A cable icing thickness grading overrun early warning method based on a flexible strain sensor

Abstract

The application relates to an over-limit early warning method for cable icing thickness grading based on a flexible strain sensor and relates to the technical field of overhead cable disease monitoring. A flexible strain sensor is fixedly attached to a laying point of an overhead cable, temperature compensation is performed by considering temperature drift effects, and overhead cable strain signals are measured and corrected in real time; continuous strain signals are collected, low-frequency noise in the strain signals is removed through wavelet denoising, and ice load mutation information is reserved; an ice layer-strain coupling finite element model is used to obtain icing thickness and overhead cable strain data, an icing thickness prediction model based on PINN is constructed by taking cable mechanical relations as prior information; according to existing finite element model data, a statistical probability-based icing thickness grading standard is constructed by using a probability quantile method through random sampling, risk levels are divided according to quantile numbers, and grading over-limit early warning is realized. The flexible strain sensor can be used as a medium to effectively monitor and early warn the icing thickness of the overhead cable.

Description

Technical Field

[0001] This invention relates to the field of overhead cable defect monitoring technology, specifically a method for graded early warning of excessive cable icing thickness based on flexible strain sensors. Background Technology

[0002] Telecommunications and power are the lifelines of economic and social development. Overhead cables are highly susceptible to icing when operating in cold and high-altitude regions. When freezing rain, sleet, or supersaturated rime ice occurs, ice rapidly accumulates on the cables, significantly increasing their equivalent diameter and weight. This leads to a series of structural safety problems, such as increased mid-span deflection, increased local stress, and reduced clearance. In severe icing conditions, it can even cause cable jumpers, hardware damage, tower overturning, or entire cable sections to break, making it a key risk factor for power grid operation during winter.

[0003] However, current power grid icing monitoring methods still have significant shortcomings: traditional image recognition or laser scanning methods are easily limited by ambient light, wind and snow obstruction, and equipment installation location, making it difficult to continuously and stably output high-precision data. Monitoring methods based on mechanical response, such as vibration and tension methods, rely on changes in the overall structural state, making it difficult to perform localized and quantitative measurements of icing thickness. Furthermore, traditional sensors are generally rigid structures, which are mismatched with the deformation of flexible cables, making them prone to detachment or damage under large displacement and curvature deformation conditions, thus failing to achieve synchronous sensing of actual icing changes on the cable surface. These problems make current power grid icing monitoring technology unable to meet the engineering requirements of high flexibility, high fit, and real-time continuous sensing.

[0004] Flexible sensors, with their bendable, stretchable, and highly deformable material properties, are more suitable for curved and dynamically deformable structures like overhead cables compared to traditional rigid sensors. Once attached to the cable surface, the flexible sensor deforms synchronously with the cable, without affecting its mechanical properties. It can also sensitively detect the thickness, distribution, and trend of icing in the early stages of icing, enabling continuous and quantitative monitoring. Furthermore, flexible sensors are lightweight, environmentally resistant, and flexible in deployment, significantly improving monitoring reliability in complex climatic environments. Therefore, real-time and accurate monitoring of icing thickness on overhead cables using flexible sensors is of significant engineering importance for preventing power grid disasters, providing early warnings, and enhancing the resilience of transmission systems. Summary of the Invention

[0005] To address the shortcomings of the prior art, this invention provides a method for graded early warning of excessive icing thickness in overhead cables based on flexible strain sensors. It uses flexible strain sensors as a medium to construct an icing thickness prediction model based on PINN, and combines quantiles to classify risk levels, thereby achieving effective monitoring and early warning of icing thickness in overhead cables.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for graded early warning of excessive cable icing thickness based on a flexible strain sensor, comprising the following steps:

[0007] S1. Flexible Strain Sensor Deployment and Signal Acquisition: Flexible strain sensors are selected and fixed along the length of the overhead cable at the installation point. Temperature compensation is implemented to account for sensor temperature drift. The axial and bending strain signals of the overhead cable are measured in real time using a correction formula. To determine the actual strain of the cable, the formula is as follows: This indicates the actual strain value of the cable. This indicates the change in sensor resistance. This represents the change in sensor resistance due to the temperature drift effect. This indicates the initial value of the sensor resistance. Indicates sensor sensitivity. This represents the strain transfer coefficient of the adhesive layer. This indicates that the angle between the flexible strain sensor and the overhead cable is taken into account.

[0008] S2. Data Time-Frequency Denoising: Acquire continuous strain signals, use wavelet denoising method to remove low-frequency noise from the strain signals, retain ice load abrupt change information, and output the reconstructed and filtered signal sequence. ;

[0009] S3. Finite Element Data Acquisition and PINN Prediction Model Construction: Ice thickness and overhead cable strain data are acquired using an ice-strain coupled finite element model. Based on the cable's mechanical relationship as prior information, a PINN-based ice thickness prediction model is constructed. Specifically:

[0010] S3.1 Data Acquisition and Division: By consulting the specifications, the correlation data between ice accretion and overhead cables is obtained. The ice thickness is converted into an equivalent additional gravity load. An ice layer-strain coupled finite element model is constructed to obtain the ice thickness and strain data of overhead cables. The data is then divided into training set and validation set.

[0011] S3.2 Constructing the PINN network: Setting the input of the PINN network ,in, For the span of overhead cables, The diameter of the overhead cable. Set the output of the PINN network to the initial tension of the overhead cable. ,in, The predicted value represents the thickness of the ice layer. The parameter is Neural networks;

[0012] S3.3 Constructing a Physical Information Model: The displacement field predicted by the PINN network should satisfy the vertical force balance of the overhead cable, and a physical information model should be constructed. In the formula, The icing thickness calculated for the physical information model. The elastic modulus of overhead cable material. The cross-sectional area of ​​the overhead cable. The gravity load per unit length caused by icing;

[0013] S3.4 Constructing the Loss Function: The loss function consists of a weighted sum of data observations and physical constraints, expressed as follows: ,in, For data observation items, For physical constraints, and These are the weight parameters for the data observation term and the physical constraint term, respectively. The total number of observation points. For the PINN network in the 1st Predicted values ​​at each observation point For overhead cables in the first The true value input at each observation point Let the residual function be the function of the first observation point. For the first The residual function for each observation point, the residual function , The physical information model represents the first time. Observed values ​​at each observation point;

[0014] S4. Establish multi-level monitoring thresholds for ice thickness: Based on existing finite element model data, a statistical probability-based grading standard for ice thickness is constructed using random sampling and probability quantile method. Risk levels are divided according to quantiles to achieve graded over-limit early warning.

[0015] Furthermore, in step S1, the change in sensor resistance under the temperature drift effect is described by a linear model, i.e. In the formula, Indicates the sensor temperature drift coefficient. This indicates the temperature difference between the sensor installation time and the measurement time.

[0016] Furthermore, in step S1, the strain transfer coefficient of the adhesive layer To ensure stable adhesion, the value should be selected within the range of 0.9 to 1.0, depending on the material properties of the adhesive layer and the overhead cable.

[0017] Furthermore, step S2 specifically includes:

[0018] S2.1 Signal Discretization: The continuous strain signal is discretized to obtain a length of... The original strain discrete sequence, single sampling In the formula, Indicates the first The original values ​​of each sampling point , Indicates the first The actual strain value of the cable obtained from each sampling point;

[0019] S2.2, Sliding window smoothing: Define the length as... A sliding window is used, and the arithmetic mean is calculated for each sliding window to obtain the smoothed signal. In the formula, Indicates the first Filtering results for each sampling point Indicates the number of seconds counting backwards from the current moment. The original values ​​of each sampling point;

[0020] S2.3 Signal Reconstruction: Reconstruct the filtered output sequence, represented as follows: .

[0021] Furthermore, in step S3.2, the PINN network adopts a fully connected feedforward neural network, with 4 neurons in the input layer, 3 hidden layers, and 32 neurons in a single layer. The activation function is the tanh function, and the output layer outputs one neuron, which represents the ice thickness.

[0022] Furthermore, step S4 specifically includes:

[0023] S4.1 Obtain discrete data on icing thickness based on the output of the PINN network. Considering sample uncertainty, perform a finite number of samplings with replacement on the discrete data to form a set. , Represents discrete data in a set. Indicates the number of samples, and is based on the set. Construct its empirical distribution function , Describe the probability measure, and then define the quantile. satisfy In the formula, Indicates at the quantile level The corresponding icing thickness grading threshold, Indicates the infimum, guaranteeing This is the minimum thickness value that first reached this quantile level. The values ​​are 0.65, 0.7, 0.85, and 0.95.

[0024] S4.2. Divide the risk into five levels from low to high: normal or low risk range. Mild warning zone Moderate warning zone Severe warning zone and extreme risk range , To monitor the icing thickness output by the PINN network.

[0025] Compared with existing technologies, the beneficial effects of this invention are as follows: Based on the large deformation characteristics of overhead cables, this invention uses flexible strain sensors with strong deformation capabilities as a medium to accurately collect strain information of overhead cables under extreme conditions in real time. Low-frequency information is removed by wavelet denoising, and an ice-strain coupling model is obtained by constructing a finite element model. Considering cable mechanics as prior information, an ice thickness prediction model based on physical information neural networks is built. Based on existing finite element model data, a statistical probability-based ice thickness grading standard is constructed by random sampling and using the probability quantile method. This helps to grade and warn of ice thickness of overhead cables under extreme conditions, thereby providing effective protection for the safe operation and maintenance of overhead cables during the operation cycle. Attached Figure Description

[0026] Figure 1 This is a flowchart of the method of the present invention;

[0027] Figure 2 This is a schematic diagram of a flexible strain sensor that can be used in the method of the present invention;

[0028] Figure 3 This is a schematic diagram of the layout of the flexible strain sensor in the method of the present invention;

[0029] Figure 4 This is a schematic diagram of the continuously sampled signal in the method of the present invention;

[0030] Figure 5 This is a schematic diagram of the signal after wavelet denoising in the method of the present invention;

[0031] Figure 6 This is a schematic diagram of the ice layer-strain coupled finite element model in the method of this invention;

[0032] Figure 7 This is a schematic diagram of the PINN network architecture in the method of this invention;

[0033] Figure 8 This is a schematic diagram of the multi-level monitoring results of ice thickness in the method of the present invention. Detailed Implementation

[0034] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the invention, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0035] To address the issue of monitoring icing on overhead cables due to extreme weather conditions, this invention leverages the deformable and bendable characteristics of overhead cables. Utilizing flexible strain sensors as the sensing medium, it achieves real-time strain acquisition of the overhead cables. Low-frequency noise is removed using wavelet denoising, and a finite element model is constructed to obtain an ice-strain coupling model. Considering cable mechanics as prior information, an ice thickness prediction model based on a physical information neural network is built. Through random sampling, a probability quantile method is employed to construct a statistical probability-based ice thickness grading standard, enabling multi-level threshold construction. Risk levels are then categorized based on these thresholds to facilitate the safe operation of overhead cables.

[0036] like Figures 1-8 As shown, a method for graded early warning of excessive cable icing thickness based on flexible strain sensors is presented, and its process combines... Figure 1 As shown, it includes the following steps:

[0037] S1. Deployment and signal acquisition of flexible strain sensors;

[0038] Select flexible strain sensors, combined with Figure 2 As shown, it is pasted and fixed along its length to the surface of the overhead cable laying point, combined with... Figure 3 As shown, temperature compensation is performed considering the sensor's temperature drift effect, and the axial and bending strain signals of the overhead cable are measured in real time. Specifically:

[0039] Considering the angle between the flexible strain sensor and the overhead cable Then we have:

[0040] (1)

[0041] Considering the strain transfer coefficient of the adhesive layer caused by bonding, we have:

[0042] (2)

[0043] Based on the relationship between the resistance change and strain of the flexible strain sensor, the strain measurement value of the sensor is calculated, as follows:

[0044] (3)

[0045] Considering the sensor temperature drift effect, and using a linear model to describe the resistance baseline change, we have:

[0046] (4)

[0047] Based on the above corrections, the actual value of cable strain is expressed as:

[0048] (5)

[0049] In the formula, This indicates the strain measurement value of the sensor. This indicates the actual strain value of the cable. This represents the strain transfer coefficient of the adhesive layer. It is taken within the range of 0.9 to 1.0, depending on the material properties of the adhesive layer and the overhead cable, to ensure stable bonding. This indicates the change in sensor resistance. This indicates the initial value of the sensor resistance. Indicates sensor sensitivity. This represents the change in sensor resistance due to the temperature drift effect. Indicates the sensor temperature drift coefficient. This indicates the temperature difference between the sensor installation time and the measurement time.

[0050] S2, Data Time-Frequency Denoising;

[0051] Collect continuous strain signals, combined with Figure 4 As shown, time-frequency denoising processing was performed on the strain signal, and wavelet denoising was used to remove low-frequency noise (mainly temperature interference) from the strain signal. Figure 5 As shown, information on abrupt changes in ice load is preserved, specifically:

[0052] S2.1 Signal Discretization

[0053] Continuous sampling is performed using a flexible strain sensor. The continuous strain signal measured by the flexible strain sensor is then discretized to obtain a length of... The original discrete strain sequence, where the original value of a single sample is represented as:

[0054] (6)

[0055] In the formula, Indicates the first The original values ​​of each sampling point , Indicates the first The actual strain value of the cable obtained from each sampling point;

[0056] S2.2, Smooth Sliding Window

[0057] For the original strain discrete sequence, a length of is defined. Using a sliding window, the arithmetic mean is calculated for each sliding window, resulting in the smoothed signal represented as:

[0058] (7)

[0059] In the formula, Indicates the first Filtering results for each sampling point Indicates the number of seconds counting backwards from the current moment. The original values ​​of each sampling point;

[0060] S2.3 Signal Reconstruction

[0061] The reconstructed filtered output sequence is represented as follows:

[0062] (8)

[0063] In the formula, This represents the reconstructed and filtered signal sequence.

[0064] S3. Finite element method for data acquisition and PINN prediction model construction;

[0065] Ice thickness and overhead cable strain data were obtained using an ice-strain coupled finite element model. Using the cable's mechanical relationship as prior information, an ice thickness prediction model based on a Physical Information Neural Network (PINN) was constructed. Specifically:

[0066] S3.1 Data Acquisition and Division

[0067] By consulting relevant standards (such as the "Technical Specification for Design of Overhead Transmission Lines with Heavy Ice Covering"), the correlation data between icing and overhead cables was obtained. The icing thickness was then converted into an equivalent additional gravity load, and an ice-strain coupled finite element model was constructed. Figure 6 As shown, the ice thickness and strain data of the overhead cable were obtained, and the data were divided into training set and validation set according to the ratio (e.g., 8:2).

[0068] S3.2, Constructing the PINN Network

[0069] Configure the inputs to the PINN network as follows:

[0070] (9)

[0071] In the formula, For the span of overhead cables, The diameter of the overhead cable. The initial tension of the overhead cable. The weight per unit length of overhead cable, The design parameters for the mid-span sag are determined by the design parameters of the overhead cable.

[0072] The output of the PINN network is configured as follows:

[0073] (10)

[0074] In the formula, The predicted value represents the thickness of the ice layer. The parameter is The neural network.

[0075] The PINN network employs a fully connected feedforward neural network, combined with Figure 7 As shown, the input layer has 4 neurons, the hidden layer has 3 layers, the number of neurons in a single layer is 32, the activation function is the tanh function, and the output layer outputs one neuron (ice thickness).

[0076] S3.3 Constructing a Physical Information Model

[0077] The displacement field predicted by the PINN network should satisfy the vertical force balance of the overhead cable. The physical information model is constructed as follows:

[0078] (11)

[0079] In the formula, The icing thickness calculated for the physical information model. The elastic modulus of overhead cable material. The cross-sectional area of ​​the overhead cable. This refers to the gravity load per unit length caused by ice accumulation.

[0080] S3.4 Constructing the loss function

[0081] The loss function consists of a weighted sum of data observations and physical constraints, where:

[0082] The data observations are represented as follows:

[0083] (12)

[0084] In the formula, For data observation items, The total number of observation points. For the PINN network in the 1st Predicted values ​​at each observation point For overhead cables in the first The actual values ​​input at each observation point.

[0085] The residual function is defined as follows:

[0086] (13)

[0087] In the formula, Represents the residual function. The physical information model represents the first time. The observed values ​​at each observation point.

[0088] The physical constraint term is derived from the residual function and is expressed as follows:

[0089] (14)

[0090] In the formula, For physical constraints, Let the residual function be the function of the first observation point. For the first The residual function of each observation point.

[0091] The loss function is then expressed as follows:

[0092] (15)

[0093] In the formula, For loss function, and These are the weight parameters for the data observation term and the physical constraint term, respectively.

[0094] S4. Establish multi-level monitoring thresholds for ice cover thickness under extreme freezing rain disaster conditions, combined with... Figure 8 As shown, based on existing finite element model data, a statistical probability-based grading standard for icing thickness is constructed using random sampling and the probability quantile method. Specifically:

[0095] S4.1 Obtain discrete data on icing thickness based on the output of the PINN network. Considering sample uncertainty, perform a finite number of samplings with replacement on the discrete data (sampling 10% of the data, not exceeding 2000 samples) to form a set. , means as follows:

[0096] (16)

[0097] In the formula, Represents discrete data in a set. Indicates the number of samples.

[0098] Based on sets Its empirical distribution function is constructed as follows:

[0099] (17)

[0100] In the formula, It represents a probability measure.

[0101] Define quantiles satisfy:

[0102] (18)

[0103] In the formula, Indicates at the quantile level The corresponding icing thickness grading threshold, Indicates the infimum, guaranteeing This is the minimum thickness value that first reached this quantile level. The values ​​are 0.65, 0.7, 0.85 and 0.95.

[0104] S4.2 Based on the four values ​​of this quantile, the risk caused by icing thickness is divided into five levels from low to high: normal or low risk range. Mild warning zone Moderate warning zone Severe warning zone and extreme risk range , To monitor the icing thickness output by the PINN network.

[0105] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0106] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for graded early warning of excessive cable icing thickness based on flexible strain sensors, characterized in that: Includes the following steps: S1. Flexible Strain Sensor Deployment and Signal Acquisition: Flexible strain sensors are selected and fixed along the length of the overhead cable at the installation point. Temperature compensation is implemented to account for sensor temperature drift. The axial and bending strain signals of the overhead cable are measured in real time using a correction formula. To determine the actual strain of the cable, the formula is as follows: This indicates the actual value of cable strain. This indicates the change in sensor resistance. This represents the change in sensor resistance due to the temperature drift effect. This indicates the initial value of the sensor resistance. Indicates sensor sensitivity. This represents the strain transfer coefficient of the adhesive layer. This indicates that the angle between the flexible strain sensor and the overhead cable is taken into account. S2. Data Time-Frequency Denoising: Acquire continuous strain signals, use wavelet denoising method to remove low-frequency noise from the strain signals, retain ice load abrupt change information, and output the reconstructed and filtered signal sequence. ; S3. Finite Element Data Acquisition and PINN Prediction Model Construction: Ice thickness and overhead cable strain data are acquired using an ice-strain coupled finite element model. Based on the cable's mechanical relationship as prior information, a PINN-based ice thickness prediction model is constructed. Specifically: S3.1 Data Acquisition and Division: Obtain the correlation data between ice accretion and overhead cables by consulting the specifications, convert the ice thickness into an equivalent additional gravity load, construct an ice layer-strain coupled finite element model, obtain the ice thickness and strain data of overhead cables, and divide the data into training set and validation set; S3.2 Constructing the PINN network: Setting the input of the PINN network ,in, For the span of overhead cables, The diameter of the overhead cable. Set the output of the PINN network to the initial tension of the overhead cable. ,in, The predicted value represents the thickness of the ice layer. The parameter is Neural networks; S3.3 Constructing a Physical Information Model: The displacement field predicted by the PINN network should satisfy the vertical force balance of the overhead cable, and a physical information model should be constructed. In the formula, The icing thickness calculated for the physical information model. The elastic modulus of overhead cable material. The cross-sectional area of ​​the overhead cable. The gravity load per unit length caused by icing; S3.4 Constructing the Loss Function: The loss function consists of a weighted sum of data observations and physical constraints, expressed as follows: ,in, For data observation items, For physical constraints, and These are the weight parameters for the data observation term and the physical constraint term, respectively. The total number of observation points. For the PINN network in the 1st Predicted values ​​at each observation point For overhead cables in the first The true value input at each observation point Let the residual function be the function of the first observation point. For the first The residual function at each observation point, the residual function , The physical information model represents the first time. Observed values ​​at each observation point; S4. Establish multi-level monitoring thresholds for icing thickness: Based on existing finite element model data, a statistical probability-based icing thickness classification standard is constructed using random sampling and probability quantile method. Risk levels are divided according to quantiles to achieve graded over-limit early warning.

2. The method for graded early warning of excessive cable icing thickness based on a flexible strain sensor according to claim 1, characterized in that: In step S1, the change in sensor resistance under the temperature drift effect is described by a linear model, i.e. In the formula, Indicates the sensor temperature drift coefficient. This indicates the temperature difference between the sensor installation time and the measurement time.

3. The method for graded early warning of excessive cable icing thickness based on a flexible strain sensor according to claim 1, characterized in that: In step S1, the strain transfer coefficient of the adhesive layer To ensure stable adhesion, the value should be selected within the range of 0.9 to 1.0, depending on the material properties of the adhesive layer and the overhead cable.

4. The method for graded early warning of excessive cable icing thickness based on a flexible strain sensor according to claim 1, characterized in that: Step S2 specifically includes: S2.1 Signal Discretization: The continuous strain signal is discretized to obtain a length of... The original strain discrete sequence, single sampling In the formula, Indicates the first The original values ​​of each sampling point , Indicates the first The actual strain value of the cable obtained from each sampling point; S2.2, Sliding window smoothing: Define the length as... A sliding window is used, and the arithmetic mean is calculated for each sliding window to obtain the smoothed signal. In the formula, Indicates the first Filtering results for each sampling point Indicates the number of seconds counting backwards from the current moment. The original values ​​of each sampling point; S2.3 Signal Reconstruction: Reconstruct the filtered output sequence, represented as follows: .

5. The method for graded early warning of excessive cable icing thickness based on a flexible strain sensor according to claim 1, characterized in that: In step S3.2, the PINN network adopts a fully connected feedforward neural network, with 4 neurons in the input layer, 3 hidden layers, and 32 neurons in a single layer. The activation function is the tanh function, and the output layer outputs one neuron, which represents the ice thickness.

6. The method for graded early warning of excessive cable icing thickness based on a flexible strain sensor according to claim 1, characterized in that: Step S4 specifically includes: S4.1 Obtain discrete data on icing thickness based on the output of the PINN network. Considering sample uncertainty, perform a finite number of samplings with replacement on the discrete data to form a set. , Represents discrete data in a set. Indicates the number of samples, and is based on the set. Construct its empirical distribution function , Describe the probability measure, and then define the quantile. satisfy In the formula, Indicates at the quantile level The corresponding icing thickness grading threshold, Indicates the infimum, guaranteeing This is the minimum thickness value that first reached this quantile level. The values ​​are 0.65, 0.7, 0.85, and 0.

95. S4.

2. Divide the risk into five levels from low to high: normal or low risk range. Mild warning zone Moderate warning zone Severe warning zone and extreme risk range , To monitor the icing thickness output by the PINN network.

Images