Iced conductor resistance coefficient prediction method and equipment based on machine learning

Through machine learning-based methods, integrating multiple models and optimization technologies, the problem of inaccurate prediction of the resistance coefficient of ice-covered conductors in the prior art is solved, and the safety and reliability of the power system are improved.

CN118940912APending Publication Date: 2024-11-12ELECTRIC POWER SCI RES INST OF STATE GRID XINJIANG ELECTRIC POWER CO LTD +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411142538.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-20
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

The prior art cannot accurately predict the resistance coefficient of ice-covered conductors under different wind speeds, ice-cover thickness and wind attack angles, resulting in uncertainty and safety risks in the design and maintenance of the power system.

Method used

Using a machine learning-based approach, data preprocessing and feature enhancement are performed by obtaining data from multiple channels (sensor systems, simulations and experiments), and multiple machine learning models (linear regression, random forest regression, gradient enhancement regression) are selected for integration. Learn and hyperparameter optimization to generate an integrated model to predict the resistance coefficient of the ice-covered conductor.

Benefits of technology

It improves the robustness and generalization ability of the model, accurately predicts the resistance changes of ice-covered wires, reduces the risks of wire damage and fracture, improves the safety and reliability of power transmission, and provides optimization suggestions for power system design and maintenance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118940912A_ABST
    Figure CN118940912A_ABST
Patent Text Reader

Abstract

The invention relates to an iced conductor resistance coefficient prediction method and equipment based on machine learning. The method comprises the following steps: acquiring iced conductor resistance coefficient data under the conditions of different wind speeds, icing thicknesses and wind attack angles from multiple channels; carrying out preprocessing and feature enhancement on the obtained resistance coefficient data of the iced conductor; selecting a plurality of machine learning models for model learning and optimization, and then integrating the plurality of models to generate an integrated model; and inputting the actual wind speed, the icing thickness and the wind attack angle into the trained and optimized integrated model, and outputting a prediction result of the resistance coefficient of the icing conductor. Compared with the prior art, the method has the advantages of realizing accurate prediction of the resistance coefficient of the iced conductor, improving the safety and reliability of power transmission and the like.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of ice-covered conductor resistance, and in particular to a method and device for predicting an ice-covered conductor resistance coefficient based on machine learning. Background Art

[0002] Ice-covered conductors are a common and serious problem in power systems. In cold weather, ice forms on the surface of conductors, increasing the weight and mechanical stress of the conductors, causing them to sag or even break. In addition, ice coating significantly increases the air resistance of the conductors, affecting the efficiency and stability of power transmission. Ice-covered conductors are more likely to swing and vibrate in strong winds, increasing the risk of conductor damage and breakage. These problems not only affect the reliability of power supply, but may also lead to serious safety accidents.

[0003] Machine learning has been widely used in many fields, including prediction and optimization problems in power systems. By learning from historical data and experimental data, machine learning models can capture complex nonlinear relationships and provide high-precision prediction results. In the prediction of the drag coefficient of ice-covered conductors, traditional empirical formulas and simple physical models are difficult to cope with variable environmental conditions and complex dynamic characteristics. Machine learning methods can combine multiple input parameters, such as wind speed, ice thickness, and wind attack angle, and improve the prediction accuracy and robustness of the model through ensemble learning and hyperparameter optimization. Using machine learning technology, not only can the drag coefficient of ice-covered conductors under different conditions be accurately predicted, but also optimization suggestions can be provided for the design and maintenance of power systems to improve the reliability and safety of the system.

[0004] How to accurately predict the resistance coefficient of ice-covered conductors based on machine learning has become a technical problem that needs to be solved. Summary of the invention

[0005] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a method and device for predicting the drag coefficient of ice-covered conductors based on machine learning, so as to solve the problem that the prior art cannot accurately predict the drag coefficient of ice-covered conductors under different wind speeds, ice thicknesses and wind attack angles.

[0006] The purpose of the present invention can be achieved by the following technical solutions:

[0007] According to one aspect of the present invention, a method for predicting the drag coefficient of an ice-covered conductor based on machine learning is provided, the method comprising the following steps:

[0008] Step S1, obtaining ice-covered conductor resistance coefficient data under different wind speeds, ice thicknesses and wind attack angles from multiple channels;

[0009] Step S2, preprocessing and feature enhancement of the acquired ice-covered conductor resistance coefficient data;

[0010] Step S3: Based on the result of step S2, multiple machine learning models are selected to perform model learning and optimization respectively, and then the multiple models are integrated to generate an integrated model;

[0011] Step S4: input the actual wind speed, ice thickness and wind attack angle into the integrated model trained and optimized in step S3, and output the prediction result of the drag coefficient of the ice-covered conductor.

[0012] Preferably, the ice-covered conductor resistance coefficient data includes tension value, wind speed, ice thickness, wind attack angle and corresponding resistance coefficient;

[0013] The multiple channels include sensor systems, simulations and experiments.

[0014] Preferably, the preprocessing includes cleaning, removing noise and outliers, and standardizing the data.

[0015] Preferably, the feature enhancement includes:

[0016] Step S2-1, extracting original features, wherein the original features include wind speed, ice thickness, wind attack angle, wire radius, wire tension value and aerodynamic drag coefficient;

[0017] Step S2-2, constructing new features based on the original features, wherein the new features include the square of the wind speed, the sine and cosine values ​​of the wind attack angle, and the product of the wind speed and the ice thickness.

[0018] Preferably, in step S3, the process of selecting multiple machine learning models to perform model learning respectively includes:

[0019] Step S3-1: Select a linear regression model and use the least squares method to fit the model parameters. The formula is:

[0020] y=β0+β1x1+β2x2+β3x3+β4x4+β5x5;

[0021] Among them, y is the drag coefficient of the ice-covered conductor, x1 is the wind speed, x2 is the ice thickness, x3 is the wind attack angle, x4 is the conductor radius, and x5 is the conductor tension value;

[0022] Step S3-2: Select the random forest regression model and use an ensemble of multiple decision trees for regression. The hyperparameters set include the number of trees, the depth of the tree, and the minimum number of sample splits.

[0023] Step S3.3: Select the gradient boosting regression model and fit the residual by gradually improving the performance of the weak learner. The hyperparameters set include the learning rate, the number of trees, and the depth of the tree.

[0024] More preferably, the generation of the integrated model is specifically:

[0025] The prediction results of the three regression models are weighted averaged to obtain the prediction results of the integrated model, which are as follows:

[0026] y semble =w1y lr +w2y rf +w3y gbr ;

[0027] Among them, y semble is the prediction result of the integrated model, y lr ,y rf ,y gbr They are the prediction results of linear regression, random forest regression and gradient boosting regression respectively. w1, w2 and w3 are weighted coefficients, and they satisfy w1+w2+w3=1.

[0028] More preferably, the optimization process of the multiple machine learning models includes:

[0029] The training set is used to train each machine learning model. The training of the random forest model and the gradient boosting regression model adjusts the model parameters by optimizing the loss function. The formula of the loss function is:

[0030]

[0031] Where n is the number of samples, y i is the true value of the i-th sample, is the predicted value of the i-th sample;

[0032] For random forest models and gradient boosting regression models, use grid search or random search for hyperparameter optimization;

[0033] Training the integrated model means continuously optimizing the weights of each machine learning model to obtain the optimal weights.

[0034] Preferably, step S3 further comprises evaluating the performance of the integrated model using a cross-validation method, comprising:

[0035] a) Evaluate the ensemble model on the validation set and use the optimal weights to calculate the prediction value of the ensemble model;

[0036] b) Use k-fold cross validation to evaluate the performance of the integrated model. Specifically, calculate the MSE of each fold, find the average MSE, and evaluate the performance of the integrated model.

[0037] Preferably, the step S4 further comprises drawing a 3D surface variation diagram of the drag coefficient of the ice-covered conductor, the wind speed, and the ice thickness, and analyzing the characteristic relationship between the drag coefficient of the ice-covered conductor, the wind speed, and the ice thickness.

[0038] According to another aspect of the present invention, an electronic device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the method described above is implemented when the processor executes the program.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] 1) The present invention obtains ice-covered conductor resistance coefficient data from multiple channels, and through data preprocessing and feature enhancement, integrates multiple machine learning models for learning and hyperparameter optimization, effectively enhancing the robustness and generalization ability of the model, and reducing the overfitting problem that may be caused by a single model. Therefore, the prediction result of the ice-covered conductor resistance coefficient is automatically output based on the trained and optimized model, which can accurately reflect the changes in ice-covered conductor resistance under different wind speeds, ice thicknesses and wind attack angles, thereby providing optimization suggestions for the design and maintenance of power systems, reducing the risk of conductor damage and breakage caused by icing, and improving the safety and reliability of power transmission.

[0041] 2) The present invention constructs new features based on the extracted features to enhance the expressiveness of the model; uses the cross-validation method to evaluate the performance of the integrated model in the validation set, and adjusts the hyperparameters to obtain the optimal model, thereby improving the prediction accuracy, robustness and generalization ability of the model.

[0042] 3) The present invention utilizes sensor systems, experiments and simulations to obtain the drag coefficient data of ice-covered conductors under different wind speeds, ice thicknesses and wind attack angles through multiple channels, thereby improving the utilization efficiency of data and the training effect of the model.

[0043] 4) The present invention can be applied to the optimization design and operation strategy in the field of power transmission to reduce energy consumption and improve safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a schematic flow chart of the method for predicting the resistance coefficient of ice-covered conductor in the present invention;

[0045] FIG2( a ) is a schematic diagram of the structure of a sensor system for collecting ice-covered conductor related data in the present invention;

[0046] FIG2( b ) is a schematic diagram of the sensor in FIG2( a );

[0047] Figure 3 It is a schematic diagram of the simulation of uniform turbulent field in the present invention;

[0048] Figure 4 It is a schematic diagram of the cross section of the ice-covered conductor in the simulation software of the present invention;

[0049] Figure 5 A schematic diagram of data features extracted in the present invention;

[0050] Figure 6 A schematic diagram of the process of extracting and dividing features of pre-processed data in the present invention;

[0051] Figure 7 A schematic diagram of the flow of model training and ensemble learning in the present invention;

[0052] Figure 8 This is a schematic diagram of the hyperparameter optimization process of a single model in the present invention;

[0053] Fig. 9 It is a schematic diagram of the cross-validation process in the present invention;

[0054] Fig.10 This is a schematic diagram of the hyperparameter optimization process of the random forest regression model in the present invention;

[0055] Fig.11 This is a schematic diagram of the hyperparameter optimization process of the gradient boosting regression model in the present invention;

[0056] Fig.12 This is a K-fold cross validation diagram of the model in the present invention;

[0057] Fig.13 It is a schematic diagram of the process of predicting the drag coefficient by loading the model in the present invention;

[0058] Fig.14 It is a 3D surface variation diagram between the drag coefficient and the parameters in the example of the present invention. DETAILED DESCRIPTION

[0059] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0060] This embodiment relates to a method for predicting the drag coefficient of an ice-covered conductor based on machine learning, which solves the problem in the prior art that the drag coefficient of an ice-covered conductor under different wind speeds, ice thicknesses and wind attack angles cannot be accurately predicted.

[0061] like Figure 1 , the method comprises the following steps:

[0062] Step S1: using sensor systems, simulations, experiments and other methods, collect ice-covered wire drag coefficient data under different wind speeds, ice thicknesses and wind attack angles. The data include: tension value, wind speed, ice thickness, wind attack angle and corresponding drag coefficient.

[0063] Step S2: preprocess and enhance the collected data, including cleaning, removing noise and outliers. Standardize the data, extract and construct new features, such as the square of wind speed (m / s), ice thickness (mm), etc., to enhance the expression ability of the model;

[0064] Step S3, select the machine learning models of Linear Regression, Random Forest Regression, and Gradient Boosting Regression for ensemble learning, use weighted averaging as the ensemble method, and prepare for model training;

[0065] Step S4: Divide the preprocessed data set into a training set and a validation set. Use the training set to train the model and adjust the model parameters by optimizing the loss function. The formula of the loss function is:

[0066]

[0067] Where n is the number of samples, y i is the true value of the i-th sample, is the predicted value of the ith sample.

[0068] Step S5: Use the validation set to evaluate the model performance and adjust the hyperparameters to obtain the best model. Use the cross-validation method to improve the robustness and generalization ability of the model.

[0069] Step S6: Input parameters such as wind speed, ice thickness and wind attack angle in actual application, and use the trained model to predict the drag coefficient of ice-covered wires. Generate prediction results for further analysis and optimization suggestions.

[0070] The specific operations of step S1 include:

[0071] Use sensors to collect relevant data such as the tension value and wind speed of ice-covered wires;

[0072] The aerodynamic coefficient data of the conductor is collected by means of wind tunnel experiments;

[0073] Fluid analysis simulation software was used to set different ice thicknesses, wire diameters, and wind attack angles for simulation experiments to obtain drag coefficient data.

[0074] In step S1, as shown in FIG. 2(a) and FIG. 2(b), a sensor system is used to collect relevant data of the ice-covered conductor, including parameters such as time, ice thickness, air temperature, and tension value.

[0075] like Figure 3As shown in the figure, a wind tunnel experiment was carried out using a uniform turbulence field simulation. The measured data showed that the wind speed at which dancing occurs is mostly concentrated between 7 and 15 m / s. Therefore, the wind field takes a uniform flow with a wind speed of 10 m / s and a turbulence degree of 5% to collect the aerodynamic coefficients of the conductor under different ice thicknesses and different wind attack angles.

[0076] like Figure 4 As shown in the figure, the green cylinder model is the initial shape of the conductor without ice. The conductor length is set to 200mm and the diameter is set to 20mm. Ice models of different thicknesses and shapes are added to the surface of the green cylindrical conductor model to simulate the actual working conditions of the conductor ice. In the simulation software, the wind attack angle is set to 0°, and the wind parameters are changed. The wind speed U(t) is taken as 5m / s, 7.5m / s, 10m / s, 12.5m / s, 15m / s, 17.5m / s and 20m / s respectively. The ice thickness is taken as 5mm, 10mm, 15mm and 20mm respectively. The resistance coefficients of conductors with different ice thicknesses under different working conditions are shown in Table 1:

[0077] Table 1

[0078]

[0079] The specific operations of step S2 for preprocessing and enhancing the collected data are as follows:

[0080] Step S2.1: Check missing values, fill in missing values ​​with the mean, and delete duplicate data;

[0081] Step S2.2: Detect outliers using the box plot method, mark outliers as missing values ​​and fill or delete them;

[0082] Step S2.3: Calculate the mean and standard deviation of each feature, and standardize each feature using the Z-score standardization method;

[0083] Step S2.4: Extract original features: including wind speed (v), ice thickness (h), wind attack angle (α), wire radius (r) and wire tension value (T), and construct new features: including the square of wind speed (v 2 ), the sine and cosine of the wind attack angle (sin(α), cos(α)), and the product of wind speed and ice thickness (v*h).

[0084] The acquired data is preloaded and feature extracted in the model, and the obtained partial preload initial data is shown in Table 2. The features to be extracted for the initial data include wind_speed: wind speed (m / s), ice_thickness: ice thickness (mm), attack_angle: wind attack angle (degrees), tension: wire tension value (N), drag_coefficient: aerodynamic drag coefficient (Cd), and the extracted features are as follows Figure 5 shown.

[0085] Table 2

[0086] Temperature humidity Wind speed wind direction Tensile force Air pressure Tensile force benchmark Ice thickness -6 97.9 2.9 298 3.2 859.1 1.6 7.04 -5.6 98.1 4.1 295 3.2 860.1 1.6 6.99 -5.6 97.9 3.9 231 3.4 858.9 8 7.1 -5.5 97.7 3.7 244 3.4 859.1 8 7.8 -5.3 97.9 4.1 279 3.2 857.1 1.6 7.92 -10.9 47 0.6 294 7.8 860.1 8 7.06 -10.9 44.9 1.1 248 3.2 859.1 1.6 6.89 -10.7 45.5 0.6 206 3.4 859.1 1.6 7.25 -10.6 47.7 0.9 221 3.2 858.9 1.6 7.83 -10.7 47.4 1 231 3.2 857.9 1.6 7.92 -10.3 50.4 2.09 209 1.5 858.5 1.6 7.79 -7.9 54 2.09 302 7.8 857.1 8 6.55 -8.39 58.9 1.8 205 8 856.7 1.6 6.91 -5.3 46 2.09 294 1.6 856.1 8 5.66 -5.3 46.2 1.4 196 1.7 854.7 1.6 5.5

[0087] The preprocessed dataset is feature enhanced and divided into a training set and a validation set with a ratio of 75% and 25%. Figure 6 shown.

[0088] The specific method of step S3 selecting multiple machine learning models for ensemble learning is as follows:

[0089] Step S3.1: Select the linear regression model, the formula is:

[0090] y=β0+β1x1+β2x2+β3x3+β4x4+β5x5;

[0091] Among them, y is the drag coefficient, x1 is the wind speed, x2 is the ice thickness, x3 is the wind attack angle, x4 is the wire radius, and x5 is the wire tension value. The least squares method is used to fit the model parameters;

[0092] Step S3.2: Select the random forest regression model and use an ensemble of multiple decision trees for regression. The number of trees is n, and hyperparameters such as tree depth and minimum sample split number are optimized through cross-validation.

[0093] Step S3.3: Select the gradient boosting regression model, gradually improve the performance of the weak learner (decision tree), fit the residual, set the hyperparameters such as learning rate, number of trees, and depth of the tree, and optimize through cross-validation.

[0094] Step S3.4: Take the weighted average of the prediction results of the three models. The weighted formula is:

[0095] y semble =w1y lr +w2y rf +w3y gbr ;

[0096] Among them, y semble is the prediction result of the integrated model, y lr ,y rf ,ygbr They are the prediction results of linear regression, random forest regression and gradient boosting regression respectively; w1, w2, w3 are weighting coefficients, satisfying w1+w2+w3=1.

[0097] like Figure 7 As shown, linear regression, random forest regression, and gradient boosting regression model training are performed respectively and integrated learning is performed.

[0098] The first step is to use the LinearRegression() function to initialize the linear regression model, and then use the .fit() method in the model to train the model. After that, prediction is made on the divided validation set and the mean square error (MSE) on the validation set is calculated.

[0099] The second step is to use the RandomForestRegressor() function to initialize the random forest regression model, where the initial parameters are set as follows:

[0100] n_estimators (the number of random forest trees) is 100,

[0101] max_depth (maximum depth) uses the default value of None (nodes are split until a leaf node contains fewer than min_samples_split samples),

[0102] random_state (random number seed) is set to 42,

[0103] Then use the .fit() method in the model to train the model, then make predictions on the divided validation set and calculate the mean square error (MSE) on the validation set.

[0104] The third step is to use the GradientBoostingRegressor() function to initialize the gradient boosting regression model, where the parameters are set as follows:

[0105] n_estimators (the number of random forest trees) is 100,

[0106] max_depth (maximum depth) uses 3,

[0107] The learning_rate is set to 0.1.

[0108] random_state (random number seed) is set to 42.

[0109] Then use the .fit() method in the model to train the model, then make predictions on the divided validation set and calculate the mean square error (MSE) on the validation set.

[0110] After the linear regression, random forest regression, and gradient boosting regression model training is completed, the prediction value of a single model on the validation set is calculated, such as Figure 8 Define the weighted average function weighted_ensemble() and set the initial weight value to 1 / 3, set the constraints and weight range to (0,1), and optimize the best weight for model integration.

[0111] In step S4, the specific operations of adjusting the model parameters by optimizing the loss function and hyperparameter optimization are as follows:

[0112] Step S4.1: Perform model training of linear regression model, random forest model and gradient boosting regression model on the training set respectively, and adjust the model parameters by optimizing the loss function. The formula of the loss function is:

[0113]

[0114] Where n is the number of samples, y i is the true value of the i-th sample, is the predicted value of the ith sample.

[0115] Step S4.2: For random forest and gradient boosting regression models, perform hyperparameter optimization using grid search or random search.

[0116] Step S4.3: Perform weighted average ensemble model training, set the initial w1, w2, w3 to 1 / 3, and continuously optimize the weights to obtain the optimal weights.

[0117] like Fig. 9 As shown in Figure 1, linear regression models usually do not require hyperparameter tuning, but we can evaluate their performance through cross-validation. Through cross-validation, the average MSE is calculated to evaluate the best model.

[0118] For the random forest regression model, GridSearchCV can be used for hyperparameter optimization, such as Fig.10 , including: First define the hyperparameter grid of the random forest regression model as n_estimators: [50, 100, 200], max_depth: [None, 10, 20, 30], min_samples_split: [2, 5, 10], min_samples_leaf: [1, 2, 4]. Initialize grid search and hyperparameter search, obtain the best parameters and model and evaluate the best model on the validation set.

[0119] For the gradient boosting regression model, GridSearchCV is also used for hyperparameter optimization, such as Fig.11, including: defining the hyperparameter grid of the gradient boosting regression model as n_estimators: [50, 100, 200], learning_rate: [0.01, 0.1, 0.2], max_depth: [3, 5, 7], min_samples_split: [2, 5, 10], min_samples_leaf: [1, 2, 4]. Initialize grid search and hyperparameter search, obtain the best parameters and model and evaluate the best model on the validation set.

[0120] Step S5: Use cross-validation method to improve the robustness and generalization ability of the model. The specific method is as follows:

[0121] Step S5.1: Evaluate the ensemble model on the validation set and use the optimal weights to calculate the prediction value of the ensemble model.

[0122] Step S5.2: Use k-fold cross validation to evaluate the model performance.

[0123] like Fig.12 As shown, k-fold cross validation is used to evaluate model performance to ensure the robustness and generalization ability of the model. Initialize the model, define k in k-fold cross validation as 5, random state as 42, perform cross validation, calculate the MSE of each fold, and take the negative value (because cross_val_score is negative by default).

[0124] The specific operations of step S6 are as follows:

[0125] Step S6.1: Load the trained optimal model, define the input parameters and generate prediction results.

[0126] like Fig.13 As shown, use joblib.load to load the previously saved linear regression, random forest regression, and gradient boosting regression models.

[0127] The models are saved in linear_regression_model.pkl, random_forest_model.pkl, and gradient_boosting_model.pkl files.

[0128] Create a dictionary input_data containing input parameters such as wind speed, wire tension value, ice thickness, and wind angle of attack. Convert the dictionary to a DataFrame so that the model can be used directly.

[0129] For each model (linear regression, random forest regression, and gradient boosting regression), use the predict method to make predictions and get their respective prediction results.

[0130] Use the optimal weights optimal_weights obtained in advance to perform weighted averaging on the prediction results of each model to obtain the final integrated prediction result y_pred_ensemble.

[0131] Step S6.2: Draw a 3D surface variation diagram of the drag coefficient, wind speed, and ice thickness, and analyze the characteristic relationship between the drag coefficient and the parameters.

[0132] like Fig.14 As shown in the figure, for the integrated prediction results, we can use the 3D surface change diagram to intuitively explore the characteristic relationship between the drag coefficient and the parameters. Through the changes in the surface, we can draw the following conclusions:

[0133] Relationship between drag coefficient and wind speed: The surface graph shows that as wind speed increases, the drag coefficient increases first and then stabilizes. This indicates that at low wind speeds, wind resistance increases rapidly; when wind speed increases to a certain level, the growth rate of the drag coefficient slows down and approaches a limit value.

[0134] The relationship between the drag coefficient and the thickness of the ice cover: As can be seen from the figure, the drag coefficient increases with the increase of the thickness of the ice cover. This means that the thicker the ice cover, the less smooth the wire surface is, and the greater the aerodynamic drag.

[0135] The overall shape of the surface: The surface presents an upward convex shape, with the highest point roughly in the middle area, indicating that under a certain combination of wind speed and ice thickness, the drag coefficient will reach its maximum value.

[0136] This invention provides an innovative solution, which effectively improves the efficiency and accuracy of drag coefficient measurement and analysis and has good practicality.

[0137] The electronic device of the present invention includes a central processing unit (CPU), which can perform various appropriate actions and processes according to computer program instructions stored in a read-only memory (ROM) or loaded from a storage unit into a random access memory (RAM). In the RAM, various programs and data required for device operation can also be stored. The CPU, ROM and RAM are connected to each other via a bus. An input / output (I / O) interface is also connected to the bus.

[0138] Multiple components in the device are connected to the I / O interface, including: input units, such as keyboards, mice, etc.; output units, such as various types of displays, speakers, etc.; storage units, such as disks, optical disks, etc.; and communication units, such as network cards, modems, wireless communication transceivers, etc. The communication unit allows the device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunication networks.

[0139] The processing unit performs the various methods and processes described above, such as methods S1 to S6. For example, in some embodiments, methods S1 to S6 may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on the device via a ROM and / or a communication unit. When the computer program is loaded into the RAM and executed by the CPU, one or more steps of methods S1 to S6 described above may be performed. Alternatively, in other embodiments, the CPU may be configured to execute methods S1 to S6 in any other appropriate manner (e.g., by means of firmware).

[0140] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0141] The program code for implementing the method of the present invention can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code can be executed entirely on the machine, partially on the machine, partially on the machine as a stand-alone software package and partially on a remote machine, or entirely on a remote machine or server.

[0142] In the context of the present invention, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0143] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for predicting the resistance coefficient of ice-covered conductors based on machine learning, characterized in that: The method comprises the following steps: Step S1, obtaining ice-covered conductor resistance coefficient data under different wind speeds, ice thicknesses and wind attack angles from multiple channels; Step S2, preprocessing and feature enhancement of the acquired ice-covered conductor resistance coefficient data; Step S3: Based on the result of step S2, multiple machine learning models are selected to perform model learning and optimization respectively, and then the multiple models are integrated to generate an integrated model; Step S4: input the actual wind speed, ice thickness and wind attack angle into the integrated model trained and optimized in step S3, and output the prediction result of the drag coefficient of the ice-covered conductor.

2. The method for predicting the resistance coefficient of an ice-covered conductor based on machine learning according to claim 1 is characterized in that: The ice-covered conductor resistance coefficient data includes tension value, wind speed, ice thickness, wind attack angle and corresponding resistance coefficient; The multiple channels include sensor systems, simulations and experiments.

3. The method for predicting the resistance coefficient of ice-covered conductors based on machine learning according to claim 1 is characterized in that: The preprocessing includes cleaning, removing noise and outliers, and standardizing the data.

4. The method for predicting the resistance coefficient of an ice-covered conductor based on machine learning according to claim 1 is characterized in that: The feature enhancements include: Step S2-1, extracting original features, wherein the original features include wind speed, ice thickness, wind attack angle, wire radius, wire tension value and aerodynamic drag coefficient; Step S2-2, constructing new features based on the original features, wherein the new features include the square of the wind speed, the sine and cosine values ​​of the wind attack angle, and the product of the wind speed and the ice thickness.

5. The method for predicting the resistance coefficient of ice-covered conductor based on machine learning according to claim 1 is characterized in that: In step S3, the process of selecting multiple machine learning models to perform model learning respectively includes: Step S3-1: Select a linear regression model and use the least squares method to fit the model parameters. The formula is: y=β0+β1x1+β2x2+β3x3+β4x4+β5x5; Among them, y is the drag coefficient of the ice-covered conductor, x1 is the wind speed, x2 is the ice thickness, x3 is the wind attack angle, x4 is the conductor radius, and x5 is the conductor tension value; Step S3-2: Select the random forest regression model and use an ensemble of multiple decision trees for regression. The hyperparameters set include the number of trees, the depth of the tree, and the minimum number of sample splits. Step S3.3: Select the gradient boosting regression model and fit the residual by gradually improving the performance of the weak learner. The hyperparameters set include the learning rate, the number of trees, and the depth of the tree.

6. The method for predicting the resistance coefficient of ice-covered conductors based on machine learning according to claim 5 is characterized in that: The generation integration model is specifically: The prediction results of the three regression models are weighted averaged to obtain the prediction results of the integrated model, which are as follows: y semble =w1y lr +w2y rf +w3y gbr ; Among them, y semble is the prediction result of the integrated model, y lr ,y rf ,y gbr They are the prediction results of linear regression, random forest regression and gradient boosting regression respectively. w1, w2 and w3 are weighted coefficients, and they satisfy w1+w2+w3=1.

7. The method for predicting the resistance coefficient of ice-covered conductors based on machine learning according to claim 5 is characterized in that: The optimization process of the multiple machine learning models includes: The training set is used to train each machine learning model. The training of the random forest model and the gradient boosting regression model adjusts the model parameters by optimizing the loss function. The formula of the loss function is: Where n is the number of samples, y i is the true value of the i-th sample, is the predicted value of the i-th sample; For random forest models and gradient boosting regression models, use grid search or random search for hyperparameter optimization; Training the integrated model means continuously optimizing the weights of each machine learning model to obtain the optimal weights.

8. The method for predicting the resistance coefficient of ice-covered conductors based on machine learning according to claim 1, characterized in that: Step S3 also includes evaluating the performance of the integrated model using a cross-validation method, including: a) Evaluate the ensemble model on the validation set and use the optimal weights to calculate the prediction value of the ensemble model; b) Use k-fold cross validation to evaluate the performance of the integrated model. Specifically, calculate the MSE of each fold, find the average MSE, and evaluate the performance of the integrated model.

9. The method for predicting the resistance coefficient of ice-covered conductors based on machine learning according to claim 1, characterized in that: The step S4 also includes drawing a 3D surface variation diagram of the drag coefficient of the ice-covered conductor, wind speed, and ice thickness, and analyzing the characteristic relationship between the drag coefficient of the ice-covered conductor, wind speed, and ice thickness.

10. An electronic device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the program, the method according to any one of claims 1 to 9 is implemented.