Ice melting decision evaluation method and system based on machine learning

Abstract

The embodiment of the invention provides an ice melting decision evaluation method and system based on machine learning, and relates to the technical field of power grid ice melting. According to the method, the multi-source heterogeneous icing state parameter set can be established after the multi-source monitoring data is acquired. And performing weighted low-rank sparse cooperative decomposition to obtain an icing dominant mode matrix, and then based on the icing dominant mode matrix, in combination with a particle swarm cross validation joint optimization strategy, determining an optimal ice melting power gear. And calculating ice melting parameters according to the optimal ice melting power gear, and setting the ice melting parameters to an ice melting device. According to the method, through icing state evaluation, multi-source data fusion and optimal ice melting power gear control, the icing grade can be calculated according to the multi-source monitoring data, ice melting power calculation is optimized, the ice melting device is driven and controlled to operate under the optimal ice melting parameters, the accuracy and response speed of icing state evaluation can be improved, and the ice melting efficiency is improved. And the over-fusion or under-fusion risk is reduced.

Description

Technical Field

[0001] This application relates to the field of power grid de-icing technology, and in particular to a machine learning-based de-icing decision evaluation method and system. Background Technology

[0002] In mountainous power transmission areas during cold seasons, the surfaces of power transmission lines and other equipment are prone to icing. Icing increases the load on these devices, affecting conductor tension, sag, and tower load, potentially leading to accidents such as line breaks and tower collapses, thus compromising the continuity of power supply. To obtain timely and accurate information on the icing status of transmission lines and ensure power supply continuity and public safety, it is necessary to monitor and issue alarms for the icing status in power transmission areas.

[0003] Due to factors such as manpower, weather windows, and blind spots in observation, manual ice observation or fixed threshold alarm methods are insufficient for a comprehensive real-time assessment of icing conditions. Therefore, decision-making technology can be used to perform data detection and condition assessment of icing. Based on the decision matrix, de-icing equipment can be deployed to perform de-icing operations on transmission equipment, thereby improving de-icing efficiency and reducing the risks to power grid safety.

[0004] However, because the decision-making process for ice melting relies on fixed thresholds or manual experience to determine the ice melting power level, it lacks dynamic perception of ice thickness, sudden weather changes, and load disturbances. This leads to over-melting or under-melting during the ice melting process, resulting in inaccurate power level settings for the ice melting device, which can easily cause misjudgments, energy waste, and incomplete ice removal. Furthermore, the decision-making process for ice melting lacks sensitivity to multi-source monitoring data. The real-time collected meteorological, stress, and other multi-source data suffer from inconsistencies in dimensions, high noise levels, and numerous outliers, making it difficult to effectively integrate and extract robust features for the ice melting decision-making process. Consequently, the accuracy of decision-making drops significantly under extreme weather conditions. Summary of the Invention

[0005] In view of this, embodiments of this application provide a machine learning-based ice-melting decision evaluation method and system to solve the problem of low decision accuracy in the ice-melting decision-making process.

[0006] According to a first aspect of this application, a machine learning-based ice-melting decision evaluation method is provided, the method comprising: Acquire multi-source monitoring data, which includes multi-source meteorological parameters and multi-source transmission line stress parameters; A multi-source heterogeneous icing state parameter set is established based on the multi-source monitoring data. The multi-source heterogeneous icing state parameter set includes an icing state assessment factor set, an icing factor weight vector, an icing state evaluation set, and an icing state judgment model. The multi-source heterogeneous icing state parameter set is subjected to weighted low-rank sparse collaborative decomposition to obtain the icing dominant mode matrix; The optimal ice-melting power level is determined based on the ice-dominant mode matrix; the optimal ice-melting power level is the optimal hyperparameter locked by using the ice-dominant mode matrix and a joint optimization strategy of particle swarm cross-validation. Calculate the ice-melting parameters based on the optimal ice-melting power setting, and set the ice-melting parameters to the ice-melting device.

[0007] In some embodiments, a multi-source heterogeneous icing state parameter set is established based on the multi-source monitoring data, including: Collect driving factors from the multi-source monitoring data, wherein the driving factors are data from the multi-source monitoring data that can cause icing on transmission lines; An icing state assessment factor set is constructed based on the driving factors, and the icing state assessment factor set includes a numerical set of multiple types of driving factors over multiple time periods; A weight vector for icing factors is set, which includes the influence weight of the driving factors; the influence weight is used to quantify the severity of the driving factors' impact on icing. An icing status evaluation set is established based on a preset icing risk level, and the icing status evaluation set includes multiple icing status levels. The icing factor weight vector and the icing state membership matrix are fuzzily synthesized to obtain the icing level evaluation vector; the icing state membership matrix is ​​used to describe the membership relationship of the driving factors to the icing state level. An icing state assessment model is established based on the icing level assessment vector. The icing state assessment model is used to select the icing state level corresponding to the maximum value in the icing level assessment vector as the target icing state level.

[0008] In some embodiments, a weighted low-rank sparse collaborative decomposition is performed on the multi-source heterogeneous icing state parameter set to obtain the icing dominant mode matrix, including: A state parameter matrix is ​​constructed based on the multi-source heterogeneous icing state parameter set, and the state parameter matrix includes multiple icing state parameters; The icing state parameters of different data types in the state parameter matrix are standardized to generate a standardized matrix; The normalized matrix is ​​decomposed into an icing dominant mode matrix and an icing noise matrix; The icing dominant mode matrix and the icing noise matrix are iteratively optimized using the augmented Lagrange multiplier method.

[0009] In some embodiments, the icing state parameters of different data types in the state parameter matrix are standardized to generate a standardized matrix, including: Extract the icing state parameters from the state parameter matrix; Calculate the maximum and minimum values ​​of the column containing the icing state parameter; A first difference is calculated based on the icing state parameters and the minimum value of the column, and a second difference is calculated based on the maximum value of the column and the minimum value of the column; Generate a standardized parameter, wherein the standardized parameter is the ratio of the first difference to the second difference.

[0010] In some embodiments, the augmented Lagrange multiplier method is used to iteratively optimize the icing dominant mode matrix and the icing noise matrix, including: Obtain the target's icing status level; The iteration weights are dynamically adjusted based on the target icing state level. The iterative constraints and convergence conditions are set. The iterative constraints include the sum of the ice-dominant mode matrix and the ice-noise matrix being equal to the normalized matrix. The convergence conditions include the change in the ice-dominant mode matrix and the ice-noise matrix being less than a preset change threshold. Initialize the icing dominant mode matrix, the icing noise matrix, and the Lagrange multiplier matrix; The ice-dominant mode matrix and the ice-noise matrix are solved based on the iterative weights and the iterative constraints, and the ice-dominant mode matrix, the ice-noise matrix, and the Lagrange multiplier matrix are updated according to the convergence conditions. When the icing dominant mode matrix and the icing noise matrix satisfy the convergence condition, the icing dominant mode matrix is ​​output.

[0011] In some embodiments, determining the optimal ice-melting power level based on the ice-dominant mode matrix includes: The ice-covered dominant mode matrix is ​​used as sample data and input into the support vector machine classification model. Define a label vector for the classification model. The label vector includes multiple label values, which are used to characterize the melting power required by the melting device under any icing state level in the icing state evaluation set. Initialize a particle swarm, wherein the particles in the particle swarm are used to represent the parameter combination of the regularization parameter and the radial basis kernel parameter of the classification model; The process of optimizing the parameters of the classification model using the target icing state level is used to iteratively update the position and velocity of the particles, and the classification model is trained using the combination of the particle parameters. When the update rate and position change converge during training, or when the number of iterations reaches the maximum number of iterations, the globally optimal particle is output. Using the globally optimal particle, a binary classification support vector machine is trained for each level, and the ice melting power level is obtained using the decision function; Based on the ice-melting power level, calculate the decision values ​​of all classifiers, and determine the optimal ice-melting power level based on the decision values.

[0012] In some embodiments, the process of optimizing the parameters of the classification model using the target icing state level to iteratively update the position and velocity of particles, and training the classification model using the combination of the particle parameters, includes: For each particle, a five-fold cross-validation mechanism is used to verify the classification performance of each support vector machine classification model; Calculate the fitness value, and update the historical best position and global best position of each particle based on the fitness value; Calculate the change in update rate and position; If the changes in the update velocity and position do not converge, and / or the number of iterations does not reach the maximum number of iterations, the process of optimizing the parameters of the classification model using the target icing state level is repeated to iteratively update the position and velocity of the particles.

[0013] In some embodiments, the globally optimal particle is used to train a binary classification support vector machine for each level, and the decision function is used to obtain the ice melting power level, including: The globally optimal particle is used as an input sample to the trained support vector machine classifier; Calculate the Euclidean distance between the input samples; A kernel function is constructed based on the Euclidean distance and a preset width control parameter, wherein the width control parameter is used to control the width of the kernel function. Obtain the Lagrange multipliers and the bias term; A decision function is constructed based on the input sample, the Lagrange multiplier, the label vector, the kernel function, and the bias term; The required ice-melting power level is obtained using the decision function.

[0014] In some embodiments, the decision values ​​of all classifiers are calculated based on the ice-melting power level, and the optimal ice-melting power level is determined based on the decision values, including: Perform multi-class classification decision on the icing dominant mode matrix; Calculate the decision values ​​of all classifiers for the ice-dominant mode matrix; Obtain the ice-melting power level corresponding to the maximum decision value to generate the optimal ice-melting power level.

[0015] According to a second aspect of this application, a machine learning-based ice-melting decision evaluation system is provided, the system comprising: The data acquisition module is used to acquire multi-source monitoring data, which includes multi-source meteorological parameters and multi-source transmission line stress parameters. The icing status assessment module is used to establish a multi-source heterogeneous icing status parameter set based on the multi-source monitoring data. The multi-source heterogeneous icing status parameter set includes an icing status assessment factor set, an icing factor weight vector, an icing status evaluation set, and an icing status judgment model. The multi-source data source fusion module is used to perform weighted low-rank sparse collaborative decomposition on the multi-source heterogeneous icing state parameter set to obtain the icing dominant mode matrix. The ice-melting power level control module is used to determine the optimal ice-melting power level based on the ice-covering dominant mode matrix; the optimal ice-melting power level is the optimal hyperparameter locked by using the ice-covering dominant mode matrix and a joint optimization strategy of particle swarm cross-validation. The ice-melting parameter output module is used to calculate the ice-melting parameters according to the optimal ice-melting power level, and to set the ice-melting parameters to the ice-melting device.

[0016] According to a third aspect of this application, a computer device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor executes the program to implement the above-described machine learning-based ice melting decision evaluation method.

[0017] According to a fourth aspect of this application, a storage medium is provided that stores a computer program thereon, which, when executed by a processor, implements the above-described machine learning-based ice-melting decision evaluation method.

[0018] By employing the above technical solutions, embodiments of this application provide a machine learning-based ice-melting decision-making assessment method and system. The method, after acquiring multi-source monitoring data, establishes a multi-source heterogeneous icing state parameter set based on the multi-source monitoring data. Then, it performs a weighted low-rank sparse collaborative decomposition on the multi-source heterogeneous icing state parameter set to obtain the icing dominant mode matrix. Based on the icing dominant mode matrix, and combined with a particle swarm optimization cross-validation joint optimization strategy, it determines the optimal ice-melting power level. Finally, it calculates the ice-melting parameters based on the optimal ice-melting power level and sets the ice-melting parameters to the ice-melting device. This method, through icing state assessment, multi-source data fusion, and optimal ice-melting power level control, calculates the icing level based on multi-source monitoring data, optimizes the ice-melting power calculation, and drives the ice-melting device to operate under the optimal ice-melting parameters. This improves the accuracy and response speed of icing state assessment and reduces the risk of over-melting or under-melting.

[0019] The proposed method establishes a three-level fusion assessment of icing status based on factors, weights, and membership, outputting an icing level vector in a single step, enabling interpretable and quantifiable determination of icing severity. The method also employs weighted low-rank sparse collaborative decomposition, decomposing the standardized multi-source heterogeneous matrix into an icing-dominant model matrix and an icing noise matrix, and dynamically adjusting the weight matrix to achieve robust noise reduction and collaborative extraction of global and local features. Furthermore, the method utilizes a particle swarm optimization (PSO) cross-validation joint optimization strategy to search for SVM hyperparameters online, quickly locking the globally optimal setting and achieving precise adaptive icing melting control under complex disturbance environments.

[0020] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below. Attached Figure Description

[0021] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a schematic diagram illustrating the connection relationship between the electronic device and the ice-melting device provided in an embodiment of this application; Figure 2 A schematic diagram of the ice-melting decision evaluation method based on machine learning provided in an embodiment of this application; Figure 3 This is a schematic diagram illustrating the process of establishing a multi-source heterogeneous icing state parameter set provided in an embodiment of this application. Figure 4 This is a schematic diagram of the weighted low-rank sparse collaborative decomposition process provided in an embodiment of this application; Figure 5 This is a schematic diagram of the DC ice-melting device provided in the embodiments of this application; Figure 6 This is a schematic diagram of the structure of a machine learning-based ice-melting decision evaluation system provided in an embodiment of this application. Detailed Implementation

[0022] The present application will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present application can be combined with each other.

[0023] In this embodiment of the application, the de-icing decision evaluation refers to the decision-making process of obtaining multi-source detection data by detecting detection parameters in the power transmission area, analyzing the multi-source detection data, and making decisions to determine the de-icing parameters for controlling the de-icing device.

[0024] Ice-melting decision assessment can be applied to the equipment maintenance process of transmission lines by power transmission organizations. In mountainous transmission areas during cold seasons, the surfaces of transmission lines and other equipment are prone to icing, which increases the load on these devices, affecting conductor tension, sag, and tower load, potentially leading to accidents such as line breaks and tower collapses, thus compromising the continuity of power supply. Therefore, to obtain timely and accurate information on the icing status of transmission lines and ensure power supply continuity and public safety, it is necessary to monitor and issue alarms regarding the icing status of transmission areas.

[0025] In some embodiments, the icing condition of the power transmission area can be visually inspected through manual inspections. When visual inspection determines that the icing condition of the power transmission equipment is severe, de-icing devices can be activated to carry out de-icing operations to alleviate the icing condition of the power transmission equipment.

[0026] In some embodiments, data quantification can also be performed by combining visual detection results. That is, by capturing visual images and performing image recognition on the visual images, the ice thickness of the power transmission equipment in the image can be determined. A fixed ice thickness threshold is then set. When the ice thickness reaches the ice thickness threshold, an alarm signal can be generated to trigger the ice-melting device to perform ice-melting operations.

[0027] Due to limitations imposed by human resources, weather windows, and blind spots in observation, manual ice observation or fixed threshold alarm methods are insufficient for achieving a comprehensive, real-time assessment of icing conditions. Therefore, in some embodiments, classification decision-making technology can be used to perform data detection and status assessment of icing conditions. Based on the decision matrix, de-icing equipment can be deployed to perform de-icing operations on transmission equipment, thereby improving de-icing efficiency and reducing the risks to power grid safety.

[0028] For example, we can obtain the first icing risk parameters for M substations, and for each substation, obtain N second icing risk parameters and N target icing risk parameters. Then, we generate a target risk matrix based on the target icing risk parameters, and based on the target risk matrix, objective function, and constraints, we obtain a decision matrix and allocate de-icing devices to the substations according to the decision matrix. Since the target icing risk parameters represent the icing risk reduction achieved by allocating de-icing devices to the substations, and the objective function refers to maximizing the total de-icing benefit, the decision matrix obtained through classification decision-making based on the target risk matrix, objective function, and constraints can represent the optimal de-icing decision scheme.

[0029] However, because the decision-making process for ice melting relies on fixed thresholds or manual experience to determine the ice melting power level, it lacks dynamic perception of ice thickness, sudden weather changes, and load disturbances. This leads to over-melting or under-melting during the ice melting process, resulting in inaccurate power level settings for the ice melting device, which can easily cause misjudgments, energy waste, and incomplete ice removal. Furthermore, the decision-making process for ice melting lacks sensitivity to multi-source monitoring data. The real-time collected meteorological, stress, and other multi-source data suffer from inconsistencies in dimensions, high noise levels, and numerous outliers, making it difficult to effectively integrate and extract robust features for the ice melting decision-making process. Consequently, the accuracy of decision-making drops significantly under extreme weather conditions.

[0030] To address the issue of low decision-making accuracy in the ice-melting decision-making process, this application provides a machine learning-based ice-melting decision-making evaluation method in some embodiments. This method can be deployed as a control program on a field terminal, real-time accessing and fusing multi-source monitoring data such as meteorological data and conductor tension data to calculate the icing level and optimize and accelerate the calculation of ice-melting power, thereby driving the ice-melting device to operate at the optimal ice-melting power level. By fusing multi-source monitoring data from different meteorological factors and stress levels, the method maintains robust decision-making under complex weather conditions, sensor errors, and sudden load changes. This method can provide core technical support for icing monitoring and ice-melting decision-making for transmission lines in winter, improving the accuracy and response speed of icing status assessment, reducing the risk of over-melting or under-melting, and providing reliable decision-making tools for the safe and stable operation of the power grid and public safety.

[0031] The method can be applied to electronic devices with data processing capabilities. These electronic devices include, but are not limited to, computers, servers, mobile terminals, smart wearable devices, and industrial control machines. For ease of description, this application embodiment uses an electronic device as the execution subject of the method. It should be understood that the method can also be applied to other types of execution subjects, which are not illustrated in this application embodiment. Figure 1 As shown, the electronic device can establish a communication connection with the ice-melting device, and can execute a machine learning-based ice-melting decision-making and evaluation method by configuring an ice-covering state processing unit, a multi-source data fusion unit, and an optimal ice-melting power level control unit in the electronic device. Figure 2 As shown, the method includes: S101. Obtain multi-source monitoring data.

[0032] When implementing de-icing decisions, multi-source monitoring data can be acquired first. This multi-source monitoring data includes multi-source meteorological parameters and multi-source transmission line stress parameters. Since the de-icing decision-making and evaluation method can be deployed as a control program in electronic devices such as field terminals, the electronic devices can access and integrate multi-source monitoring data, including meteorological parameters and transmission line stress parameters, in real time when acquiring the data. For example, meteorological parameters may include temperature, humidity, wind speed, and rainfall (snowfall). Transmission line stress parameters may include conductor tension, insulator string tension, and insulator string tilt angle.

[0033] Multi-source monitoring data can be obtained through real-time monitoring of power transmission areas from multiple data sources. For example, multiple temperature sensors, humidity sensors, wind speed sensors, stress sensors, and other detection devices can be installed in the power transmission area. Electronic devices can also connect to the meteorological data platform server via a network. Multiple detection devices and servers can provide the electronic devices with detection data of different structures, meaning the monitoring data is multi-source and heterogeneous. Therefore, when acquiring multi-source monitoring data, the electronic devices can send data acquisition requests to multiple detection devices and servers, enabling these devices and servers to provide feedback on meteorological parameters and power transmission line stress parameters to the electronic devices.

[0034] Because data monitoring of the data source is a lengthy process, and the multi-source monitoring data within the data source has a time sequence, which affects the icing status of power transmission equipment, in some embodiments, electronic devices can acquire monitoring data in stages and by type when acquiring multi-source monitoring data. After acquiring the multi-source monitoring data, the data is segmented according to a preset monitoring cycle for classification and decision-making.

[0035] S102. Establish a set of multi-source heterogeneous icing state parameters based on multi-source monitoring data.

[0036] After acquiring multi-source monitoring data, electronic equipment can assess the icing status of the transmission area based on the established icing status assessment unit. Therefore, a multi-source heterogeneous icing status parameter set can be established based on the multi-source monitoring data. This multi-source heterogeneous icing status parameter set includes an icing status assessment factor set, an icing factor weight vector, an icing status evaluation set, and an icing status judgment model.

[0037] The icing status assessment unit can construct a four-level fuzzy evaluation system of "factor-weight-evaluation-membership" to map multi-source monitoring data such as temperature, humidity, wind speed, and conductor stress into four icing levels: no icing, slight icing, moderate icing, and severe icing. This can help alleviate the problem of high misjudgment rate of the threshold method in scenarios with fuzzy boundaries.

[0038] like Figure 3As shown, in some embodiments, when establishing a multi-source heterogeneous icing state parameter set based on multi-source monitoring data, driving factors in the multi-source monitoring data can be collected first, and an icing state assessment factor set can be constructed based on the driving factors. The driving factors are data from the multi-source monitoring data that can cause icing of transmission lines, and the icing state assessment factor set includes a set of values ​​for multiple types of driving factors over multiple time periods.

[0039] For example, when establishing a set of factors for assessing the icing status of transmission lines, the main driving factors that cause icing of transmission lines can be collected, and a set of factors for assessing the icing status of transmission lines can be established. U : U ={ u 1, u 2, ..., u m}; in, u 1, u 2, ..., u m These represent the values ​​of various driving factors over multiple time periods; m This indicates the number of key driving factors. For example, when temperature, humidity, wind speed, and conductor stress are identified as the four essential key factors for assessing icing risk, it can be determined that... m =4.

[0040] When establishing a multi-source heterogeneous icing state parameter set, an icing factor weight vector can also be set. This icing factor weight vector includes the influence weights of driving factors, which quantify the severity of the driving factors' impact on icing. For example, in determining the icing state of transmission lines, the factors in the icing state assessment factor set show significant differences in their ability to reflect the severity of icing. Therefore, when establishing the icing factor weight vector, an icing factor weight vector can be set. A This is used to quantify the severity of the impact of various factors on icing; the icing factor weight vector. A It can be represented as: A ={ a 1, a 2, ..., a m}; in, a 1, a 2, ..., a m The weights of the above-mentioned main driving factors can be represented by: a i The first in the representation i Each influence weight, such as a iIt is a value greater than 0 and less than 1, and a 1, a 2, ..., a m The sum of 1 and 2 equals 1.

[0041] When establishing a multi-source heterogeneous icing state parameter set, an icing state evaluation set can also be established based on a preset icing risk level. This icing state evaluation set includes multiple icing state levels. For example, the icing state evaluation set V can be used to evaluate the icing state of transmission line conductors, i.e.: V ={ v 1, v 2, ..., v n}; in, v 1, v 2, ..., v n These represent the possible icing status assessment results. For example, based on the actual icing risk level, the assessment results can include four icing status levels: n =4, v A value of 1 indicates an icing status level of no icing. v The 2 indicates a light icing level; v The 3 indicates a low-level icing condition. v A score of 4 indicates a severe icing condition.

[0042] After setting the weight vector of icing factors, an icing state assessment model can be established. That is, electronic devices can perform fuzzy synthesis of the icing factor weight vector and the icing state membership matrix to obtain an icing level assessment vector. The icing state membership matrix describes the membership relationship between driving factors and the icing state level.

[0043] For example, when establishing an icing condition assessment model, the weight vector of icing factors can be first set. A membership matrix of icing state R Fuzzy synthesis is performed to obtain the icing level assessment vector B, i.e.: B = A ⊙ R ={ b 1, b 2, ..., b n}; Where ⊙ represents the weighted average matrix multiplication operation; R The membership matrix represents the icing state and is used to describe the set of factors for assessing the icing state. U Evaluation set of driving factors on icing statusV The membership relationships of various icing states are determined by the icing state membership matrix. R This allows for a comprehensive quantification of the impact of these factors on the icing state. Therefore, the membership matrix of the icing state... R It can be represented as:

[0044] in, r ij Membership matrix of icing state R The matrix elements in the set represent the th element in the icing state assessment factor set U. i The element values ​​of each driving factor affect the icing state evaluation set. V The Middle j The degree of membership of each icing state indicator, such as r ij It is a value that takes the value in the interval [0, 1].

[0045] Then, an icing state assessment model is established based on the icing level assessment vector. This icing state assessment model is used to select the icing state level corresponding to the maximum value in the icing level assessment vector as the target icing state level. For example, by establishing the icing state assessment model, the icing level assessment vector can be determined. B The maximum value in b k And query the maximum value. b k Corresponding rating set level v k As the target icing state, the target icing state v k The selection of can be expressed by the following formula:

[0046] Based on the above formula, an evaluation vector can be derived from the icing level assessment. B Find the index corresponding to the maximum value of the icing status assessment value. k And mapped to the icing state evaluation set. V The first k Each element is used as the target icing state. v k .

[0047] It is evident that by establishing an icing condition assessment unit in electronic devices, an icing condition evaluation model can be built for multi-source meteorological parameters and conductor stress values, thereby achieving a quantitative assessment of the icing condition. This helps to solve the problems of parameter ambiguity and uncertainty in the assessment process and improves the accuracy of the assessment results.

[0048] S103. Perform weighted low-rank sparse collaborative decomposition on the multi-source heterogeneous icing state parameter set to obtain the icing dominant mode matrix.

[0049] After establishing a multi-source heterogeneous icing state parameter set, the electronic device can call the multi-source data source fusion unit and use weighted low-rank sparse collaborative decomposition to split the multi-source heterogeneous icing state parameter set to obtain the icing dominant mode matrix and the icing noise matrix, which are used to represent the low-rank principal components and sparse anomalies, respectively.

[0050] like Figure 4 As shown, in some embodiments, when an electronic device performs a weighted low-rank sparse cooperative decomposition of a multi-source heterogeneous icing state parameter set to obtain the icing dominant mode matrix, it can first construct a state parameter matrix based on the multi-source heterogeneous icing state parameter set. The state parameter matrix includes multiple icing state parameters.

[0051] For example, a state parameter matrix can be formed from N m-dimensional icing state parameter data. X The state parameter matrix is ​​as follows: R N×m Scale parameter matrix. State parameter matrix. X It can include N×m Each matrix element represents an icing state parameter. x ij Icing state parameters x ij It can represent the state parameter matrix X The Middle i The first (row) of data j Individual (column) values.

[0052] After constructing the state parameter matrix, the icing state parameters of different data types in the state parameter matrix can be standardized to generate a standardized matrix. In some embodiments, when standardizing the icing state parameters of different data types in the state parameter matrix, the icing state parameters can be extracted from the state parameter matrix, and the maximum and minimum values ​​of the columns containing the icing state parameters can be calculated. Then, a first difference is calculated based on the icing state parameters and the minimum values, and a second difference is calculated based on the maximum and minimum values, thereby generating standardized parameters and forming a standardized matrix. The standardized parameter is the ratio of the first difference to the second difference.

[0053] For example, icing state parameters of different data types in the state parameter matrix can be standardized based on the maximum and minimum values ​​of each column, thereby standardizing the icing state parameters. x ij Mapping to the interval [0, 1], then performing a standardization operation, i.e.:

[0054] in, x ij norm Indicates standardized parameters; x ij Indicates icing state parameters; min ( x j () represents the minimum value of the j-th parameter. max ( x j ) indicates the first j The maximum value of each parameter. After standardization, the state parameter matrix... X Updated to a standardized matrix X norm .

[0055] After generating a normalized matrix through normalization, the normalized matrix can be decomposed into an icing dominant mode matrix and an icing noise matrix. For example, the normalized matrix... X norm Decomposed into the icing dominant mode matrix L and icing noise matrix S Among them, the dominant mode matrix of icing L Normal data constitutes the majority of the representative data; icing noise matrix S This represents outliers or noise in the data. Therefore, X norm The decomposition form can be expressed as: X norm = L + S .

[0056] The augmented Lagrange multiplier method is then used to iteratively optimize the icing dominant mode matrix and the icing noise matrix. In some embodiments, when using the augmented Lagrange multiplier method to iteratively optimize the icing dominant mode matrix and the icing noise matrix, the target icing state level can be obtained first, and the iteration weights can be dynamically adjusted based on the target icing state level.

[0057] Because data from different icing levels affect the dominant icing mode matrix L and icing noise matrix S The contributions differ; for example, in cold, windy weather, data with high icing levels contain more outliers. Therefore, the dominant icing mode matrix can be solved. L and icing noise matrix S This will lead to more accurate decision-making results.

[0058] Solving the dominant mode matrix of icing L and icing noise matrix S At that time, an icing status level can be introduced. bk Dynamic adjustment of weights W i Dynamically adjust weights W i Make the icing noise matrix S The penalty is stronger, making the decomposed ice-dominant mode matrix more effective. L More accurate.

[0059] Then set iterative constraints and convergence conditions. The iterative constraints include the sum of the ice-dominant mode matrix and the ice-noise matrix being equal to the normalized matrix; the convergence conditions include the change in the ice-dominant mode matrix and the ice-noise matrix being less than a preset change threshold.

[0060] For example, the dominant mode matrix of icing L and icing noise matrix S The constraint calculation is shown in the following formula, where the constraint condition is the icing dominant mode matrix. L and icing noise matrix S Completely equal to X norm ,Right now:

[0061]

[0062] Where W is the weight matrix, which is related to the icing level; b k This is the rating value for the icing level. α This is the adjustment coefficient; λ To weigh the parameters; ⊙ represents element-wise multiplication; Represents nuclear norm operations; This represents the L1 norm operation.

[0063] Then, by initializing the ice-dominant mode matrix, ice-noise matrix, and Lagrange multiplier matrix, and solving the ice-dominant mode matrix and ice-noise matrix based on iterative weights and iterative constraints, the ice-dominant mode matrix, ice-noise matrix, and Lagrange multiplier matrix can be updated according to the convergence condition. When the ice-dominant mode matrix and ice-noise matrix satisfy the convergence condition, the ice-dominant mode matrix is ​​output.

[0064] For example, the icing dominant mode matrix can be iteratively optimized by introducing the augmented Lagrange multiplier method. L and icing noise matrix S The solution can be obtained by initializing the settings. L =0, S =0, and the Lagrange multiplier matrix Y =0. Then the icing dominant mode matrix can be updated according to the following formula. L :

[0065] in, The square root operation represents the sum of the squares of all elements in a matrix. μ This represents the penalty parameter, used to control the constraint conditions. X norm =L+S The strictness of it, μ >0.

[0066] Similarly, the icing noise matrix can be updated according to the following formula. S :

[0067] And, update the Lagrange multiplier matrix according to the following formula. Y :

[0068] Repeat the above steps until the convergence condition is met, i.e., the dominant icing mode matrix is ​​obtained. L and icing noise matrix S The change is less than the preset threshold:

[0069] Among them, the term used to balance the low-rank terms and sparse terms The weights; This is the preset tolerance; for example, the tolerance can be set to 10. -6 .

[0070] Therefore, in order to robustly and interpretably extract global and local features from multi-source heterogeneous icing state parameters, and to more accurately and completely identify and separate normal meteorological load patterns and anomalous change details under different icing levels, a multi-source data fusion unit is used to perform weighted low-rank sparse collaborative decomposition on temperature, humidity, wind speed, and conductor stress values ​​to obtain the icing dominant mode matrix. L It can preserve the overall meteorological trend while suppressing sensor noise and sudden disturbances, providing more accurate input for subsequent determination of the optimal ice melting power level.

[0071] S104. Determine the optimal ice melting power level based on the dominant icing mode matrix.

[0072] After obtaining the dominant icing mode matrix, the optimal ice-melting power level can be determined based on the dominant icing mode matrix. This optimal ice-melting power level is the optimal hyperparameter locked using the dominant icing mode matrix combined with a particle swarm optimization (PSO) cross-validation joint optimization strategy. Five-fold cross-validation is a model evaluation method that randomly and uniformly divides all training data into five non-overlapping parts, and performs multiple rounds of training and validation.

[0073] In real-world ice-melting scenarios, icing state parameters are affected by sensor errors, sudden weather disturbances, and load abrupt changes. This makes it difficult for fixed-parameter Support Vector Machines (SVMs) to provide robust power level classifications based on icing levels, and they are prone to getting trapped in local optima. Therefore, electronic devices can utilize a globally optimal ice-melting power level control unit. This unit adaptively uses the dominant icing mode matrix L and combines it with a particle swarm optimization-cross-validation joint optimization strategy to quickly identify the optimal hyperparameters, enabling accurate determination of the optimal ice-melting power level under complex disturbance environments.

[0074] In some embodiments, when determining the optimal ice-melting power level based on the ice-dominant mode matrix, the ice-dominant mode matrix can be first input as sample data into the support vector machine classification model, and then a label vector can be defined for the classification model. The label vector includes multiple label values, which are used to characterize the ice-melting power required by the ice-melting device under any ice-melting state level in the ice-melting state evaluation set.

[0075] Support Vector Machines (SVMs) are binary classification models that can implicitly increase the dimensionality of data through kernel functions, making it linearly separable. Electronic devices can improve the decision-making process of SVMs using particle swarm optimization and multi-source array fusion to obtain the ice-dominant model matrix. This allows for the training of a corresponding classifier for each ice-melting power level, enabling intelligent decision-making regarding ice melting power.

[0076] For example, the dominant icing mode matrix L can be used as sample data for the SVM classification model. Simultaneously, a label vector y is defined for the SVM classification model, i.e., y∈{1,2,…,n}, corresponding to the melting power required by the DC de-icing device under the icing state assessment results defined in the icing state assessment set V.

[0077] The particle swarm is then initialized, where the particles in the swarm represent the parameter combinations of the regularization parameters and radial basis function kernel parameters of the classification model. Each particle can represent a set of possible parameter combinations. For example, during particle swarm initialization, a total of N randomly generated particles are uniformly distributed in the parameter space. Then each particle... pk Represents a combination of parameters ( C k , γ k ), initial velocity v k It is 0. Among them, C k It is the regularization parameter of SVM, used to control the complexity of the classifier. γ k These are the kernel parameters of the Radial Basis Function (RBF), used to control the nonlinearity of the decision boundary.

[0078] After initializing the particle swarm, the parameters of the classification model can be optimized using the target icing state level. This involves iteratively updating the particle positions and velocities, and then training the classification model using the combined particle parameters. In other words, the icing state is utilized. v k The SVM parameter optimization process involves iteratively updating the particle positions and velocities, training an SVM model using the parameter combinations for each particle, cross-validating each SVM to evaluate the model's classification performance, calculating the fitness value, and continuously updating the historical best position and global best position for each particle. The fitness value quantifies the impact of particle parameter combinations on the classifier's results; a higher fitness value indicates better classification performance.

[0079] In some embodiments, during the training of the classification model, a five-fold cross-validation mechanism can be used for each particle to verify the classification performance of each support vector machine classification model. The fitness value is then calculated, and the historical best position and global best position of each particle are updated based on the fitness value. By calculating the change in update velocity and position and comparing it with preset convergence conditions, if the change in update velocity and position exceeds the threshold set in the convergence conditions, indicating that convergence has not yet occurred, or the maximum number of iterations has not been reached, the particle position and velocity can be iteratively updated by repeatedly executing the process of optimizing the classification model parameters using the target icing state level.

[0080] For example, you can set the maximum number of iterations. T =100; C k The value range is

[10] . -3 10 3 ]; γ k The value range is

[10] . -5 10 2 ]; c1=c2=2. Then for each particle ( C k , γ k(This can be achieved by) training an SVM using 5-fold cross-validation and calculating the classification accuracy as the fitness value. The formula for calculating the fitness value is:

[0081]

[0082]

[0083] in, β For adjustment coefficients, w i As a weight for icing level, v ki The icing status of the i-th data point is calculated based on the set of factors for assessing line icing status. b ki This represents the confidence level of the icing level, which is used to flexibly adjust the weight of the classification error penalty. This is an indicator function; the function value is 1 if the classification is correct, and 0 otherwise.

[0084] The globally optimal particle is output when the update rate and position changes converge during training, or when the maximum number of iterations is reached. That is, the update rate is used to determine the optimal particle. v k and location p k The process continues until the maximum number of iterations T or the changes in update velocity and position converge, yielding the globally optimal particle. g best =( C' , γ' ),speed v k and location p k It can be represented as:

[0085]

[0086] in, w Weights based on motion inertia; c 1 and c 2 is the learning factor. r 1 and r 2 represents the introduced randomness factor used to avoid getting trapped in the global optimum, hence it is taken as... r 1, r 2 U (0, 1). p best This represents the optimal combination of parameters reached during the particle's iterative process. gbest This represents the globally optimal combination of hyperparameters.

[0087] If update speed v k and location p k If the change in the rate of change does not converge or does not reach the maximum number of iterations T, then repeat the above steps to continue iterating until the update rate... v k and location p k The change in the quantity of change has not converged, and / or the maximum number of iterations T has been reached, resulting in the globally optimal particle. g best =( C' , γ' ).

[0088] Then, using the globally optimal particle, a binary classification support vector machine is trained for each level, and the melting power level is obtained using a decision function. In some embodiments, the globally optimal particle can be used as an input sample to the trained support vector machine classifier, and the Euclidean distance between the input samples is calculated. A kernel function is then constructed based on the Euclidean distance and a preset width control parameter. The width control parameter controls the width of the kernel function.

[0089] Next, obtain the Lagrange multipliers and bias terms, and then construct a decision function based on the input samples, Lagrange multipliers, label vector, kernel function, and bias terms, thereby using the decision function to obtain the required ice melting power level.

[0090] For example, local optimal particles can be used ( C' , γ' For each gear level, train a binary classification SVM and utilize the decision function. f c ( x The required ice-melting power level is obtained. The decision function is... f c ( x ) is represented as:

[0091] Where x is the input sample to be predicted; α i (c) Let be a Lagrange multiplier, representing the training samples in the c-th classifier. x i The importance weights of the decision boundary can be obtained through training; b (c) This is a bias term used to ensure that the classification hyperplane is located in the optimal position; K ( x i, x ) is the kernel function, which can be represented as:

[0092] In the formula, K ( x i , x j ) represents sample points x i and x j The kernel function values ​​between; exp is the exponential function; γ' This represents the width control parameter, which is a parameter of the kernel function; Represents sample points x i and x j The square of the Euclidean distance between them.

[0093] Then the temporary binary label of the c-th classifier in the decision function y i (c) The conversion formula is:

[0094] Next, based on the ice-melting power level, the decision values ​​of all classifiers are calculated to determine the optimal ice-melting power level. In some embodiments, when determining the optimal ice-melting power level, multi-classification decision-making can be performed on the ice-dominant mode matrix, and the decision values ​​of all classifiers for the ice-dominant mode matrix can be calculated. Then, the ice-melting power level corresponding to the maximum decision value is obtained to generate the optimal ice-melting power level.

[0095] For example, to predict the icing level of new data, after assessing the icing status of new multi-source monitoring data and fusing multiple data sources, the corresponding dominant icing mode matrix can be obtained. L new To perform multi-class classification decisions and calculate the decision values ​​of all classifiers. f c ( L new The final optimal ice-melting power level. Input p It can be calculated using the following formula:

[0096] It is evident that the optimal ice-melting power level control unit can employ a two-layer optimization method of "particle swarm optimization-cross-validation" to train multiple SVM decision generators and output the power level of the ice-melting device corresponding to the maximum decision value in real time. This can solve the problems of fixed-parameter SVMs easily getting trapped in local extrema and the coarse division of ice-melting power levels.

[0097] S105. Calculate the ice-melting parameters based on the optimal ice-melting power setting, and set the ice-melting parameters to the ice-melting device.

[0098] After determining the optimal de-icing power level, de-icing parameters can be calculated based on the optimal de-icing power level and then set to the de-icing device. The de-icing device can be a DC step-down chopper type de-icing device used to perform de-icing operations on transmission lines.

[0099] The main circuit structure of the DC de-icing device is as follows: Figure 5 As shown. The de-icing device may include: a head-end transformer, an AC / DC rectifier, an IGBTBuck chopper, and de-icing output terminals. For example, the head-end transformer can be a three-phase oil-immersed step-down transformer, with a zero-excitation voltage regulator ±2×2.5 on the high-voltage side, a star-connected low-voltage side, and a neutral point grounded through a small resistor. The AC / DC rectifier can use a three-phase full-bridge diode rectifier module, with four sets of 600A modules connected in parallel to improve redundancy; the DC bus is connected in parallel with four x 6800μF support capacitors, and the ripple factor is ≤3%. The IGBTBuck chopper can be based on a single-tube step-down topology, using a planar gate IGBT, with anti-parallel diodes in the same package; the switching frequency is 1kHz, the output is a continuously adjustable current of 0-3000A, and the driver board integrates fiber optic triggering and hardware overcurrent shutdown dual protection. The copper busbar of the de-icing output terminal has a cross-sectional area of ​​120mm×10mm, a silver-plated surface, and an insulation spacing of 60mm; it is equipped with a DC fast disconnect switch and a Hall current sensor to ensure line safety.

[0100] The ice-melting device can receive the output of the optimal ice-melting power level control unit and calculate the ice-melting current that should be used in the device. It directly passes a controllable DC current into the inside of the wire to be melted, and the generated Joule heat makes the wire itself heat up evenly, thereby quickly melting the surface ice.

[0101] By applying the technical solutions of the above embodiments, the machine learning-based ice-melting decision-making and evaluation method described in the above embodiments can use a factor set, weight vector, membership matrix, and weighted average synthesis as the overall architecture and process when evaluating the icing state. In the multi-source data fusion process, a weighted low-rank-sparse collaborative decomposition model, dynamic adjustment of weights according to the icing state, and its augmented Lagrange multiplier solution algorithm are used. Furthermore, in controlling the optimal ice-melting power level, the global optimization of SVM hyperparameters, the weighted fitness function, and the multi-class decision-making strategy for new samples can be improved to use power-controlled DC ice-melting devices to melt ice on the lines.

[0102] The proposed method establishes a three-level fusion assessment of icing status based on factors, weights, and membership, outputting an icing level vector in a single step, enabling interpretable and quantifiable determination of icing severity. The method also employs weighted low-rank sparse collaborative decomposition, decomposing the standardized multi-source heterogeneous matrix into an icing-dominant model matrix and an icing noise matrix, and dynamically adjusting the weight matrix to achieve robust noise reduction and collaborative extraction of global and local features. Furthermore, the method utilizes a particle swarm optimization (PSO) cross-validation joint optimization strategy to search for SVM hyperparameters online, quickly locking the globally optimal setting and achieving precise adaptive icing melting control under complex disturbance environments.

[0103] The DC step-down chopper de-icing device driven by the evaluation results in the above embodiments can output precise DC current as needed, realizing directional, timed, and power-controlled de-icing operations. After training and calibration, the method can be directly deployed on various transmission lines to complete real-time evaluation of icing conditions and de-icing execution without manual intervention, significantly reducing operation and maintenance costs and power outage time.

[0104] In some embodiments, as a specific implementation of the machine learning-based ice-melting decision evaluation method in the above embodiments, some embodiments of this application also provide a machine learning-based ice-melting decision evaluation system, such as... Figure 6 As shown, the system includes: The data acquisition module is used to acquire multi-source monitoring data, which includes multi-source meteorological parameters and multi-source transmission line stress parameters. The icing status assessment module is used to establish a multi-source heterogeneous icing status parameter set based on multi-source monitoring data. The multi-source heterogeneous icing status parameter set includes an icing status assessment factor set, an icing factor weight vector, an icing status evaluation set, and an icing status judgment model. The multi-source data fusion module is used to perform weighted low-rank sparse collaborative decomposition on the multi-source heterogeneous icing state parameter set to obtain the icing dominant mode matrix. The ice-melting power level control module is used to determine the optimal ice-melting power level based on the ice-dominant mode matrix; the optimal ice-melting power level is the optimal hyperparameter locked by using the ice-dominant mode matrix and a joint optimization strategy of particle swarm cross-validation. The ice-melting parameter output module is used to calculate the ice-melting parameters based on the optimal ice-melting power level, and to set the ice-melting parameters to the ice-melting device.

[0105] By applying the technical solutions of the above embodiments, the ice-melting decision-making and evaluation system based on machine learning described in the above embodiments, after the data acquisition module acquires multi-source monitoring data, the ice-covering state evaluation module establishes a multi-source heterogeneous ice-covering state parameter set based on the multi-source monitoring data. The multi-source data source fusion module then performs weighted low-rank sparse collaborative decomposition on the multi-source heterogeneous ice-covering state parameter set to obtain the ice-covering dominant mode matrix. The ice-melting power level control module then determines the optimal ice-melting power level based on the ice-covering dominant mode matrix and a particle swarm cross-validation joint optimization strategy, so that the ice-melting parameter output module can calculate the ice-melting parameters according to the optimal ice-melting power level and set the ice-melting parameters to the ice-melting device. The system can calculate the ice-covering level based on multi-source monitoring data and optimize the ice-melting power calculation through ice-covering state evaluation, multi-source data fusion, and optimal ice-melting power level control, driving and controlling the ice-melting device to operate under the best ice-melting parameters, which can improve the accuracy and response speed of ice-covering state evaluation and reduce the risk of over-melting or under-melting.

[0106] It should be noted that other corresponding descriptions of the functional units involved in the machine learning-based ice melting decision evaluation system provided in the embodiments of this application can be found in the corresponding descriptions in the machine learning-based ice melting decision evaluation method provided in the above embodiments, and will not be repeated here.

[0107] This application also provides a computer device, specifically a personal computer, server, network device, etc. The computer device includes a bus, processor, memory, and communication interface, and may also include input / output interfaces and a display device. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the computer device stores location information. The network interface of the computer device is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the steps in the various method embodiments.

[0108] Those skilled in the art will understand that the structure of the computer device described above is only a partial structure related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. A specific computer device may include more or fewer components, or combine certain components, or have different component arrangements.

[0109] In one embodiment, a computer-readable storage medium is also provided, which may be non-volatile or volatile, having stored thereon a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0110] In one embodiment, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0111] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0112] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the methods described above.

[0113] Any references to memory, database, or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc.

[0114] Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can take many forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0115] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchain. The processors involved in the embodiments provided in this application may be, but are not limited to, general-purpose processors, graphics processors, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc.

[0116] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0117] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A machine learning-based method for evaluating ice-melting decisions, characterized in that, The method includes: Acquire multi-source monitoring data, which includes multi-source meteorological parameters and multi-source transmission line stress parameters; A multi-source heterogeneous icing state parameter set is established based on the multi-source monitoring data. The multi-source heterogeneous icing state parameter set includes an icing state assessment factor set, an icing factor weight vector, an icing state evaluation set, and an icing state judgment model. The multi-source heterogeneous icing state parameter set is subjected to weighted low-rank sparse collaborative decomposition to obtain the icing dominant mode matrix; The optimal ice-melting power level is determined based on the ice-dominant mode matrix; the optimal ice-melting power level is the optimal hyperparameter locked by using the ice-dominant mode matrix and a joint optimization strategy of particle swarm cross-validation. Calculate the ice-melting parameters based on the optimal ice-melting power setting, and set the ice-melting parameters to the ice-melting device.

2. The method according to claim 1, characterized in that, A multi-source heterogeneous icing state parameter set is established based on the multi-source monitoring data, including: Collect driving factors from the multi-source monitoring data, wherein the driving factors are data from the multi-source monitoring data that can cause icing on transmission lines; An icing state assessment factor set is constructed based on the driving factors, and the icing state assessment factor set includes a numerical set of multiple types of driving factors over multiple time periods; A weight vector for icing factors is set, which includes the influence weight of the driving factors; the influence weight is used to quantify the severity of the driving factors' impact on icing. An icing status evaluation set is established based on a preset icing risk level, and the icing status evaluation set includes multiple icing status levels. The icing factor weight vector and the icing state membership matrix are fuzzily synthesized to obtain the icing level evaluation vector; the icing state membership matrix is ​​used to describe the membership relationship of the driving factors to the icing state level. An icing state assessment model is established based on the icing level assessment vector. The icing state assessment model is used to select the icing state level corresponding to the maximum value in the icing level assessment vector as the target icing state level.

3. The method according to claim 1, characterized in that, The weighted low-rank sparse collaborative decomposition of the multi-source heterogeneous icing state parameter set is performed to obtain the icing dominant mode matrix, including: A state parameter matrix is ​​constructed based on the multi-source heterogeneous icing state parameter set, and the state parameter matrix includes multiple icing state parameters; The icing state parameters of different data types in the state parameter matrix are standardized to generate a standardized matrix; The normalized matrix is ​​decomposed into an icing dominant mode matrix and an icing noise matrix; The icing dominant mode matrix and the icing noise matrix are iteratively optimized using the augmented Lagrange multiplier method.

4. The method according to claim 3, characterized in that, The icing state parameters of different data types in the state parameter matrix are standardized to generate a standardized matrix, including: Extract the icing state parameters from the state parameter matrix; Calculate the maximum and minimum values ​​of the column containing the icing state parameter; A first difference is calculated based on the icing state parameters and the minimum value of the column, and a second difference is calculated based on the maximum value of the column and the minimum value of the column; Generate a standardized parameter, wherein the standardized parameter is the ratio of the first difference to the second difference.

5. The method according to claim 3, characterized in that, The augmented Lagrange multiplier method is used to iteratively optimize the icing dominant mode matrix and the icing noise matrix, including: Obtain the target's icing status level; The iteration weights are dynamically adjusted based on the target icing state level. The iterative constraints and convergence conditions are set. The iterative constraints include the sum of the ice-dominant mode matrix and the ice-noise matrix being equal to the normalized matrix. The convergence conditions include the change in the ice-dominant mode matrix and the ice-noise matrix being less than a preset change threshold. Initialize the icing dominant mode matrix, the icing noise matrix, and the Lagrange multiplier matrix; The ice-dominant mode matrix and the ice-noise matrix are solved based on the iterative weights and the iterative constraints, and the ice-dominant mode matrix, the ice-noise matrix, and the Lagrange multiplier matrix are updated according to the convergence conditions. When the icing dominant mode matrix and the icing noise matrix satisfy the convergence condition, the icing dominant mode matrix is ​​output.

6. The method according to claim 1, characterized in that, Determining the optimal ice-melting power level based on the aforementioned dominant icing mode matrix includes: The ice-covered dominant mode matrix is ​​used as sample data and input into the support vector machine classification model. Define a label vector for the classification model. The label vector includes multiple label values, which are used to characterize the melting power required by the melting device under any icing state level in the icing state evaluation set. Initialize a particle swarm, wherein the particles in the particle swarm are used to represent the parameter combination of the regularization parameter and the radial basis kernel parameter of the classification model; The process of optimizing the parameters of the classification model using the target icing state level is used to iteratively update the position and velocity of the particles, and the classification model is trained using the combination of the particle parameters. When the update rate and position change converge during training, or when the number of iterations reaches the maximum number of iterations, the globally optimal particle is output. Using the globally optimal particle, a binary classification support vector machine is trained for each level, and the ice melting power level is obtained using the decision function; Based on the ice-melting power level, calculate the decision values ​​of all classifiers, and determine the optimal ice-melting power level based on the decision values.

7. The method according to claim 6, characterized in that, The process of optimizing the parameters of the classification model using the target icing state level to iteratively update the position and velocity of particles, and training the classification model using the combination of particle parameters, includes: For each particle, a five-fold cross-validation mechanism is used to verify the classification performance of each support vector machine classification model; Calculate the fitness value, and update the historical best position and global best position of each particle based on the fitness value; Calculate the change in update rate and position; If the changes in the update velocity and position do not converge, and / or the number of iterations does not reach the maximum number of iterations, the process of optimizing the parameters of the classification model using the target icing state level is repeated to iteratively update the position and velocity of the particles.

8. The method according to claim 6, characterized in that, Using the globally optimal particle, a binary classification support vector machine is trained for each level, and the melting power level is obtained using the decision function, including: The globally optimal particle is used as an input sample to the trained support vector machine classifier; Calculate the Euclidean distance between the input samples; A kernel function is constructed based on the Euclidean distance and a preset width control parameter, wherein the width control parameter is used to control the width of the kernel function. Obtain the Lagrange multipliers and the bias term; A decision function is constructed based on the input sample, the Lagrange multiplier, the label vector, the kernel function, and the bias term; The required ice-melting power level is obtained using the decision function.

9. The method according to claim 6, characterized in that, Based on the ice-melting power level, calculate the decision values ​​of all classifiers, and determine the optimal ice-melting power level based on the decision values, including: Perform multi-class classification decision on the icing dominant mode matrix; Calculate the decision values ​​of all classifiers for the ice-dominant mode matrix; Obtain the ice-melting power level corresponding to the maximum decision value to generate the optimal ice-melting power level.

10. A machine learning-based ice-melting decision-making and evaluation system, characterized in that, The system includes: The data acquisition module is used to acquire multi-source monitoring data, which includes multi-source meteorological parameters and multi-source transmission line stress parameters. The icing status assessment module is used to establish a multi-source heterogeneous icing status parameter set based on the multi-source monitoring data. The multi-source heterogeneous icing status parameter set includes an icing status assessment factor set, an icing factor weight vector, an icing status evaluation set, and an icing status judgment model. The multi-source data source fusion module is used to perform weighted low-rank sparse collaborative decomposition on the multi-source heterogeneous icing state parameter set to obtain the icing dominant mode matrix. The ice-melting power level control module is used to determine the optimal ice-melting power level based on the ice-covering dominant mode matrix; the optimal ice-melting power level is the optimal hyperparameter locked by using the ice-covering dominant mode matrix and a joint optimization strategy of particle swarm cross-validation. The ice-melting parameter output module is used to calculate the ice-melting parameters according to the optimal ice-melting power level, and to set the ice-melting parameters to the ice-melting device.

Images