An overhead ground wire ice thickness monitoring device and method

Abstract

The application discloses an overhead ground wire ice thickness monitoring device and method, which comprises the following steps: plastic cover plates and bottom plates are arranged at two ends of a cylindrical metal shell, metal plates are arranged in the cylindrical metal shell, the metal plates and the cylindrical metal shell form an electrode, and equivalent capacitances are formed with three-phase transmission lines; a clamping groove is arranged through the centers of the plastic cover plates, the metal plates and the bottom plates, an overhead ground wire is arranged in the clamping groove, and the overhead ground wire and the metal plates are kept equipotential through conductive connection; an electric field array probe module is arranged on the metal plates below the clamping groove, the electric field array probe module is used for collecting power frequency electric field intensity signals below the overhead ground wire, a monitoring mainboard converts the collected power frequency electric field intensity signals into electric field intensity amplitudes, and the ice thickness of the overhead ground wire is predicted through a DWT-CNN-ELM algorithm. The electric field coupling principle is used to predict the ice thickness of the overhead ground wire, the sensor does not need to be powered off when being installed on the overhead ground wire, and the on-site implementation and the later maintenance are facilitated.

Description

Technical Field

[0001] This invention relates to the field of power system anti-icing and disaster reduction technology, specifically to a device and method for monitoring the ice thickness of overhead ground wires. Background Technology

[0002] The problem of icing on overhead power lines persists. Untimely icing control can lead to serious disasters such as line breaks and tower collapses, jeopardizing the safe and stable operation of the power grid and causing significant economic losses. Numerous studies have shown that most power outages caused by ice storms are due to excessively heavy icing on overhead conductors, pulling towers to the ground and reducing the power supply to meet demand-side loads. The ground wire of an overhead line is located at the top of the tower, where the surrounding electric field is weak, and it does not generate Joule heat from current. Therefore, the ground wire is more prone to icing, and the damage caused by ground wire icing is even more severe. Ground wire icing can reduce the distance between the ground wire and the conductor. Under the influence of wind deflection and other factors, the ground wire and conductor may collide and discharge, causing phase-to-phase or ground-to-ground short circuits. During asynchronous icing, the jumping of the ground wire can also result in insufficient electrical clearance between the ground wire and the conductor, causing short circuits. Excessive icing increases the tension on the ground wire, potentially leading to strand breakage or wire breakage. Meanwhile, the galloping or vibration of the ground wire caused by icing will subject the fittings to repeated impact and friction, resulting in wear, deformation, and breakage, such as loosening or falling off of clamps and cracking of insulators, affecting the connection and insulation performance of the line. Therefore, effective monitoring of the icing thickness of the ground wire is crucial.

[0003] Extensive research has been conducted both domestically and internationally on icing monitoring of overhead lines, resulting in several typical sensing technologies for icing thickness monitoring. However, these technologies primarily focus on measuring the icing thickness of conductors, with few applications involving overhead ground wires, and each has its own shortcomings.

[0004] The rotating cylindrical conductor volumetric ice sensing method simulates wire icing by rotating multiple cylindrical conductors, measuring the tensile force changes of each conductor after icing, and calculating the equivalent ice thickness of the wire through fitting. This method requires the additional installation of a multi-conductor volumetric ice generator, which needs to be rotated, resulting in a complex structure and high cost. In high-humidity weather conditions in areas with medium to heavy icing, the mechanical structure is prone to aging and damage. Under extremely harsh icing conditions, the rotating conductors may freeze and fail.

[0005] The conductor tension sensor method replaces the ball-head ring of the insulator with a tension sensor. By measuring the change in tension on the suspension insulator string, combined with the tilt angle and wind deflection angle of the suspension insulator string measured by an angle sensor, and meteorological parameters such as wind speed and direction measured by a meteorological sensor, the icing condition of the conductor is calculated based on mechanical principles and relevant formulas. This method can obtain real-time changes in conductor load. However, the tension sensor is under heavy load for a long time, and the measured values ​​may be offset, resulting in a gradual decrease in accuracy over the years. Furthermore, it is difficult to repair and replace because installation and maintenance require power outages and removal of the insulator string.

[0006] The simulated conductor monitoring method involves suspending a simulated conductor of the same type and material as the actual power line conductor on the transmission tower. This simulated conductor is subjected to the same icing conditions as the transmission line conductor. By measuring the weight change of the simulated conductor using a tension sensor within the monitoring unit, the ice thickness is determined, thus allowing for an accurate estimation of the overall load and ice thickness of the transmission line conductor. This method requires an additional simulated conductor and a built-in tension sensor. However, the tension sensor is under constant heavy load, which can lead to inaccuracies in the measured values ​​and a gradual decline in accuracy over time. Furthermore, under severe icing conditions, both the simulated conductor and the tension sensor are prone to freezing and failure.

[0007] The ice dielectric capacitance sensing method involves installing three equally spaced parallel electrodes on a conductor and applying an external high-frequency excitation source to measure the ice layer capacitance, thereby reflecting the ice thickness. This method requires an insulating layer at the location where the monitoring unit is fixed on the conductor and necessitates an additional external power supply, making installation cumbersome and unstable.

[0008] The aforementioned ice thickness monitoring technologies still suffer from common problems such as complex device structure, susceptibility to failure in extreme environments, need for power outages for installation, and inconvenience in subsequent maintenance. Summary of the Invention

[0009] To address the problems existing in the prior art, this invention provides a device and method for monitoring the ice thickness of overhead ground wires. It uses the principle of electric field coupling to predict the ice thickness of overhead ground wires. Installing sensors on overhead ground wires does not require power outages, which facilitates on-site implementation and subsequent maintenance.

[0010] To achieve the above technical objectives, the present invention adopts the following technical solution:

[0011] An overhead ground wire icing thickness monitoring device includes: a cylindrical metal shell, metal electrodes, an electric field array probe module, a monitoring main board, a lithium battery, and a slot. The cylindrical metal shell has a plastic cover plate and a bottom plate at its two ends, respectively. The metal electrodes are located inside the cylindrical metal shell, forming an electrode with the cylindrical metal shell and creating equivalent capacitances with the three-phase power transmission lines. The slot passes through the center of the plastic cover plate, the metal electrodes, and the bottom plate. An overhead ground wire is installed in the slot, and the overhead ground wire and the metal electrodes are electrically connected to maintain equipotential. An electric field array probe module is located on the metal electrodes below the slot. This module is used to collect the power frequency electric field strength signal below the overhead ground wire. The monitoring main board converts the collected power frequency electric field strength signal into an electric field strength amplitude, and uses the electric field strength amplitude to predict the icing thickness of the overhead ground wire using the DWT-CNN-ELM algorithm. The lithium battery powers the monitoring main board.

[0012] Furthermore, a temperature and humidity sensor is embedded in the base plate. The temperature and humidity sensor is used to collect the temperature and humidity information of the environment at regular intervals and transmit the collected temperature and humidity information to the monitoring motherboard. The monitoring motherboard is used to determine whether the temperature and humidity information meets the icing conditions. If so, the monitoring motherboard wakes up the power frequency electric field strength signal acquisition of the electric field array probe module. Otherwise, the electric field array probe module does not work.

[0013] Furthermore, the monitoring motherboard includes: an amplification and filtering module, a wireless transmission module, an MCU module, an inverter module, and an icing thickness prediction module;

[0014] The electric field array probe module is used to convert the power frequency electric field intensity signal collected below the overhead ground wire into a voltage signal and transmit it to the amplification and filtering module;

[0015] The amplification and filtering module is used to amplify the voltage signal, filter out interference, and transmit it to the ADC port of the MCU module.

[0016] The MCU module is used to convert the voltage signal acquired by the ADC port into the electric field strength amplitude.

[0017] The ice thickness prediction module is used to predict the ice thickness of the overhead ground wire by using the DWT-CNN-ELM algorithm to calculate the electric field strength amplitude.

[0018] The wireless transparent transmission module is used to transmit the predicted icing thickness of the overhead ground wire to the remote user terminal.

[0019] The inverter module is connected to the lithium battery and is used to convert the DC power of the lithium battery to power the amplification and filtering module, the wireless transmission module, the MCU module, and the icing thickness prediction module.

[0020] Furthermore, the specific process of predicting the icing thickness of the overhead ground wire using the DWT-CNN-ELM algorithm based on the electric field intensity amplitude is as follows:

[0021] i: The electric field intensity amplitude is subjected to piecewise cubic Hermite interpolation, wavelet decomposition, wavelet denoising and wavelet reconstruction through discrete wavelet transform fitting algorithm to generate a continuous electric field amplitude distribution curve.

[0022] ii: Based on the generated electric field amplitude distribution curve, select indicators that can characterize the differences in electric field amplitude distribution curves, and determine the indicators related to ice thickness based on the selected indicators;

[0023] iii: Input the indicators related to ice thickness into the convolutional neural network for feature extraction and output the feature matrix;

[0024] iv: Input the feature matrix into the extreme learning machine to predict the icing thickness of the overhead ground wire.

[0025] Furthermore, the specific process for determining the indicators related to ice thickness based on the screened indicators is as follows: determine the electric field amplitude distribution curve under the actual ice thickness, screen out the indicators that can characterize the differences in the electric field amplitude distribution curve, calculate the Pearson correlation coefficient between the screened indicators and the actual ice thickness, and take the indicators with an absolute value of Pearson correlation coefficient greater than 0.95 as the indicators related to ice thickness.

[0026] Furthermore, the convolutional neural network includes: an input layer, a convolutional layer, a ReLU activation layer, an average pooling layer, and a fully connected layer;

[0027] The input layer is used to normalize the indicators related to ice thickness;

[0028] The convolutional layer is used to extract local features from the normalized metrics and generate a multi-channel feature map.

[0029] The ReLU activation layer is used to introduce a non-linear activation function into the multi-channel feature map to obtain a feature map;

[0030] The average pooling layer downsamples the feature map to generate a pooled feature map.

[0031] The fully connected layer is used to flatten the pooled feature map and extract features through linear transformation and nonlinear activation, outputting a feature matrix.

[0032] Furthermore, the size of the convolutional kernel of the convolutional layer is set to... The step size is set to 1, where, This represents a number of indicators related to icing thickness. This indicates rounding down to the nearest integer.

[0033] Furthermore, the prediction process for the icing thickness of the overhead ground wire is as follows:

[0034]

[0035] in, Indicates the icing thickness of overhead ground wires. This represents the number of hidden layer neurons in the Extreme Learning Machine. express index, Indicates the first hidden layer The weights between each neuron node and the output layer of the extreme learning machine Indicates the first hidden layer Each neuron node and the weight vector of the input layer of the extreme learning machine Indicates the first hidden layer Each neuron node and the unit bias vector of the input layer of the extreme learning machine Represents the characteristic matrix, This represents the activation function. This indicates the transpose operation.

[0036] Furthermore, the extreme learning machine needs to be trained before it can predict the icing thickness of overhead ground wires. The process is as follows:

[0037] Using the actual icing thickness of overhead ground wires as the label of the Extreme Learning Machine (ELM), and the corresponding feature matrix as the input of the ELM, the icing thickness is predicted. If the residual norm between the predicted icing thickness and the actual icing thickness is greater than a set threshold, the weights between the hidden and output layers of the ELM need to be optimized based on the hidden layer output matrix and the actual icing thickness.

[0038]

[0039] in, This represents the weights between the hidden and output layers of the optimized Extreme Learning Machine. Indicates the actual icing thickness. Represents the hidden layer output matrix of the Extreme Learning Machine The generalized inverse matrix, , This indicates the number of samples used for training. This represents the number of hidden layer neurons in the Extreme Learning Machine.

[0040] Furthermore, the present invention also provides a method for monitoring the icing thickness of overhead ground wires, comprising the following steps:

[0041] Step S1: Collect temperature and humidity information of the environment where the overhead ground wire is located at regular intervals. Only when the collected temperature and humidity information meets the icing conditions, collect the power frequency electric field intensity signal below the overhead ground wire and convert the power frequency electric field intensity signal into electric field intensity amplitude.

[0042] Step S2: Generate a continuous electric field amplitude distribution curve by fitting the electric field intensity amplitude using a discrete wavelet transform algorithm;

[0043] Step S3: Based on the generated electric field amplitude distribution curve, screen out the indicators that can characterize the differences in the electric field amplitude distribution curve, and determine the indicators related to the ice thickness based on the screened indicators.

[0044] Step S4: Input the indicators related to ice thickness into the convolutional neural network for feature extraction and output the feature matrix;

[0045] Step S5: Input the feature matrix into the extreme learning machine to predict the icing thickness of the overhead ground wire.

[0046] Compared with the prior art, the present invention has the following advantages: The overhead ground wire icing thickness monitoring device and method of the present invention adopts a cylindrical metal shell, which allows the overhead ground wire icing monitoring device to have the same aerodynamic structure as the overhead ground wire, and basically eliminates the influence of edge effect on the icing thickness and icing shape on both sides of the overhead ground wire icing monitoring device; the electric field array probe module is set inside the cylindrical metal shell, so even if the outside of the overhead ground wire icing monitoring device is iced, it will not affect the detection of electric field intensity amplitude, avoiding failure of the icing thickness monitoring device due to severe icing; the DWT-CNN-ELM algorithm is used to quickly and accurately predict the icing thickness of the overhead ground wire through the electric field coupling principle, thereby reflecting the degree of icing of the transmission line. The installation of the icing thickness monitoring device on the overhead ground wire does not require power outage, which is convenient for on-site implementation and subsequent maintenance. Compared with the existing methods, there are no problems such as mechanical fatigue and stress failure, which can ensure the stability and reliability of long-term monitoring. Attached Figure Description

[0047] Figure 1 This is a cross-sectional view of the overall structure of the overhead ground wire ice thickness monitoring device of the present invention;

[0048] Figure 2 This is a schematic diagram of the monitoring motherboard in this invention;

[0049] Figure 3 This is a schematic diagram illustrating the relevant theoretical principles of the present invention;

[0050] Figure 4 This is a flowchart illustrating the prediction of icing thickness of overhead ground wires using the DWT-CNN-ELM algorithm in this invention.

[0051] Figure 5This is a schematic diagram of the electric field amplitude distribution curves under different ice thicknesses. Detailed Implementation

[0052] The technical solution of the present invention will be further explained and described below with reference to the accompanying drawings.

[0053] like Figure 1 This is a cross-sectional view of the overall structure of the overhead ground wire icing thickness monitoring device of the present invention. The device includes: a cylindrical metal shell 1, a plastic cover plate 2, a metal electrode plate 3, an electric field array probe module 4, a monitoring main board 5, a base plate 6, a lithium battery 8, and a card slot 9. The cylindrical metal shell 1 serves as a shield to protect the internal circuitry. Simultaneously, the cylindrical metal shell 1 allows the overhead ground wire icing monitoring device to have the same aerodynamic structure as the overhead ground wire, and essentially eliminates the influence of edge effects on the icing thickness and shape on both sides of the device. The cylindrical metal shell 1 has a plastic cover plate 2 and a base plate 6 at its two ends, respectively. The outer surface of the plastic cover plate 2 is coated with a superhydrophobic coating to prevent ice accumulation. The cylindrical metal shell 1 contains a metal electrode plate 3, which is integrated with the cylindrical metal shell 1. An electrode is formed, which, together with the three-phase power transmission line, forms an equivalent capacitance, causing the electric field lines generated by the electric field strength to converge on the surface of the metal electrode plate 3. The slot 9 passes through the center of the plastic cover plate 2, the metal electrode plate 3, and the base plate 6. An overhead ground wire is laid in the slot 9, and the overhead ground wire and the metal electrode plate 3 are electrically connected to maintain equipotential. An electric field array probe module 4 is provided on the metal electrode plate 3 located below the slot 9. The electric field array probe module 4 consists of several spatial electric field probes arranged vertically at equal intervals, which are used to collect the power frequency electric field strength signal below the overhead ground wire. The distance between the plastic cover plate 2 and the metal electrode plate 3 is greater than the length of the spatial electric field probe. The monitoring main board 5 converts the collected power frequency electric field strength signal into an electric field strength amplitude, and uses the electric field strength amplitude to predict the icing thickness of the overhead ground wire using the DWT-CNN-ELM algorithm. The lithium battery 8 supplies power to the monitoring main board 5.

[0054] This invention houses the electric field array probe module 4 within a cylindrical metal casing 1. Even if the overhead ground wire icing monitoring device is externally covered with ice, it will not affect the detection of the electric field strength amplitude, thus avoiding failure of the icing thickness monitoring device due to severe icing. By using the electric field coupling principle and the DWT-CNN-ELM algorithm, the icing thickness of the overhead ground wire is predicted to reflect the degree of icing on the transmission line. Installing the icing thickness monitoring device on the overhead ground wire does not require power outages, facilitating on-site implementation and subsequent maintenance. Compared with existing methods, it avoids problems such as mechanical fatigue and stress failure, ensuring long-term stable and reliable monitoring.

[0055] A temperature and humidity sensor 7 is embedded in the base plate 6. The temperature and humidity sensor 7 is used to collect ambient temperature and humidity information at regular intervals and transmit the collected information to the monitoring main board 5. The monitoring main board 5 is used to determine whether the temperature and humidity information meets the icing conditions. If so, the monitoring main board 5 activates the electric field array probe module 4 to collect the power frequency electric field strength signal. Otherwise, the electric field array probe module 4 does not work, thus ensuring safe operation in a low-power mode without considering the problem of failure under extreme conditions. This meets the corresponding requirements of the current digital and intelligent operation and maintenance of the power system and the "strong" smart grid for ice prevention and disaster reduction. Through intelligent monitoring of the ice thickness of the overhead ground wire, this device can be combined with the existing ice melting device in the future. When the ice thickness exceeds a certain value, the ice melting device will be activated to remove the ice from the line.

[0056] like Figure 2 The monitoring motherboard 5 in this invention includes: an amplification and filtering module, a wireless transmission module, an MCU module, an inverter module, and an icing thickness prediction module;

[0057] The electric field array probe module 4 converts the power frequency electric field intensity signal collected below the overhead ground wire into a millivolt-level voltage signal and transmits it to the amplification and filtering module;

[0058] The amplification and filtering module is used to amplify the voltage signal, filter out interference, and transmit it to the ADC port of the MCU module;

[0059] The MCU module converts the voltage signal acquired by the ADC port into the electric field strength amplitude;

[0060] The icing thickness prediction module uses the DWT-CNN-ELM algorithm to predict the icing thickness of the overhead ground wire based on the electric field intensity amplitude.

[0061] The wireless pass-through module is used to transmit the predicted icing thickness of the overhead ground wire to a remote user terminal.

[0062] The inverter module is connected to the lithium battery 8 and converts the DC power from the lithium battery 8 to power the amplification and filtering module, the wireless transmission module, the MCU module, and the icing thickness prediction module.

[0063] like Figure 3 The overhead ground wire icing thickness monitoring device of the present invention can be stably installed on the overhead ground wire. The overhead ground wire plays a role in introducing lightning current into the ground in the power system; therefore, under normal circumstances, no current flows through the conductor, and the conductor potential is zero. Since the metal plate 3 of the overhead ground wire icing thickness monitoring device is made of metal, the device is equivalent to grounding. During normal operation of the transmission line, the overhead conductor 10 is suspended from the crossarm by the insulator string 11, and three-phase alternating current flows through it. During the power transmission process of the overhead conductor 10, the three-phase transmission conductors will respectively form coupling capacitances with the overhead ground wire. , ,in, This represents the equivalent dielectric constant between the transmission conductor and the overhead ground wire. This represents the equivalent area of ​​the virtual plates of the coupling capacitor. This represents the distance between the two virtual plates of the coupling capacitor. Since both are constants, the coupling capacitor... With equivalent dielectric constant It is directly proportional. According to relevant electromagnetic field theories, the electric field distribution around an overhead ground wire is proportional to... The magnitude of the coupling capacitance is closely related to the electric field distribution around the overhead ground wire. When there is ice on the overhead ground wire, the dielectric constant of the ice is 3.2, which is greater than that of air. Therefore, the presence of ice will change the magnitude of the coupling capacitance. The thicker the ice, the greater the change in the value of the coupling capacitance, and the more severe the distortion of the measured electric field strength amplitude. This provides a rationale for using the electric field distribution method to monitor the ice thickness.

[0064] In one technical solution of the present invention, such as Figure 4 The specific process of predicting the icing thickness of overhead ground wires using the DWT-CNN-ELM algorithm based on the electric field strength amplitude is as follows:

[0065] i: The electric field intensity amplitude is subjected to piecewise cubic Hermite interpolation, wavelet decomposition, wavelet denoising, and wavelet reconstruction through the Discrete Wavelet Transform (DWT) fitting algorithm to generate a continuous electric field amplitude distribution curve. The DWT fitting algorithm can decompose the nonlinear and non-stationary electric field intensity amplitude into high-frequency and low-frequency components. The wavelet denoising and reconstruction steps reduce noise from other interferences and improve the smoothness of the electric field amplitude distribution curve, thus achieving accurate fitting of the nonlinear and non-stationary electric field.

[0066] ii: Based on the generated electric field amplitude distribution curves, select indicators that can characterize the differences in electric field amplitude distribution curves, and determine the indicators related to icing thickness based on the selected indicators. Specifically:

[0067] The electric field amplitude distribution curve was determined under the actual icing thickness, and indicators that could characterize the differences in the electric field amplitude distribution curve were selected. The Pearson correlation coefficient between the selected indicators and the actual icing thickness was calculated.

[0068]

[0069] in, Indicates the first filter The Pearson correlation coefficients of these indicators with the actual icing thickness The sample size representing the actual icing thickness. express index, Indicates the first The first sample The size of each indicator, Indicates the first The average of the values ​​of each indicator. Indicates the first The actual icing thickness of each sample This represents the mean true ice thickness for all samples.

[0070] The range of values ​​for the Pearson correlation coefficient is as follows: ,like , indicating the first j There is a strong correlation between these indicators and icing thickness. , indicating the first j Several indicators showed a correlation with icing thickness. Indicators with an absolute Pearson correlation coefficient exceeding 0.95 were considered to be correlated with icing thickness.

[0071] iii: Input the indicators related to ice thickness into the convolutional neural network (CNN) for feature extraction and output the feature matrix. Through hierarchical feature extraction mode and multi-channel feature mapping method, multi-dimensional digital features can be extracted from the indicators related to ice thickness in a short time, and the generalization ability of the CNN can be greatly improved. Specifically, the convolutional neural network includes: input layer, convolutional layer, ReLU activation layer, average pooling layer and fully connected layer.

[0072] The input layer is used to normalize the metrics related to ice thickness and remove the dimensions.

[0073] Convolutional layers are used to extract local features from normalized metrics and generate multi-channel feature maps. In one embodiment of this invention, the size of the convolutional kernel is set to... The step size is set to 1, where, This represents a number of indicators related to icing thickness. This indicates rounding down to the nearest integer.

[0074] ReLU activation layers are used to introduce non-linear activation functions into multi-channel feature maps to enhance the non-linear expression and expansion capabilities of convolutional layers, alleviate the gradient vanishing problem, and obtain feature maps.

[0075] The average pooling layer downsamples the feature map to generate a pooled feature map.

[0076] Fully connected layers are used to flatten the pooled feature maps and extract features through linear transformations and nonlinear activations, outputting a feature matrix.

[0077] iv: Input the feature matrix into the Extreme Learning Machine (ELM) to predict the icing thickness of the overhead ground wire. Using the Extreme Learning Machine (ELM) simplifies the complex parameter tuning process of traditional neural networks, greatly enhances operability, and enables accurate prediction of the icing thickness of the overhead ground wire.

[0078] The prediction process for the icing thickness of overhead ground wires in this invention is as follows:

[0079]

[0080] in, Indicates the icing thickness of overhead ground wires. This represents the number of hidden layer neurons in the Extreme Learning Machine. express index, Indicates the first hidden layer The weights between each neuron node and the output layer of the extreme learning machine Indicates the first hidden layer Each neuron node and the weight vector of the input layer of the extreme learning machine Indicates the first hidden layer Each neuron node and the unit bias vector of the input layer of the extreme learning machine Represents the characteristic matrix, This represents the activation function. This indicates the transpose operation.

[0081] In one technical solution of the present invention, the Extreme Learning Machine is a feedforward neural network with strong generalization ability, which respectively trains the first hidden layer... Each neuron node and the weight vector of the input layer of the extreme learning machine With the hidden layer Each neuron node and the unit bias vector of the input layer of the extreme learning machine To make the appropriate settings for the Extreme Learning Machine, and Once set, its value will not change. Then, it is necessary to adjust the value of the hidden layer. The weights between each neuron node and the output layer of the extreme learning machine The initial settings are configured and then updated and adjusted during training. The specific training process is as follows:

[0082] Using the actual icing thickness of overhead ground wires as the label of the Extreme Learning Machine (ELM), and the corresponding feature matrix as the input of the ELM, the icing thickness is predicted. If the residual norm between the predicted icing thickness and the actual icing thickness is greater than a set threshold, the weights between the hidden and output layers of the ELM need to be optimized using the least squares method based on the hidden layer output matrix and the actual icing thickness.

[0083]

[0084] in, This represents the weights between the hidden and output layers of the optimized Extreme Learning Machine. Indicates the actual icing thickness. Represents the hidden layer output matrix of the Extreme Learning Machine The generalized inverse matrix, , This indicates the number of samples used for training. This represents the number of hidden layer neurons in the Extreme Learning Machine.

[0085] In one technical solution of the present invention, a method for monitoring the icing thickness of overhead ground wires is also provided, comprising the following steps:

[0086] Step S1: Collect temperature and humidity information of the environment where the overhead ground wire is located at regular intervals. Only when the collected temperature and humidity information meets the icing conditions, collect the power frequency electric field intensity signal below the overhead ground wire and convert the power frequency electric field intensity signal into electric field intensity amplitude.

[0087] Step S2: Generate a continuous electric field amplitude distribution curve by fitting the electric field intensity amplitude using a discrete wavelet transform algorithm;

[0088] Step S3: Based on the generated electric field amplitude distribution curve, screen out the indicators that can characterize the differences in the electric field amplitude distribution curve, and determine the indicators related to the ice thickness based on the screened indicators.

[0089] Step S4: Input the indicators related to ice thickness into the convolutional neural network for feature extraction and output the feature matrix;

[0090] Step S5: Input the feature matrix into the extreme learning machine to predict the icing thickness of the overhead ground wire.

[0091] This invention constructs the DWT-CNN-ELM algorithm, which can predict the ice thickness of overhead ground wires simply, quickly, and accurately, laying a solid foundation for smart grid operation and maintenance, ice disaster prevention, and stability control.

[0092] Example

[0093] This embodiment simulates a 220kV overhead icing ground wire line. By constructing relevant models such as towers, overhead conductors, overhead ground wires, and suspended insulator strings, and setting the icing thickness monitoring device of this invention on the overhead ground wire, when the temperature and humidity sensor 7 senses that the external temperature has reached the conditions for icing, the monitoring main board 5 is activated, and the electric field probe array module 4 uses three equally spaced vertically arranged spatial electric field probes to start monitoring the electric field amplitude around the icing.

[0094] like Figure 5 From the electric field amplitude distribution curve, it can be found that when there is ice on the overhead ground line, as the vertical distance between the monitoring point and the overhead ground line increases, the electric field amplitude shows a "peak" shape of first rising and then falling. Moreover, as the thickness of the ice on the overhead ground line increases, the electric field amplitude also continues to rise overall. Furthermore, through the analysis of the curve, it can be found that the maximum value is basically concentrated at a vertical distance of 46mm from the overhead ground line. Therefore, it can be deduced that the steepness, standard deviation, and maximum slope of the curve will also show significant differences in value. Considering the equality of electric displacement, the electric field strength is inversely proportional to the dielectric constant. Therefore, under this condition, the electric field strength inside the ice will decrease, forcing charges to accumulate on the surface of the ice. Furthermore, due to the unevenness of the ice surface, the radius of curvature of the ice ridges at the interface between the ice and the air will decrease. These factors will all contribute to the increase in the electric field at the interface between the ice and the air. Since this invention collects the power frequency electric field strength signal below the overhead ground wire, i.e., the electric field at the interface between the ice and the air, rather than the electric field inside the ice, the measured electric field strength amplitude should also increase with the increase of the ice thickness. This provides a theoretical basis for the simulation results.

[0095] against Figure 5 Based on the electric field amplitude distribution curve, six types of indicators can be initially selected. These six indicators are: the degree of abrupt change, etc. Maximum slope Standard deviation Average slope of waveform Waveform skewness and waveform kurtosis .

[0096] degree of mutation ,

[0097] maximum slope ,

[0098] Standard deviation ,

[0099] average slope of waveform ,

[0100] Waveform skewness ,

[0101] waveform kurtosis ,

[0102] in, The maximum value of the electric field amplitude is represented by the electric field amplitude distribution curve. The minimum value of the electric field amplitude is represented by the electric field amplitude distribution curve. The average value of the electric field amplitude represents the electric field amplitude distribution curve. This indicates the number of segments in the electric field amplitude distribution curve. express index, This represents the length of each segment of the electric field amplitude distribution curve. Indicates the first The electric field amplitude of the segment, The horizontal axis of the electric field amplitude distribution curve is... The slope at that time Represents numbers that approach 0.

[0103] The correlation between the above indicators and ice thickness was calculated using the Pearson correlation coefficient to determine the degree of abrupt change. Maximum slope Standard deviation Average slope of waveform Both showed a high correlation with icing thickness, while waveform skewness... With waveform kurtosis There is a weak correlation between waveform skewness and icing thickness, therefore the waveform skewness is... With waveform kurtosis Remove.

[0104] Set the size of the input layer of the convolutional neural network (CNN) to... 4 represents the number of input metrics. This represents the width and height of the input layer; sets the kernel size of the convolutional layer to 2 and the stride to 1; sets the window size of the average pooling layer to... Set the output feature count of the fully connected layer to 2. Adjust the abrupt change level. Maximum slope Standard deviation Average slope of waveform The resulting index matrix is ​​input into a convolutional neural network (CNN) for feature extraction.

[0105] The Extreme Learning Machine (ELM) is configured with 4 neurons in the input layer, 100 neurons in the hidden layer, and 1 neuron in the output layer. The weight vectors and unit bias vectors between the input and hidden layers are adjusted accordingly. The ELM is then trained using the training set. Features extracted by the Convolutional Neural Network (CNN) are input into the trained ELM to predict the icing thickness of overhead ground wires.

[0106] when , 、 , At that time, the predicted ice thickness of the overhead ground wire was 24.6288 mm, while the actual ice thickness was 25 mm. It can be seen that the difference between the predicted and actual ice thickness is small. Therefore, the overhead ground wire ice thickness monitoring device and method of the present invention have a good ice thickness monitoring effect.

[0107] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A device for monitoring the icing thickness of overhead ground wires, characterized in that, include: The system comprises a cylindrical metal casing, metal electrodes, an electric field array probe module, a monitoring motherboard, a lithium battery, and a slot. The cylindrical metal casing has plastic cover plates at both ends and a bottom plate at the other. The metal electrodes, together with the casing, form an electrode that generates equivalent capacitance with the three-phase power transmission lines. The slot extends through the center of the plastic cover plate, metal electrodes, and bottom plate, and contains an overhead ground wire. The ground wire and metal electrodes are electrically connected to maintain equipotential. An electric field array probe module is located on the metal electrodes below the slot. This module collects the power frequency electric field strength signal below the overhead ground wire. The monitoring motherboard converts the collected signal into an electric field strength amplitude and uses the DWT-CNN-ELM algorithm to predict the icing thickness of the overhead ground wire. The lithium battery powers the monitoring motherboard. The specific process of predicting the icing thickness of overhead ground wires using the DWT-CNN-ELM algorithm based on the electric field strength amplitude is as follows: i: The electric field intensity amplitude is subjected to piecewise cubic Hermite interpolation, wavelet decomposition, wavelet denoising and wavelet reconstruction through discrete wavelet transform fitting algorithm to generate a continuous electric field amplitude distribution curve; ii: Based on the generated electric field amplitude distribution curve, select indicators that can characterize the differences in electric field amplitude distribution curves, and determine the indicators related to ice thickness based on the selected indicators; iii: Input the indicators related to ice thickness into the convolutional neural network for feature extraction and output the feature matrix; iv: Input the feature matrix into the extreme learning machine to predict the icing thickness of the overhead ground wire; The specific process for determining the indicators related to ice thickness based on the screened indicators is as follows: determine the electric field amplitude distribution curve under the actual ice thickness, screen out the indicators that can characterize the difference in the electric field amplitude distribution curve, calculate the Pearson correlation coefficient between the screened indicators and the actual ice thickness, and take the indicators with an absolute value of Pearson correlation coefficient greater than 0.95 as the indicators related to ice thickness.

2. The overhead ground wire icing thickness monitoring device according to claim 1, characterized in that, The base plate is embedded with a temperature and humidity sensor, which is used to collect the temperature and humidity information of the environment at regular intervals and transmit the collected temperature and humidity information to the monitoring motherboard. The monitoring motherboard is used to determine whether the temperature and humidity information meets the icing conditions. If so, the monitoring motherboard wakes up the power frequency electric field strength signal acquisition of the electric field array probe module. Otherwise, the electric field array probe module does not work.

3. The overhead ground wire icing thickness monitoring device according to claim 1, characterized in that, The monitoring motherboard includes: an amplification and filtering module, a wireless transmission module, an MCU module, an inverter module, and an icing thickness prediction module; The electric field array probe module is used to convert the power frequency electric field intensity signal collected below the overhead ground wire into a voltage signal and transmit it to the amplification and filtering module; The amplification and filtering module is used to amplify the voltage signal, filter out interference, and transmit it to the ADC port of the MCU module. The MCU module is used to convert the voltage signal acquired by the ADC port into the electric field strength amplitude. The ice thickness prediction module is used to predict the ice thickness of the overhead ground wire by using the DWT-CNN-ELM algorithm to calculate the electric field strength amplitude. The wireless transparent transmission module is used to transmit the predicted icing thickness of the overhead ground wire to the remote user terminal. The inverter module is connected to the lithium battery and is used to convert the DC power of the lithium battery to power the amplification and filtering module, the wireless transmission module, the MCU module, and the icing thickness prediction module.

4. The overhead ground wire icing thickness monitoring device according to claim 1, characterized in that, The convolutional neural network includes: an input layer, a convolutional layer, a ReLU activation layer, an average pooling layer, and a fully connected layer; The input layer is used to normalize the indicators related to ice thickness; The convolutional layer is used to extract local features from the normalized metrics and generate a multi-channel feature map. The ReLU activation layer is used to introduce a non-linear activation function into the multi-channel feature map to obtain a feature map; The average pooling layer downsamples the feature map to generate a pooled feature map. The fully connected layer is used to flatten the pooled feature map and extract features through linear transformation and nonlinear activation, outputting a feature matrix.

5. The overhead ground wire icing thickness monitoring device according to claim 4, characterized in that, The size of the convolutional kernel of the convolutional layer is set to The step size is set to 1, where, This represents a number of indicators related to icing thickness. This indicates rounding down to the nearest integer.

6. The overhead ground wire icing thickness monitoring device according to claim 1, characterized in that, The prediction process for the icing thickness of the overhead ground wire is as follows: in, Indicates the icing thickness of overhead ground wires. This represents the number of hidden layer neurons in the Extreme Learning Machine. express index, Indicates the first hidden layer The weights between each neuron node and the output layer of the extreme learning machine Indicates the first hidden layer Each neuron node and the weight vector of the input layer of the extreme learning machine Indicates the first hidden layer Each neuron node and the unit bias vector of the input layer of the extreme learning machine Represents the characteristic matrix, This represents the activation function. This indicates the transpose operation.

7. The overhead ground wire icing thickness monitoring device according to claim 1, characterized in that, The extreme learning machine needs to be trained before it can predict the icing thickness of overhead ground wires. The process is as follows: Using the actual icing thickness of overhead ground wires as the label of the Extreme Learning Machine (ELM), and the corresponding feature matrix as the input of the ELM, the icing thickness is predicted. If the residual norm between the predicted icing thickness and the actual icing thickness is greater than a set threshold, the weights between the hidden and output layers of the ELM need to be optimized based on the hidden layer output matrix and the actual icing thickness. in, This represents the weights between the hidden and output layers of the optimized Extreme Learning Machine. Indicates the actual icing thickness. Represents the hidden layer output matrix of the Extreme Learning Machine The generalized inverse matrix, , This indicates the number of samples used for training. This represents the number of hidden layer neurons in the Extreme Learning Machine.

8. A monitoring method using the overhead ground wire icing thickness monitoring device according to any one of claims 1-7, characterized in that, Includes the following steps: Step S1: Collect temperature and humidity information of the environment where the overhead ground wire is located at regular intervals. Only when the collected temperature and humidity information meets the icing conditions, collect the power frequency electric field intensity signal below the overhead ground wire and convert the power frequency electric field intensity signal into electric field intensity amplitude. Step S2: Generate a continuous electric field amplitude distribution curve by fitting the electric field intensity amplitude using a discrete wavelet transform algorithm; Step S3: Based on the generated electric field amplitude distribution curve, screen out the indicators that can characterize the differences in the electric field amplitude distribution curve, and determine the indicators related to the ice thickness based on the screened indicators. Step S4: Input the indicators related to ice thickness into the convolutional neural network for feature extraction and output the feature matrix; Step S5: Input the feature matrix into the extreme learning machine to predict the icing thickness of the overhead ground wire.

Images