Method and device for predicting icing thickness of power transmission line and computer equipment

Abstract

The invention relates to a power transmission line icing thickness prediction method and device and computer equipment. The method comprises the following steps: acquiring a prediction time span of an icing thickness, and determining an icing stage corresponding to the prediction time span; according to the micro-meteorological information and the physical monitoring information of the icing stage, respectively determining micro-meteorological characteristics and physical monitoring characteristics; inputting the micrometeorological features and the physical monitoring features into a trained stage prediction model to obtain an icing growth rate prediction value of the icing stage; and determining an icing thickness prediction value of the prediction time span according to the icing growth rate prediction value. By adopting the method, the icing thickness of different icing stages can be accurately predicted, and the accuracy of predicting the icing thickness of the power transmission line is improved.

Description

Technical Field

[0001] The present application relates to the field of electric power technology, and in particular to a method, device and computer equipment for predicting ice thickness on transmission lines. Background Art

[0002] Icing on transmission lines can lead to catastrophic consequences such as line swaying, conductor breakage, and tower toppling, causing large-scale power outages and seriously endangering the reliability of power supply and the integrity of transmission infrastructure. It is one of the serious natural disasters facing the power system and poses a major threat to the safe and stable operation of the power grid.

[0003] At present, research on transmission line icing prediction models is mainly divided into three categories: physical mechanism models, statistical calculation models, and intelligent calculation models. Among them, many important parameters in physical mechanism models are difficult to measure through conventional meteorological observations, which limits their practical application; statistical calculation models need to meet many statistical assumptions and find it difficult to consider the dynamic impact of environmental factors, resulting in inaccurate model predictions and limited scope of application; in addition, intelligent calculation models (such as neural networks and support vector machines) usually directly use environmental factors to predict ice thickness, and the prediction accuracy is not high.

[0004] Therefore, the current transmission line icing prediction technology has the problem of low prediction accuracy. Summary of the Invention

[0005] Based on this, it is necessary to provide a method, device, computer equipment, computer-readable storage medium and computer program product for predicting ice thickness of transmission lines that can improve prediction accuracy in order to address the above technical problems.

[0006] In a first aspect, the present application provides a method for predicting ice thickness on a transmission line, comprising:

[0007] Obtaining a predicted time span of ice thickness, and determining an ice stage corresponding to the predicted time span;

[0008] determining micrometeorological characteristics and physical monitoring characteristics respectively according to the micrometeorological information and physical monitoring information during the icing stage;

[0009] Inputting the micro-meteorological characteristics and the physical monitoring characteristics into a trained stage prediction model to obtain a predicted value of ice cover growth rate in the ice cover stage;

[0010] A predicted value of ice thickness for the predicted time span is determined according to the predicted value of ice growth rate.

[0011] In a second aspect, the present application further provides a device for predicting ice thickness on a transmission line, comprising:

[0012] an acquisition module, configured to acquire a predicted time span of ice thickness and determine an ice stage corresponding to the predicted time span;

[0013] a determination module, configured to determine micrometeorological characteristics and physical monitoring characteristics respectively according to the micrometeorological information and physical monitoring information during the icing stage;

[0014] A prediction module, configured to input the micrometeorological characteristics and the physical monitoring characteristics into a trained stage prediction model to obtain a predicted value of ice growth rate in the ice accumulation stage;

[0015] A calculation module is used to determine a predicted ice thickness value for the predicted time span based on the predicted ice growth rate value.

[0016] In a third aspect, the present application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0017] Obtaining a predicted time span of ice thickness, and determining an ice stage corresponding to the predicted time span;

[0018] determining micrometeorological characteristics and physical monitoring characteristics respectively according to the micrometeorological information and physical monitoring information during the icing stage;

[0019] Inputting the micro-meteorological characteristics and the physical monitoring characteristics into a trained stage prediction model to obtain a predicted value of ice cover growth rate in the ice cover stage;

[0020] A predicted value of ice thickness for the predicted time span is determined according to the predicted value of ice growth rate.

[0021] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the following steps are implemented:

[0022] Obtaining a predicted time span of ice thickness, and determining an ice stage corresponding to the predicted time span;

[0023] determining micrometeorological characteristics and physical monitoring characteristics respectively according to the micrometeorological information and physical monitoring information during the icing stage;

[0024] Inputting the micro-meteorological characteristics and the physical monitoring characteristics into a trained stage prediction model to obtain a predicted value of ice cover growth rate in the ice cover stage;

[0025] A predicted value of ice thickness for the predicted time span is determined according to the predicted value of ice growth rate.

[0026] In a fifth aspect, the present application further provides a computer program product, comprising a computer program, which, when executed by a processor, implements the following steps:

[0027] Obtaining a predicted time span of ice thickness, and determining an ice stage corresponding to the predicted time span;

[0028] determining micrometeorological characteristics and physical monitoring characteristics respectively according to the micrometeorological information and physical monitoring information during the icing stage;

[0029] Inputting the micro-meteorological characteristics and the physical monitoring characteristics into a trained stage prediction model to obtain a predicted value of ice cover growth rate in the ice cover stage;

[0030] A predicted value of ice thickness for the predicted time span is determined according to the predicted value of ice growth rate.

[0031] The above-mentioned transmission line ice thickness prediction method, device, computer equipment, computer-readable storage medium and computer program product obtain the predicted time span of the ice thickness, determine the ice stage corresponding to the predicted time span, and determine the micrometeorological characteristics and physical monitoring characteristics respectively according to the micrometeorological information and physical monitoring information of the ice stage. The micrometeorological characteristics and physical monitoring characteristics are input into the trained stage prediction model to obtain the ice growth rate prediction value of the ice stage, and determine the ice thickness prediction value of the predicted time span according to the ice growth rate prediction value. The physical laws in the ice process and the time accumulation effect of ice thickness can be combined to predict the ice thickness for different ice stages under the predicted time span, accurately predict the ice thickness of different ice stages, and improve the accuracy of ice thickness prediction of transmission lines. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments of the present application or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying any creative work.

[0033] Figure 1 1 is a flow chart of a method for predicting ice thickness on a transmission line according to an embodiment;

[0034] Figure 2 A schematic diagram of a process for constructing a transmission line ice thickness prediction model in one embodiment;

[0035] Figure 3 Schematic diagram of a flow chart of a method for predicting ice thickness on a transmission line in another embodiment;

[0036] Figure 4 This is a structural block diagram of a device for predicting ice thickness on a transmission line in one embodiment;

[0037] Figure 5 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0039] It should be noted that the terms "first," "second," and the like in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or precedence. It should be understood that the numbers used in this manner are interchangeable where appropriate so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of apparatus and methods consistent with certain aspects of the present disclosure as detailed in the appended claims.

[0040] In an exemplary embodiment, Figure 1 As shown, a method for predicting ice thickness on transmission lines is provided. This embodiment uses the method applied to a terminal as an example for illustration. It is understood that the method can also be applied to a server, or to a system including a terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0041] Step S101: obtaining a predicted time span of ice thickness and determining an ice stage corresponding to the predicted time span.

[0042] Ice thickness refers to the maximum vertical thickness of ice formed when water vapor or raindrops freeze and adhere to the conductor surface under severe cold conditions. The prediction time span refers to the length of time for ice thickness predictions, for example, the next 12 hours for short-term predictions, the next 24 hours for medium-term predictions, and the next 48 hours for long-term predictions. Ice stages can refer to different stages of ice development, such as the growth phase, the stabilization phase, and the melting phase.

[0043] In a specific implementation, a predicted time span may be input into the terminal, and the terminal divides the input predicted time span into multiple ice covering stages according to a pre-set setting.

[0044] For example, for short-term forecasts, if the forecast time span is the next 12 hours, the next 1 to 5 hours can be determined as the growth period and the next 5 to 12 hours can be determined as the stable period according to the pre-settings; for long-term forecasts, if the forecast time span is the next 48 hours, the next 1 to 5 hours can be determined as the growth period, the next 5 to 36 hours can be determined as the stable period, and the next 36 to 48 hours can be determined as the melting period according to the pre-settings.

[0045] Step S102 : determining micrometeorological characteristics and physical monitoring characteristics respectively according to the micrometeorological information and physical monitoring information during the icing stage.

[0046] Micrometeorological information refers to high-precision meteorological observation data collected on a small scale, including but not limited to ambient temperature, relative humidity, wind speed, wind direction, precipitation, solar radiation intensity, and atmospheric pressure. Physical monitoring information refers to various monitoring data collected through physical means such as sensors and instruments, including but not limited to physical parameters such as conductor load and wind deflection. Micrometeorological characteristics are the characteristic data of micrometeorological information. Physical monitoring characteristics are the characteristic data of physical monitoring information.

[0047] In a specific implementation, the terminal can obtain the micro-meteorological information and physical monitoring information of each icing stage, perform feature extraction on the micro-meteorological information and the physical monitoring information, and obtain micro-meteorological features and physical monitoring features respectively.

[0048] For example, for long-term predictions, micrometeorological information and physical monitoring information can be collected for each stage of the growth period, stable period and melting period respectively, and the micrometeorological characteristics and physical monitoring characteristics of each stage can be obtained through feature extraction.

[0049] Step S103: Input the micro-meteorological characteristics and physical monitoring characteristics into the trained stage prediction model to obtain the predicted value of ice cover growth rate in the ice cover stage.

[0050] The trained stage prediction model may be a pre-trained prediction model corresponding to the icing stage. The icing growth rate prediction value refers to the predicted icing growth rate.

[0051] The prediction model may be a joint model of a bidirectional gated recurrent unit (BiGRU) and a multi-kernel relevance vector machine (MKRVM).

[0052] In the specific implementation, the terminal can train the prediction model for each ice covering stage in advance to obtain the trained stage prediction model for each ice covering stage, input the extracted micrometeorological characteristics and physical monitoring characteristics into the corresponding trained stage prediction model, and obtain the predicted value of ice covering growth rate for each ice covering stage.

[0053] For example, the stage prediction models corresponding to the growth period, stable period and melting period can be trained in advance, and the micrometeorological characteristics and physical monitoring characteristics of the growth period can be input into the trained stage prediction model of the growth period to obtain the predicted value of the ice cover growth rate in the growth period; the micrometeorological characteristics and physical monitoring characteristics of the stable period can be input into the trained stage prediction model of the stable period to obtain the predicted value of the ice cover growth rate in the stable period; the micrometeorological characteristics and physical monitoring characteristics of the melting period can be input into the trained stage prediction model of the melting period to obtain the predicted value of the ice cover growth rate in the melting period.

[0054] Step S104: determining a predicted ice thickness value for a predicted time span according to the predicted ice growth rate.

[0055] The predicted ice thickness value may be the predicted ice thickness.

[0056] In a specific implementation, the terminal can calculate the ice thickness prediction value according to the ice growth rate prediction value for each ice covering stage, and integrate the calculated ice thickness prediction values for each ice covering stage to obtain the ice thickness prediction value for the entire prediction time span.

[0057] For example, for long-term predictions, the predicted ice thickness for the next 1 to 5 hours can be obtained based on the initial ice thickness and the predicted ice growth rate in the growth period; the predicted ice thickness for the next 5 to 36 hours can be obtained based on the initial ice thickness and the predicted ice growth rate in the stable period; and the predicted ice thickness for the next 36 to 48 hours can be obtained based on the initial ice thickness and the predicted ice growth rate in the melting period. By connecting the obtained ice thickness prediction values, the entire prediction time span, that is, the predicted ice thickness for the next 1 to 48 hours, can be obtained.

[0058] It should be noted that the predicted ice thickness values at overlapping time points between adjacent icing stages can be obtained by weighted averaging. For example, the predicted ice thickness value for the next 5 hours can be the weighted average of the predicted ice thickness value for the next 5 hours in the growth period and the predicted ice thickness value for the next 5 hours in the stable period. The predicted ice thickness value for the next 36 hours can be the weighted average of the predicted ice thickness value for the next 36 hours in the stable period and the predicted ice thickness value for the next 36 hours in the melting period.

[0059] The above-mentioned method for predicting the ice thickness of transmission lines obtains the predicted time span of the ice thickness, determines the ice stage corresponding to the predicted time span, and determines the micrometeorological characteristics and physical monitoring characteristics respectively according to the micrometeorological information and physical monitoring information of the ice stage. The micrometeorological characteristics and physical monitoring characteristics are input into the trained stage prediction model to obtain the predicted value of the ice growth rate in the ice stage, and the predicted value of the ice thickness for the predicted time span is determined according to the predicted value of the ice growth rate. The physical laws in the icing process and the time accumulation effect of the ice thickness can be combined to predict the ice thickness for different icing stages under the predicted time span, accurately predict the ice thickness in different icing stages, and improve the accuracy of the ice thickness prediction of the transmission line.

[0060] It is understandable that in actual applications, the prediction model to be trained can be determined in advance, and then the stage prediction model to be trained can be determined based on the prediction model to be trained, and the stage prediction model to be trained can be trained to obtain the above-mentioned trained stage prediction model. In an exemplary embodiment, before the above-mentioned step S101, it can also specifically include: obtaining the parameter space of the prediction model to be trained; performing a first search on the parameter space based on the Sparrow Search Algorithm (SSA) to obtain a first parameter combination; performing a second search on the first parameter combination based on the Adaptive Parallel Jaya Algorithm (APJA) to obtain a second parameter combination; and determining the prediction model to be trained based on the second parameter combination.

[0061] The parameter space refers to the space consisting of all parameters of the unfiltered BiGRU and MKRVM joint model. The first search is a coarse search performed using SSA. The first parameter combination is the parameter combination obtained through the coarse search. The second search is a fine search performed using APJA. The second parameter combination is the parameter combination obtained through the fine search. The prediction model to be trained can be an untrained BiGRU and MKRVM joint model that has undergone parameter optimization to select the optimal parameter combination.

[0062] In the specific implementation, the terminal can obtain the parameter space of the prediction model to be trained, first use SSA to perform a first search on the parameter space to obtain a first parameter combination, and then use APJA to perform a second search on the first parameter combination to obtain a second parameter combination, and use the second parameter combination as the model parameters of the prediction model to be trained.

[0063] For example, the terminal can obtain the parameter space of the BiGRU and MKRVM joint model to be trained, first use SSA to perform a global search on the entire parameter space to determine the potential high-quality area in the parameter space, and then use APJA to perform a fine search on the potential high-quality area to determine the final optimal parameter combination, and use the optimal parameter combination as the model parameters of the BiGRU and MKRVM joint model to be trained.

[0064] In this embodiment, by obtaining the parameter space of the prediction model to be trained, a first search is performed on the parameter space based on SSA to obtain a first parameter combination, a second search is performed on the first parameter combination based on APJA to obtain a second parameter combination, and the prediction model to be trained is determined according to the second parameter combination. The accuracy and generation efficiency of the prediction model can be increased through a secondary search.

[0065] In an exemplary embodiment, after determining the prediction model to be trained according to the second parameter combination as described above, it may further specifically include: obtaining the micrometeorological sample values, physical monitoring sample values and ice thickness sample values of the transmission line; determining the sample icing stage according to the micrometeorological sample values, physical monitoring sample values and ice thickness sample values; determining the stage prediction model to be trained according to the sample icing stage and the prediction model to be trained; inputting the micrometeorological sample values and physical monitoring sample values into the stage prediction model to be trained to obtain the stage prediction results of the stage prediction model to be trained; training the stage prediction model to be trained according to the difference between the stage prediction results and the stage icing growth rate to obtain a trained stage prediction model; the stage icing growth rate is obtained according to the ice thickness sample values of the sample icing stage.

[0066] The micrometeorological sample value refers to the micrometeorological information used as training samples during model training. The physical monitoring sample value refers to the physical monitoring information used as training samples during model training. The ice thickness sample value refers to the ice thickness used as the sample label during model training. The sample ice stage refers to the ice stage corresponding to the micrometeorological sample value, physical monitoring sample value, and ice thickness sample value.

[0067] The stage prediction model to be trained may be a prediction model for a specified icing stage that requires training. The stage prediction result may be a predicted icing growth rate for the specified icing stage. The stage icing growth rate may be an icing growth rate calculated based on ice thickness sample values for the specified icing stage.

[0068] In a specific implementation, the micro-meteorological sample values, physical monitoring sample values and ice thickness sample values of the transmission line can be collected and input into the terminal. The terminal determines the sample icing stage corresponding to the micro-meteorological sample values, physical monitoring sample values and ice thickness sample values, adjusts the model parameters of the prediction model to be trained according to the sample icing stage, and obtains the stage prediction model to be trained corresponding to the sample icing stage. The terminal can input the micro-meteorological sample values and physical monitoring sample values into the stage prediction model to be trained to obtain the stage prediction result. The terminal can also calculate the stage icing growth rate based on the ice thickness sample value of the sample icing stage, and adjust the stage prediction model to be trained according to the difference between the stage prediction result and the stage icing growth rate until the difference is less than the preset threshold, and the final stage prediction model is used as the trained stage prediction model.

[0069] In practical applications, according to the characteristics of different stages in the entire icing process, a stage identification evaluation function S(t) can be constructed to divide the icing into stages. The specific formula is:

[0070]

[0071] Among them, Δh(t) represents the rate of change of ice thickness, which can be calculated based on the ice thickness sample value. f(T(t), H(t)) is the comprehensive influence function of temperature T and humidity H, corresponding to the micrometeorological sample value. g(Q(t), θ(t)) is the physical indicator function of the comprehensive load Q and wind angle θ, corresponding to the physical monitoring sample value. α, β, and γ are the weight coefficients of each part.

[0072] According to the value of S(t), the icing process can be divided into three stages, and the sample icing stage is obtained. The specific formula is:

[0073]

[0074] Among them, τ1=10 and τ2=2.5 are empirical thresholds, which can be determined through historical data training.

[0075] For the optimized BiGRU and MKRVM joint model, the feature set, time step, and model weights are adjusted according to the different icing stages. For example, during the icing growth period, sample features focus on temperature, relative humidity, and wind speed, using a smaller time step for accumulation, and increasing the weight of the BiGRU in the joint model. During the icing stability period, sample features focus on wind speed, wind direction, and conductor load, using a larger time step for accumulation, and balancing the weights of the BiGRU and MKRVM in the joint model. During the icing melt period, sample features focus on temperature and solar radiation intensity, taking into account negative growth rates for accumulation, and increasing the weight of the MKRVM in the joint model. This results in the trained stage prediction models corresponding to different icing stages.

[0076] For each icing stage, the filtered sample features are input into the stage prediction model to be trained to obtain the predicted value of the icing growth rate predicted by the stage prediction model to be trained, and the icing growth rate is calculated according to the ice thickness sample value of the corresponding icing stage. The predicted value of the icing growth rate predicted by the stage prediction model to be trained is compared with the icing growth rate calculated according to the ice thickness sample value, and the model parameters are adjusted according to the difference between the two until the difference is less than the preset threshold, so as to obtain the trained stage prediction model corresponding to the icing stage.

[0077] In this embodiment, by obtaining the micrometeorological sample values, physical monitoring sample values and ice thickness sample values of the transmission line, the sample icing stage is determined according to the micrometeorological sample values, physical monitoring sample values and ice thickness sample values, and the stage prediction model to be trained is determined according to the sample icing stage and the prediction model to be trained. The micrometeorological sample values and the physical monitoring sample values are input into the stage prediction model to be trained to obtain the stage prediction results of the stage prediction model to be trained. According to the difference between the stage prediction results and the stage icing growth rate, the stage prediction model to be trained is trained to obtain a trained stage prediction model. The stage icing growth rate is obtained according to the ice thickness sample values of the sample icing stage. The prediction model can be trained separately for each icing stage to obtain a prediction model that is adapted to the icing stage. The use of this model can improve the accuracy of the transmission line icing thickness prediction.

[0078] In an exemplary embodiment, the above-mentioned prediction model to be trained is obtained by fusing a bidirectional gated recurrent unit and a multi-core correlation vector machine; the above-mentioned step of determining the stage prediction model to be trained based on the sample icing stage and the prediction model to be trained may specifically include: determining the first weight of the bidirectional gated recurrent unit and the second weight of the multi-core correlation vector machine based on the sample icing stage; adjusting the bidirectional gated recurrent unit in the prediction model to be trained according to the first weight, and adjusting the multi-core correlation vector machine in the prediction model to be trained according to the second weight to obtain the stage prediction model to be trained.

[0079] The first weight may be the weight of BiGRU in the BiGRU and MKRVM joint model, and the second weight may be the weight of MKRVM in the BiGRU and MKRVM joint model.

[0080] In the specific implementation, the terminal can determine the first weight of BiGRU in the joint model and the second weight of MKRVM according to the sample icing stage, use the first weight to adjust the prediction result of BiGRU, and use the second weight to adjust the prediction result of MKRVM to form a stage prediction model to be trained.

[0081] For example, in the prediction model to be trained, BiGRU and MKRVM can adopt a parallel structure, and the test data are input into BiGRU and MKRVM respectively, and the prediction results of BiGRU can be obtained. And the prediction results of MKRVM The weight β of BiGRU and the weight 1-β of MKRVM are determined according to the sample icing stage. For example, a larger β can be set in the growth period, and a smaller 1-β can be set accordingly; β and 1-β can be balanced in the stable period; a larger 1-β can be set in the melting period, and a smaller β can be set accordingly. The prediction results of BiGRU can then be adjusted according to the determined β. Adjust the MKRVM predictions based on the determined 1-β The prediction results of BiGRU and MKRVM are added together to obtain the prediction result of the stage prediction model to be trained:

[0082]

[0083] In this embodiment, by determining the first weight of the bidirectional gated recurrent unit and the second weight of the multi-core correlation vector machine according to the sample icing stage, adjusting the bidirectional gated recurrent unit in the prediction model to be trained according to the first weight, and adjusting the multi-core correlation vector machine in the prediction model to be trained according to the second weight, a stage prediction model to be trained is obtained. The bidirectional learning ability of BiGRU can be used to simultaneously capture the forward and backward time series features of the sequence data, thereby improving the ability to recognize long-term and short-term dependencies. At the same time, MKRVM is introduced, and the polynomial kernel function of the global response characteristics and the Gaussian kernel function of the local response characteristics are combined to establish a more adaptive icing growth rate prediction model. The weights of BiGRU and MKRVM are adjusted for different icing stages. The time accumulation effect can also be considered in the prediction process to improve the accuracy of the prediction results.

[0084] In an exemplary embodiment, the above-mentioned step S104 may specifically include: determining the initial ice thickness of each ice covering stage; obtaining the stage thickness prediction value of each ice covering stage based on the initial ice covering thickness and the ice covering growth rate prediction value; and fusing the stage thickness prediction values to obtain the ice covering thickness prediction value of the predicted time span.

[0085] The initial ice thickness may be the ice thickness at the start of the ice covering stage, and the stage thickness prediction value may be the ice thickness prediction value at each moment of the ice covering stage.

[0086] In the specific implementation, for each icing stage, the terminal can obtain the initial icing thickness, determine the icing thickness growth value at each moment of the icing stage according to the icing growth rate prediction value, add the icing thickness growth value to the initial icing thickness, and obtain the icing thickness prediction value at each moment of the icing stage, that is, the stage thickness prediction value, and connect the stage thickness prediction values of each icing stage to obtain the icing thickness prediction value of the entire prediction time span.

[0087] For example, you can establish a time accumulation calculation formula

[0088]

[0089] Among them H t represents the ice thickness at time t, is the initial ice thickness at time t-t0, V i is the ice cover growth rate at time i, T i is the data sampling time interval, t0 is the time span parameter. For short-term prediction, t0 can be set to 24 (corresponding to 12 hours, sampling interval is 30 minutes); for medium-term prediction, t0 can be set to 48 (corresponding to 24 hours); for long-term prediction, t0 can be set to 96 (corresponding to 48 hours).

[0090] In this embodiment, by determining the initial ice thickness of each ice covering stage, the stage thickness prediction value of each ice covering stage is obtained according to the initial ice covering thickness and the ice covering growth rate prediction value, and the stage thickness prediction values are fused to obtain the ice covering thickness prediction value of the predicted time span. The cumulative effect can be calculated separately according to the ice covering stage (growth period, stable period, melting period) to ensure the accuracy of predictions at different stages.

[0091] In an exemplary embodiment, the above-mentioned step of fusing the stage thickness prediction values to obtain the ice thickness prediction value of the prediction time span may specifically include: obtaining a first prediction value of the first stage and a second prediction value of the second stage; the first stage is adjacent to the second stage, and there is an overlapping time point between the first stage and the second stage; at the overlapping time point, performing a weighted summation on the first prediction value and the second prediction value to obtain the ice thickness prediction value at the overlapping time point.

[0092] The first stage and the second stage may be two adjacent ice covering stages. The first predicted value may be a stage thickness predicted value at a time point when the first stage overlaps. The second predicted value may be a stage thickness predicted value at a time point when the second stage overlaps.

[0093] In the specific implementation, after obtaining the stage thickness prediction values of two adjacent icing stages, if there is an overlapping time point between the two icing stages, the stage thickness prediction values of each icing stage can be weighted and summed for the overlapping time point as the icing thickness prediction value at the overlapping time point.

[0094] For example, if the predicted thickness value for the next 1 to 5 hours during the growth period and the predicted thickness value for the next 5 to 36 hours during the stable period are obtained, and the overlapping time point is the next 5 hours, then the predicted ice thickness value for the next 5 hours can be the weighted average of the predicted thickness value for the next 5 hours during the growth period and the predicted thickness value for the next 5 hours during the stable period.

[0095] In this embodiment, by obtaining the first prediction value of the first stage and the second prediction value of the second stage, the first stage is adjacent to the second stage, and there is an overlapping time point between the first stage and the second stage, at the overlapping time point, the first prediction value and the second prediction value are weightedly summed to obtain the ice thickness prediction value at the overlapping time point, which can smooth the ice thickness prediction values of adjacent ice stages and increase the accuracy of the ice thickness prediction values at the overlapping time points.

[0096] In order to facilitate those skilled in the art to have a deeper understanding of the embodiments of the present application, a specific example will be used for illustration below.

[0097] In order to meet the differentiated anti-icing needs of the power grid in different icing stages, this application takes into account the characteristics of different icing stages. By organically combining physical laws, time accumulation effects and intelligent algorithms, an ice thickness prediction model that conforms to physical laws and takes time accumulation characteristics into account is constructed, achieving accurate prediction of different icing stages. The transmission line ice thickness prediction model proposed in this application mainly includes:

[0098] (1) Physical guidance model construction. First, the force analysis of the transmission tower line is carried out, and a comprehensive load calculation model under icing conditions is established. The physical relationship between ice thickness, wind deflection angle and comprehensive load is systematically analyzed. Based on this relationship, a loss function that conforms to the laws of physics is constructed to guide the training process of the neural network. Subsequently, a two-layer BiGRU network structure is constructed, and its bidirectional learning ability is used to simultaneously capture the forward and backward time series features of the sequence data, significantly improving the ability to identify long-term and short-term dependencies. At the same time, the multi-core correlation vector machine technology is introduced into the system, combining the polynomial kernel function of the global response characteristics and the Gaussian kernel function of the local response characteristics to establish a more adaptive ice growth rate prediction model.

[0099] (2) Integration of temporal accumulation effects. By analyzing in detail the correlation between ice cover duration and ice cover thickness, an ice cover accumulation model based on the time span t0 is established. This model combines the initial ice cover thickness and the predicted growth rate to accurately calculate the ice cover thickness at different time points in the future. In addition, based on the characteristics of different stages of ice cover development, a phased prediction strategy is designed to dynamically adjust the prediction parameters and weights to improve the pertinence and accuracy of predictions at different stages.

[0100] Figure 2 A flow chart of the construction process of the transmission line ice thickness prediction model is provided. Figure 2 , the model building process mainly includes:

[0101] Step S201: Data input and preprocessing. This model first acquires key data required for icing prediction from multiple data sources. This data primarily includes four types: micrometeorological data, physical monitoring data, historical ice thickness data, and time series information. Micrometeorological data is collected by meteorological stations or sensors installed near conductors; physical monitoring data is obtained through a monitoring system installed on transmission towers; and historical icing data and time information are extracted from historical grid operation records.

[0102] For micrometeorological data, this model focuses on seven key influencing factors: ambient temperature (T), relative humidity (H), wind speed (Ws), wind direction (Wd), precipitation (P), solar radiation intensity (Sr), and atmospheric pressure (Ap). Physical monitoring data focuses on physical parameters such as the conductor's combined load (Q) and wind deflection angle (θ).

[0103] For the ice thickness value of transmission lines, the variable window sliding median method is used to detect and process abnormal values:

[0104]

[0105] Among them, x i represents the thickness of ice to be processed, x' i Indicates the ice thickness after treatment, W i Represents x i A sliding window centered on is the standard deviation of the data in the window, k is an adjustable threshold coefficient (adjusted between 2.5 and 3.0), and median indicates the median.

[0106] Based on the characteristics of different stages in the entire icing process, this model introduces a multi-parameter comprehensive evaluation method to divide the icing stage. By analyzing the change rate of ice thickness and the trend of micro-meteorological characteristics, a stage identification evaluation function is constructed:

[0107]

[0108] Among them, Δh(t) represents the rate of change of ice thickness, f(T(t), H(t)) is the comprehensive influence function of temperature and humidity, g(Q(t), θ(t)) is the physical index function of comprehensive load and wind angle, and α, β, and γ are the weight coefficients of each part.

[0109] According to the value of S(t), the icing process is divided into three stages:

[0110]

[0111] τ1 = 10 and τ2 = 2.5 are empirical thresholds determined through historical data training. This detailed stage division lays the foundation for subsequent targeted predictions.

[0112] The temporal accumulation characteristics of ice cover require that attention be paid to the growth rate of ice thickness rather than its absolute value. This model uses a time window adaptive difference method to process ice thickness data and calculate the ice growth rate:

[0113] V(t)=(h(t)-h(t-Δt)) / Δt·ω(t),

[0114] Where h(t) represents the ice thickness at time t, Δt is the sampling time interval, and ω(t) is the adaptive weight coefficient used to eliminate the influence of small fluctuations:

[0115] ω(t)=1 / (1+exp(-λ·|Δh(t)|)),

[0116] Where λ = 5 is the sensitivity parameter, and |Δh(t)| is the absolute value of the change in ice thickness. When the ice thickness changes significantly, ω(t) approaches 1, preserving the true change information. When the change is minimal, ω(t) approaches 0, effectively suppressing noise interference.

[0117] Through adaptive difference processing, the ice growth rate data set {V(t1), V(t2), ..., V(t N )}, which will serve as the key output variable for subsequent modeling.

[0118] Taking into account the differences in the dimensions and distributions of different input features, this model uses a distribution-adaptive normalization method to normalize each feature. Two normalization strategies are designed for different data characteristics:

[0119] For features x that are approximately normally distributed (such as temperature, pressure, etc.), standardization is performed using the standard score (Z-score):

[0120]

[0121] Where μ is the feature mean and σ is the standard deviation. For features x with non-normal distribution or outliers (such as wind speed, precipitation, etc.), the improved minimum-maximum (Min-Max) normalization is used:

[0122]

[0123] where x min and x max are the minimum and maximum values of the feature, respectively, and ∈=0.01 is used to avoid disturbances caused by extreme values.

[0124] In addition, in order to capture the changing features x at different time scales, a multi-scale sliding window feature construction method is introduced:

[0125]

[0126] By constructing the mean of different window sizes w (30 minutes, 2 hours, 12 hours) and standard deviation Enhance the model's ability to perceive changes at multiple time scales.

[0127] Next, this model further enhances the feature expression capability through spatiotemporal correlation analysis. First, the time lag feature is constructed:

[0128] x (lag) (t)=[x(t),x(t-Δt),x(t-2Δt),...,x(t-KΔt)],

[0129] Where k is the lag order, and the optimal lag order is determined by autocorrelation analysis:

[0130]

[0131] Select Make | R k The maximum K value that is significantly non-zero is taken as the lag order.

[0132] In addition, considering the spatial correlation of transmission line icing, the data of adjacent monitoring points are introduced as auxiliary features:

[0133] X spatial (t)=[x0(t),α1x1(t),α2x2(t),...,α M x M (t)],

[0134] Where x0(t) is the characteristic vector of the target monitoring point, x i (t) is the feature vector of the i-th adjacent monitoring point, α i is the spatial weight coefficient, which is inversely proportional to the distance between monitoring points:

[0135]

[0136] where d i is the distance from the i-th monitoring point to the target point, d0 is the normalization constant, η = 0.1 is the spatial attenuation coefficient, and M is the number of adjacent monitoring points considered.

[0137] Step S202: Feature selection and parameter optimization. After data preprocessing, the model is presented with a multidimensional feature space, including the original micrometeorological features and the multi-scale features constructed through preprocessing. To reduce model complexity and improve generalization, this method uses a modified random forest algorithm to assess the importance of each feature to ice growth rate. The specific steps are as follows:

[0138] First, construct the feature set F = {F1, F2, ..., F n}, contains all candidate features. Define the feature importance evaluation function:

[0139]

[0140] Among them, N T =100 is the number of trees in the random forest, ΔMSE j (F i ) indicates that in the jth tree, the feature F i The increase in mean square error after random permutation:

[0141]

[0142] here Represents the feature F i Results after random permutation.

[0143] To enhance the stability of the model, a weighted feature importance score is introduced:

[0144] WImp(F i )=Imp(F i )·[1+ρ·Corr(F i ,V)],

[0145] Where Corr(F i ,V) represents feature F i Pearson correlation coefficient with ice cover growth rate V, ρ = 0.5 is the correlation weighting coefficient.

[0146] According to the feature importance score, all features are sorted in descending order to obtain the sorted set F (1) ,F (2) ,...,F (n) , where WImp(F (1) )≥WImp(F (2))≥...≥WImp(F (n) Finally, select the top k features whose cumulative importance reaches the preset threshold:

[0147]

[0148] Where η = 0.9 is the importance threshold, which means that the selected feature set contains about 90% of the information.

[0149] After feature selection in the above steps, the features are input into the BiGRU and MKRVM models. The model's parameter optimization ideas are as follows:

[0150] The hybrid optimization algorithm adopts a two-stage strategy of "coarse search-fine search": first, the sparrow search algorithm SSA is used to perform global exploration to determine the potential high-quality areas in the parameter space; then the adaptive parallel Jaya algorithm APJA is used to perform a fine search in the potential high-quality areas to determine the final optimal parameter combination.

[0151] The objective function of parameter optimization is defined as:

[0152] f(Θ)=MSE(Θ)+μ·COMP(Θ),

[0153] Where Θ represents the set of model parameters, MSE(Θ) is the mean square error of cross-validation, COMP(Θ) is the model complexity penalty term, and μ = 0.01 is the trade-off coefficient.

[0154] The parameters to be optimized include:

[0155] The BiGRU network parameters are: batch size: [16, 64]; number of hidden layer units: [32, 128]; learning rate: [0.0001, 0.01]; dropout rate (Dropout): [0.1, 0.5]; number of network layers: [1, 3];

[0156] Multi-kernel correlation vector machine parameters: kernel function weight coefficient: [0,1]; Gaussian kernel width parameter: [0.1,20]; polynomial kernel parameters: [0.01,1], [1,20], [1,3].

[0157] The optimization algorithm adopts a two-stage strategy:

[0158] (1) Use SSA to perform global exploration and identify potential high-quality parameter areas;

[0159] (2) Use Jaya algorithm to perform fine search in high-quality areas.

[0160] The core update rule of the SSA algorithm is based on the foraging and defense behavior of sparrows, while the Jaya algorithm updates parameters based on the principle of approaching the optimal solution and moving away from the worst solution:

[0161]

[0162] in represents the jth variable of the ith solution in the tth iteration, r1 and r2 are random numbers in the interval [0,1], and represent the jth variable of the optimal solution and the worst solution in the current population respectively.

[0163] After parameter optimization, the optimal BiGRU network structure and MKRVM model configuration were determined. The BiGRU network includes both forward and backward directions to capture the bidirectional dependencies of time series data:

[0164]

[0165] where h t represents the hidden state output at time t, and represent the forward and backward hidden states respectively.

[0166] The MKRVM model uses a combined kernel function that combines a polynomial kernel for global response with a Gaussian kernel for local response:

[0167] K(x,x i )=αK G (x,x i )+(1-α)K P (x,x i ),

[0168] Where α is the kernel function weight coefficient, x represents the input feature vector, x i Represents the feature vector of the i-th training sample.

[0169] The Gaussian kernel and polynomial kernel are:

[0170]

[0171] in is the Gaussian kernel width parameter, and r1, r2, and r3 are polynomial kernel parameters. These three parameters are optimized using a hybrid optimization algorithm (a combination of the Sparrow Search Algorithm (SSA) and the Jaya algorithm). Once the optimal values are determined, they remain fixed during the actual prediction process and are no longer automatically updated. represents the vector dot product.

[0172] Step S203: Physics-guided model construction: To introduce physical laws into model training, it is first necessary to establish a force analysis model of the transmission line under ice conditions.

[0173] When ice-covered, transmission lines are primarily subjected to three types of loads: the conductor's own weight, the longitudinal load caused by the weight of the ice, and the lateral load caused by wind pressure. A force analysis model for transmission lines under ice-covered conditions was constructed to extract key physical laws:

[0174] (1) The deadweight load per unit length of conductor is:

[0175] q line =G·9.8×10 -3 ,

[0176] Where G is the mass of the conductor per unit length (kg / km).

[0177] (2) Calculate the unit ice load of the equivalent ice-covered conductor:

[0178] q ice =ρ ice ·π[(r+d) 2 -r 2 ]·9.8×10 -3 ,

[0179] Where r is the wire radius (mm), d is the ice thickness (mm), ρ ice is the equivalent ice density (g / cm 3 ), take 0.9.

[0180] (3) Calculate the resultant load in the vertical direction:

[0181] q v =q line +q ice ,

[0182] (4) Calculate the conductor's comprehensive load Q based on the wind deflection angle θ:

[0183]

[0184] (5) can be simplified to the expression of ice thickness d:

[0185]

[0186] Implement a loss function based on physical consistency to guide model training:

[0187] (1) Define the traditional mean square error loss function:

[0188]

[0189] in is the ice thickness predicted by the model, y t is the actual ice thickness, and n is the number of samples.

[0190] (2) Construct physical consistency judgment conditions:

[0191] Case ①: Δθ(t)≤0 and Δd(t)≤0, theoretically ΔQ≤0;

[0192] Case ②: Δθ(t) ≥ 0 and Δd(t) ≥ 0, theoretically ΔQ ≥ 0;

[0193] Case ③: Other cases

[0194] Where Δθ(t) = |θ(t)| - |θ(t-1)| represents the change in wind angle, Δd(t) = d(t) - d(t-1) represents the change in ice thickness, and ΔQ = Q(t) - Q(t-1) represents the change in comprehensive load.

[0195] (3) Implementing physical consistency loss function:

[0196]

[0197] (4) Define the RULE function: This function is used to quantify the degree of violation of physical laws.

[0198]

[0199] (5) Calculate the average physical consistency loss:

[0200]

[0201] (6) Constructing a comprehensive loss function:

[0202]

[0203] in is a balancing factor used to adjust the weight of experience loss and physical loss. std() represents the standard deviation function.

[0204] The Bidirectional Gated Recurrent Unit (BiGRU) network construction includes:

[0205] (1) Set network structure parameters:

[0206] batch_size: batch size, optimized value (between 16 and 64); hidden_units: number of hidden layer units, optimized value (between 32 and 128); num_layers: number of network layers, optimized value (between 1 and 3); dropout_rate: Dropout rate, optimized value (between 0.1 and 0.5); learning_rate: learning rate, optimized value (between 0.0001 and 0.01).

[0207] (2) Implement the BiGRU calculation process: forward GRU unit calculation; backward GRU unit calculation; bidirectional hidden state merging; output layer prediction.

[0208] (3) GRU unit core calculation steps: update gate, reset gate calculation; candidate hidden state calculation; final hidden state update.

[0209] The MKRVM model for physical booting includes:

[0210] Set MKRVM parameters: α: kernel function weight coefficient, optimized value (between 0 and 1); rg: Gaussian kernel width parameter, optimized value (between 0.1 and 20); r1: polynomial kernel first parameter, optimized value (between 0.01 and 1); r2: polynomial kernel second parameter, optimized value (between 1 and 20); r3: polynomial kernel third parameter, optimized value (between 1 and 3).

[0211] Implement the combined kernel function:

[0212] K C (x,x i )=αK G (x,x i )+(1-α)K P (x,x i ),

[0213] Where x represents the input feature vector, x i Represents the feature vector of the i-th training sample.

[0214] Implement the Gaussian kernel function:

[0215]

[0216] Among them ‖xx i ‖ represents the Euclidean distance between eigenvectors.

[0217] Implement the polynomial kernel function:

[0218]

[0219] where x·x i represents the dot product of eigenvectors.

[0220] Compute MKRVM prediction output:

[0221]

[0222] Where M is the number of related vectors, ω i is the corresponding weight, and ω0 is the bias term.

[0223] The physics-guided training process includes: performing training of the physics-guided model.

[0224] (1) Prepare training data: load the input features after feature selection → load the target ice growth rate → load the comprehensive load and wind angle data;

[0225] (2) Model parameter initialization: Initialize the BiGRU network using optimized hyperparameters → set MKRVM parameters;

[0226] (3) Execute the batch training cycle: forward propagation calculates the predicted value → calculates the empirical loss → calculates the physical consistency loss → calculates the comprehensive loss → back propagation updates the parameters;

[0227] (4) Execute the early stopping strategy: monitor the performance of the validation set → terminate the training when the conditions are met.

[0228] Model fusion implementation includes: performing BiGRU and MKRVM model fusion.

[0229] Calculate the fusion weight β:

[0230]

[0231] where MSE MKRVM and MSE BiGRU are the mean square errors of the two models on the validation set.

[0232] Generate fusion prediction results:

[0233]

[0234] in and These are the prediction results of the two models respectively.

[0235] Step S204: Integration of time accumulation effect. Establishing the time accumulation calculation formula:

[0236]

[0237] Among them H t represents the ice thickness at time t, is the initial ice thickness at time t-t0, V i is the ice cover growth rate at time i, T i is the data sampling time interval, and t0 is the time span parameter.

[0238] Set the time span parameter t0: short-term prediction: t0 is set to 24 (corresponding to 12 hours, sampling interval is 30 minutes); medium-term prediction: t0 is set to 48 (corresponding to 24 hours); long-term prediction: t0 is set to 96 (corresponding to 48 hours).

[0239] Implement segmented cumulative calculation: calculate the cumulative effect separately according to the ice cover stage (growth period, stable period, melting period) to ensure the accuracy of predictions at different stages.

[0240] Stage-adaptive prediction strategy: achieve differentiated predictions for different stages of ice cover development.

[0241] Growth period prediction strategy: Feature focus: temperature, relative humidity, wind speed; Accumulation parameter: use smaller time step accumulation; Model weight: increase BiGRU network weight β;

[0242] Stable period prediction strategy: Feature focus: wind speed, wind direction, conductor load; Accumulation parameter: use a larger time step accumulation; Model weight: BiGRU and MKRVM weight balance;

[0243] Melting period prediction strategy: Feature focus: temperature, solar radiation intensity; Accumulation parameter: Consider negative growth rate for accumulation; Model weight: Increase the MKRVM model weight (1-β).

[0244] Automatic phase switching: According to the phase identification function S(t) defined above, the prediction strategy that best suits the current icing phase is dynamically selected during the prediction process.

[0245] Integration of growth rate and thickness prediction: Integrate ice cover growth rate prediction with thickness prediction:

[0246] Initial thickness determination: Use the most recent observation value as the initial thickness When observations are not available, historical contemporaneous averages are used for estimation.

[0247] Growth rate prediction process: Use BiGRU network to predict ice cover growth rate V at future time points BiGRU , use the MKRVM model to predict the ice cover growth rate V at future time points MKRVM The fusion weight β is then applied to calculate the final growth rate forecast:

[0248] V final =β·V BiGRU +(1-β)·V MKRVM ,

[0249] Thickness accumulation calculation: According to the predicted growth rate sequence V final , the time accumulation formula is applied to calculate the future ice thickness, and physical consistency constraints are considered during the calculation process to ensure that the prediction results conform to physical laws.

[0250] Adaptive time window adjustment: Implement adaptive adjustment of the prediction time window:

[0251] t0=t 0,base ·exp(-γ w·||V recent ||),

[0252] where t 0,base is the base window size (fixed value 24), γ w is the window adjustment coefficient (fixed value 0.5), V recent is the average absolute value of the ice cover growth rate in the recent period.

[0253] Adaptive adjustment of window step size: when the growth rate changes dramatically, reduce the prediction step size; when the growth rate changes slowly, increase the prediction step size; step size adjustment formula:

[0254] Δt=Δt base ·(1+δ·exp(-σ·var(V recent ))),

[0255] where Δt base is the basic step size (fixed value 1), δ and σ are adjustment coefficients (fixed to 0.5 and 5.0 respectively), var(V recent ) is the variance of the recent growth rate.

[0256] Forecast accuracy evaluation indicators: Design evaluation indicators for time-accumulated predictions.

[0257] Time-weighted mean square error:

[0258] where w i =exp(-κ·i) is the time weight, which decays as the prediction step increases. κ is the decay coefficient (fixed value 0.05);

[0259] Cumulative error rate:

[0260] Maximum error time indicator: This indicator indicates the point in time when the forecast error is the largest.

[0261] Multi-timescale integration and optimization. Build a multi-scale forecast ensemble: short-term forecast: 6 hours ahead, 30-minute step; medium-term forecast: 24 hours ahead, 1-hour step; long-term forecast: 72 hours ahead, 3-hour step.

[0262] Integration of multi-scale prediction results:

[0263] For overlapping time points, a weighted average strategy is used:

[0264] in is the weight of the k-th time scale prediction, which is inversely proportional to the prediction error.

[0265] Smoothing of prediction results: Applying a sliding average filter to smooth the results

[0266] Where m is the half-width of the smoothing window (fixed value 2).

[0267] Next, we will enhance the model's generalization capabilities under different meteorological conditions:

[0268] Meteorological condition classification: Mild icing conditions: temperature -3℃ to 0℃, relative humidity 85% to 90%; moderate icing conditions: temperature -8℃ to -3℃, relative humidity 90% to 95%; heavy icing conditions: temperature below -8℃, relative humidity above 95%;

[0269] Conditional adaptability parameter adjustment:

[0270] where ξ c is the adjustment coefficient corresponding to meteorological condition c (mild: 0.8, moderate: 1.0, severe: 1.2).

[0271] Dynamically adjust ice density ρ according to temperature and humidity conditions ice :

[0272] ρ ice =ρ base ·(1+η T ·T+η H ·H), where ρ base The basic ice density (fixed value 0.9g / cm 3 ), η T and η H are the temperature and humidity influence coefficients (fixed values -0.01 and 0.005), respectively.

[0273] Step S205: Model implementation and verification. Integrate the above modules into a complete prediction system:

[0274] (1) System architecture design:

[0275] Data collection layer: real-time collection module for micro-meteorological data and physical monitoring data;

[0276] Data processing layer: feature extraction, data preprocessing, and data storage modules;

[0277] Model calculation layer: physical guidance and time accumulation prediction core module;

[0278] Result display layer: prediction result visualization and warning information push module;

[0279] (2) Model integration process:

[0280] Input feature data stream: input preprocessed multi-source feature data into the prediction system;

[0281] Prediction calculation process: perform BiGRU and MKRVM prediction calculations and perform fusion;

[0282] Time accumulation calculation: perform time accumulation thickness calculation based on growth rate prediction results;

[0283] Output result stream: Generate multi-time-scale ice thickness prediction results;

[0284] (3) Model calling interface design:

[0285] Initialization interface: initialize(config_params), load model parameters;

[0286] Prediction interface: predict(input_features,forecast_horizon), performs prediction calculation;

[0287] Update interface: update(new_data), real-time update of model input data;

[0288] Evaluation interface: evaluate(ground_truth), evaluates prediction accuracy.

[0289] Build real-time prediction processes to support continuous operation monitoring:

[0290] (1) Real-time data acquisition and preprocessing: Collect micro-meteorological data at fixed intervals (30 minutes) → Collect transmission line physical parameter data → Execute data preprocessing process, including outlier processing and feature normalization;

[0291] (2) Sliding window prediction mechanism: Maintain a fixed-length historical data window (length 2t_0) → After each data update, the window is slid and a new prediction calculation is triggered → the prediction result is updated, overwriting the old prediction value;

[0292] (3) Prediction result self-correction mechanism: Compare the previous prediction result with the actual observation value, calculate the prediction deviation → adjust the prediction parameters according to the deviation, such as the model fusion weight β;

[0293] (4) Apply the bias correction function to adjust subsequent prediction results:

[0294]

[0295] Among them, bias recent is the average relative deviation of the most recent forecasts, is the correction factor (fixed value 0.7).

[0296] Implement early warning decision support based on prediction results

[0297] (1) Icing warning levels: Level 1 warning (blue): predicted ice thickness 5-10 mm; Level 2 warning (yellow): predicted ice thickness 10-15 mm; Level 3 warning (orange): predicted ice thickness 15-25 mm; Level 4 warning (red): predicted ice thickness > 25 mm.

[0298] (2) Generation of warning information: Ice thickness warning: Determine the warning level based on the predicted ice thickness; Ice growth rate warning: Determine the risk growth rate based on the ice growth rate; Ice duration warning: Evaluate the cumulative risk based on the expected duration of ice.

[0299] (3) De-icing strategy recommendations: Generate corresponding de-icing operation recommendations for different warning levels, provide decision support information on the best de-icing time and method, and generate a sorted list of priority processing sections.

[0300] Application example: This example uses a 500kV cable line in a certain province, with a conductor model of LGJ-400 / 35 and a conductor radius of r = 13.8mm. From tower 37 to tower 38 in the mountainous section, a tower-based micro-meteorological station and an online icing monitoring device collect data including temperature, humidity, wind speed, wind direction, ice thickness, wind deflection, and comprehensive load at a frequency of 30 minutes.

[0301] First, we collect continuous data from January 10 to January 20, 2023. The following is a sample of some raw data:

[0302] Table 1. Sample graph of raw data

[0303]

[0304] Then the data is cleaned: data that does not meet the basic conditions for icing (temperature > 0°C or relative humidity < 85%) are eliminated, and outliers such as sudden changes in wind speed and integrated load are processed; the sliding median method is used to smooth the ice thickness data.

[0305] Next, perform first-order difference processing to calculate the ice thickness growth rate:

[0306] Table 2 Examples of ice thickness growth rates

[0307] Time point Ice thickness (mm) Growth rate (mm / h) 01-10 08:00 0 01-10 08:30 01-10 09:00 0.2 0.5 0.4 0.6 ... ... ...

[0308] Then, each feature is normalized so that its value range is [0, 1].

[0309] The second step is feature selection and parameter optimization. First, the random forest algorithm is used to evaluate the importance of each feature to the ice growth rate:

[0310] Table 3 Example of feature importance

[0311] feature Importance score temperature 0.382 relative humidity 0.245 wind speed 0.196 wind direction 0.083 Comprehensive load change rate 0.058 Wind angle 0.036

[0312] According to the principle that the cumulative importance reaches 90%, temperature, relative humidity and wind speed are selected as the main input features.

[0313] Next, we used a hybrid optimization algorithm to determine the optimal parameters. BiGRU network parameters: batch size (batch_size): 32; number of hidden units (hidden_units): 64; learning rate (learning_rate): 0.0008; dropout rate (dropout_rate): 0.3; number of network layers (num_layers): 2; MKRVM parameters: kernel weight coefficient (α): 0.65; Gaussian kernel width parameter (rg): 5.8; polynomial kernel parameters (r1, r2, r3): 0.08, 6.5, 2.

[0314] Step 3: Physical guidance model construction. According to the conductor parameters, an ice covering physical model is established: conductor radius r = 13.8 mm, conductor mass per unit length G = 1435 kg / km, and ice density ρ_ice = 0.9 g / cm 3 .

[0315] Calculate the self-weight load per unit length: q_line = 1435 × 9.8 × 10^-3 = 14.063 kN / km.

[0316] Analyze the ice cover data from January 12 to 14 and extract the physical consistency rules:

[0317] Table 4 Physical consistency example table

[0318]

[0319]

[0320] According to the physical consistency rule, the loss function is constructed: empirical loss: MSE = 0.0086; physical consistency loss: PHYloss = 0.0034; balance factor λ = 1.24; comprehensive loss: f = 0.0086 + 1.24 × 0.0034 = 0.0128;

[0321] The training was performed using data from January 10th to 15th.

[0322] Step 4: Integration of time accumulation effect. According to the historical icing characteristics of the cable line, the following settings are made: time span parameter t0 = 24 (12 hours); basic window size t0,base = 24; window adjustment coefficient γ ω =0.5;

[0323] Then use BiGRU and MKRVM to predict the ice growth rate on January 16. Based on the growth rate prediction results, calculate the predicted ice thickness on January 16:

[0324] Table 5 Example of ice thickness prediction values

[0325] Time point Initial thickness Cumulative growth Predicted thickness Actual thickness error 01-1608:00 8.5mm 0.00mm 8.50mm 8.50mm 0.00mm 01-16 08:30 0 0.10mm 8.60mm 8.65mm 0.05mm 01-16 09:00 0 0.15mm 8.75mm 8.82mm 0.07mm ... ... ... ... ... ... 01-16 20:00 0 3.86mm 12.36mm 12.58mm 0.22mm

[0326] Based on the icing development trend, January 16 was identified as a period of increasing icing, and the prediction strategy was adjusted: features focused on temperature and relative humidity; the time step was reduced to 30 minutes; and the BiGRU weight β was increased to 0.7.

[0327] Step 5: Model implementation and verification: Generate ice cover prediction results for three time scales: short-term prediction (next 6 hours): average error 0.18mm, maximum error 0.31mm, physical inconsistency 3.5%; medium-term prediction (next 24 hours): average error 0.42mm, maximum error 0.86mm, physical inconsistency 6.3%; long-term prediction (next 72 hours): average error 1.05mm, maximum error 2.15mm, physical inconsistency 8.7%.

[0328] Finally, the forecast results were compared with the actual observations from January 16 to 17: Time Weighted Mean Square Error (TMSE): 0.0843; Cumulative Error Rate (CER): 3.86%; Maximum Error Time (METI): 05:30 on January 17;

[0329] Compared with traditional methods: traditional physical model: TMSE = 0.3267, CER = 12.45%; traditional machine learning model: TMSE = 0.1582, CER = 7.23%; this model: TMSE = 0.0843, CER = 3.86%; the prediction accuracy is improved by more than 46%, and the physical consistency is significantly improved.

[0330] In one embodiment, Figure 3 As shown, a method for predicting ice thickness of transmission lines is provided. The method is described by taking the application of the method to a terminal as an example, and includes the following steps:

[0331] Step S301: Obtain a parameter space of a prediction model to be trained, perform a first search on the parameter space based on a sparrow search method to obtain a first parameter combination, perform a second search on the first parameter combination based on an adaptive parallel search method to obtain a second parameter combination, and determine the prediction model to be trained based on the second parameter combination:

[0332] Step S302: Acquire micro-meteorological sample values, physical monitoring sample values, and ice thickness sample values of the transmission line; determine the sample ice stage based on the micro-meteorological sample values, physical monitoring sample values, and ice thickness sample values; determine the stage prediction model to be trained based on the sample ice stage and the prediction model to be trained; input the micro-meteorological sample values and physical monitoring sample values into the stage prediction model to be trained to obtain a stage prediction result of the stage prediction model to be trained; train the stage prediction model to be trained based on the difference between the stage prediction result and the stage ice growth rate to obtain a trained stage prediction model; the stage ice growth rate is obtained based on the ice thickness sample values of the sample ice stage;

[0333] Step S303: Obtain the predicted time span of ice thickness, determine the ice stage corresponding to the predicted time span, determine micrometeorological characteristics and physical monitoring characteristics based on the micrometeorological information and physical monitoring information of the ice stage, input the micrometeorological characteristics and physical monitoring characteristics into the trained stage prediction model, and obtain a predicted value of ice growth rate for the ice stage;

[0334] Step S304: determine the initial ice thickness of each ice covering stage, obtain the stage thickness prediction value of each ice covering stage based on the initial ice covering thickness and the ice covering growth rate prediction value, and fuse the stage thickness prediction values to obtain the ice covering thickness prediction value of the prediction time span.

[0335] In a specific implementation, the terminal can obtain the parameter space of the prediction model to be trained, first use SSA to perform a first search on the parameter space to obtain a first parameter combination, and then use APJA to perform a second search on the first parameter combination to obtain a second parameter combination, and use the second parameter combination as the model parameters of the prediction model to be trained, collect the micro-meteorological sample values, physical monitoring sample values and ice thickness sample values of the transmission line and input them into the terminal, the terminal determines the sample icing stage corresponding to the micro-meteorological sample values, physical monitoring sample values and ice thickness sample values, adjusts the model parameters of the prediction model to be trained according to the sample icing stage, obtains the stage prediction model to be trained corresponding to the sample icing stage, inputs the micro-meteorological sample values and physical monitoring sample values into the stage prediction model to be trained, obtains the stage prediction result, and adjusts the model parameters of the prediction model to be trained according to the sample icing stage according to the sample icing stage. The stage ice growth rate is calculated based on the thickness sample value. The stage prediction model to be trained is adjusted according to the difference between the stage prediction result and the stage ice growth rate until the difference is less than the preset threshold. The final stage prediction model is used as the trained stage prediction model, and the prediction time span is divided into multiple ice stages. The micrometeorological information and physical monitoring information of each ice stage are feature extracted to obtain the micrometeorological features and physical monitoring features respectively. The micrometeorological features and physical monitoring features are input into the trained stage prediction model corresponding to the ice stage to obtain the ice growth rate prediction value of each ice stage. The stage thickness prediction value is calculated based on the initial ice thickness and ice growth rate prediction value of each ice stage. The stage thickness prediction values of each ice stage are integrated to obtain the ice thickness prediction value of the entire prediction time span.

[0336] The above-mentioned method for predicting the ice thickness of transmission lines determines the prediction model to be trained, obtains the trained stage prediction model of each icing stage according to the prediction model to be trained, predicts the ice growth rate of each icing stage by the trained stage prediction model, and then calculates the predicted ice thickness value. It can combine the physical laws in the icing process and the time accumulation effect of ice thickness to predict the ice thickness for different icing stages under the prediction time span, accurately predict the ice thickness in different icing stages, and improve the accuracy of ice thickness prediction of transmission lines.

[0337] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0338] Based on the same inventive concept, embodiments of the present application also provide a transmission line ice thickness prediction device for implementing the aforementioned transmission line ice thickness prediction method. The solution provided by this device is similar to the solution described in the aforementioned method. Therefore, the specific limitations of one or more embodiments of the transmission line ice thickness prediction device provided below can be found in the limitations of the transmission line ice thickness prediction method described above and will not be further elaborated here.

[0339] In an exemplary embodiment, Figure 4 As shown, a device for predicting ice thickness of a transmission line is provided, comprising: an acquisition module 401, a determination module 402, a prediction module 403 and a calculation module 404, wherein:

[0340] An acquisition module 401 is configured to acquire a predicted time span of ice thickness and determine an ice stage corresponding to the predicted time span;

[0341] A determination module 402 is configured to determine micrometeorological characteristics and physical monitoring characteristics based on the micrometeorological information and physical monitoring information during the icing stage;

[0342] A prediction module 403 is configured to input the micrometeorological characteristics and the physical monitoring characteristics into a trained stage prediction model to obtain a predicted value of ice cover growth rate in the ice cover stage;

[0343] The calculation module 404 is configured to determine a predicted ice thickness value for the predicted time span according to the predicted ice growth rate value.

[0344] In an exemplary embodiment, the above-mentioned transmission line ice thickness prediction device also includes a training module for obtaining a parameter space of a prediction model to be trained; performing a first search on the parameter space based on a sparrow search method to obtain a first parameter combination; performing a second search on the first parameter combination based on an adaptive parallel search method to obtain a second parameter combination; and determining the prediction model to be trained based on the second parameter combination.

[0345] In an exemplary embodiment, the above-mentioned training module is also used to obtain the micrometeorological sample values, physical monitoring sample values and ice thickness sample values of the transmission line; determine the sample icing stage according to the micrometeorological sample values, the physical monitoring sample values and the ice thickness sample values; determine the stage prediction model to be trained according to the sample icing stage and the prediction model to be trained; input the micrometeorological sample values and the physical monitoring sample values into the stage prediction model to be trained to obtain the stage prediction result of the stage prediction model to be trained; train the stage prediction model to be trained according to the difference between the stage prediction result and the stage icing growth rate to obtain the trained stage prediction model; the stage icing growth rate is obtained according to the ice thickness sample value of the sample icing stage.

[0346] In an exemplary embodiment, the above-mentioned training module is also used to determine the first weight of the bidirectional gated recurrent unit and the second weight of the multi-core correlation vector machine according to the sample icing stage; adjust the bidirectional gated recurrent unit in the prediction model to be trained according to the first weight, and adjust the multi-core correlation vector machine in the prediction model to be trained according to the second weight to obtain the stage prediction model to be trained.

[0347] In an exemplary embodiment, the calculation module 404 is further used to determine the initial ice thickness of each of the ice covering stages; obtain the stage thickness prediction value of each of the ice covering stages based on the initial ice covering thickness and the ice covering growth rate prediction value; and fuse the stage thickness prediction values to obtain the ice covering thickness prediction value of the prediction time span.

[0348] In an exemplary embodiment, the above-mentioned calculation module 404 is also used to obtain a first prediction value of the first stage and a second prediction value of the second stage; the first stage is adjacent to the second stage, and there is an overlapping time point between the first stage and the second stage; at the overlapping time point, the first prediction value and the second prediction value are weightedly summed to obtain the ice thickness prediction value at the overlapping time point.

[0349] Each module in the aforementioned transmission line ice thickness prediction device can be implemented in whole or in part through software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in hardware form, or stored in a computer device memory in software form, so that the processor can call and execute the corresponding operations of each module.

[0350] In an exemplary embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as shown in FIG. Figure 5 As shown. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. The processor, the memory and the input / output interface are connected via a system bus, and the communication interface, the display unit and the input device are connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, a mobile cellular network, near field communication (NFC) or other technologies. When the computer program is executed by the processor, a method for predicting the ice thickness of a transmission line is implemented. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device casing, or an external keyboard, touchpad or mouse.

[0351] Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0352] In one embodiment, a computer device is further provided, including a memory and a processor. The memory stores a computer program, and the processor implements the steps in the above method embodiments when executing the computer program.

[0353] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0354] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0355] 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, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0356] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile memory and volatile memory. Non-volatile memory may 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 may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processor involved in the various embodiments provided herein may be, but are not limited to, a general-purpose processor, a central processing unit (CPU), a graphics processing unit (GPU), a digital signal processor (DSP), a programmable logic unit (PLC), a data processing logic unit based on quantum computing, an artificial intelligence (AI) processor, and the like.

[0357] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, 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 application.

[0358] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A method for predicting ice thickness of transmission lines, characterized in that: The method comprises: Obtaining a predicted time span of ice thickness, and determining an ice stage corresponding to the predicted time span; determining micrometeorological characteristics and physical monitoring characteristics respectively according to the micrometeorological information and physical monitoring information during the icing stage; Inputting the micro-meteorological characteristics and the physical monitoring characteristics into a trained stage prediction model to obtain a predicted value of ice cover growth rate in the ice cover stage; A predicted value of ice thickness for the predicted time span is determined according to the predicted value of ice growth rate.

2. The method according to claim 1, characterized in that The method further comprises: Obtain the parameter space of the prediction model to be trained; Performing a first search on the parameter space based on a sparrow search method to obtain a first parameter combination; Performing a second search on the first parameter combination based on an adaptive parallel search method to obtain a second parameter combination; The prediction model to be trained is determined according to the second parameter combination.

3. The method according to claim 2, characterized in that After determining the prediction model to be trained according to the second parameter combination, the method further includes: Obtaining micro-meteorological sample values, physical monitoring sample values, and ice thickness sample values of the transmission line; Determining a sample icing stage according to the micrometeorological sample value, the physical monitoring sample value, and the ice thickness sample value; Determining a stage prediction model to be trained according to the sample ice coverage stage and the prediction model to be trained; Inputting the micrometeorological sample values and the physical monitoring sample values into the stage prediction model to be trained to obtain a stage prediction result of the stage prediction model to be trained; According to the difference between the stage prediction result and the stage icing growth rate, the stage prediction model to be trained is trained to obtain the trained stage prediction model; the stage icing growth rate is obtained according to the ice thickness sample value of the sample icing stage.

4. The method according to claim 3, characterized in that The prediction model to be trained is obtained by fusing a bidirectional gated recurrent unit and a multi-core relevance vector machine; The step of determining a stage prediction model to be trained based on the sample ice-covered stage and the prediction model to be trained includes: Determining a first weight of the bidirectional gated recurrent unit and a second weight of the multi-core relevance vector machine according to the sample icing stage; The bidirectional gated recurrent unit in the prediction model to be trained is adjusted according to the first weight, and the multi-core relevance vector machine in the prediction model to be trained is adjusted according to the second weight to obtain the stage prediction model to be trained.

5. The method according to claim 1, wherein Determining the predicted ice thickness value for the predicted time span based on the predicted ice growth rate includes: determining an initial ice thickness in each of the ice accumulation stages; Obtaining a stage thickness prediction value for each of the ice covering stages according to the initial ice covering thickness and the ice covering growth rate prediction value; The stage thickness prediction values are fused to obtain the ice thickness prediction value of the prediction time span.

6. The method according to claim 5, characterized in that The step of fusing the stage thickness prediction values to obtain the ice thickness prediction value for the prediction time span includes: Obtaining a first prediction value of a first stage and a second prediction value of a second stage; the first stage is adjacent to the second stage, and there is an overlapping time point between the first stage and the second stage; At the overlapping time point, a weighted sum is performed on the first prediction value and the second prediction value to obtain a predicted value of ice thickness at the overlapping time point.

7. A device for predicting ice thickness of transmission lines, characterized in that: The device comprises: an acquisition module, configured to acquire a predicted time span of ice thickness and determine an ice stage corresponding to the predicted time span; a determination module, configured to determine micrometeorological characteristics and physical monitoring characteristics respectively according to the micrometeorological information and physical monitoring information during the icing stage; A prediction module, configured to input the micrometeorological characteristics and the physical monitoring characteristics into a trained stage prediction model to obtain a predicted value of ice growth rate in the ice accumulation stage; A calculation module is used to determine a predicted ice thickness value for the predicted time span based on the predicted ice growth rate value.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images