A method and system for predicting ice thickness of railway contact network

Abstract

The present application relates to the technical field of railway contact networks, and provides a method and system for predicting ice thickness of railway contact networks. The method comprises: inputting meteorological environment parameters into a machine learning model, using the machine learning model to dynamically calibrate the ice medium parameters based on the meteorological environment parameters, and obtaining calibrated ice medium parameters; obtaining an ice growth coefficient based on the meteorological environment parameters, the physical characteristic parameters of the conductor, and the calibrated ice medium parameters; and inputting the calibrated ice medium parameters and the ice growth coefficient into a Makkonen model to obtain an ice thickness prediction value. The technical solution provided by the present application improves the accuracy of ice thickness prediction under complex meteorological conditions by dynamically calibrating key ice parameters through machine learning and optimizing the calculation logic of the Makkonen model, thereby solving the problem of prediction failure caused by insufficient environmental adaptability of traditional static models.

Description

Technical Field

[0001] The present application relates to the technical field of railway contact networks, and in particular to a method and system for predicting ice thickness of railway contact networks. Background Art

[0002] The railway catenary system is a core component of the electrified railway's traction power supply system. Composed of exposed conductors such as contact wires and catenary cables, it transmits electrical energy to electric locomotives. In cold and humid regions (such as mountainous areas and snowy areas), ice easily forms on the contact wire surface, causing conductor vibration, pantograph damage, and even wire breakage and pole collapse, posing a serious threat to operating safety. For example, high ice thickness can cause abnormal current flow to trains, while excessive conductor vibration amplitude can cause derailments. Therefore, accurate prediction of ice thickness is a core requirement for ensuring safe all-weather railway operations and provides a direct basis for decision-making on ice and snow melting.

[0003] Currently, three types of technologies are mainly used to monitor the ice thickness of railway contact networks: (1) manual inspection and track inspection vehicles identify ice through visual inspection or on-board equipment, but accessibility is poor in high-altitude mountainous areas and the response is delayed; (2) image recognition is combined with infrared thermal imaging technology, and cameras are used to capture the ice morphology, but the false alarm rate is high in rainy and foggy weather, and it is difficult to quantify the thickness; (3) the static Makkonen model transplanted by the industry directly applies the empirical formula of the transmission line, completely ignoring the unique structure of the contact line. For example, the diameter of the contact line is only 8-15mm, which is much smaller than the transmission line in the power industry, resulting in inaccurate thickness prediction.

[0004] However, existing solutions have the following core limitations: first, static parameters fail. The traditional Makkonen model uses a fixed liquid water content formula and a constant ice surface temperature assumption, which cannot adapt to the complex microclimate environment of the contact network; second, the lack of physical characteristics and the accumulation of open-loop errors result in large interference errors, which in turn cause the prediction results to deviate seriously from reality. Summary of the Invention

[0005] The present application provides a method and system for predicting the ice thickness of a railway contact network, which is used to solve the problem of low accuracy in predicting the ice thickness of a railway contact network in the prior art and improve the accuracy of ice thickness prediction under complex meteorological conditions.

[0006] In a first aspect, the present application provides a method for predicting ice thickness of a railway contact network, comprising:

[0007] Obtain meteorological environment parameters, conductor physical characteristic parameters and icing medium parameters of the contact wire area in the railway overhead contact network;

[0008] Inputting the meteorological environment parameters into a machine learning model, and dynamically calibrating the icing medium parameters based on the meteorological environment parameters using the machine learning model to obtain calibrated icing medium parameters;

[0009] Obtaining an ice growth coefficient based on the meteorological environment parameters, the conductor physical characteristic parameters, and the calibrated ice medium parameters;

[0010] The calibrated ice-covered medium parameters and the ice-covered growth coefficient are input into the Makkonen model to obtain a predicted ice-covered thickness value.

[0011] Optionally, the meteorological environment parameters include humidity, ambient temperature and wind speed;

[0012] The step of inputting the meteorological environment parameters into a machine learning model, and dynamically calibrating the icing medium parameters based on the meteorological environment parameters using the machine learning model to obtain calibrated icing medium parameters includes:

[0013] When the humidity is greater than a preset humidity threshold and the ambient temperature is less than or equal to a preset temperature threshold, the temperature, humidity, and wind speed time series in the historical icing data of the railway contact network are analyzed using the machine learning model, wherein the machine learning model is a hybrid neural network model, and the hybrid neural network model includes a bidirectional long short-term memory network layer and a convolutional neural network layer, which are used to respectively extract the long-term dependence characteristics and local mutation characteristics of the temperature, humidity, and wind speed time series in the historical icing data;

[0014] Optimizing the dynamic weight parameters corresponding to humidity, ambient temperature, and wind speed in a preset joint correction model of humidity, ambient temperature, and wind speed based on the long-term dependence characteristics and local mutation characteristics;

[0015] After the dynamic weight parameter optimization is completed, the optimized humidity, ambient temperature and wind speed joint correction model is used to calculate the calibrated ice-covered medium parameters.

[0016] Optionally, the optimizing the dynamic weight parameters corresponding to humidity, ambient temperature and wind speed in a preset joint correction model of humidity, ambient temperature and wind speed according to the long-term dependence characteristics and the local mutation characteristics includes:

[0017] According to the duration feature of the humidity being greater than the preset humidity threshold in the long-term dependency feature, adjusting the dynamic weight parameter corresponding to the humidity according to a preset exponential decay rule;

[0018] According to the moisture accumulation feature in the long-term dependency feature where the temperature is less than a preset temperature threshold, a dynamic weight parameter corresponding to the temperature is calculated using an S-type function;

[0019] Based on the coupling characteristics between the temperature drop and the wind speed in the local mutation characteristics, a dynamic weight parameter corresponding to the wind speed is calculated.

[0020] Optionally, the conductor physical characteristic parameters include conductor diameter, contact line surface roughness coefficient and conductor surface radiation parameter, and the ice growth coefficient includes collision coefficient, freezing coefficient and capture coefficient;

[0021] The generating of the ice growth coefficient based on the meteorological environment parameter, the conductor physical characteristic parameter, and the calibrated ice medium parameter includes:

[0022] Calculating a collision coefficient using a preset collision coefficient correction function based on the wire diameter and the calibrated ice-covered medium parameters;

[0023] Based on the ambient temperature, the ice surface temperature and the conductor surface radiation parameter, a freezing coefficient is obtained through a balance relationship between radiation heat transfer and convection heat transfer;

[0024] A preset cube root function is used to quantify the attenuation effect of the contact line surface roughness coefficient on the water droplet capture efficiency to obtain the capture coefficient.

[0025] Optionally, the freezing coefficient is obtained based on the ambient temperature, the ice surface temperature, and the conductor surface radiation parameter by a balance between radiation heat transfer and convection heat transfer, including:

[0026] Calculating the heat radiation power of the conductor surface using the Stefan-Boltzmann law based on the conductor surface radiation parameters;

[0027] Calculating a forced convection heat transfer coefficient based on the wind speed and the ambient temperature, and calculating a convection heat transfer power based on the forced convection heat transfer coefficient;

[0028] Based on the heat radiation power of the conductor surface and the convection heat transfer power, a heat balance analysis is performed on the ice surface to generate a freezing coefficient.

[0029] Optionally, inputting the calibrated icing medium parameters and the icing growth coefficient into a Makkonen model to obtain an icing thickness prediction value includes:

[0030] Based on the collision coefficient, freezing coefficient and capture coefficient, combined with wind speed and calibrated ice medium parameters, the ice mass growth equation in the Makkonen model is used to solve the ice mass growth per unit time.

[0031] Convert the ice mass increment to the equivalent cylindrical ice thickness;

[0032] The dancing amplitude of the contact line is obtained, and the ice thickness of the equivalent cylinder is adjusted according to the dancing amplitude to obtain a predicted value of the ice thickness.

[0033] Optionally, after inputting the calibrated icing medium parameters and the icing growth coefficient into a Makkonen model to obtain a predicted value of icing thickness, the method further includes:

[0034] When the absolute deviation between the actual ice thickness value and the predicted ice thickness value is greater than a preset deviation threshold, the prediction error gradient is calculated using a gradient descent algorithm;

[0035] Based on the prediction error gradient, a partial derivative optimization calculation is performed on the weight corresponding to the wire diameter in the collision coefficient correction function to obtain the collision coefficient correction function after parameter adjustment.

[0036] In a second aspect, the present application provides a railway contact network ice thickness prediction system, comprising:

[0037] An acquisition module is used to obtain meteorological environment parameters, conductor physical characteristic parameters and icing medium parameters of the area where the contact wire in the railway contact network is located;

[0038] a calibration module, configured to input the meteorological environment parameters into a machine learning model, and dynamically calibrate the icing medium parameters based on the meteorological environment parameters using the machine learning model to obtain calibrated icing medium parameters;

[0039] A generating module, configured to generate an ice growth coefficient based on the meteorological environment parameters, the conductor physical characteristic parameters, and the calibrated ice medium parameters;

[0040] The prediction module is used to input the calibrated ice-covered medium parameters and the ice-covered growth coefficient into the Makkonen model to obtain an ice-covered thickness prediction value.

[0041] In a third aspect, the present application provides a computing device comprising a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a method for predicting the ice thickness of a railway contact network as described in any one of the first aspects.

[0042] In a fourth aspect, the present application provides a computer storage medium storing a computer program. When the computer program is executed by a computer, it implements a method for predicting the ice thickness of a railway contact network as described in any one of the first aspects.

[0043] In the present application, a method for predicting ice thickness of a railway contact network is provided, which includes: obtaining meteorological environment parameters, conductor physical characteristic parameters and ice medium parameters of the area where the contact line in the railway contact network is located; inputting the meteorological environment parameters into a machine learning model, and using the machine learning model to dynamically calibrate the ice medium parameters based on the meteorological environment parameters to obtain calibrated ice medium parameters; obtaining an ice growth coefficient based on the meteorological environment parameters, the conductor physical characteristic parameters and the calibrated ice medium parameters; inputting the calibrated ice medium parameters and the ice growth coefficient into a Makkonen model to obtain a predicted value of ice thickness.

[0044] In this application, non-contact sensors and indirect monitoring data are integrated to capture dynamic meteorological parameters such as temperature, humidity, and wind speed, as well as physical characteristics such as conductor diameter and surface state, in real time, while ensuring the stability of the contact network structure. This addresses the limited sensor deployment problem of traditional solutions. Historical icing data is used to train the model, and the liquid water content fitting coefficient is replaced from a fixed value to a dynamic weight parameter that is jointly corrected by humidity, temperature, and wind speed. The ice surface thermodynamic equilibrium factor is introduced to correct the freezing coefficient in real time, eliminating the prediction deviation of the static empirical formula in low temperature and high humidity environments. Based on the dynamically calibrated medium parameters, real-time meteorological data, and conductor characteristics, the calculation logic of the collision coefficient and capture coefficient is optimized, and a real-time interaction mechanism between the meteorological environment, conductors, and icing medium is established to improve the physical coupling accuracy of the icing growth model. The calibrated icing medium parameters and icing growth coefficient are input into the improved Makkonen model, and the prediction error is corrected in a closed-loop manner through a gradient descent algorithm combined with laser ranging feedback data. The final output is the ice thickness value, which achieves early warning and reduces the hysteresis and mechanical interference risk of traditional mechanical indirect monitoring methods.

[0045] Furthermore, when the humidity is greater than the preset humidity threshold and the ambient temperature is less than or equal to the preset temperature threshold, the long-term dependence characteristics and local mutation characteristics of the temperature, humidity and wind speed time series in the historical icing data are extracted through a hybrid neural network model, and based on this, the weight parameters of the temperature, humidity and wind joint correction model are dynamically optimized: the humidity weight is adjusted according to the exponential decay rule according to the duration of high humidity to enhance the cumulative effect of continuous high humidity; the temperature weight is calculated based on the low-temperature moisture accumulation characteristics through the S-type function to accurately quantify the critical nonlinear characteristics of the phase change near the freezing point; the wind speed weight is dynamically calculated based on the temperature drop and wind speed coupling characteristics to capture the instantaneous impact of wind speed mutation on liquid water transport. By accurately characterizing the meteorological coupling mechanism through the fusion of long-term and short-term features, using exponential decay weights to enhance the cumulative effect of continuous high humidity, and using S-type function weights to quantify the critical phase change nonlinearity of the freezing point, the accuracy of medium parameters in low temperature and high humidity scenarios is improved; with the help of convolutional neural networks, local mutation characteristics are captured, and wind speed weights are adjusted in real time to avoid the lag in liquid water content prediction caused by traditional averaging processing; while maintaining the physical interpretability of the joint correction model, on the basis of machine learning optimization, the defect of low safety warning reliability of the pure black box model is overcome.

[0046] These and other aspects of the present application will become more readily apparent from the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0048] Figure 1 A flow chart of a method for predicting ice thickness of a railway contact network provided in an embodiment of the present application;

[0049] Figure 2 A schematic diagram of the structure of a railway contact network ice thickness prediction system provided in an embodiment of the present application;

[0050] Figure 3 A schematic diagram of the structure of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0051] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0052] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this document or may be executed in parallel. The serial numbers of the operations, such as 11, 12, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any order of execution. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to being different types.

[0053] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0054] In order to solve the problem of low accuracy in railway catenary ice thickness prediction caused by the lack of early warning, mechanical interference error and lack of meteorological coupling in the prior art, an embodiment of the present application provides a railway catenary ice thickness prediction method, which adopts the following ideas: meteorological environment parameters, conductor physical characteristic parameters and initial icing medium parameters are acquired in real time through multi-source non-contact sensing technology to construct a dynamic data input layer; a hybrid neural network model is designed to deeply explore the temperature, humidity and wind time series coupling rules in historical icing data, and a bidirectional long-short-term memory network is used to extract long-term dependent features and a convolutional neural network is used to capture local mutation features, and the weight parameters of the joint correction model are dynamically optimized to achieve real-time calibration of the icing medium parameters; the calibrated icing medium parameters and the icing growth coefficient determined by the meteorological environment parameters and conductor characteristics are input into the Makkonen model to reconstruct the calculation logic of the collision coefficient, freezing coefficient and capture coefficient; finally, through closed-loop feedback of the measured ice thickness value and the predicted ice thickness value, a gradient descent algorithm is used to dynamically correct the growth coefficient deviation, forming an iterative optimization mechanism of monitoring, calibration, prediction and verification, to achieve millimeter-level accuracy in ice thickness prediction, thereby enabling accurate dynamic warning.

[0055] Figure 1 A flow chart of a method for predicting ice thickness of a railway contact network provided in an embodiment of the present application is shown as follows: Figure 1 As shown, the method includes:

[0056] S11. Obtain meteorological environment parameters, conductor physical characteristic parameters, and icing medium parameters of the area where the contact wire in the railway contact network is located.

[0057] Among them, the railway contact network refers to the overhead wire system that transmits electricity to electric locomotives in electrified railways, including contact wires, load-bearing cables and tension compensation devices. The contact wire refers to the exposed wire in the railway contact network that is in direct sliding contact with the pantograph, and its icing state directly affects the safety of power supply. Meteorological environmental parameters refer to dynamic data reflecting the microclimate conditions of the contact wire, including temperature, humidity, wind speed and liquid water content. The physical characteristic parameters of the conductor refer to the parameters that describe the inherent properties of the contact wire, including conductor diameter, surface roughness coefficient and surface radiation heat dissipation characteristics. The icing medium parameters characterize the key physical quantities of the icing formation process, mainly referring to the liquid water content available for freezing in the air.

[0058] In an embodiment of the present application, first, non-contact meteorological sensors deployed in the contact network area are used to collect temperature, humidity and wind speed data in real time as meteorological environment parameters; secondly, millimeter-wave radar or pre-buried optical fiber sensors are used to capture the conductor diameter, surface roughness coefficient and radiation parameters as conductor physical characteristic parameters; at the same time, ice-covered medium parameters such as liquid water content are obtained based on laser rangefinders or electric field inversion technology, and finally a multi-source dynamic monitoring data set is formed.

[0059] S12. Input the meteorological environment parameters into the machine learning model, and use the machine learning model to dynamically calibrate the icing medium parameters based on the meteorological environment parameters to obtain the calibrated icing medium parameters.

[0060] The machine learning model uses a hybrid architecture of a bidirectional long-short-term memory network and a convolutional neural network to extract long-term dependencies and local mutation characteristics from meteorological time series data. Dynamic calibration involves optimizing the calculation logic for adjusting the parameters of the icing medium through weighted parameters based on real-time meteorological data and historical icing records. The calibrated icing medium parameters refer to the liquid water content values ​​corrected by the machine learning model, eliminating the environmental adaptability flaws of static empirical formulas.

[0061] In an embodiment of the present application, the real-time collected temperature, humidity, and wind speed time series data are first input into the hybrid neural network model; secondly, the humidity weight is optimized according to the exponential decay rule based on the high humidity duration characteristics, the temperature weight is calculated by the S-type function combined with the low-temperature moisture accumulation characteristics, and the wind speed weight is dynamically adjusted based on the temperature drop-wind speed coupling characteristics; finally, the liquid water content is recalculated using the optimized temperature, humidity, and wind joint correction model, and the calibrated ice-covered medium parameters are output.

[0062] S13. Based on the meteorological environment parameters, the physical characteristic parameters of the conductor and the calibrated ice medium parameters, an ice growth coefficient is obtained.

[0063] Among them, the ice growth coefficient refers to a set of physical parameters that quantify the ice growth rate, including the collision efficiency coefficient, freezing efficiency coefficient and water droplet capture coefficient.

[0064] In an embodiment of the present application, first, the collision coefficient is calculated using a collision coefficient correction function based on the mapping relationship between the wire diameter and the calibrated liquid water content; secondly, based on the ambient temperature, ice surface temperature and wire surface radiation parameters, the Stefan-Boltzmann law is used to calculate the thermal radiation power, and the convective heat transfer power is obtained in combination with the forced convection heat transfer coefficient, and the freezing coefficient is generated through thermal balance analysis; finally, based on the attenuation effect of the contact line surface roughness coefficient on the water droplet capture efficiency, the capture coefficient is quantified using the cube root function, and finally integrated into the ice growth coefficient.

[0065] S14. Input the calibrated icing medium parameters and icing growth coefficient into the Makkonen model to obtain a predicted value of icing thickness.

[0066] The Makkonen model is a thermodynamic-hydrodynamic equation that describes the growth of ice mass on conductors. Its inputs are the parameters of the icing medium and the ice growth coefficient, and its output is the ice mass increment. The predicted ice thickness is the geometric thickness value converted from the ice mass increment based on the assumption of an equivalent cylindrical ice layer. It can also be referred to as the final output after correction for the dancing amplitude.

[0067] In an embodiment of the present application, the collision coefficient, freezing coefficient, capture coefficient and calibrated liquid water content are first input into the ice mass growth equation in the Makkonen model to solve the ice mass increment per unit time; secondly, the ice mass increment is converted into an equivalent cylindrical ice thickness; then, the equivalent ice thickness is dynamically corrected in combination with the dancing amplitude measured by the millimeter-wave radar; finally, the ice thickness prediction value is output, and the accuracy is verified by closed-loop verification of laser ranging measured data.

[0068] Here's a specific example: On an electrified railway crossing a high-altitude mountainous area, a catenary section encountered freezing rain. A meteorological station installed near the tracks first collected real-time meteorological data showing an ambient temperature of -3°C, a relative humidity of 90%, and a wind speed of 8 m / s. Pre-embedded fiber optic sensors also measured the catenary wire diameter as 15 mm, a surface roughness coefficient of 0.2, and a heat dissipation parameter of 0.85. Millimeter-wave radar detected a concentration of 0.2 g / m³ of freezable water droplets in the air. This data was fed into an intelligent analysis system for processing. For example, the system's long-term analysis unit detected that humidity levels exceeding 85% had persisted for five hours. The rapid response unit also captured an unusual fluctuation: a sudden increase of 3 m / s in wind speed accompanied by a 2°C drop in temperature within two minutes. Based on these characteristics, the computational model weights were dynamically adjusted, calibrating the original water drop concentration of 0.2 g / m³ to a more accurate 0.25 g / m³. Next, the key parameters for ice formation were calculated: based on the corresponding relationship between a 15mm diameter conductor and a water droplet content of 0.25g / m³, the collision coefficient of 0.6 was calculated using the collision efficiency formula. Combined with the ambient temperature of -3°C, the ice surface temperature of -5°C, and the conductor heat dissipation parameter of 0.85, the freezing coefficient of 0.7 was calculated using the principle of heat inflow and outflow balance. Considering the weakening effect of the surface roughness of 0.2 on water droplet adhesion, the capture coefficient of 0.85 was calculated using the cube root function. Substituting the above parameters into the ice growth model for prediction, for example, the ice mass increment per second = 0.6 × 0.85 × 8 × 0.25 × 0.7 = 0.15 grams. After 5 minutes (300 seconds), the accumulated mass = 0.15 × 300 = 45 grams, which translates to an ice thickness of 0.8 mm. At this point, the millimeter-wave radar detected a 150 mm wave motion of the conductor. Based on the rule of "compensating 5% thickness for every 300 mm increase in wave motion," a correction is applied: for example, 0.8 × (1 + 0.05 × 150 ÷ ​​300) = 0.85 mm. Thirty minutes after the prediction, the laser rangefinder measured an ice thickness of 0.9 mm, a deviation of only 0.05 mm from the predicted value of 0.85 mm, well below the 1 mm error threshold, validating the model's accuracy. The system continuously monitored the entire process, and because 0.85 mm did not reach the 5 mm ice melt threshold, no alarm was triggered.

[0069] By executing S11 to S14, the embodiment of the present application improves the accuracy of ice medium parameters under complex meteorological conditions by integrating non-invasive sensor data with a machine learning dynamic calibration mechanism; optimizes the calculation of ice growth coefficients by combining thermodynamic equilibrium to solve the physical distortion caused by the traditional model ignoring the temperature change of the ice surface; and finally achieves a highly adaptable prediction of ice thickness within the Makkonen framework, reducing the risk of warning failure caused by environmental coupling effects and mechanical interference.

[0070] In a possible embodiment, the meteorological environment parameters include humidity, ambient temperature and wind speed.

[0071] The meteorological environment parameters are input into the machine learning model, and the machine learning model is used to dynamically calibrate the icing medium parameters based on the meteorological environment parameters to obtain the calibrated icing medium parameters, including:

[0072] Step 121: When the humidity is greater than a preset humidity threshold and the ambient temperature is less than or equal to a preset temperature threshold, a machine learning model is used to analyze the temperature, humidity, and wind speed time series in the historical icing data of the railway contact network. The machine learning model is a hybrid neural network model, which includes a bidirectional long short-term memory network layer and a convolutional neural network layer, and is used to respectively extract the long-term dependence characteristics and local mutation characteristics of the temperature, humidity, and wind speed time series in the historical icing data.

[0073] The icing medium parameter can be liquid water content. The preset temperature threshold and humidity threshold are defined by railway specifications, with the minimum and maximum humidity values ​​set, respectively, to trigger the icing prediction mechanism. Historical icing data, including a database of past meteorological time series and measured ice thickness, is used to train machine learning models. A temperature time series is a chronological collection of ambient temperature monitoring values ​​that reflects temperature trends. A humidity time series is a continuously recorded relative humidity dataset that characterizes the continuity of water supply. A wind speed time series is a time series of historical wind speeds used to analyze sudden airflow patterns. A bidirectional long-short-term memory network layer is a network layer capable of forward and reverse parsing of time series data, capturing long-term dependencies. A convolutional neural network layer uses convolutional kernels to scan local data and extract short-term sudden change features. Long-term dependency features are abstract expressions that reflect the persistent patterns of changes in meteorological parameters, such as the length of periods of high humidity. Local sudden change features are features that describe sudden fluctuations in meteorological parameters, such as a sudden increase in wind speed accompanied by a drop in temperature.

[0074] In an embodiment of the present application, it is first determined whether the real-time collected humidity exceeds a preset humidity threshold and whether the ambient temperature is not higher than a preset temperature threshold; when both conditions are met, the hybrid neural network model is activated; secondly, the temperature time series, humidity time series, and wind speed time series in the historical icing database are input into the model; then, the long-term periodic dependency characteristics of the meteorological parameters are extracted through a bidirectional long-short-term memory network layer, and the short-term local mutation characteristics are captured through a convolutional neural network layer; finally, a meteorological coupling feature vector that integrates long-term and short-term characteristics is output.

[0075] Step 122: Optimize the dynamic weight parameters corresponding to humidity, ambient temperature, and wind speed in the preset humidity, ambient temperature, and wind speed joint correction model based on the long-term dependency characteristics and the local mutation characteristics.

[0076] The combined temperature, humidity, and wind correction model dynamically weights and integrates the physical equations for temperature, humidity, and wind speed to calculate liquid water content. Dynamic weight parameters are model coefficients that adjust in real time based on meteorological conditions and quantify the contribution of each parameter to ice cover.

[0077] In the embodiment of the present application, the high humidity duration component in the long-term dependence feature is first analyzed, and the humidity weight is enhanced according to the exponential decay rule to strengthen the impact of persistent high humidity; secondly, based on the low-temperature moisture accumulation component in the long-term dependence feature, the temperature weight is calculated using an S-type function to quantify the nonlinear phase change effect near the freezing point; then, according to the temperature drop-wind speed coupling component in the local mutation feature, the wind speed weight is dynamically adjusted to respond to instantaneous meteorological mutations; finally, the coordinated optimization of the three types of dynamic weight parameters in the temperature, humidity and wind joint correction model is completed.

[0078] Step 123: After the dynamic weight parameter optimization is completed, the calibrated icing medium parameters are calculated using the optimized humidity, ambient temperature and wind speed joint correction model.

[0079] Among them, in the embodiment of the present application, the dynamic weight parameters optimized in step 122 are first loaded into the temperature, humidity and wind joint correction model; secondly, the real-time collected temperature, humidity and wind speed data are input; then the exponential term in the liquid water content expression is recalculated based on the weight distribution mechanism; finally, the calibrated icing medium parameters that eliminate static deviations are output.

[0080] The following is a specific example: the humidity in the contact line area is monitored to be 90%, exceeding the preset humidity threshold of 85%, and the ambient temperature drops to -3°C, meeting the trigger condition of ≤0°C. The hybrid neural network model is activated to analyze the temperature, humidity and wind speed time series in the historical database. The bidirectional long short-term memory network layer extracts the high humidity feature lasting for 5 hours as a long-term dependent feature. The convolutional neural network layer identifies the local mutation pattern of a sudden increase of 3m / s in wind speed accompanied by a temperature drop of 2°C within 3 minutes. Parameter optimization is then performed based on the extracted features. For example, based on the duration of high humidity for 5 hours, the exponential decay rule is used to calculate the dynamic weight corresponding to humidity to be 0.7. Based on the accumulated moisture feature of a temperature drop of 2°C, the dynamic weight corresponding to temperature is output as 0.6 through the S-type function. Combined with the temperature drop-wind speed coupling feature, the wind speed suddenly changes by 3m / s. According to the linear relationship, the weight corresponding to wind speed is 0.8. Finally, the optimized dynamic weight is loaded into the temperature, humidity and wind joint correction model, and the real-time meteorological data is input to recalculate the liquid water content, and the calibrated ice medium parameter of 0.25g / m is output. 3 .

[0081] By executing steps 121 to 123, the embodiment of the present application focuses on key meteorological scenarios through an intelligent triggering mechanism, uses a hybrid neural network to deeply mine the long-term and short-term meteorological coupling laws in historical data, and dynamically optimizes the weight distribution logic of the joint correction model, effectively solving the medium parameter distortion problem of the traditional static model in low temperature, high humidity and meteorological mutation scenarios, and improving the environmental adaptability of icing prediction.

[0082] In one possible embodiment, step 122, optimizing the dynamic weight parameters corresponding to humidity, ambient temperature, and wind speed in a preset joint correction model of humidity, ambient temperature, and wind speed based on the long-term dependency characteristics and the local mutation characteristics, includes:

[0083] Step a1: According to the duration feature of the humidity being greater than the preset humidity threshold in the long-term dependency feature, the dynamic weight parameter corresponding to the humidity is adjusted according to a preset exponential decay rule.

[0084] The preset humidity threshold is a minimum humidity threshold set based on ice-covered water vapor conditions, typically above 85%. The duration feature reflects the cumulative duration that humidity continuously exceeds the threshold and is used to quantify the persistent impact of high humidity environments. The preset exponential decay rule is a weight adjustment function that increases linearly with time in the initial phase and then decreases exponentially to simulate saturation effects.

[0085] In an embodiment of the present application, first, the total duration data of the humidity continuously exceeding the preset humidity threshold is extracted from the long-term dependent features; secondly, according to the size of the duration data, the humidity weight parameter is dynamically adjusted using a preset exponential decay function, wherein the longer the duration, the greater the initial weight enhancement amplitude, but it shows a decay trend as the duration increases; finally, the updated humidity dynamic weight parameter value is output.

[0086] Step a2: Calculate the dynamic weight parameter corresponding to the temperature using an S-type function based on the moisture accumulation feature in the long-term dependency feature where the temperature is less than a preset temperature threshold.

[0087] The moisture accumulation characteristic describes the concentration of unfrozen water vapor in low-temperature environments and is positively correlated with the duration of subfreezing temperatures. The sigmoid function, a mathematical function shaped like an S-curve, is used to smoothly map moisture accumulation values ​​to a range between zero and one, accurately capturing the nonlinear weight changes near the freezing point.

[0088] In an embodiment of the present application, first, the continuous low temperature data below the preset temperature threshold in the long-term dependent characteristics is analyzed to quantify the moisture accumulation effect in this temperature range; secondly, the moisture accumulation value is input into the S-type function for nonlinear mapping, so that the temperature weight presents a smooth transition characteristic in the critical region of the freezing point; finally, the temperature dynamic weight parameter is generated to match the phase change process.

[0089] It should be noted that the embodiments of the present application do not specifically limit the content of the preset exponential decay rule in the above process and the expression of the S-type function.

[0090] Step a3: Calculate the dynamic weight parameter corresponding to the wind speed based on the coupling characteristics between the temperature drop and the wind speed in the local mutation characteristics.

[0091] Temperature drop refers to the magnitude of the temperature decrease per unit time, reflecting the intensity of the cold air invasion. The coupling characteristic characterizes the synergistic effect of temperature drop and wind speed change, quantifying the strength of the interaction by multiplying the rates of change of the two.

[0092] In an embodiment of the present application, first, the temperature drop and wind speed surge events that occur simultaneously in the local mutation characteristics are identified, and the correlation coefficient of the change amplitudes of the two is extracted; secondly, a proportional relationship model between the temperature drop gradient and the wind speed increment is established, and the wind speed weight parameter is dynamically calculated based on this; finally, the real-time wind speed dynamic weight value in response to the meteorological mutation is output.

[0093] The following is a specific example: First, based on long-term dependency characteristics, it is identified that humidity has exceeded the preset threshold of 85% for five consecutive hours. The humidity weight is then increased from a baseline value of 0.5 to 0.7 using an exponential decay rule. Secondly, based on the moisture accumulation characteristic of temperatures remaining below freezing for three hours, the moisture accumulation value is mapped to a temperature weight of 0.6 using a sigmoid function. Subsequently, a localized sudden change characteristic, a 4°C temperature drop accompanied by a 5 m / s wind speed surge within 10 minutes, is detected. A wind speed weight of 0.8 is calculated based on the proportional model of the temperature drop gradient and wind speed increment. Finally, the optimized dynamic weight parameters (0.7, 0.6, 0.8) are loaded into the joint correction model to complete the medium parameter calibration.

[0094] By executing steps a1 to a3, the embodiment of the present application quantifies the cumulative effect of continuous high humidity through duration characteristics, uses the S-type function to analyze the phase change characteristics of moisture in the critical zone of the freezing point, combines the temperature drop-wind speed coupling mechanism to capture meteorological mutation events, realizes the physical interpretability optimization of dynamic weight parameters, and effectively improves the calculation accuracy of medium parameters in complex meteorological scenarios.

[0095] In a possible embodiment, S13, the conductor physical characteristic parameters include the conductor diameter, the contact line surface roughness coefficient and the conductor surface radiation parameter, and the ice growth coefficient includes the collision coefficient, the freezing coefficient and the capture coefficient.

[0096] Based on meteorological environment parameters, conductor physical characteristic parameters and calibrated icing medium parameters, an ice growth coefficient is generated, including:

[0097] Step 131 : Calculate the collision coefficient using a preset collision coefficient correction function based on the conductor diameter and the calibrated ice-covered medium parameters.

[0098] The conductor diameter refers to the geometric dimensions of the contact wire's cross-section, which directly affects the cross-sectional area of ​​the water droplet's impact. The contact wire surface roughness coefficient is a dimensionless parameter that quantifies the degree of surface roughness on the conductor, ranging from zero to one. A larger value indicates a rougher surface. The conductor surface radiation parameters describe the conductor's thermal radiation capacity, including surface emissivity and emissivity. The collision coefficient reflects the efficiency of water droplets impacting the conductor and is equal to the ratio of the actual impacting water volume to the theoretical maximum impacting water volume.

[0099] In an embodiment of the present application, the physical diameter data of the contact wire and the liquid water content parameters calibrated by machine learning are first obtained; secondly, the product of the wire diameter and the liquid water content is input into a preset collision coefficient correction function, which calculates the basic collision efficiency by multiplying the wire diameter by the liquid water content and then dividing it by an empirical constant; then, a wire surface curvature correction factor is introduced to smooth and optimize the basic value; finally, a collision coefficient value that dynamically adapts to meteorological conditions is output.

[0100] Step 132: Based on the ambient temperature, the ice surface temperature, and the conductor surface radiation parameters, the freezing coefficient is obtained through the balance relationship between radiation heat transfer and convection heat transfer.

[0101] Among them, the freezing coefficient refers to a physical quantity that characterizes the freezing ratio of the impacting water droplets, which is determined by the thermal equilibrium state of the ice surface. The capture coefficient refers to the actual water droplet capture efficiency affected by surface characteristics, which is equal to the ratio of the captured water volume to the impacting water volume. The preset collision coefficient correction function refers to an equation established based on fluid dynamics, and the input is the product term of the wire diameter and the liquid water content. The ice surface temperature refers to the real-time temperature value of the ice-covered surface of the wire, which is obtained through thermal imaging or pre-buried optical fiber sensors. Radiation heat transfer refers to the physical process of heat transfer in the form of electromagnetic waves, and the power is proportional to the fourth power of the surface temperature. Convective heat transfer refers to the heat transfer caused by fluid flow, and the power depends on the wind speed and temperature difference. The equilibrium relationship is the thermodynamic steady-state condition for the thermal radiation power of the wire surface to be equal to the convective heat transfer power.

[0102] In an embodiment of the present application, the ambient temperature, the real-time temperature of the ice surface, and the emissivity parameters of the conductor surface are first read synchronously; secondly, the thermal radiation power of the conductor surface is calculated according to the Stefan-Boltzmann law, which is equal to the surface emissivity multiplied by the fourth power of the absolute temperature and then multiplied by the Stefan constant; at the same time, the forced convection heat transfer coefficient is calculated based on the wind speed and the ambient temperature, and the convection heat transfer power is obtained; then, a balance equation of the thermal radiation power and the convection heat transfer power is established to solve the freezing efficiency of the ice surface under the thermodynamic steady state; and finally, the freezing coefficient that eliminates the heat transfer error is output.

[0103] Step 133: Use a preset cube root function to quantify the attenuation effect of the contact line surface roughness coefficient on the water droplet capture efficiency to obtain a capture coefficient.

[0104] The preset cube root function is a mathematical function that takes the form of one minus the cube root of the roughness coefficient and is used to quantify the interception effect of a rough surface on water flow. The droplet capture efficiency guideline is the ratio of the number of droplets actually trapped to the total number of impacting droplets. The attenuation effect refers to the physical effect of surface roughness causing droplet detachment, which increases nonlinearly with increasing roughness.

[0105] In an embodiment of the present application, the measured value of the surface roughness coefficient of the contact line is first extracted; secondly, the roughness coefficient is input into a preset cube root function, which calculates the basic attenuation rate by subtracting the cube of the roughness coefficient from one; then, the attenuation rate is dynamically compensated in combination with the wind speed gradient; finally, the capture coefficient representing the actual water droplet capture efficiency is output.

[0106] The following is a specific example: First, based on a wire diameter of 15mm and a calibrated liquid water content of 0.25g / m³, a collision coefficient correction function is used to calculate a basic collision efficiency of 0.7. After curvature correction, a collision coefficient of 0.68 is output. Next, the ambient temperature of -3°C, the ice surface temperature of -5°C, and the emissivity of 0.85 are read to calculate the thermal radiation power of 5.0W / m². Combined with a wind speed of 8m / s, the convective heat transfer power is calculated to be 5.0W / m², generating a freezing coefficient of 0.72 in thermal equilibrium. The surface roughness coefficient of 0.2 is then extracted, and the basic attenuation rate of 0.42 is calculated using the cube root function. After compensating for the wind speed gradient, the capture coefficient is output as 0.83, which is finally integrated into the ice growth coefficient set.

[0107] By executing steps 131 to 133, the embodiment of the present application integrates the dynamic characteristics of the conductor size and the medium parameters through the collision coefficient correction function, uses the thermal balance relationship to accurately solve the freezing coefficient, and combines the cube root function to quantify the nonlinear attenuation of the surface roughness on the capture efficiency, thereby realizing the coordinated optimization of the physical mechanism of the ice growth coefficient and real-time data.

[0108] In one possible embodiment, step 133, based on the ambient temperature, the ice surface temperature, and the conductor surface radiation parameter, obtaining the freezing coefficient by balancing the radiation heat transfer and the convection heat transfer, includes:

[0109] Step b1: Based on the conductor surface radiation parameters, the Stefan-Boltzmann law is used to calculate the conductor surface thermal radiation power.

[0110] The Stefan-Boltzmann law is a physical law that describes the relationship between blackbody radiation power and temperature. Its formula is: thermal radiation power equals the emissivity multiplied by the Stefan constant multiplied by the fourth power of the absolute temperature. The thermal radiation power of a conductor's surface is the amount of energy dissipated by the conductor per unit area and per unit time through thermal radiation. It is determined by both the surface emissivity and the temperature.

[0111] In the embodiment of the present application, the surface emissivity parameters of the conductor and the real-time temperature data of the ice surface are first obtained. Then, according to the Stefan-Boltzmann law, the surface emissivity is multiplied by the Stefan constant and then by the fourth power of the absolute temperature of the ice surface to calculate the thermal radiation power per unit area of ​​the conductor surface. Finally, the thermal radiation power value that quantifies the radiation heat dissipation capacity is output.

[0112] Step b2: Calculate the forced convection heat transfer coefficient based on the wind speed and the ambient temperature, and calculate the convection heat transfer power based on the forced convection heat transfer coefficient.

[0113] The forced convection heat transfer coefficient reflects the heat transfer capacity of forced air flow. It is proportional to wind speed and is used to calculate convection heat transfer. Convection heat transfer power refers to the heat transfer power caused by air flow and is equal to the convection heat transfer coefficient multiplied by the product of the temperature difference and the surface area.

[0114] In an embodiment of the present application, the real-time wind speed and ambient temperature data are first read; secondly, the forced convection heat transfer coefficient is calculated using the Nusselt empirical formula based on the correlation between the wind speed and the wire diameter; the convection heat transfer coefficient is then multiplied by the difference between the ambient temperature and the ice surface temperature, and then multiplied by the wire surface area to obtain the convection heat transfer power; and finally, a power value representing the air flow heat transfer effect is output.

[0115] Step b3: Based on the heat radiation power and convection heat transfer power of the conductor surface, a heat balance analysis is performed on the ice surface to generate a freezing coefficient.

[0116] Among them, thermal balance analysis refers to solving the energy conservation state of stable ice surface temperature by establishing an equation that thermal radiation power is equal to convective heat transfer power.

[0117] In an embodiment of the present application, first, a thermodynamic equilibrium equation is established in which the thermal radiation power is equal to the convective heat transfer power; secondly, the energy conservation relationship under the thermal steady state of the ice surface in the equation is solved; then, the ratio of the actual frozen water volume to the theoretical maximum frozen water volume is used as the freezing efficiency; and finally, the freezing coefficient that eliminates the heat transfer error is output.

[0118] The following is a specific example: First, the Stefan-Boltzmann law is used to calculate the thermal radiation power of the conductor surface based on the emissivity and ice surface temperature. Next, the forced convection heat transfer coefficient is calculated using the Nusselt formula based on the wind speed and ambient temperature. The convection heat transfer power is then calculated by combining the temperature difference and the surface area. Finally, a balance equation is established, where the thermal radiation power equals the convection heat transfer power. The steady-state freezing efficiency is then solved and the freezing coefficient is output.

[0119] By executing steps b1 to b3, the embodiment of the present application establishes a thermodynamic steady-state equation for the ice surface by accurately quantifying the energy exchange process between thermal radiation and forced convection, thereby solving the distortion in the calculation of the freezing coefficient caused by the traditional model assuming a constant ice surface temperature, and improving the physical rationality of ice cover prediction under the wet growth model.

[0120] In a possible embodiment, S14, inputting the calibrated icing medium parameters and the icing growth coefficient into the Makkonen model to obtain a predicted value of icing thickness, includes:

[0121] Step 141: Based on the collision coefficient, freezing coefficient, and capture coefficient, combined with wind speed and calibrated ice medium parameters, the ice mass growth equation in the Makkonen model is used to solve the ice mass growth per unit time.

[0122] The ice mass growth equation, the core equation of the Makkonen model, describes the rate of ice mass accumulation per unit time by calculating the mass increment equal to the collision coefficient multiplied by the capture coefficient multiplied by the freezing coefficient multiplied by the liquid water content multiplied by the wind speed. The ice mass growth per unit time is the amount of ice accumulated per minute or second on the surface of the guide line, reflecting the ice growth rate.

[0123] In an embodiment of the present application, the dynamically optimized collision coefficient, freezing coefficient, and capture coefficient are first obtained, and the real-time wind speed and calibrated liquid water content parameters are read synchronously; secondly, the above parameters are input into the Makkonen ice mass growth equation, which obtains the ice mass growth per unit time by multiplying the collision coefficient by the capture coefficient by the freezing coefficient and then by the liquid water content by the wind speed; finally, the physical value of the quantified ice rate is output.

[0124] Step 142: Convert the ice-covered mass increment into an equivalent cylindrical ice-covered thickness.

[0125] The ice mass increment refers to the specific numerical result of the ice mass growth per unit time. The equivalent cylindrical ice thickness refers to the geometric thickness calculated by volumetric inversion after simplifying the irregular ice cover into a cylindrical ice layer model that uniformly wraps the conductor.

[0126] In an embodiment of the present application, the ice volume increment is first calculated based on the ice mass increment divided by the ice density; secondly, based on the contact line length and the equivalent cylindrical model, the ice volume increment is divided by the product of the conductor circumference and the ice layer thickness, and the equivalent cylindrical ice thickness is inversely solved; finally, a geometric thickness representation value is output.

[0127] Step 143: Obtain the dancing amplitude of the contact line, adjust the ice thickness of the equivalent cylinder according to the dancing amplitude, and obtain a predicted ice thickness value.

[0128] Among them, the dancing amplitude refers to the maximum vibration displacement of the contact line in the vertical or horizontal direction, which is captured in real time by millimeter wave radar or fiber optic sensor.

[0129] In an embodiment of the present application, first, the dancing amplitude data of the contact line is monitored in real time by millimeter-wave radar or embedded optical fiber sensors; secondly, a negative correlation correction model between the dancing amplitude and the centrifugal shedding effect of the ice layer is established, and the larger the amplitude, the smaller the thickness correction coefficient; finally, the equivalent ice thickness is multiplied by the correction coefficient, and the ice thickness prediction value that dynamically adapts to the influence of dancing is output.

[0130] The following is a specific example: First, the Makkonen equation is used to calculate the ice mass increase per unit time based on the collision coefficient, freezing coefficient, and capture coefficient, combined with wind speed and liquid water content. Next, the mass increase is converted to volume increment based on ice density, and the equivalent cylindrical ice thickness is calculated based on the conductor circumference. The millimeter-wave radar vibration amplitude is then obtained, and the equivalent thickness is adjusted using a negative correlation model between amplitude and correction coefficient, ultimately outputting a predicted ice thickness.

[0131] By executing steps 141 to 143, the embodiment of the present application accurately quantifies the ice mass accumulation process through physical equations, uses an equivalent geometric model to achieve observable conversion of mass to thickness, and innovatively introduces a dynamic correction mechanism for dancing amplitude to solve the problem of prediction deviation caused by traditional methods that ignore mechanical vibration and lead to ice shedding, thereby improving the reliability of thickness prediction under complex working conditions.

[0132] In a possible embodiment, after step 131, inputting the calibrated icing medium parameters and the icing growth coefficient into the Makkonen model to obtain the predicted icing thickness value, the method further includes:

[0133] Step c1: When the absolute deviation between the measured ice thickness value and the predicted ice thickness value is greater than a preset deviation threshold, the prediction error gradient is calculated using a gradient descent algorithm.

[0134] The measured ice thickness refers to the geometric thickness of the ice layer on the contact line measured directly by a laser ranging sensor or millimeter-wave radar. The absolute deviation refers to the absolute value of the difference between the predicted ice thickness and the measured value, quantifying the magnitude of the model prediction error. The preset deviation threshold refers to the maximum allowable error value set according to the monitoring accuracy requirements, which is used to trigger the model correction mechanism. The gradient descent algorithm refers to an optimization method that iteratively calculates the partial derivatives of the loss function with respect to the model parameters and updates the parameters along the negative gradient direction. The prediction error gradient refers to the set of partial derivatives of the loss function with respect to each parameter in the collision coefficient correction function, indicating the direction of parameter adjustment.

[0135] In an embodiment of the present application, first, the actual value of the ice thickness is obtained by a laser ranging sensor, and the difference between it and the predicted value is calculated to obtain an absolute deviation; secondly, it is determined whether the deviation exceeds a preset deviation threshold; if it exceeds, the gradient descent algorithm is activated, and the partial derivatives of the loss value with respect to all parameters in the collision coefficient correction function are calculated using the predicted error as the loss function; finally, the predicted error gradient vector representing the direction of error change is output.

[0136] Step c2: performing partial derivative optimization calculation on the weight corresponding to the wire diameter in the collision coefficient correction function based on the prediction error gradient to obtain the collision coefficient correction function after parameter adjustment.

[0137] Partial derivative optimization is the process of mathematically deriving and numerically updating specific parameters based on the error gradient component. The parameter-adjusted collision coefficient correction function is the fluid dynamics equation updated iteratively via error feedback, with the wire diameter weight coefficient adaptively corrected.

[0138] In an embodiment of the present application, the partial derivative value of the wire diameter weight component is first extracted from the predicted error gradient vector; secondly, the wire diameter weight parameter is adjusted in the direction of error reduction according to the positive and negative sign and size of the partial derivative value; then the weight term in the collision coefficient correction function is updated; finally, the parameter-adjusted collision coefficient correction function that eliminates systematic deviations is output.

[0139] The following is a specific example: First, laser ranging is used to obtain the measured ice thickness. If the absolute deviation from the predicted value exceeds a preset threshold, a gradient descent algorithm is activated to solve for the predicted error gradient. Next, the partial derivative of the wire diameter weight is extracted, and the weight parameter is adjusted along the negative gradient direction to update the collision coefficient correction function. Finally, the new function is used to recalculate the collision coefficient, bringing subsequent predictions closer to the measured value.

[0140] By executing steps c1 to c2, the embodiment of the present application triggers a gradient optimization mechanism through real-time deviation, accurately locates the systematic error of the wire diameter weight in the collision coefficient function, realizes adaptive correction of parameters, effectively reduces the risk of prediction inaccuracy caused by changes in the physical properties of the wire, and improves the long-term stability of the model.

[0141] Optionally, the method further includes obtaining a measured value of ice thickness on the railway contact network, calculating a predicted value of liquid water content based on the measured value of ice thickness, calculating an error between the predicted value and the actual value of liquid water content, and adjusting network parameters of the hybrid neural network model using a backpropagation algorithm based on the error to achieve iterative optimization of dynamic weight parameters.

[0142] In actual applications, due to the requirements for operating safety and structural stability of the contact network, it is not suitable to install detection equipment directly on the contact line. Considering non-contact, compensation area installation equipment, return line installation equipment, or contact line and return line prefabrication, the detection of contact line icing, dancing, and temperature parameters is more in line with field applications and feasibility requirements. Specifically, this embodiment can consider the use of fusion sensor monitoring, which can be divided into three technical categories:

[0143] (1) Direct monitoring method: machine vision recognition and monitoring of contact line dancing and icing, thermal imaging detection of contact line temperature, millimeter wave radar detection of contact line dancing and dancing amplitude.

[0144] (2) Auxiliary monitoring method: Wind speed, ambient temperature, snow depth and rainfall monitoring weather station technology, through the monitoring of the environment where the contact network is located, assists direct monitoring sensing to achieve accurate measurement.

[0145] (3) Indirect monitoring method: 1) The embodiment of the present application explores the use of prefabricated load-bearing cables and pre-buried optical fiber sensors to detect contact line vibration (i.e., dancing), contact line temperature, and force, and deduce the contact line icing status; explores the use of monitoring equipment (e.g., inclinometers, microwave ice detectors, temperature sensors, etc.) suspended on the return line to monitor the return line icing and infer the contact line icing status. 2) Electric field inversion and reconstruction: The electric field sensor is used to analyze the changes in the electric field around the contact line, invert the conductor voltage and the surrounding electric field, and extract the electrical characteristic values ​​that can characterize the conductor sag changes for dynamic analysis to determine whether icing has occurred. An early warning system is established, and an icing warning is issued if the sag changes. In addition, if the contact line dances, the surrounding electric field will undergo drastic periodic changes. By monitoring the electric field changes within a cycle, a judgment logic is established. If dancing occurs, the system will issue an early warning or automatically process according to the preset rules.

[0146] Optionally, this embodiment provides a technology for estimating the thickness of ice on the contact line by monitoring contact line forces and weight displacement based on finite element analysis. To maintain the stability of the existing catenary structure, a tensile stress detection sensor is installed between the compensation pulley and weight string of the contact line tension compensation device, and a displacement monitoring device is installed on the contact line weight. By detecting changes in tension and displacement (B value), ice thickness can be estimated based on data such as contact line length, deadweight, the transmission coefficient of the compensation pulley, the distance from the center anchor to the compensator, the linear expansion coefficient of the contact line load-bearing cable, current temperature, and upper cross-section width. Pre-set software simulation analyzes the impact of ice on the compensation displacement b value of the contact line compensation device. An approximate formula for calculating ice thickness is developed based on contact line geometric parameters and its accuracy verified through experiments. As ice thickness increases, the compensation displacement of the contact line compensation device increases continuously. Based on the contact line compensation displacement, ice thickness can be calculated relatively accurately. Alternatively, a more direct method for simultaneously detecting contact wire tension, deformation, temperature, and vibration is to prefabricate the contact wire. This involves embedding sensing optical fibers during the contact wire production process. Before embedding, photolithography is used to create a high-density sensor chain. Depending on the control equipment requirements, up to 100,000 sensors can be connected in series to monitor contact wire icing and vibration. Based on this principle, fiber optic sensors can also be embedded in the catenary or return wires for indirect detection.

[0147] The embodiment of the present application monitors the ice thickness and dancing amplitude of the contact line, and proposes an alarm threshold based on the impact of these two amplitudes on the mechanics of the contact network support structure and the safety of power supply. According to domestic recorded cases of contact network dancing, the vibration frequency of the contact network when dancing is usually 0.5 to 2 Hz, and the amplitude is generally 300 to 1000 mm. The vibration direction includes the up and down jumping of the conductor, the left and right swing, and the movement of an elliptical trajectory in the direction of the conductor cross section. Its manifestation is generally a first-order standing wave. The dancing phenomenon generally occurs in the wind season from the end of November to February of the following year. The wind speed during dancing is about 5-20 m / s, the temperature is -5 to 5°C, and the conductor is not covered with ice or has a small amount of ice.

[0148] Research in the high-voltage power grid field believes that conductor galloping is a unique mechanical movement caused by large-amplitude, low-frequency oscillations of the conductors under certain geographical and meteorological conditions. It is a type of aerodynamic instability. It is a low-frequency (0.1-3 Hz), large-amplitude (5-300 times the conductor diameter) self-excited vibration phenomenon caused by uneven transmission line conductors and wind action after icing.

[0149] The causes of galloping are very complex, and the various factors influence and restrict each other. The causes of catenary galloping can be summarized into the following three aspects:

[0150] (1) Catenary icing: When there is rain, snow, sleet, or freezing rain, the conductors are more likely to be covered with ice. Under the influence of continuous wind, the ice accumulates and forms eccentric ice. This ice changes the performance of the conductor in two aspects: first, the smooth surface prevents wind from entering the wire strands, and the wind does not cause disturbance when passing through the conductor, and does not form turbulence; second, the eccentric shape of the ice cross section is like an airfoil, which allows the wind to pass through the conductor in a stable laminar flow, thus providing the conditions for the conductor to obtain a force in the vertical direction.

[0151] (2) Contact network dancing: Wind is the most basic condition for contact network dancing and the energy source for contact network dancing. The wind speed when contact network dancing occurs is generally 5-20m / s, and the wind direction is nearly perpendicular to the conductor. In open areas without shielding (plains, passes, viaducts), the wind is less affected by the terrain friction coefficient and the wind unevenness coefficient is low. When a uniform, continuous, and stable external force can be achieved, contact network dancing is more likely to occur.

[0152] (3) Suspension structure and wire characteristics: The type, tension, sag, span and structural parameters of the wire are the internal factors that cause wire dancing. Unreasonable combination of wire parameters has a great relationship with wire dancing.

[0153] Railway electric traction power supply design specifications require that the catenary ice thickness used in design should be determined based on meteorological records and operational experience along the line, with integer values ​​of 0, 5, 10, 15, or 20 mm. The alarm threshold is 50% of the ice thickness on the catenary. The temperature during icing should be calculated as -5°C. The wind speed during icing should be calculated as 10 m / s in all areas, except in areas with strong winds and heavy ice, where it can be calculated as 15 m / s. The ice density is calculated as 0.9 g / cm.

[0154] The conditions for contact wire icing are: first, the exterior of the wires, where frozen water droplets are present, and the surface temperature of the wires must be below 0°C; second, a high relative humidity, typically above 85%, serves as the source of water for the conductor ice; and third, sufficient wind speed, typically above 5 m / s, captures moisture and mist in the air, which condenses into ice on the wires. The severity of contact wire icing depends on ambient temperature, wind speed, water droplets, conductor area, and impact factor. Temperature affects the cooling of water droplets, while wind speed and conductor area affect the impact factor, and these factors are somewhat coupled.

[0155] Based on existing experience, the monitoring values ​​of this embodiment of the present application are preliminarily designed as follows: (1) The detection accuracy of the maximum amplitude of the contact line (including the positive feeder) dance is no more than 3 cm, and the amplitude of the dance during the alarm is about 30 cm; (2) The monitoring accuracy of the contact line (including the positive feeder) ice coverage is no more than 1 mm, and the alarm threshold is 5 mm; (3) The embodiment of the present application intends to adopt a monitoring solution that integrates multiple sensors, while optimizing machine vision recognition and radar point cloud data calculation algorithms to improve monitoring accuracy and reduce alarm thresholds.

[0156] Figure 2 A schematic diagram of a railway contact network ice thickness prediction system provided in an embodiment of the present application is shown in FIG. Figure 2 As shown, the system includes:

[0157] The acquisition module 21 is used to obtain meteorological environment parameters, conductor physical characteristic parameters and ice-covered medium parameters of the area where the contact wire in the railway contact network is located.

[0158] The calibration module 22 is used to input meteorological environment parameters into the machine learning model, and use the machine learning model to dynamically calibrate the icing medium parameters based on the meteorological environment parameters to obtain calibrated icing medium parameters.

[0159] The generating module 23 is used to generate an ice growth coefficient based on meteorological environment parameters, conductor physical characteristic parameters and calibrated ice medium parameters.

[0160] The prediction module 24 is used to input the calibrated ice medium parameters and ice growth coefficient into the Makkonen model to obtain a predicted value of ice thickness.

[0161] Figure 2 The railway contact network ice thickness prediction system can be executed Figure 1 The implementation principles and technical effects of the railway catenary ice thickness prediction method described in the illustrated embodiment are not further elaborated. The specific manner in which each module and unit performs operations in the railway catenary ice thickness prediction system in the aforementioned embodiment has been described in detail in the relevant embodiments of the method and will not be further elaborated here.

[0162] In one possible design, Figure 2 A railway catenary ice thickness prediction system of the embodiment shown can be implemented as a computing device, such as Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32 .

[0163] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32 .

[0164] The processing component 32 is configured to obtain meteorological parameters, conductor physical parameters, and icing medium parameters for the area in which the contact wire of the railway overhead line is located. The meteorological parameters are input into a machine learning model, and the machine learning model is used to dynamically calibrate the icing medium parameters based on the meteorological parameters to obtain calibrated icing medium parameters. An icing growth coefficient is obtained based on the meteorological parameters, conductor physical parameters, and calibrated icing medium parameters. The calibrated icing medium parameters and icing growth coefficient are input into a Makkonen model to obtain a predicted ice thickness.

[0165] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above method.

[0166] The storage component 31 is configured to store various types of data to support operations on the terminal. The storage component can be implemented by any type of volatile or non-volatile memory device, or a combination thereof, such as random access memory (RAM), static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0167] Of course, a computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0168] The input / output interface provides an interface between the processing component and the peripheral interface module, which can be an output device, an input device, etc.

[0169] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0170] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0171] The present application also provides a computer storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 1 A method for predicting ice thickness of a railway contact network according to the illustrated embodiment.

[0172] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0173] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0174] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0175] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for predicting ice thickness of railway contact network, characterized in that: include: Obtain meteorological environment parameters, conductor physical characteristic parameters and icing medium parameters of the contact wire area in the railway overhead contact network; Inputting the meteorological environment parameters into a machine learning model, and dynamically calibrating the icing medium parameters based on the meteorological environment parameters using the machine learning model to obtain calibrated icing medium parameters; Obtaining an ice growth coefficient based on the meteorological environment parameters, the conductor physical characteristic parameters, and the calibrated ice medium parameters; Inputting the calibrated icing medium parameters and the icing growth coefficient into the Makkonen model to obtain a predicted value of icing thickness; The meteorological environment parameters include humidity, ambient temperature and wind speed; The step of inputting the meteorological environment parameters into a machine learning model, and dynamically calibrating the icing medium parameters based on the meteorological environment parameters using the machine learning model to obtain calibrated icing medium parameters includes: When the humidity is greater than a preset humidity threshold and the ambient temperature is less than or equal to a preset temperature threshold, the temperature, humidity, and wind speed time series in the historical icing data of the railway contact network are analyzed using the machine learning model, wherein the machine learning model is a hybrid neural network model, and the hybrid neural network model includes a bidirectional long short-term memory network layer and a convolutional neural network layer, which are used to respectively extract the long-term dependence characteristics and local mutation characteristics of the temperature, humidity, and wind speed time series in the historical icing data; Optimizing the dynamic weight parameters corresponding to humidity, ambient temperature, and wind speed in a preset joint correction model of humidity, ambient temperature, and wind speed based on the long-term dependence characteristics and local mutation characteristics; After the dynamic weight parameter optimization is completed, the optimized humidity, ambient temperature and wind speed joint correction model is used to calculate the calibrated ice medium parameters; After inputting the calibrated icing medium parameters and the icing growth coefficient into the Makkonen model to obtain the predicted icing thickness value, the method further includes: When the absolute deviation between the actual ice thickness value and the predicted ice thickness value is greater than a preset deviation threshold, the prediction error gradient is calculated using a gradient descent algorithm; Based on the prediction error gradient, a partial derivative optimization calculation is performed on the weight corresponding to the wire diameter in the collision coefficient correction function to obtain the collision coefficient correction function after parameter adjustment.

2. The method according to claim 1, characterized in that The optimization of the dynamic weight parameters corresponding to the humidity, ambient temperature and wind speed in the preset humidity, ambient temperature and wind speed joint correction model according to the long-term dependence characteristics and the local mutation characteristics includes: According to the duration feature of the humidity being greater than the preset humidity threshold in the long-term dependency feature, adjusting the dynamic weight parameter corresponding to the humidity according to a preset exponential decay rule; According to the moisture accumulation feature in the long-term dependency feature where the temperature is less than a preset temperature threshold, a dynamic weight parameter corresponding to the temperature is calculated using an S-type function; Based on the coupling characteristics between the temperature drop and the wind speed in the local mutation characteristics, a dynamic weight parameter corresponding to the wind speed is calculated.

3. The method according to claim 1, characterized in that The physical characteristic parameters of the conductor include the conductor diameter, the surface roughness coefficient of the contact line and the radiation parameter of the conductor surface; the ice growth coefficient includes the collision coefficient, the freezing coefficient and the capture coefficient; The generating of the ice growth coefficient based on the meteorological environment parameter, the conductor physical characteristic parameter, and the calibrated ice medium parameter includes: Calculating a collision coefficient using a preset collision coefficient correction function based on the wire diameter and the calibrated ice-covered medium parameters; Based on the ambient temperature, the ice surface temperature and the radiation parameters of the conductor surface, the freezing coefficient is obtained through the balance relationship between radiation heat transfer and convection heat transfer; A preset cube root function is used to quantify the attenuation effect of the contact line surface roughness coefficient on the water droplet capture efficiency to obtain the capture coefficient.

4. The method according to claim 3, characterized in that The freezing coefficient is obtained based on the ambient temperature, the ice surface temperature and the conductor surface radiation parameter through the balance relationship between radiation heat transfer and convection heat transfer, including: Calculating the heat radiation power of the conductor surface using the Stefan-Boltzmann law based on the conductor surface radiation parameters; Calculating a forced convection heat transfer coefficient based on the wind speed and the ambient temperature, and calculating a convection heat transfer power based on the forced convection heat transfer coefficient; Based on the heat radiation power of the conductor surface and the convection heat transfer power, a heat balance analysis is performed on the ice surface to generate a freezing coefficient.

5. The method according to claim 3, characterized in that The step of inputting the calibrated icing medium parameters and the icing growth coefficient into a Makkonen model to obtain an icing thickness prediction value includes: Based on the collision coefficient, freezing coefficient and capture coefficient, combined with wind speed and calibrated ice medium parameters, the ice mass growth equation in the Makkonen model is used to solve the ice mass growth per unit time. Convert the ice mass increment to the equivalent cylindrical ice thickness; The dancing amplitude of the contact line is obtained, and the ice thickness of the equivalent cylinder is adjusted according to the dancing amplitude to obtain a predicted value of the ice thickness.

6. A railway contact network ice thickness prediction system, characterized in that: Executing the method for predicting ice thickness of a railway contact network according to claim 1 comprises: An acquisition module is used to obtain meteorological environment parameters, conductor physical characteristic parameters and icing medium parameters of the area where the contact wire in the railway contact network is located; a calibration module, configured to input the meteorological environment parameters into a machine learning model, and dynamically calibrate the icing medium parameters based on the meteorological environment parameters using the machine learning model to obtain calibrated icing medium parameters; A generating module, configured to generate an ice growth coefficient based on the meteorological environment parameters, the conductor physical characteristic parameters, and the calibrated ice medium parameters; The prediction module is used to input the calibrated ice-covered medium parameters and the ice-covered growth coefficient into the Makkonen model to obtain an ice-covered thickness prediction value.

7. A computing device, characterized in that It comprises a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a railway contact network ice thickness prediction method as described in any one of claims 1 to 5.

8. A computer storage medium, characterized in that A computer program is stored, and when the computer program is executed by a computer, a method for predicting the ice thickness of a railway contact network as claimed in any one of claims 1 to 5 is implemented.

Images