Power transmission line icing prediction method, system and equipment based on satellite data assimilation and medium

Abstract

The invention discloses a power transmission line icing prediction method, system and device based on satellite data assimilation and a medium, and relates to the technical field of ice and snow monitoring, and the method comprises the steps: obtaining satellite microwave radiance observation data, fusing the observation data and a numerical mode background field through data assimilation, generating grid meteorological data, and carrying out the prediction of the icing of a power transmission line. The method comprises the following steps: processing gridding meteorological data by using a spatial-temporal feature joint modeling technology, reconstructing meteorological data of continuous time steps into a spatial-temporal coupled four-dimensional tensor, synchronously extracting spatial-temporal dimension dynamic evolution features through a three-dimensional convolution kernel, and fusing historical states and current features based on a gating mechanism to obtain a spatial-temporal feature fusion model; and outputting a spatial distribution prediction result of the icing thickness of the power transmission corridor. According to the method, the high-precision meteorological field is obtained, the time-space dynamic characteristics of icing evolution are synchronously extracted, the key meteorological factors are screened in combination with an improved gating mechanism and attention, and accurate closed-loop simulation of power transmission corridor icing is realized based on iterative prediction of physical constraints, so that the accuracy and reliability of icing early warning are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ice and snow monitoring, and in particular to a power transmission line icing prediction method, system, device and medium based on satellite data assimilation. BACKGROUND

[0002] In recent years, with the intensification of global climate change, extreme weather events have occurred frequently, and power transmission line icing has become one of the main natural disasters threatening the safe operation of the power system. Traditional power transmission line icing prediction techniques mainly rely on ground meteorological observation station data and empirical statistical models, such as regression analysis or machine learning methods based on temperature, humidity, wind speed and other parameters. However, ground observation stations are sparsely distributed and difficult to cover complex terrain areas, resulting in insufficient data representation. At the same time, empirical models are difficult to accurately depict the complex interaction of microphysical processes and terrain effects in the atmospheric boundary layer. With the development of satellite remote sensing technology, passive remote sensing devices such as microwave radiometers can provide large-scale and continuous meteorological observation data. However, existing satellite data applications are mostly limited to simple inversion or qualitative analysis, and have not yet achieved deep integration with numerical models. In addition, traditional time series prediction models such as LSTM and GRU have limited feature extraction and long-range dependence capture capabilities when dealing with high-dimensional spatio-temporal data, making it difficult to meet the fine-grained prediction needs of power transmission corridors.

[0003] The existing technology has the following significant deficiencies: First, in terms of data acquisition, the spatial resolution of ground observation networks is insufficient and the update frequency is low, which cannot capture the microscale meteorological element changes in the key icing formation area. Although satellite data covers a wide range, the raw radiance data without assimilation has systematic errors and noise interference, which will lead to accuracy loss if directly used for icing prediction. Second, in terms of model construction, existing methods mostly use two-dimensional convolution or simple time series networks to separate spatial and temporal features, ignoring the three-dimensional dynamic evolution characteristics of meteorological elements in the icing process, resulting in insufficient physical characterization of the icing growth mechanism. More importantly, traditional prediction models lack adaptive fusion mechanisms for historical states and current features, making it difficult to accurately describe the nonlinear characteristics of the icing accumulation process. Their prediction time is usually no more than 24 hours and the spatial resolution is coarse, which cannot provide fine-grained decision support for power transmission line deicing. SUMMARY

[0004] In view of the above existing problems, the present application provides a power transmission line icing prediction method, system, device and medium based on satellite data assimilation, to solve the problem of energy waste caused by selecting the wrong energy storage method in the prior art.

[0005] To solve the above technical problems, a power transmission line icing prediction method based on satellite data assimilation is proposed, which includes,

[0006] Satellite microwave radiance observation data are acquired, and grid meteorological data are generated by data assimilation of the observation data and a numerical model background field; the grid meteorological data are processed by using a spatiotemporal feature joint modeling technology, the meteorological data at continuous time steps are reconstructed into a spatiotemporal coupling four-dimensional tensor, and spatiotemporal dimension dynamic evolution features are synchronously extracted through a three-dimensional convolution kernel; a historical state and a current feature are fused based on a gating mechanism, and a spatial distribution prediction result of power transmission corridor icing thickness is output.

[0007] As a preferred scheme of the power transmission line icing prediction method based on satellite data assimilation, the ring data assimilation comprises minimizing an original radiance input target function by using a direct assimilation strategy;

[0008] A radiance and model variable physical mapping is established through a radiative transfer model;

[0009] Multi-source observation data are subjected to spatiotemporal standardization and multilevel quality screening.

[0010] As a preferred scheme of the power transmission line icing prediction method based on satellite data assimilation, the direct assimilation strategy comprises dynamically correcting a scanning channel system error by using a statistical model;

[0011] A background field outlier detection and terrain sheltering filtering three-level criterion is used;

[0012] A standardized observation increment vector is generated to constrain a model state update.

[0013] As a preferred scheme of the power transmission line icing prediction method based on satellite data assimilation, the spatiotemporal feature joint modeling technology comprises integrating continuous time data along a time dimension sliding window;

[0014] A spatiotemporal coupling feature flow is transmitted through a vertical fusion branch;

[0015] Gating calculation and a three-dimensional convolution kernel are jointly applied to feature extraction.

[0016] As a preferred scheme of the power transmission line icing prediction method based on satellite data assimilation, the gating mechanism comprises inputting a pure time sequence feature transmitted by an input branch to a hidden state memory, outputting and inputting data of a vertical fusion branch to jointly participate in three-dimensional convolution gating calculation, and weighting and screening output features through a time channel attention mechanism;

[0017] The gating mechanism formula is represented as:

[0018]

[0019] W 3Dis the 3D convolution kernel weight, * is the spatio-temporal convolution operation, i t is the input gate value at time t, f t is the forget gate value at time t, o t is the output gate value at time t, g t is the candidate state generation gate, X t is the input meteorological data at time t, H t-1 is the hidden state at the previous time, M t is the vertical fusion feature flow, and sigma is a Sigmoid activation function, and b is a bias term.

[0020] As a preferred scheme of the power transmission line icing prediction method based on satellite data assimilation provided by the application, wherein: the time channel attention mechanism comprises: global average pooling and global maximum pooling are performed on the channel dimension spliced time sequence feature, a channel level weight matrix is generated through a shared weight multi-layer perception, and the weighted feature is input into a single-layer three-dimensional deconvolution network to reconstruct the future icing state;

[0021] The time channel attention mechanism formula is expressed as:

[0022] F = Concat (H1, H2,..., H T ) Z = sigma (MLP (GAP (F)) + MLP (GMP (F)))

[0023]

[0024] Wherein, F is the channel dimension spliced time sequence feature tensor, H1, H2,..., H T is the hidden state output by the ST-LSTM at time 1-T, T is the total time step, Concat is a channel dimension splicing operation, GAP is global average pooling, GMP is global maximum pooling, MLP is a shared weight multi-layer perception, sigma is a Sigmoid activation function, Z is a channel attention weight matrix, is channel multiplication, is the weighted time sequence feature tensor.

[0025] As a preferred scheme of the power transmission line icing prediction method based on satellite data assimilation provided by the application, wherein: the single-layer three-dimensional deconvolution network reconstructs the future icing state, which comprises: the current icing thickness distribution is fed back to the input end in real time as a three-dimensional boundary condition, and the nonlinear action of the inversion layer and the water vapor flux is analyzed through a spatio-temporal coupling weight matrix to generate a 72-hour icing spatial distribution prediction sequence in a cycle.

[0026] The beneficial effects of the preferred technical scheme are that the future icing state is reconstructed through a single-layer three-dimensional deconvolution network, the spatial resolution of the prediction result is maintained, the detail loss caused by traditional upsampling is avoided, the effect of outputting an icing distribution map with the same resolution as the input is achieved, and the refined operation and maintenance demand is met; and the current icing thickness distribution is fed back to the input end as a three-dimensional boundary condition in real time, the dynamic self-correction of the prediction process is realized, and the long-time prediction accuracy is significantly improved.

[0027] As a preferred scheme of the power transmission line icing prediction system based on satellite data assimilation, it comprises a satellite data assimilation module, a spatio-temporal feature joint modeling module, a time channel attention module and an iterative extrapolation prediction module.

[0028] The satellite data assimilation module is used to generate high-precision gridded meteorological data by fusing satellite raw radiance observation data and numerical model background fields, and to avoid inversion error transmission by using a direct assimilation strategy, and to ensure data reliability in combination with three-level quality control.

[0029] The spatio-temporal feature joint modeling module is used to reconstruct the meteorological grid data at consecutive time steps into a four-dimensional tensor, to synchronously extract spatio-temporal coupling features through an improved ST-LSTM unit, and to explicitly capture short-time mutations and regional correlations of icing evolution through a double-branch design combined with a 3D convolution gating mechanism.

[0030] The time channel attention module is used to concatenate the channel dimension of the historical time series features, to compress the features through double-path pooling, and to learn the channel-level time correlation by using a shared weight multi-layer perception machine to generate a weight matrix to adaptively strengthen key meteorological factors and suppress noise data.

[0031] The iterative extrapolation prediction module is used to input the weighted spatio-temporal features into a single-layer 3D deconvolution network to perform closed-loop prediction with a step of 6 hours, and to analyze the nonlinear interaction of inversion layer and water vapor flux through a spatio-temporal coupling weight matrix to feed back the current icing distribution as a three-dimensional boundary condition in real time, so as to realize the iterative extrapolation of the icing spatial distribution within 72 hours.

[0032] A computer device comprises a memory and a processor, the memory stores a computer program, and the processor implements the steps of the method for predicting power transmission line icing based on satellite data assimilation when executing the computer program.

[0033] A computer readable storage medium stores a computer program, and the processor implements the steps of the method for predicting power transmission line icing based on satellite data assimilation when executing the computer program.

[0034] The present application has the following beneficial effects: the satellite data assimilation technology directly fuses the original radiance and the numerical model background field, generates high-precision gridded meteorological data by using dynamic error correction and three-level quality control, solves the problems of sparse traditional station observation and accumulated inversion error, and improves the spatial coverage of the meteorological field; the spatiotemporal feature joint modeling technology is used to reconstruct the continuous meteorological data into a four-dimensional tensor, the spatiotemporal dynamic features are synchronously extracted by using a three-dimensional convolution kernel, the spatial structure information is explicitly transmitted by combining the vertical fusion branch, the training efficiency of the traditional model is improved, and the sudden freezing rain can be responded within 30 minutes; based on the improved ST-LSTM gating mechanism, the physical parameters of the icing accumulation process are quantitatively controlled by using the input gate and the forgetting gate, the water vapor channel weight in the key period of the cold wave is improved by combining the double-path pooling screening of the time channel attention mechanism; by using the iterative extrapolation prediction framework, the nonlinear action of the inversion layer and the water vapor flux is analyzed in a closed-loop feedback manner combined with the spatiotemporal coupling weight matrix, so that the prediction error is reduced, and reliable technical support is provided for the disaster prevention decision of the power grid. BRIEF DESCRIPTION OF DRAWINGS

[0035] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0036] Figure 1 The overall flowchart of the power transmission line icing prediction method based on satellite data assimilation provided by an embodiment of the present application is shown.

[0037] Figure 2 The system scheme flowchart of the power transmission line icing prediction system based on satellite data assimilation provided by an embodiment of the present application is shown. DETAILED DESCRIPTION

[0038] In order to make the above-mentioned objects, features and advantages of the present application more apparent and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should be within the protection scope of the present application.

[0039] Embodiment 1, refer to Figure 1 For an embodiment of the present application, a power transmission line icing prediction method based on satellite data assimilation is provided, which comprises:

[0040] S100: Obtain satellite microwave radiance observation data, and generate gridded meteorological data by data assimilation of the observation data and a numerical model background field.

[0041] S200: Process the gridded meteorological data using a spatiotemporal feature joint modeling technique, reconstruct the meteorological data at consecutive time steps into a spatiotemporal coupled four-dimensional tensor, and synchronously extract dynamic evolution features in the spatiotemporal dimensions by a three-dimensional convolution kernel.

[0042] S300: Output a spatial distribution prediction result of the icing thickness of the power transmission corridor based on a gating mechanism that fuses historical states and current features.

[0043] It should be noted that obtaining satellite microwave radiance observation data and using data assimilation to fuse the observation data and a numerical model background field achieves generation of high-precision gridded meteorological data, overcoming the defects of sparse traditional ground observation sites and large noise of satellite raw data; through direct assimilation strategy, radiation transmission model, spatiotemporal standardization and quality screening, four-dimensional tensor reconstruction and three-dimensional convolution, gating mechanism, spatiotemporal convolution, time channel attention mechanism, shared weight multilayer perception, single-layer three-dimensional deconvolution network, real-time feedback of current icing thickness, spatiotemporal coupling weight matrix analysis, and cyclic generation of 72-hour prediction sequence, optimal fusion of observation data and numerical model, nonlinear physical mapping, abnormal data elimination, spatiotemporal dynamic evolution feature extraction, adaptive modeling of the icing accumulation process, deep analysis of the three-dimensional meteorological structure, key time sequence feature screening, nonlinear modeling of channel relationships, high-resolution retention of prediction results, dynamic self-correction, explicit modeling of the icing physical mechanism, and long-time-scale continuous prediction are achieved, which significantly improves the precision, reliability, comprehensiveness, physical consistency, mechanism matching, feature quality, spatial details, long-time accuracy, interpretability, and anti-icing decision support capability of the icing prediction, effectively reducing the risk of power outages caused by ice disasters.

[0044] Embodiment 2, with reference to Figure 1 As a second embodiment of the present application, the embodiment provides a power line icing prediction method based on satellite data assimilation, comprising:

[0045] In step S100, the obtaining of the satellite microwave radiance observation data comprises obtaining microwave radiance observation data of a Fengyun-3 satellite MWHS-2 detector at 183.31 GHz and 118.75 GHz frequency bands.

[0046] In the present application embodiment, in step S100, the data assimilation comprises steps S101-S105:

[0047] S101: Fusion of a 0.25°×0.25° resolution GDAS / FNL numerical model background field using a direct assimilation strategy.

[0048] S102: Establishing a physical mapping between the radiance and the model variables through a CRTM radiative transfer model;

[0049] S103: Correcting the scanning channel system error by using a dynamic statistical model;

[0050] S104: Performing three-level quality control of background field outlier detection, channel consistency verification, and terrain shielding filtering;

[0051] S105: Generating 3 km gridded temperature, humidity, and wind speed meteorological data based on a three-dimensional variational scheme incremental form objective function minimization; the objective function is defined as:

[0052]

[0053] wherein x is an analysis field, x b is a background field, B is a background field error covariance matrix, y is an observation field, H is an observation operator of a nonlinear operator of an observation space, and R is an observation field error covariance matrix.

[0054] It should be noted that the three-dimensional variational scheme incremental form objective function minimization is expressed as:

[0055] x = x b + δx

[0056] d = y - H(x b )

[0057]

[0058] wherein δx is an analysis increment, δx b is a background field increment initial estimate, T is a transpose operation, d is an observation increment vector, and H(x b ) is a simulation value of the observation operator on the background field.

[0059] In an optional implementation, the data assimilation further includes generating a 50-member ensemble simulation atmospheric state probability distribution, calculating a Kalman gain by comparing satellite brightness temperature observations with each member simulation brightness temperature, and updating an ensemble analysis field through linear minimum variance estimation, and outputting probabilistic meteorological grid data.

[0060] In another optional implementation, the data assimilation further includes training a U-Net network to learn a mapping relationship from historical satellite brightness temperature to meteorological field, inputting real-time brightness temperature data, and directly outputting temperature / humidity grid by the network, and introducing a physical constraint loss function (thermodynamic equation residual term) to generate high-resolution meteorological field, and skipping traditional assimilation iterative calculation.

[0061] In the embodiments of the present application, in step S200, the spatio-temporal feature joint modeling technology comprises steps S201-S203:

[0062] S201: cutting meteorological grid data with 20 time steps x 5 sliding steps, inputting an ST-LSTM unit, a left branch LSTM transmitting time sequence features, and a right branch 3D convolution extracting spatial features;

[0063] S202: transmitting spatio-temporal coupling feature flow through a vertical fusion branch;

[0064] S203: applying a gating calculation and a three-dimensional convolution kernel to feature extraction.

[0065] Further, the process of the ST-LSTM recurrent unit improved based on the 3D convolution is represented as:

[0066] C t ′=tanh(W c *[X t ,M t ])

[0067] i t =σ(W i *[H t-1 ,C t-1 ,M t ])

[0068] f t =σ(W f *[H t-1 ,C t-1 ,M t ])

[0069] C t =f t ⊙C t-1 +i t ⊙C t ′

[0070] o t =σ(W o *[H t-1 ,C t ,M t ])

[0071] H t =o t ⊙tanh(W h *[C t ,M t ])

[0072] Wherein, * is a 3D convolution operation, σ is a Sigmoid activation function, ⊙ is a Hadamard product, tanh is a hyperbolic tangent function, W cand W h is a three-dimensional convolution kernel weight matrix, C t is a candidate memory state, i t is an input gate, f t is a forget gate, C t is an updated memory state, o t is an output gate, H t is a hidden state, X t is a current time input, H t-1 is a previous time hidden state, C t-1 is a previous time memory state, M t is a vertically fused feature stream, W i is a weight matrix of an input gate, W f is a weight matrix of an update gate, W o is a weight matrix of an output gate.

[0073] In an optional implementation, in step S200, the spatio-temporal feature joint modeling technique further includes reconstructing the meteorological grid into a [batch, time step, height, width, channel] tensor, replacing the fully connected operation with a 2D convolution within the ConvLSTM unit, sliding the convolution kernel to extract local spatial features, using LSTM gate memory time evolution, and outputting a jump connection to retain shallow details.

[0074] In another optional implementation, in step S200, the spatio-temporal feature joint modeling technique can further include extracting short-time (less than 6 hours) spatial features using a 3D CNN, inputting the feature sequence into a time Transformer encoder, using a multi-head attention mechanism to capture long-term dependence (more than 24 hours), and using a decoder to fuse spatio-temporal features to output an icing probability map.

[0075] In step S300, the gating mechanism includes steps S301-S303.

[0076] S301: The input branch delivers pure time sequence features to the hidden state memory.

[0077] S302: The vertically fused branch outputs are used together with the input data to participate in three-dimensional convolution gating calculation.

[0078] S303: The output features are weighted and filtered through a time channel attention mechanism.

[0079] The gating mechanism is represented by the following formula:

[0080]

[0081] wherein, W 3D is a 3D convolution kernel weight, * is a spatio-temporal convolution operation, i tf is the input gate value at time t. t Let o be the forgetting gate value at time t. t Let g be the output gate value at time t. t X is a gate for generating candidate states. t H represents the input meteorological data at time t. t-1 M is the hidden state from the previous moment. t σ represents the vertically fused feature flow, where σ is the Sigmoid activation function and b is the bias term.

[0082] Furthermore, in this embodiment of the application, in step S303, the temporal channel attention mechanism includes steps A1 to A3:

[0083] A1: Perform global average pooling and global max pooling on the temporal features concatenated along the channel dimensions;

[0084] A2: Generate a channel-level weight matrix using a shared-weight multilayer perceptron;

[0085] A3: Input weighted features into a single-layer 3D deconvolutional network to reconstruct the future icing state;

[0086] The formula for the time-channel attention mechanism is expressed as:

[0087] F = Concat(H1,H2,…,H) T )

[0088] Z=σ(MLP(GAP(F))+MLP(GMP(F)))

[0089]

[0090] Among them, the temporal feature tensors concatenated along the F-channel dimension, H1, H2, ..., H T Let Z represent the hidden states output by the ST-LSTM at times 1 to T, where T is the total number of time steps, Concat is the concatenation operation along the channel dimension, GAP is global average pooling, GMP is global max pooling, MLP is a multilayer perceptron with shared weights, σ ​​is the sigmoid activation function, and Z is the channel attention weight matrix. For channel multiplication, This is the weighted temporal feature tensor.

[0091] In an optional implementation, in step S303, the temporal channel attention mechanism further includes calculating the feature autocorrelation matrix along the time dimension, generating time step attention weights by Softmax normalization, and weighted fusing the features of each time step to directly input the fully connected layer to predict the ice thickness.

[0092] In another optional implementation, in step S303, the time channel attention mechanism can further include modeling the meteorological grid points as graph nodes (node = grid point), defining edges connecting adjacent grid points within 50 km, learning the attention coefficients between nodes by the GAT layer, aggregating the spatial neighborhood information, and outputting the weighted graph features.

[0093] In the embodiments of the present application, in step A3, the single-layer three-dimensional deconvolution network reconstructing the future icing state includes steps B1-B3:

[0094] B1: feeding back the current icing thickness distribution as a three-dimensional boundary condition to the input end in real time;

[0095] B2: analyzing the nonlinear action of the inversion layer and the water vapor flux through a spatiotemporal coupling weight matrix;

[0096] B3: cyclically generating a 72-hour icing spatial distribution prediction sequence.

[0097] In an optional implementation, in step A3, the single-layer three-dimensional deconvolution network reconstructing the future icing state further includes predicting the icing thickness change rate by using a CNN, inputting the change rate into a physical equation, learning the coefficients by the network, using the equation as a residual constraint, and iteratively calculating the future icing state.

[0098] In another optional implementation, in step A3, the single-layer three-dimensional deconvolution network reconstructing the future icing state can further include inputting the current meteorological field into a generator, outputting a 6-hour future icing distribution, comparing the spatiotemporal continuity of the real sequence and the generated sequence by a discriminator, optimizing the iterative prediction ability of the generator through adversarial learning, and cyclically generating a 72-hour icing animation sequence.

[0099] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application but not limit the present application. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalently replaced without departing from the spirit and scope of the present application, and all modifications and equivalent replacements should be included in the scope of the claims of the present application.

[0100] Embodiment 3, refer to Figure 2 As a third embodiment of the present application, the embodiment provides a power line icing prediction system based on satellite data assimilation, including a satellite data assimilation module, a spatiotemporal feature joint modeling module, a time channel attention module, and an iterative extrapolation prediction module.

[0101] The satellite data assimilation module is configured to generate high-precision gridded meteorological data by fusing satellite raw radiance observation data and numerical model background fields, avoid inversion error transmission by using a direct assimilation strategy, and ensure data reliability by combining three-level quality control.

[0102] The spatio-temporal feature joint modeling module is configured to reconstruct the meteorological grid data of continuous time steps into a four-dimensional tensor, synchronously extract spatio-temporal coupling features through an improved ST-LSTM unit, and explicitly capture short-time mutations and regional relevance of icing evolution through a double-branch design combined with a 3D convolution gating mechanism.

[0103] The time channel attention module is configured to perform channel dimension splicing on historical time sequence features, compress features through double-path pooling, and generate a weight matrix to adaptively strengthen key meteorological factors by learning channel-level time sequence correlation through a shared weight multi-layer perception machine, thereby suppressing noise data.

[0104] The iterative extrapolation prediction module is configured to input the weighted spatio-temporal features into a single-layer 3D deconvolution network to perform closed-loop prediction in a rolling manner with a step of 6 hours, and analyze the nonlinear interaction between inversion layer and water vapor flux through a spatio-temporal coupling weight matrix, thereby feeding back the current icing distribution as a three-dimensional boundary condition in real time to realize iterative extrapolation of the icing spatial distribution within 72 hours.

[0105] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present application, and all should be covered in the scope of the claims of the present application.

[0106] Embodiment 4, which is different from the first three embodiments, is as follows:

[0107] If the functions are realized in the form of software function units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the present application or the parts that essentially contribute to the prior art or parts of the technical solutions can be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0108] The logic and / or steps represented in the flow diagrams and / or otherwise described herein, for example, can be considered as a sequence of executable instructions, and can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. Just by way of example, a computer-readable medium can be any device or apparatus that can store and convey instructions for execution by the instruction execution system, apparatus, or device. With respect to the present description, a "computer-readable medium" can be any means that can contain, store, communicate, propagate or transport the program for use by or in connection with the instruction execution system, apparatus, or device.

[0109] More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection (electronic) having one or more wires, a portable computer diskette (magnetic), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can also be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, for example, via an optical scanner, then compiled, interpreted, or otherwise processed, and stored in a computer memory in a form that can be later executed by a computer.

[0110] It should be understood that aspects of the application can be implemented in hardware, software, firmware or combinations thereof. In the above embodiments, various steps or methods can be implemented in software or firmware that is stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, and in another embodiment, any of the following technologies can be used: a combination of discrete logic circuits having logic gates for implementing logic functions upon an application of data signals, application specific integrated circuits having logic gates, field programmable gate arrays (FPGA), or other components, in any suitable combination.

Claims

1. A method for predicting icing on a power transmission line based on assimilation of satellite data, characterized in that: The application relates to a satellite data assimilation method and system for power transmission corridor icing prediction. The application comprises the following steps: Satellite microwave radiation observation data are acquired, and the observation data are fused with a background field of a numerical model through data assimilation to generate grid meteorological data; The grid meteorological data are processed by using a space-time feature joint modeling technology, the meteorological data at continuous time steps are reconstructed into a space-time coupled four-dimensional tensor, and a three-dimensional convolution kernel is used to synchronously extract dynamic evolution features in the space-time dimensions; 2. The satellite data assimilated power line icing prediction method of claim 1, wherein: Based on a gating mechanism, historical states and current features are fused to output a spatial distribution prediction result of the power transmission corridor icing thickness. The ring data assimilation comprises the following steps: A direct assimilation strategy is used to minimize an original radiation input target function process; 3. The satellite data assimilated power line icing prediction method of claim 2, wherein: A radiation transmission model is used to establish a physical mapping relationship between the radiation and model variables; Multi-source observation data are subjected to space-time standardization and multi-level quality screening. The direct assimilation strategy comprises the following steps:

4. The method for power line icing prediction based on satellite data assimilation according to claim 3, characterized in that: A dynamic statistical model is used to correct system errors of scanning channels; Three-level criteria of background field outlier detection and terrain sheltering filtering are used; A standardized observation increment vector is used to constrain the model state updating.

5. The method for power line icing prediction based on satellite data assimilation according to claim 4, characterized in that: The space-time feature joint modeling technology comprises the following steps: Sliding windows are used to integrate continuous time data along the time dimension; where W 3D is the 3D convolution kernel weight, * is the spatio-temporal convolution operation, i t is the input gate value at time t, f t is the forget gate value at time t, o t is the output gate value at time t, g t is the candidate state generation gate, X t is the input meteorological data at time t, H t-1 is the hidden state at the previous time, M t is the vertical fusion feature stream, σ is the Sigmoid activation function, and b is the bias term.

6. The satellite-based data assimilated power line icing prediction method of claim 5, wherein: A vertical fusion branch is used to transmit space-time coupled feature flow; Gating calculation and a three-dimensional convolution kernel are combined to be applied to feature extraction. F = Concat(H1, H2,..., H T ) The gating mechanism comprises the following steps: where F is the time sequence feature tensor spliced along the channel dimension, H1, H2, …, H T is the hidden state output by ST-LSTM at time 1 ~ T, T is the total number of time steps, Concat is the concatenation operation along the channel dimension, GAP is the global average pooling, GMP is the global maximum pooling, MLP is the multi-layer perceptron with shared weights, σ is the Sigmoid activation function, Z is the channel attention weight matrix, is the channel multiplication, is the weighted time sequence feature tensor.

7. The satellite-based data assimilated power line icing prediction method of claim 6, wherein: Pure time sequence features are transmitted to an implicit state memory through an input branch; 8. A power line icing prediction system based on satellite data assimilation, which applies the power line icing prediction method based on satellite data assimilation according to any one of claims 1 to 7, characterized in that, Vertical fusion branch output and input data are used to participate in three-dimensional convolution gating calculation, and a time channel attention mechanism is used to weight and screen output features; The gating mechanism is expressed by the following formula: The time channel attention mechanism comprises the following steps: Global average pooling and global maximum pooling are performed on time sequence features spliced in a channel dimension, a channel level weight matrix is generated through a shared weight multi-layer perception, and weighted features are input into a single-layer three-dimensional deconvolution network to reconstruct future icing states; The time channel attention mechanism is expressed by the following formula: Z = sigma (MLP (GAP (F)) + MLP (GMP (F))) The single-layer three-dimensional deconvolution network reconstructing the future icing state comprises the following steps: A current icing thickness distribution is used as a three-dimensional boundary condition to be fed back to an input end in real time, and a space-time coupled weight matrix is used to analyze the nonlinear action of an inversion temperature layer and a water vapor flux, so that an icing spatial distribution prediction sequence within 72 hours is generated in a cycle. The application comprises a satellite data assimilation module, a space-time feature joint modeling module, a time channel attention module and an iterative extrapolation prediction module; The satellite data assimilation module is used to generate high-precision grid meteorological data by fusing satellite original radiation observation data and a background field of a numerical model, avoid inversion error transmission by using a direct assimilation strategy, and ensure data reliability by combining three-level quality control; The space-time feature joint modeling module is used to reconstruct grid data at continuous time steps into a four-dimensional tensor, synchronously extract space-time coupled features through an improved ST-LSTM unit, and explicitly capture short-time mutations and regional correlations of icing evolution through a double-branch design and a 3D convolution gating mechanism. The time channel attention module is configured to perform channel dimension splicing on historical time sequence features, compress the features through double-path pooling, learn channel-level time sequence correlation by using a shared weight multi-layer perception, generate a weight matrix to adaptively strengthen key meteorological factors, and suppress noise data. The iterative extrapolation prediction module is configured to input the weighted spatio-temporal features into a single-layer 3D deconvolution network, perform closed-loop prediction with a step of 6 hours, analyze the nonlinear interaction between inversion layer and water vapor flux through a spatio-temporal coupling weight matrix, and feed back the current icing distribution as a three-dimensional boundary condition in real time, so as to realize the iterative extrapolation of the icing spatial distribution within 72 hours. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8 when the computer program is executed by the processor. The processor executes the computer program to realize the steps of the power transmission line icing prediction method based on satellite data assimilation in any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to realize the steps of the power transmission line icing prediction method based on satellite data assimilation in any one of claims 1 to 7.

Images