatial implicit function relationship and the time series feature data; The adjustment module is used to adjust the parameters of the ice prediction model based on the predicted ice thickness corresponding to the time to be predicted and the reference ice thickness to obtain the trained ice prediction model, wherein the predicted ice thickness corresponding to the time to be predicted is obtained based on the target time series feature data.

9. A transmission line icing prediction device, characterized in that: include: An acquisition module, configured to acquire icing data of the transmission line in response to an icing prediction instruction for the transmission line, wherein the icing data includes micro-meteorological data, terrain data and icing monitoring data; A prediction module is used to input the icing data into an icing prediction model for icing prediction to obtain the icing thickness of the transmission line corresponding to the time to be predicted, wherein the icing prediction model is trained using the icing prediction model training method according to any one of claims 1 to 6.

10. An electronic device, characterized in that: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 7.

11. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer program instructions, and when the computer program instructions are executed, the method according to any one of claims 1 to 7 is implemented.

12. A computer program product, comprising a computer program, characterized in that When the computer program is executed, the method according to any one of claims 1 to 7 is implemented.

Images