Power transmission line icing thickness prediction method and device based on LSTM model

Abstract

The invention discloses a power transmission line icing thickness prediction method and device based on an LSTM model, a storage medium and computer equipment, and the method comprises the steps: obtaining a meteorological data sequence of a preset time period corresponding to a target power transmission line, and enabling the meteorological data sequence to be constructed based on the micrometeorological data of a plurality of sampling time points, the micrometeorological data at each sampling time point comprises a plurality of target meteorological index values, and the correlation degree between the target meteorological index values and the icing thickness of the power transmission line is greater than a preset correlation degree threshold value; and according to the meteorological data sequence, generating a meteorological feature vector of the target power transmission line, and based on the meteorological feature vector, through a pre-trained LSTM model, outputting a predicted icing thickness of the target power transmission line. According to the invention, through the LSTM model, historical time sequence meteorological data are fully utilized, a complex nonlinear dynamic relationship between a plurality of target meteorological index values and the icing thickness of the power transmission line is intelligently captured, and an icing thickness prediction value in a future time period can be output in advance.

Description

Technical Field

[0001] This application relates to the field of transmission line icing prediction technology, and in particular to a method and device for predicting transmission line icing thickness based on an LSTM model, as well as a storage medium and computer equipment. Background Technology

[0002] Icing on transmission lines is a physical phenomenon determined by a combination of meteorological factors, including air temperature, relative humidity, and air velocity. Excessive icing can overload conductors, leading to line breakage and tower collapse; ice detachment can cause conductor jumping, resulting in inter-conductor discharge breakdown and unbalanced tension breakage; eccentric or uneven icing can cause conductor galloping. Therefore, timely determination of icing thickness and prevention of major disasters is a crucial issue that urgently needs to be addressed in the field of transmission line disaster prevention and mitigation.

[0003] Current research on icing of transmission lines mainly focuses on de-icing jumps, and often only analyzes the dynamic response state after de-icing occurs, lacking a method to accurately predict the icing thickness of transmission lines. Summary of the Invention

[0004] In view of this, this application provides a method and device, storage medium and computer equipment for predicting the icing thickness of transmission lines based on an LSTM model. By fully utilizing historical time-series meteorological data through the LSTM model, the complex nonlinear dynamic relationship between multiple target meteorological index values ​​and the icing thickness of transmission lines is intelligently captured. This enables the output of the predicted icing thickness for future periods in advance, providing key decision-making basis for proactively implementing disaster prevention measures such as de-icing and ice removal, and realizing the transformation from passive disaster response to proactive early warning and prevention.

[0005] According to one aspect of this application, a method for predicting icing thickness of transmission lines based on an LSTM model is provided, comprising: Obtain a meteorological data sequence for a preset time period corresponding to the target transmission line. The meteorological data sequence is constructed based on micro-meteorological data from multiple sampling time points. The micro-meteorological data from each sampling time point includes multiple target meteorological index values. The correlation between the target meteorological index and the icing thickness of the transmission line is greater than a preset correlation threshold. Based on the meteorological data sequence, a meteorological feature vector of the target transmission line is generated, and based on the meteorological feature vector, the predicted icing thickness of the target transmission line is output through a pre-trained LSTM model.

[0006] According to another aspect of this application, a transmission line icing thickness prediction device based on an LSTM model is provided, comprising: The data acquisition module is used to acquire meteorological data sequences for a preset time period corresponding to the target transmission line. The meteorological data sequences are constructed based on micro-meteorological data from multiple sampling time points. Each sampling time point's micro-meteorological data includes multiple target meteorological index values. The correlation between the target meteorological index and the icing thickness of the transmission line is greater than a preset correlation threshold. The icing thickness prediction module is used to generate a meteorological feature vector of the target transmission line based on the meteorological data sequence, and output the predicted icing thickness of the target transmission line based on the meteorological feature vector and a pre-trained LSTM model.

[0007] According to another aspect of this application, a storage medium is provided that stores a computer program thereon, which, when executed by a processor, implements the above-described method for predicting the icing thickness of transmission lines based on an LSTM model.

[0008] According to another aspect of this application, a computer device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor executes the program to implement the above-described method for predicting the icing thickness of transmission lines based on the LSTM model.

[0009] By utilizing the above technical solutions, this application provides a method and device for predicting the icing thickness of transmission lines based on an LSTM model, a storage medium, and a computer device. Through the LSTM model, it makes full use of historical time-series meteorological data to intelligently capture the complex nonlinear dynamic relationship between multiple target meteorological index values ​​and the icing thickness of transmission lines. This enables the output of predicted icing thickness values ​​for future periods in advance, providing key decision-making basis for proactively implementing disaster prevention measures such as de-icing and ice removal, and realizing the transformation from passive disaster response to proactive early warning and prevention.

[0010] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0011] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A flowchart illustrating a method for predicting icing thickness of transmission lines based on an LSTM model, as provided in an embodiment of this application, is shown. Figure 2 A schematic diagram illustrating the effect of predicting icing thickness on transmission lines according to an embodiment of this application is shown. Figure 3 This paper shows a schematic diagram of the structure of a transmission line icing thickness prediction device based on an LSTM model provided in an embodiment of this application; Figure 4 A schematic diagram of the device structure of a computer device provided in an embodiment of this application is shown. Detailed Implementation

[0012] The present application will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present application can be combined with each other.

[0013] This embodiment provides a method for predicting the icing thickness of transmission lines based on an LSTM model, such as... Figure 1 As shown, the method includes: Step 101: Obtain the meteorological data sequence for a preset time period corresponding to the target transmission line. The meteorological data sequence is constructed based on micro-meteorological data from multiple sampling time points. The micro-meteorological data from each sampling time point includes multiple target meteorological index values. The correlation between the target meteorological index and the icing thickness of the transmission line is greater than a preset correlation threshold.

[0014] Step 102: Generate a meteorological feature vector of the target transmission line based on the meteorological data sequence, and output the predicted icing thickness of the target transmission line based on the meteorological feature vector and a pre-trained LSTM model.

[0015] This application provides a method for predicting the icing thickness of transmission lines based on an LSTM (Long Short-Term Memory) model. First, the target transmission line needs to be identified. Then, a preset time period, such as the past week or month, is selected for the target transmission line. Within this preset time period, sampling is performed at certain time intervals (e.g., every hour or every half hour) to obtain micro-meteorological data of the area where the target transmission line is located at multiple sampling time points. These micro-meteorological data at sampling time points contain multiple target meteorological index values. These target meteorological indices are selected based on a correlation greater than a preset correlation threshold with the icing thickness of the transmission line. For example, target meteorological indices such as temperature, humidity, wind speed, and wind direction are closely related to the icing thickness of the transmission line. When these target meteorological indices are within a specific range, icing is more likely to occur or develop, and therefore they are selected as target meteorological indices. The meteorological data sequence constructed in this way can comprehensively and specifically reflect the meteorological factors that may affect the icing thickness of transmission lines, providing a rich data foundation for subsequent predictions.

[0016] After obtaining the meteorological data sequence, it can be processed to generate meteorological feature vectors. A meteorological feature vector is an abstract representation of the original meteorological data sequence, capable of extracting key features from it. Specifically, mathematical operations or feature extraction can be performed on multiple target meteorological index values ​​at each sampling time point, such as calculating the statistical mean, variance, maximum, and minimum values. Alternatively, more complex feature engineering methods, such as Principal Component Analysis (PCA), can be used to reduce the dimensionality of the original multidimensional meteorological data and extract the most representative feature combinations to form the meteorological feature vector.

[0017] After generating the meteorological feature vector through the above steps, it is input into the pre-trained LSTM model. LSTM is a special type of recurrent neural network (RNN) that has the ability to process sequential data and remember long-term dependencies, making it very suitable for processing time-series meteorological data. The pre-trained LSTM model has been trained on a large amount of historical meteorological data and corresponding icing thickness data, learning the complex mapping relationship between meteorological features and icing thickness. When a new meteorological feature vector is input, the LSTM model can perform a series of calculations and inferences on the input data based on its internally learned weights and parameters, ultimately outputting the predicted icing thickness of the target transmission line.

[0018] By applying the technical solution of this embodiment, and through the LSTM model, historical time-series meteorological data is fully utilized to intelligently capture the complex nonlinear dynamic relationship between multiple target meteorological index values ​​and the ice thickness of transmission lines. This enables the output of the predicted ice thickness for future periods, providing key decision-making basis for proactively implementing disaster prevention measures such as de-icing and ice removal, and realizing the transformation from passive disaster response to proactive early warning and prevention.

[0019] Optionally, before step 101, the method further includes: determining whether the predicted duration of icing thickness of the target transmission line is greater than or equal to a preset duration threshold; if so, determining an appropriate data collection time interval based on the predicted duration of icing thickness, and updating the preset time interval based on the appropriate data collection time interval, wherein the appropriate data collection time interval is greater than the data collection time interval used for conventional prediction.

[0020] In this embodiment, before acquiring the meteorological data sequence corresponding to a preset time period for the target transmission line, the predicted duration of icing thickness for the target transmission line can be determined first. Specifically, a preset duration threshold can be set, which can be determined comprehensively based on factors such as actual business needs, meteorological data change patterns, and model prediction capabilities. By determining whether the predicted duration of icing thickness is greater than or equal to this preset duration threshold, different prediction scenarios can be distinguished. When the predicted duration of icing thickness is short, the changes in meteorological data are relatively frequent and significant, and conventional data collection intervals can meet the prediction requirements; while when the predicted duration of icing thickness is long, the trend of meteorological data changes is relatively gentle, but sufficient data is still needed to reflect this trend, so special processing is required.

[0021] When the predicted duration of icing thickness for the target transmission line is determined to be greater than or equal to a preset duration threshold, an appropriate data collection interval can be determined based on the specific value of the predicted icing thickness duration. This appropriate data collection interval is longer than the data collection interval used in conventional forecasting. For long-term forecasting, excessively frequent data collection not only increases the burden of data storage and processing but may also result in data with minimal variation, offering limited assistance to long-term forecasting. Appropriately increasing the data collection interval can reduce the amount of data and improve data processing efficiency while ensuring the capture of the main trends in meteorological data. Simultaneously, based on the determined appropriate data collection interval, the preset time period is updated. For example, if the original preset time period was one month and the conventional data collection interval was one hour, and now, based on long-term forecasting needs, the data collection interval is adjusted to every two hours, then the updated preset time period can be extended accordingly to obtain a sufficient number of data points to meet the requirements of long-term forecasting.

[0022] In one specific embodiment, a mapping relationship can be set between the data acquisition time interval and the ice thickness prediction time.

[0023] The flexible data acquisition strategy described in this application embodiment can be adjusted according to actual business needs, and has strong adaptability and scalability, which can meet the needs of predicting the icing thickness of transmission lines in different scenarios.

[0024] Optionally, before step 101, the method further includes: constructing an initial micro-meteorological dataset, wherein the initial micro-meteorological dataset includes meteorological observation data from multiple consecutive historical time points, and the meteorological observation data from each historical time point includes multiple initial meteorological index values, including temperature, relative humidity, wind speed, precipitation, and atmospheric pressure; normalizing the meteorological observation data from each historical time point to obtain normalized meteorological observation data, and constructing an initial meteorological data sequence corresponding to each initial meteorological index based on the normalized meteorological observation data from each historical time point; for each historical time point, obtaining the historical time... The historical icing thickness corresponding to each transmission line sample at each time point is calculated. For each transmission line sample, an icing thickness sequence is constructed based on the historical icing thickness at each historical time point, and the correlation between the icing thickness sequence and each initial meteorological data sequence is calculated according to a preset correlation calculation formula. For each initial meteorological indicator, the final correlation between the initial meteorological indicator and the icing thickness sequence is calculated based on the correlation corresponding to each transmission line sample and the line weight. Based on the final correlation corresponding to each initial meteorological indicator and a preset correlation threshold, initial meteorological indicators with a final correlation greater than the preset correlation threshold are selected from each initial meteorological indicator as target meteorological indicators.

[0025] In this embodiment, before acquiring the meteorological data sequence corresponding to a preset time period for the target transmission line, an initial micro-meteorological dataset is first constructed. This dataset covers meteorological observation data from multiple consecutive historical time points, which constitute a time series reflecting the changes in meteorological factors over time. The meteorological observation data at each historical time point contains multiple initial meteorological index values, which may include common meteorological elements that can affect the icing thickness of transmission lines, such as temperature, relative humidity, wind speed, precipitation, and atmospheric pressure. Among these, changes in temperature can affect water vapor condensation and ice melting; relative humidity determines the water vapor content in the air and is an important condition for icing formation; wind speed can affect water vapor transport and ice adhesion; precipitation directly provides moisture; and changes in atmospheric pressure may also indirectly affect meteorological conditions.

[0026] To eliminate the impact of differences in dimensions and numerical ranges between different meteorological indicators, meteorological observation data at each historical time point can be normalized. Normalized meteorological observation data unifies the data of different indicators into a relatively standardized range, facilitating subsequent comparison and analysis. After processing, based on the normalized meteorological observation data at each historical time point, initial meteorological data sequences corresponding to each initial meteorological indicator are constructed. For example, for the initial meteorological indicator of temperature, the normalized temperature values ​​of all historical time points are arranged in chronological order to form a temperature data sequence; similarly, corresponding data sequences are constructed for other initial meteorological indicators such as relative humidity and wind speed. These initial meteorological data sequences clearly demonstrate the changes of each initial meteorological indicator over time.

[0027] For each historical time point, the historical icing thickness corresponding to each transmission line sample at that time point can also be obtained. The transmission line sample can be multiple transmission lines within the same large area. Through actual observation or reliable historical records, the historical icing thickness of these transmission line samples at each historical time point is obtained. This data reflects the actual icing situation of transmission lines under specific meteorological conditions and is an important basis for analyzing the relationship between meteorological factors and icing thickness.

[0028] Next, for each transmission line sample, an icing thickness sequence is constructed based on the historical icing thickness at each historical time point. This sequence demonstrates the trend of icing thickness over time. Then, the correlation between this icing thickness sequence and each initial meteorological data sequence is calculated using a pre-defined correlation formula. The correlation formula can be based on statistical methods, such as the Pearson correlation coefficient, to measure the degree of linear correlation between the two sequences. Through calculation, the correlation value between the initial meteorological data sequence corresponding to each initial meteorological indicator and the icing thickness sequence can be obtained. The larger the value, the stronger the correlation between the initial meteorological indicator and the icing thickness.

[0029] Since the importance of different transmission line samples may vary, for each initial meteorological indicator, the correlation between the icing thickness sequence corresponding to each transmission line sample and the initial meteorological indicator, as well as the line weight, can be considered. The line weight can be set based on factors such as the importance and geographical location of the transmission line sample. Based on the correlation degree corresponding to each transmission line sample and the line weight, the final correlation degree between the initial meteorological indicator and the icing thickness sequence is calculated. This final correlation degree comprehensively considers the influence of different transmission line samples and more accurately reflects the overall correlation between the initial meteorological indicator and icing thickness in the entire transmission line system.

[0030] Finally, based on the final correlation degree corresponding to each initial meteorological indicator and the preset correlation degree threshold, initial meteorological indicators with a final correlation degree greater than the preset correlation degree threshold are selected as target meteorological indicators. The preset correlation degree threshold can be set according to actual operational needs and data analysis results; only initial meteorological indicators with a sufficiently strong correlation to icing thickness are selected. This selection process removes initial meteorological indicators with weak correlation to icing thickness, retaining key meteorological factors and providing a more accurate data foundation for subsequent construction of meteorological data sequences and icing thickness prediction based on these target meteorological indicators.

[0031] This application's embodiments construct an initial micro-meteorological dataset and collect historical icing thickness data from multiple transmission line samples. This comprehensively considers different meteorological factors and the conditions of different transmission lines, providing rich data support for correlation analysis and improving the reliability of the analysis results. By considering line weights to calculate the final correlation between the initial meteorological indicators and the icing thickness sequence, it can more accurately reflect the correlation between the initial meteorological indicators and icing thickness in the entire transmission line system, avoiding potential biases from data from a single transmission line. Target meteorological indicators are selected based on the final correlation, removing irrelevant factors and retaining key meteorological indicators. This not only reduces the complexity of subsequent data processing and model training but also improves the accuracy and efficiency of icing thickness prediction based on these target meteorological indicators.

[0032] Optionally, in this embodiment, the preset correlation degree calculation formula is as follows: ; in, For relevance, n This represents the number of sampling time points included in the ice thickness sequence. k Indicates the first k Each sampling time point The th in the ice thickness sequence k The historical icing thickness corresponding to each sampling time point For the i-th initial meteorological data sequence, the th k The normalized initial meteorological index values ​​corresponding to each sampling time point The resolution coefficient takes values ​​in the range [0,1].

[0033] In this embodiment, the resolution coefficient This is a preset empirical parameter, ranging from 0 to 1, used to adjust the sensitivity of the calculation to the range. The above calculation process can generate a correlation degree for each initial meteorological data sequence at each sampling time point k, representing the degree of similarity between the change in the initial meteorological index value and the change in ice thickness at that sampling time point k. The closer the value is to 1, the stronger the correlation at that sampling time point. This application transforms the complex and fuzzy correlation between initial meteorological indicators and ice thickness into a quantifiable and repeatable precise mathematical measure. By comparing the similarity of sequence morphology point by point and using the resolution coefficient ρ to enhance the distinguishing ability, it provides a solid mathematical foundation for the subsequent scientific and objective selection of target meteorological indicators, making the entire feature engineering process more rigorous and interpretable, free from subjective experience judgment.

[0034] In this embodiment of the application, optionally, the line weight corresponding to each transmission line sample is determined in the following manner: For each transmission line sample, the geographical meteorological factor, line structure factor, and data quality factor corresponding to the transmission line sample are determined, and a multi-dimensional weight evaluation vector is constructed based on the geographical meteorological factor, the line structure factor, and the data quality factor. The geographical meteorological factor is calculated based on at least one of the altitude, micro-topography type, and historical icing frequency of the transmission line sample; the line structure factor is calculated based on at least one of the voltage level, conductor type, and representative span of the transmission line sample; and the data quality factor is calculated based on the completeness and / or accuracy of meteorological observation data corresponding to the transmission line sample. Based on the multi-dimensional weight evaluation vector of each transmission line sample, the mutual importance between each transmission line sample is calculated through a multi-head self-attention module to obtain the attention weight of each transmission line sample, and the attention weight is used as the line weight of the corresponding transmission line sample.

[0035] In this embodiment, for each transmission line sample, the calculation of geographic meteorological factors can be based on at least one of altitude, local micro-topography type, and historical icing frequency. Altitude significantly affects meteorological conditions such as temperature and air pressure; higher altitudes typically have lower temperatures, making icing more likely. Therefore, altitude is a crucial geographic factor influencing icing. Local micro-topography types, such as valleys, mountain peaks, and windward slopes, affect the distribution of meteorological elements like wind speed and humidity, thus influencing icing conditions. For example, windward slopes may experience more severe icing due to abundant moisture. Historical icing frequency directly reflects the frequency of icing occurrences for the transmission line sample in the past; a higher frequency indicates more favorable meteorological conditions for icing formation. The geographic meteorological factors calculated by comprehensively considering these factors can effectively reflect the impact of the geographical environment of the transmission line sample on icing.

[0036] The calculation of the line structure factor can be based on at least one of the following: voltage level, conductor type, and representative span of the transmission line sample. Different voltage levels result in variations in the electrical performance and mechanical strength of the transmission line sample, as well as its ability to withstand icing. Higher voltage levels typically have more stringent designs and can withstand thicker icing. The conductor type determines the conductor's physical characteristics, such as diameter and material. Different conductor types experience different stresses and heat dissipation under icing conditions, affecting the icing thickness. The representative span is a crucial parameter reflecting the sag characteristics of the transmission line sample. The span size affects the stress distribution on the conductor under wind and ice conditions, thus influencing the icing situation. Calculating the line structure factor using these factors reveals the impact of the transmission line sample's own structure on icing.

[0037] The data quality factor can be calculated based on the completeness and / or accuracy of meteorological observation data corresponding to the transmission line samples. The completeness of meteorological observation data reflects the integrity of data records; significant data loss can affect the accuracy of analysis of meteorological conditions and icing. The accuracy of meteorological observation data reflects the reliability of the data; inaccurate data can lead to erroneous conclusions. For example, sensor malfunctions or data transmission errors may result in inaccurate observation data. Calculating the data quality factor by comprehensively considering these two aspects can measure the quality of meteorological observation data and provide a basis for subsequent weighting assessments.

[0038] Subsequently, based on the aforementioned determined geographical and meteorological factors, line structure factors, and data quality factors, a multi-dimensional weighted evaluation vector can be constructed for each transmission line sample. This vector integrates different factors to evaluate the transmission line sample from multiple dimensions. For example, a high score on the geographical and meteorological factors for a transmission line sample indicates that its geographical environment is conducive to icing formation; a moderate score on the line structure factor indicates that its structure has a certain influence on icing; and a high score on the data quality factor means that its meteorological observation data is of good quality. Combining these factors into a multi-dimensional vector can comprehensively and holistically reflect the characteristics of the transmission line sample in terms of icing-related aspects.

[0039] Furthermore, based on the multi-dimensional weight evaluation vector of each transmission line sample, the mutual importance between the samples can be calculated using a multi-head self-attention module. The multi-head self-attention module can capture the relationships between different parts of the vector and measure the degree of correlation and importance between the transmission line samples by calculating attention weights. For example, if two transmission line samples have similar geographical and meteorological factors and their line structure factors are also related, the multi-head self-attention module can consider them to have high mutual importance and assign them a larger attention weight. The final attention weight of each transmission line sample becomes its corresponding line weight, which comprehensively considers the characteristics of the transmission line sample itself as well as its relationship with other samples.

[0040] In a specific embodiment, when calculating attention weights based on the multidimensional weight evaluation vectors of each transmission line sample, all multidimensional weight evaluation vectors can first be combined into an input matrix. This matrix is ​​then linearly transformed using h different weight matrices to obtain query, key, and value matrices. Each weight matrix corresponds to a "head" to focus on different feature subspaces. Next, the transpose of the query matrix and key matrix under a single head is calculated and scaled to obtain an attention score matrix. Each row is then subjected to a softmax function to obtain an attention weight matrix. This matrix is ​​then used to weight and sum the value matrix to obtain the single head output. The h head outputs are then concatenated and linearly transformed to obtain the final output matrix. Finally, specific information is extracted from this matrix according to actual needs, such as normalizing each row. Each row, after appropriate processing, yields the line weight of a transmission line sample relative to all other transmission line samples, reflecting the mutual importance among the transmission line samples.

[0041] This application's embodiments determine the weight evaluation factors from three dimensions: geography and meteorology, line structure, and data quality. It comprehensively considers various factors affecting icing on transmission lines, making the weight evaluation more scientific and reasonable. The multi-head self-attention module is used to calculate the attention weight, which can automatically capture the complex relationships between each transmission line sample, avoiding the subjectivity and limitations of manually setting weights, and improving the accuracy and objectivity of line weight determination.

[0042] In this embodiment of the application, optionally, the line weight corresponding to each transmission line sample is determined based on the following method: For each transmission line sample, a geographic meteorological factor, a line structure factor, and a data quality factor corresponding to the transmission line sample are determined, wherein the geographic meteorological factor is calculated based on at least one of the transmission line sample's altitude, local micro-topography type, and historical icing frequency; the line structure factor is calculated based on at least one of the transmission line sample's voltage level, conductor type, and representative span; and the data quality factor is calculated based on the completeness and / or accuracy of meteorological observation data corresponding to the transmission line sample; the geographic meteorological factor and the line structure factor are then combined. The data quality factor is standardized to obtain standardized factors. For each factor dimension, a standardized value vector of all transmission line samples in that factor dimension is constructed based on the standardized factors. The relative importance score of each transmission line sample in each factor dimension is calculated using the kernel density estimation method. For the geographic meteorological factor and the line structure factor, the relative importance score is calculated using a positively skewed kernel density function, and for the data quality factor, the relative importance score is calculated using a negatively skewed kernel density function. The relative importance score of each transmission line sample in each factor dimension is input into a pre-trained weight fusion neural network to output the line weight of each transmission line sample.

[0043] In this embodiment, for each transmission line sample, its corresponding geographic meteorological factors, line structure factors, and data quality factors can be determined. Next, these factors are standardized. Since the dimensions and numerical ranges of different factors can vary greatly—for example, altitude may range from hundreds to thousands of meters—while data completeness is a proportion between 0 and 1, directly using these raw data for subsequent calculations may lead to some factors having an excessively large impact on the results, while the effects of other factors are ignored. Standardization can unify the data of different factors to a similar numerical range, eliminating the influence of differences in dimensions and numerical ranges, making each factor comparable in subsequent analyses, and laying the foundation for constructing standardized value vectors and calculating relative importance scores.

[0044] For each factor dimension, a standardized value vector of all transmission line samples is constructed based on the standardized values ​​of each factor. For example, for the geographic meteorological factor dimension, the standardized geographic meteorological factor values ​​of all transmission line samples are arranged in order to form a vector. This vector contains the standardized information of all transmission line samples in this factor dimension, which can intuitively show the relative position and differences of each transmission line sample in this factor. This facilitates subsequent analysis of the distribution of each transmission line sample in this factor dimension using kernel density estimation methods, and then calculates the relative importance score.

[0045] Furthermore, kernel density estimation is used to calculate the relative importance score of each transmission line sample across each factor dimension. Specifically, for the geographic meteorological factor and the line structure factor, a positively skewed kernel density function can be used. This is because, to a certain extent, larger values ​​for these two factors indicate greater unfavorable conditions for the safe operation of transmission lines (e.g., high altitude, high voltage level, and conductor type unfavorable for heat dissipation), and using a positively skewed kernel density function can better highlight the importance of samples with larger values. For the data quality factor, a negatively skewed kernel density function can be used, because higher data quality (higher completeness and accuracy) is more beneficial for analyzing icing conditions on transmission lines, and using a negatively skewed kernel density function can highlight the importance of high-quality data samples. Through kernel density estimation, the relative importance of each transmission line sample in each factor dimension can be reasonably assessed based on its distribution.

[0046] Next, the relative importance score of each transmission line sample across each factor dimension is input into a pre-trained weight fusion neural network, which outputs the line weight for each transmission line sample. This pre-trained weight fusion neural network, having learned from and trained on a large number of samples, can comprehensively consider the relative importance scores across each factor dimension, uncovering the complex relationships and interactions between the factors. Through the nonlinear transformation and fusion capabilities of the neural network, the line weight for each transmission line sample can be determined more accurately. This weight comprehensively considers factors such as geographical meteorology, line structure, and data quality, and can more comprehensively and objectively reflect the importance of the transmission line sample in the entire system.

[0047] This application's embodiments determine the factors influencing the line weights of transmission line samples from three dimensions: geography and meteorology, line structure, and data quality. It comprehensively considers the natural environment, line characteristics, and data reliability, making the weight assessment more scientific and reasonable. The kernel density estimation method is used to calculate the relative importance score for each factor dimension, and different skewed kernel density functions are adopted according to the characteristics of different factors, which can more accurately reflect the importance of each transmission line sample in different factor dimensions. A pre-trained weight fusion neural network integrates the relative importance scores of each factor dimension, considering the complex relationships between factors, resulting in more accurate and objective output line weights.

[0048] Optionally, in this embodiment, step 102, "generating the meteorological feature vector of the target transmission line based on the meteorological data sequence," includes: using a sliding time window method to sequentially extract multiple meteorological data subsequences from the meteorological data sequence, wherein each meteorological data subsequence contains the values ​​of each target meteorological index at a consecutive preset number of sampling time points; for each meteorological data subsequence, extracting deep temporal features from the time domain and / or frequency domain to characterize the icing evolution pattern, wherein the deep temporal features include the statistical characteristics, trend characteristics, and periodic characteristics of each target meteorological index in the meteorological data subsequence over time; fusing the deep temporal features corresponding to each meteorological data subsequence with the values ​​of each target meteorological index at the last sampling time point in the meteorological data subsequence to obtain a fused feature vector corresponding to the meteorological data subsequence; and constructing the meteorological feature vector of the target transmission line based on the multiple fused feature vectors obtained in chronological order.

[0049] In this embodiment, the sliding time window method is a commonly used method in time series data processing. This application uses the sliding time window method to sequentially extract multiple meteorological data subsequences from a meteorological data sequence. Specifically, a fixed-length sliding window can be set, which slides sequentially along the time axis at certain step sizes. Each meteorological data subsequence contains the values ​​of target meteorological indicators for a predetermined number of consecutive sampling time points. For example, if the predetermined number is 10, each meteorological data subsequence contains the values ​​of target meteorological indicators for 10 consecutive sampling time points, such as temperature, humidity, and wind speed for 10 consecutive sampling time points. In this way, a long-term meteorological data sequence can be divided into multiple short-term meteorological data subsequences, facilitating detailed analysis of each meteorological data subsequence and capturing local meteorological change characteristics. Because icing evolution is a gradual process over time, short-term meteorological data subsequences can reflect this change more precisely.

[0050] For each extracted meteorological data subsequence, deep temporal features characterizing the evolution of icing can be extracted from the time and / or frequency domains. In the time domain, statistical features can include mean, variance, maximum, and minimum values. These statistics reflect the overall level and fluctuation of the target meteorological indicators within the subsequence. For example, the mean of a temperature subsequence reflects the average temperature over that period, while the variance reflects the degree of temperature fluctuation; both have a significant impact on icing formation and growth. Trend characteristics can be obtained by calculating the difference between target meteorological indicator values ​​at adjacent sampling time points or by fitting a trend line. For instance, the rate of temperature change over time; a sustained and rapid decrease in temperature may be more conducive to icing formation. Periodic characteristics can be analyzed by converting the time-domain signal to the frequency domain using methods such as Fourier transform to identify the periodic variation patterns of the target meteorological indicator values ​​over time. For example, meteorological conditions in certain regions may exhibit daily or annual cycles, and these periodic variations may be correlated with the periodic evolution of icing. Extracting features in the frequency domain can uncover the hidden periodic information in meteorological data more deeply, providing a more comprehensive perspective for the analysis of icing evolution patterns.

[0051] Next, the deep time-series features corresponding to each meteorological data subsequence are fused with the target meteorological index values ​​at the last sampling time point of that subsequence. The deep time-series features describe the changes in the target meteorological indicators within the subsequence from multiple perspectives, while the target meteorological index values ​​at the last sampling time point represent the real-time meteorological state at the end of the subsequence. Fusing these two pieces of information yields a more comprehensive fused feature vector reflecting the meteorological conditions of that period. For example, the deep time-series features reflect the temperature trend and fluctuations during this period, while the temperature value at the last sampling time point clarifies the current state of this trend. Combining the two allows for a more accurate characterization of meteorological features related to icing evolution. This fusion method comprehensively considers historical meteorological changes and current meteorological conditions, providing richer and more accurate information for the subsequent construction of meteorological feature vectors.

[0052] Furthermore, a meteorological feature vector for the target transmission line can be constructed based on multiple fused feature vectors obtained in chronological order. Since multiple chronologically ordered meteorological data subsequences were obtained earlier using the sliding time window method, each subsequence corresponds to a fused feature vector. Combining these fused feature vectors in chronological order forms the meteorological feature vector for the target transmission line. This meteorological feature vector not only contains meteorological characteristic information from different time periods but also reflects the changes in these characteristics over time, comprehensively and systematically reflecting the meteorological conditions and their changing patterns of the target transmission line throughout the entire observation period.

[0053] This application employs a sliding time window method to extract meteorological data subsequences, which can meticulously capture the changing characteristics of meteorological data in local time periods, avoiding the neglect of the impact of short-term meteorological changes on icing due to overall analysis. It extracts deep time-series features from the time and / or frequency domains, comprehensively considering the statistical characteristics, changing trends, and periodic patterns of the target meteorological indicators, enabling a deeper exploration of information related to icing evolution within the meteorological data. By fusing the deep time-series features with the target meteorological indicator values ​​at the last sampling time point, it takes into account both historical meteorological changes and current meteorological conditions, making the fused feature vector more representative and accurate. Constructing meteorological feature vectors in chronological order can completely present the meteorological evolution process of the target transmission line throughout the entire observation period, providing a reliable data foundation for accurately analyzing icing evolution patterns and predicting icing conditions.

[0054] Furthermore, as a refinement and extension of the specific implementation of the above embodiments, in order to fully illustrate the specific implementation process of this embodiment, a training method for an LSTM model is provided, the method comprising: First, dataset preprocessing is performed. In this stage, a training set is obtained, which may include multiple training samples. Each training sample contains the time series of each target meteorological indicator and its corresponding true icing thickness label. These training samples undergo a series of preparatory steps, including handling missing values, normalizing the values ​​to eliminate the influence of units, and organizing them according to the format required by the LSTM model, thus standardizing them for model input.

[0055] After preprocessing, the processed data can be transformed into data features and input into the LSTM model. These data features then enter the core LSTM layer. This layer consists of several LSTM units, responsible for processing the input sequence step-by-step. Through its unique gating mechanism (forget gate, input gate, output gate), it captures and remembers the long-term and short-term dependencies in the meteorological sequence, outputting a hidden state sequence. The output of the LSTM layer is then passed to the fully connected layer. This layer, located at the end of the LSTM model, is responsible for mapping the high-dimensional temporal features learned by the LSTM layer to the final prediction target dimension, i.e., outputting a specific predicted icing thickness value. The predicted value output by the LSTM model is then compared with the actual icing thickness labels of the training samples.

[0056] The difference between the two is quantified by a loss function (such as mean squared error MSE or mean absolute error MAE) to obtain the numerical value of the current model prediction error. This error value is then input into an optimization function (such as Adam or SGD optimizer), which calculates the gradient of the model parameters (including the weights and biases of LSTM layers and fully connected layers) based on the error using a backpropagation algorithm.

[0057] Based on the calculated gradients, the optimization function updates and adjusts all trainable parameters in the LSTM model, aiming to reduce the loss in the next prediction. After the parameters are updated, it is determined whether the training has reached the required number of iterations (or a more complex early stopping condition). If not, the training set is used again to start a new round of forward propagation, loss calculation, and parameter optimization, and this process is repeated.

[0058] Once the iteration stopping condition is met, the training process ends. At this point, the model parameters have been optimized to a relatively stable state, and the saved model is the pre-trained LSTM model, which can be deployed to predict icing thickness on new meteorological data sequences.

[0059] In one specific embodiment, the LSTM model comprises an input layer, an LSTM hidden layer, and an output layer. The input layer receives meteorological feature vectors, and the output layer outputs the predicted icing thickness for a specific future time period.

[0060] In one specific embodiment, a trained LSTM model is used for prediction validation. The test set used for validation consists of transmission lines from the same batch as the training dataset. Specifically, the ice thickness is reflected by the icing suspension load. The test results are as follows: Figure 2 As shown in the diagram. Analysis of the above prediction results shows that the model performs well throughout the entire ice accumulation process, closely matching the actual values. Only in the initial stage of ice accumulation are the predicted values ​​overestimated. In the final stage of ice accumulation, the predicted values ​​exhibit less noise compared to the actual values, indicating that the model tends to predict a stable ice thickness change. Furthermore, in the later stages of ice accumulation, the predicted values ​​largely overlap with the actual values, demonstrating the considerable effectiveness of the predictions. This model can be used to train models for various prediction durations and to predict ice thickness.

[0061] Furthermore, as Figure 1 In terms of specific implementation, this application provides a transmission line icing thickness prediction device based on an LSTM model, such as... Figure 3 As shown, the device includes: The data acquisition module is used to acquire meteorological data sequences for a preset time period corresponding to the target transmission line. The meteorological data sequences are constructed based on micro-meteorological data from multiple sampling time points. Each sampling time point's micro-meteorological data includes multiple target meteorological index values. The correlation between the target meteorological index and the icing thickness of the transmission line is greater than a preset correlation threshold. The icing thickness prediction module is used to generate a meteorological feature vector of the target transmission line based on the meteorological data sequence, and output the predicted icing thickness of the target transmission line based on the meteorological feature vector and a pre-trained LSTM model.

[0062] Optionally, the device further includes a determination module; the determination module is configured to: Before acquiring the meteorological data sequence for the preset time period corresponding to the target transmission line, it is determined whether the predicted duration of the icing thickness of the target transmission line is greater than or equal to a preset duration threshold. If so, then based on the predicted duration of the icing thickness, an appropriate data collection time interval is determined, and the preset time period is updated based on the appropriate data collection time interval, wherein the appropriate data collection time interval is greater than the data collection time interval used in conventional prediction.

[0063] Optionally, the device further includes a filtering module; the filtering module is used for: Before acquiring the meteorological data sequence for the preset time period corresponding to the target transmission line, an initial micro-meteorological dataset is constructed. The initial micro-meteorological dataset includes meteorological observation data from multiple consecutive historical time points. Each historical time point's meteorological observation data includes multiple initial meteorological index values, including temperature, relative humidity, wind speed, precipitation, and atmospheric pressure. The meteorological observation data at each historical time point were normalized to obtain normalized meteorological observation data. Based on the normalized meteorological observation data at each historical time point, the initial meteorological data sequence corresponding to each initial meteorological indicator was constructed. For each historical time point, obtain the historical icing thickness corresponding to each transmission line sample at that historical time point; For each transmission line sample, an ice thickness sequence is constructed based on the historical ice thickness at each historical time point, and the correlation between the ice thickness sequence and each initial meteorological data sequence is calculated according to a preset correlation calculation formula. For each initial meteorological index, the final correlation between the initial meteorological index and the icing thickness sequence is calculated based on the correlation degree corresponding to each transmission line sample and the line weight. Based on the final correlation degree corresponding to each initial meteorological indicator and the preset correlation degree threshold, the initial meteorological indicators with a final correlation degree greater than the preset correlation degree threshold are selected from the initial meteorological indicators and used as target meteorological indicators.

[0064] Optionally, the preset correlation degree calculation formula is as follows: ; in, For relevance, n This represents the number of sampling time points included in the ice thickness sequence. k Indicates the first k Each sampling time point The th in the ice thickness sequencek The historical icing thickness corresponding to each sampling time point For the i-th initial meteorological data sequence, the th k The normalized initial meteorological index values ​​corresponding to each sampling time point The resolution coefficient takes values ​​in the range [0,1].

[0065] Optionally, the line weights corresponding to each transmission line sample are determined based on the following method: For each transmission line sample, the corresponding geographic meteorological factor, line structure factor, and data quality factor are determined. Based on the geographic meteorological factor, the line structure factor, and the data quality factor, a multi-dimensional weighted evaluation vector is constructed. The geographic meteorological factor is calculated based on at least one of the transmission line sample's altitude, micro-topography type, and historical icing frequency. The line structure factor is calculated based on at least one of the transmission line sample's voltage level, conductor type, and representative span. The data quality factor is calculated based on the completeness and / or accuracy of the meteorological observation data corresponding to the transmission line sample. Based on the multidimensional weight evaluation vector of each transmission line sample, the mutual importance between each transmission line sample is calculated through a multi-head self-attention module to obtain the attention weight of each transmission line sample, and the attention weight is used as the line weight of the corresponding transmission line sample.

[0066] Optionally, the line weights corresponding to each transmission line sample are determined based on the following method: For each transmission line sample, the geographical meteorological factor, line structure factor, and data quality factor corresponding to the transmission line sample are determined. The geographical meteorological factor is calculated based on at least one of the transmission line sample's altitude, local micro-topography type, and historical icing frequency. The line structure factor is calculated based on at least one of the transmission line sample's voltage level, conductor type, and representative span. The data quality factor is calculated based on the completeness and / or accuracy of the meteorological observation data corresponding to the transmission line sample. The geographical meteorological factors, the route structure factors, and the data quality factors are standardized to obtain the standardized factors. For each factor dimension, a standardized value vector of all transmission line samples for that factor dimension is constructed based on the standardized factors. Using the kernel density estimation method, the relative importance score of each transmission line sample on each factor dimension is calculated. For the geographical and meteorological factors and the line structure factors, the relative importance score is calculated using a positively skewed kernel density function, and for the data quality factors, the relative importance score is calculated using a negatively skewed kernel density function. The relative importance score of each transmission line sample on each factor dimension is input into a pre-trained weight fusion neural network, which outputs the line weight of each transmission line sample.

[0067] Optionally, the icing thickness prediction module is used for: The sliding time window method is used to sequentially extract multiple meteorological data subsequences from the meteorological data sequence, wherein each meteorological data subsequence contains the values ​​of each target meteorological index at a consecutive preset number of sampling time points; For each meteorological data subsequence, deep temporal features for characterizing the icing evolution pattern are extracted from the time domain and / or frequency domain, wherein the deep temporal features include the statistical characteristics, trend characteristics and periodic characteristics of each target meteorological indicator in the time series within the meteorological data subsequence; The deep temporal features corresponding to each meteorological data subsequence are fused with the target meteorological index values ​​at the last sampling time point in the meteorological data subsequence to obtain the fused feature vector corresponding to the meteorological data subsequence. The meteorological feature vector of the target transmission line is constructed based on multiple fused feature vectors obtained in chronological order.

[0068] It should be noted that other corresponding descriptions of the functional units involved in the LSTM-based transmission line icing thickness prediction device provided in this application embodiment can be found in the following references. Figures 1 to 2 The corresponding descriptions in the method will not be repeated here.

[0069] This application also provides a computer device, which may specifically be a personal computer, a server, a network device, etc. Figure 4 As shown, the computer device includes a bus, a processor, memory, and a communication interface, and may also include an input / output interface and a display device. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores location information. The network interface allows communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the steps in the various method embodiments.

[0070] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0071] In one embodiment, a computer-readable storage medium is provided, which may be non-volatile or volatile, having stored thereon a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0072] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0073] 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 used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0074] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0075] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0076] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for predicting icing thickness of transmission lines based on an LSTM model, characterized in that, include: Obtain a meteorological data sequence for a preset time period corresponding to the target transmission line. The meteorological data sequence is constructed based on micro-meteorological data from multiple sampling time points. The micro-meteorological data from each sampling time point includes multiple target meteorological index values. The correlation between the target meteorological index and the icing thickness of the transmission line is greater than a preset correlation threshold. Based on the meteorological data sequence, a meteorological feature vector of the target transmission line is generated, and based on the meteorological feature vector, the predicted icing thickness of the target transmission line is output through a pre-trained LSTM model.

2. The method according to claim 1, characterized in that, Before acquiring the meteorological data sequence for a preset time period corresponding to the target transmission line, the method further includes: Determine whether the predicted duration of icing thickness of the target transmission line is greater than or equal to a preset duration threshold. If so, then based on the predicted duration of the icing thickness, an appropriate data collection time interval is determined, and the preset time period is updated based on the appropriate data collection time interval, wherein the appropriate data collection time interval is greater than the data collection time interval used in conventional prediction.

3. The method according to claim 1, characterized in that, Before acquiring the meteorological data sequence for a preset time period corresponding to the target transmission line, the method further includes: An initial micro-meteorological dataset is constructed, wherein the initial micro-meteorological dataset includes meteorological observation data from multiple consecutive historical time points, and the meteorological observation data from each historical time point includes multiple initial meteorological index values, including temperature, relative humidity, wind speed, precipitation, and atmospheric pressure; The meteorological observation data at each historical time point were normalized to obtain normalized meteorological observation data. Based on the normalized meteorological observation data at each historical time point, the initial meteorological data sequence corresponding to each initial meteorological indicator was constructed. For each historical time point, obtain the historical icing thickness corresponding to each transmission line sample at that historical time point; For each transmission line sample, an ice thickness sequence is constructed based on the historical ice thickness at each historical time point, and the correlation between the ice thickness sequence and each initial meteorological data sequence is calculated according to a preset correlation calculation formula. For each initial meteorological index, the final correlation between the initial meteorological index and the icing thickness sequence is calculated based on the correlation degree corresponding to each transmission line sample and the line weight. Based on the final correlation degree corresponding to each initial meteorological indicator and the preset correlation degree threshold, the initial meteorological indicators with a final correlation degree greater than the preset correlation degree threshold are selected from the initial meteorological indicators and used as target meteorological indicators.

4. The method according to claim 3, characterized in that, The preset correlation degree calculation formula is as follows: ； in, For relevance, n This represents the number of sampling time points included in the ice thickness sequence. k Indicates the first k Each sampling time point The th in the ice thickness sequence k The historical icing thickness corresponding to each sampling time point For the i-th initial meteorological data sequence, the th k The normalized initial meteorological index values ​​corresponding to each sampling time point The resolution coefficient takes values ​​in the range [0,1].

5. The method according to claim 3 or 4, characterized in that, The line weights corresponding to each transmission line sample are determined based on the following method: For each transmission line sample, the corresponding geographic meteorological factor, line structure factor, and data quality factor are determined. Based on the geographic meteorological factor, the line structure factor, and the data quality factor, a multi-dimensional weighted evaluation vector is constructed. The geographic meteorological factor is calculated based on at least one of the transmission line sample's altitude, micro-topography type, and historical icing frequency. The line structure factor is calculated based on at least one of the transmission line sample's voltage level, conductor type, and representative span. The data quality factor is calculated based on the completeness and / or accuracy of the meteorological observation data corresponding to the transmission line sample. Based on the multidimensional weight evaluation vector of each transmission line sample, the mutual importance between each transmission line sample is calculated through a multi-head self-attention module to obtain the attention weight of each transmission line sample, and the attention weight is used as the line weight of the corresponding transmission line sample.

6. The method according to claim 3 or 4, characterized in that, The line weights corresponding to each transmission line sample are determined based on the following method: For each transmission line sample, the geographical meteorological factor, line structure factor, and data quality factor corresponding to the transmission line sample are determined. The geographical meteorological factor is calculated based on at least one of the transmission line sample's altitude, local micro-topography type, and historical icing frequency. The line structure factor is calculated based on at least one of the transmission line sample's voltage level, conductor type, and representative span. The data quality factor is calculated based on the completeness and / or accuracy of the meteorological observation data corresponding to the transmission line sample. The geographical meteorological factors, the route structure factors, and the data quality factors are standardized to obtain the standardized factors. For each factor dimension, a standardized value vector of all transmission line samples for that factor dimension is constructed based on the standardized factors. Using the kernel density estimation method, the relative importance score of each transmission line sample on each factor dimension is calculated. For the geographical and meteorological factors and the line structure factors, the relative importance score is calculated using a positively skewed kernel density function, and for the data quality factors, the relative importance score is calculated using a negatively skewed kernel density function. The relative importance score of each transmission line sample on each factor dimension is input into a pre-trained weight fusion neural network, which outputs the line weight of each transmission line sample.

7. The method according to claim 1, characterized in that, The step of generating the meteorological feature vector of the target transmission line based on the meteorological data sequence includes: The sliding time window method is used to sequentially extract multiple meteorological data subsequences from the meteorological data sequence, wherein each meteorological data subsequence contains the values ​​of each target meteorological index at a consecutive preset number of sampling time points; For each meteorological data subsequence, deep temporal features for characterizing the icing evolution pattern are extracted from the time domain and / or frequency domain, wherein the deep temporal features include the statistical characteristics, trend characteristics and periodic characteristics of each target meteorological indicator in the time series within the meteorological data subsequence; The deep temporal features corresponding to each meteorological data subsequence are fused with the target meteorological index values ​​at the last sampling time point in the meteorological data subsequence to obtain the fused feature vector corresponding to the meteorological data subsequence. The meteorological feature vector of the target transmission line is constructed based on multiple fused feature vectors obtained in chronological order.

8. A device for predicting the icing thickness of transmission lines based on an LSTM model, characterized in that, include: The data acquisition module is used to acquire meteorological data sequences for a preset time period corresponding to the target transmission line. The meteorological data sequences are constructed based on micro-meteorological data from multiple sampling time points. Each sampling time point's micro-meteorological data includes multiple target meteorological index values. The correlation between the target meteorological index and the icing thickness of the transmission line is greater than a preset correlation threshold. The icing thickness prediction module is used to generate a meteorological feature vector of the target transmission line based on the meteorological data sequence, and output the predicted icing thickness of the target transmission line based on the meteorological feature vector and a pre-trained LSTM model.

9. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

10. A computer device, comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.

Images