Power transmission line galloping characteristic prediction and early warning system based on complex meteorological conditions

Abstract

The patent relates to a power transmission line galloping characteristic prediction and early warning system based on complex meteorological conditions. The method comprises the following steps: firstly, acquiring various meteorological parameters including wind speed, wind direction, temperature, humidity, icing thickness and the like in real time through a micro-meteorological sensor arranged along a power transmission line; secondly, analyzing and processing the meteorological data by using a deep learning algorithm and a generalized regression neural network (GRNN), and extracting correlation characteristics of meteorological factors and power transmission line galloping amplitude and frequency; then, a support vector machine (SVM) model is adopted to classify and identify the galloping risk, and a galloping risk level is generated; finally, a set of power transmission line galloping monitoring and early warning system is designed and developed, real-time acquisition of meteorological data, galloping trend prediction, risk grade evaluation and output of early warning information are realized, and safe and stable operation of the power transmission line under complex meteorological conditions is effectively guaranteed.

Description

Technical Field

[0001] The present invention relates to a power transmission line status monitoring technology in a power system, in particular to a monitoring and early warning system based on the galloping characteristics of power transmission lines under complex meteorological conditions, and belongs to the field of electrical engineering and intelligent monitoring technology. Background Art

[0002] Transmission lines are a vital component of the power system, responsible for transmitting the electricity produced by power plants to various load centers. However, under complex meteorological conditions, the safe operation of transmission lines often faces huge challenges, especially under severe weather conditions such as strong winds, low temperatures and icing, when transmission lines are prone to dancing. The so-called dancing refers to the large-scale periodic oscillations of transmission lines after being subjected to wind loads. If this phenomenon is not handled in a timely manner, it may lead to serious faults such as conductor collisions, damage to insulators, tripping, and even cause large-scale power outages, resulting in huge economic losses.

[0003] In recent years, with the rapid expansion of my country's power system, the geographical and climatic environments of power transmission lines have become increasingly complex, and the probability of galloping accidents has shown an upward trend, especially under specific meteorological conditions such as high wind speed, high humidity, low temperature and icing. Typical high-risk areas include Xinjiang and other places. Due to the complex terrain and harsh climate, the galloping of transmission lines occurs frequently, posing a serious threat to the safe and stable operation of the power grid.

[0004] According to statistics, the main causes of the galloping phenomenon include drastic changes in wind speed, increased ice thickness on the conductor surface, and sudden changes in wind direction. Increased wind speed or frequent changes in wind direction will significantly increase the amplitude and frequency of conductor galloping, while ice increases the weight of the conductor and changes in aerodynamic characteristics, making the conductor more prone to irregular vibration. The combined effect of these factors will accelerate the mechanical fatigue of the conductor and related equipment, and eventually cause line failures. Existing transmission line galloping monitoring technology mainly relies on manual inspections and regular maintenance, but manual monitoring has problems such as low efficiency, poor timeliness, and insufficient coverage. Under severe meteorological conditions, the feasibility of manual inspections is greatly limited. In addition, traditional anti-galloping measures such as installing anti-vibration hammers and adding spacers can slow down the galloping to a certain extent, but these methods cannot achieve real-time monitoring and early warning, nor can they cope with the galloping risks caused by sudden changes in meteorological conditions. Therefore, it is particularly important to study a system that can monitor and predict the galloping of transmission lines in real time under complex meteorological conditions.

[0005] With the continuous advancement of intelligent sensor technology and data processing technology, intelligent monitoring systems based on meteorological data and dancing characteristics analysis have gradually become a research hotspot. By laying micro-meteorological sensors along the transmission lines, real-time collection of meteorological data including wind speed, wind direction, temperature, humidity and ice thickness, and combining intelligent algorithms to analyze and process meteorological data, the dancing trend of transmission lines can be accurately predicted and risk warnings can be issued in advance. However, the current technology still has certain limitations, such as: unreasonable sensor layout leads to monitoring blind spots, the correlation model between meteorological data and dancing trends is not accurate enough, and the early warning mechanism does not respond in time. These shortcomings limit the further development and application of transmission line dancing monitoring technology.

[0006] Therefore, developing a monitoring and early warning system based on the galloping characteristics of power transmission lines under complex meteorological conditions can not only make up for the shortcomings of existing technologies, but also provide technical guarantees for the safe and stable operation of power transmission lines by collecting meteorological data in real time and using deep learning algorithms to predict galloping trends. At the same time, this system can effectively reduce the cost and risk of manual inspections and improve the anti-galloping capabilities of the power system under complex meteorological conditions. Summary of the invention

[0007] The purpose of the present invention is to overcome the shortcomings of the prior art and propose a monitoring and early warning system based on the galloping characteristics of transmission lines under complex meteorological conditions, so as to ensure the safe and stable operation of transmission lines under severe meteorological conditions and reduce the risk of failure caused by galloping.

[0008] To achieve the above purpose, the present invention adopts the following technical solution, a monitoring and early warning system based on the galloping characteristics of power transmission lines under complex meteorological conditions, the specific steps are:

[0009] Step 1: Micro-meteorological sensors arranged along the transmission line are used to collect meteorological data including wind speed, wind direction, temperature, humidity and ice thickness in real time;

[0010] Step 2: Using the data acquisition module, the collected meteorological data is transmitted to the back-end server for data processing;

[0011] Step 3: Use deep learning algorithm and generalized regression neural network (GRNN) model to analyze meteorological data, extract characteristic correlations between meteorological parameters and dancing amplitude and frequency, and predict dancing trend;

[0012] Step 4: Classify the galloping risk based on support vector machine (SVM), and generate the galloping risk level by analyzing the amplitude, frequency and other key indicators of the wire;

[0013] Step 5: Develop a transmission line dancing monitoring and early warning system to feed back the dancing trend and early warning results to the operator through the early warning module in real time so that appropriate protective measures can be taken.

[0014] Moreover, the step 1 has the following characteristics: scatter plots of air pressure and wind speed, air pressure and wind direction, air pressure and temperature, air pressure and humidity, wind speed and wind direction, wind speed and temperature, wind speed and humidity, wind direction and temperature, wind direction and humidity, and temperature and humidity, as shown in detail. Figure 1 As shown. The Copula theory is selected to analyze the correlation between air pressure, wind speed, wind direction, temperature and humidity. As a tool for analyzing the correlation between variables, the Copula function contains almost all the dependent information between variables. In nonlinear problems, the Copula function plays a very important role in analyzing the correlation between variables. The process of using the Copula function for correlation analysis is as follows:

[0015] (1) Kernel Density Estimation (KDE) is used to estimate the density function of each input quantity. The Gaussian function is selected as the kernel function of KDE. The expression of the Gaussian function is:

[0016]

[0017] The expression of KDE is:

[0018]

[0019] Where, X i represents a random variable, h represents the window width, and n represents the sample size.

[0020] (2) Solve the unknown parameters of the binary Copula function and construct the binary Copula function. The binary t-Copula function is selected to perform correlation analysis on meteorological factors. The distribution function of the binary t-Copula function is:

[0021]

[0022] In the formula, λ represents the degree of freedom, and They represent the inverse functions of the distribution functions respectively, and ρ represents the correlation parameter between the two distribution functions, and its value range is -1 to 1.

[0023] (3) Calculate the correlation parameters between the input quantities. The Spearman rank correlation coefficient ρ is selected to evaluate the correlation between the two meteorological elements. The formula for the Spearman rank correlation coefficient ρ derived from the Copula function is:

[0024]

[0025] According to the above method, the correlation analysis between air pressure, wind speed, wind direction, temperature and humidity is carried out. The larger the absolute value of the Spearman rank correlation coefficient ρ, the stronger the correlation. It can be found that the correlation between air pressure and wind speed is the smallest, which is only -0.0120, but the correlation between wind speed and humidity reaches -0.4286, and the correlation between air pressure and humidity reaches -0.5809. Therefore, the influence of air pressure cannot be ignored in the prediction of wind speed.

[0026] Moreover, the step 3 has the following characteristics: GRNN is a neural network with a four-layer structure, namely, an input layer, a pattern layer, a summation layer and an output layer. Its specific network structure is as follows: Figure 2 shown.

[0027] The input layer is the input of the network, which is X = [x1, x2, ... x n ], where n represents the number of input elements, also called dimension.

[0028] The number of neurons in the pattern layer is the same as that in the input layer, p i is the transfer function of the pattern layer neuron, and its calculation formula is:

[0029]

[0030] Where, X i is the learning sample corresponding to the i-th neuron, and δ is the smoothing parameter.

[0031] The output of the pattern layer is the input X and its corresponding X i The exponential form of the square of the Euclid distance between them. The calculation formula of the square of the Euclid distance is:

[0032]

[0033] The summation layer uses two transfer functions to perform arithmetic summation on the output of the pattern layer, namely s D and Nj , where the transfer function s D The output of all pattern neurons is arithmetic summed, and the connection weight between the pattern layer and each neuron is 1, s D The calculation formula is:

[0034]

[0035] Another type of transfer function s Nj The output of all pattern layers is weighted summed, that is, the output of the i-th neuron in the pattern layer is summed with the j-th element of the summation layer, and its connection weight is the j-th element in the network output Y, s Nj The calculation formula is:

[0036]

[0037] The output layer is the prediction result of the network output, which is Y=[y1,y2,…,y n ], where n represents the number of output elements, also known as the dimension. The formula for calculating the output elements is:

[0038]

[0039] The specific steps of using GRNN to perform weather forecasting on distribution network lines are as follows:

[0040] (1) Divide the training set and test set

[0041] Taking full advantage of the rapidity of GRNN training network, the method of dynamically dividing the training set is adopted. That is, when predicting the meteorological elements at a certain time t, the meteorological element data 10 days before time t is used as the training set of the network.

[0042] (2) Determine network input and output

[0043] Considering the mutual influence between air pressure, wind direction, wind speed, temperature and humidity, the input of the network will include all five factors, and then make separate predictions for air pressure, wind direction, wind speed, temperature and humidity. Since the time span of the prediction is small, the meteorological data at adjacent times of adjacent dates also contain valuable information. Therefore, the input of the network is constructed according to the following rules: Assuming that the air pressure, wind direction, wind speed, temperature or humidity at time t on a certain day is predicted, the air pressure, wind direction, wind speed, temperature and humidity at time t-1 on the same day are required, and the air pressure, wind direction, wind speed, temperature and humidity at times t-2, t-1, t, t+1 and t+2 of the previous day are also required. The air pressure, wind direction, wind speed, temperature and humidity at times t-2, t-1, t, t+1 and t+2 of the previous two days are also required. Therefore, the input of the network is:

[0044]

[0045] Among them, P t-1 , D t-1 , S t-1 , T t-1 and H t-1 They represent the air pressure, wind direction, wind speed, temperature and humidity at time t-1 of the day. and They represent the air pressure, wind direction, wind speed, temperature and humidity at time t-2 of the previous day. and They represent the air pressure, wind direction, wind speed, temperature and humidity at time t-1 of the previous day. and They represent the air pressure, wind direction, wind speed, temperature and humidity at time t of the previous day. and They represent the air pressure, wind direction, wind speed, temperature and humidity at time t+1 of the previous day. and They represent the air pressure, wind direction, wind speed, temperature and humidity at time t+2 of the previous day. and They represent the air pressure, wind direction, wind speed, temperature and humidity at time t-2 two days ago. and They represent the air pressure, wind direction, wind speed, temperature and humidity at time t-1 two days ago. and They represent the air pressure, wind direction, wind speed, temperature and humidity at time t two days ago. and They represent the air pressure, wind direction, wind speed, temperature and humidity at time t+1 two days ago. and They respectively represent the air pressure, wind direction, wind speed, temperature and humidity at time t+2 two days ago.

[0046] The output of the network is:

[0047] Y=[P,D,S,T,H]

[0048] Among them, P, D, S, T and H represent the predicted air pressure, wind direction, wind speed, temperature and humidity at time t respectively.

[0049] (3) Train the GRNN to determine the optimal prediction model parameters

[0050] A dynamic prediction model is established, that is, the network is trained once for each sample point prediction. Therefore, the optimal parameters of GRNN for each sample point prediction are only related to the prediction made this time and are not fixed.

[0051] (4) Establish a distribution network line prediction model based on GRNN

[0052] Similarly, each time a prediction is made for the air pressure, wind direction, wind speed, temperature and humidity at a certain moment, the constructed prediction model is only related to this prediction process and is not static.

[0053] Moreover, step 4 has the following characteristics: SVM is a supervised learning method, which is mainly used to solve data classification problems. It achieves the segmentation of two categories (i.e., solving the binary classification problem) by constructing a hyperplane, and further expanding it can also achieve the segmentation of multiple categories (i.e., solving the multi-classification problem).

[0054] Suppose there is a set of n data {(xi ,y i ),i=1,2,…,n}, where x i is the input vector, x i ∈R n ;y i For x i The corresponding label, that is, the category, is usually y i ∈{-1,1}. At this time, there is a hyperplane that can divide the data in the set into two categories:

[0055] ω T x+b=0

[0056] In the formula, ω=(ω1,ω2,…,ω n ) is the normal vector of the hyperplane, and b is the displacement vector of the hyperplane.

[0057] The training of the SVM classification model can be regarded as obtaining the optimal classification hyperplane, that is, obtaining the optimal solution of ω and b, which can be equivalent to a quadratic optimization problem:

[0058]

[0059] In the process of solving practical problems, not all data can be accurately divided into two categories. To solve this problem, slack variables can be introduced based on the above formula. At this time, a new quadratic optimization problem is formed:

[0060]

[0061] In the formula, a is a constant, called the penalty factor, and its size represents the degree of adjustment for erroneous samples; λi is a slack variable, which is used to adjust samples with unclear features.

[0062] In order to effectively solve the above equation, firstly, perform Lagrange transformation on it, and the transformed expression is:

[0063]

[0064] In the formula, α i and β i is the Lagrange coefficient.

[0065] Then, the optimization problem in the above formula is transformed into a dual problem through dual transformation:

[0066]

[0067] Solving it, we can get the optimal solution Thus, the optimal classification function can be obtained:

[0068]

[0069] Where b * =-1 / 2(ω * (x r +x s )), Among them, x r and x s are the support vectors for different categories of samples.

[0070] When solving nonlinear problems, a nonlinear mapping φ:R can be introduced into the support vector machine. n →F, map the original data sample to the new feature space F, and then perform operations in F. This introduces the concept of kernel function, and the optimal classification function can be expressed as:

[0071]

[0072] In the formula, K(x i ,x) represents the kernel function. Commonly used kernel functions are shown in Table 1.

[0073] Table 1

[0074]

[0075] (III) Steps to build an overhead line galloping warning model based on PSO-SVM

[0076] The calculation process of the overhead line galloping warning model based on PSO-SVM is as follows: Figure 3 As shown in the figure. By using the powerful and concise optimization ability of PSO, the optimal parameters of the SVM model are obtained, and the optimal SVM classification model is established to achieve graded early warning of the overhead line galloping risk.

[0077] ① Initialize the parameters (c, g) of SVM and encode the initialized parameters as the original particles of the PSO algorithm;

[0078] ② Initialize the number of particles N, and generate N particles by randomly perturbing the original particles (that is, initialize the particle position x i ), initialize the remaining parameters of PSO (v i ,w,c1,c2);

[0079] ③ Obtain the optimal parameters of SVM through PSO algorithm;

[0080] ④ Use the obtained optimal parameters to build an SVM classification model and conduct dancing risk classification warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Through the detailed embodiments described below, the various advantages of the present invention can be presented more clearly to researchers in the field. The accompanying drawings will serve as auxiliary explanations. The accompanying drawings are as follows:

[0082] Figure 1 It is a scatter plot between two meteorological elements;

[0083] Figure 2 It is a schematic diagram of the GRNN structure;

[0084] Figure 3 This is the calculation flow chart of the line galloping warning model based on PSO-SVM;

[0085] Figure 4 This is a flow chart of the graded warning of line galloping based on meteorological forecasts. DETAILED DESCRIPTION

[0086] The invention disclosed herein will be described in more detail below with reference to the accompanying drawings:

[0087] PSO is an intelligent optimization algorithm based on swarm intelligence theory. According to certain rules, particles in the solution space find the optimal solution through group interaction calculation. Its structure is simple and easy to implement, the calculation is convenient and fast, and the required parameters are few. Suppose there is a particle swarm with N particles in D-dimensional space. The position vector and velocity vector of the particle are represented by x and y respectively. i =(x i1 ,x i2 ,…,x iD ) and v i =(v i1 ,v i2 ,…,v iD ) indicates that pbest i =(p i1 ,p i2 ,…,p iD ) is the best position that the particle has ever experienced, gbest=(g1,g2,…,g D ) is the best position experienced by the entire group. The position and speed of each particle have a limited range. Once the speed and position exceed the limit during the optimization operation, the position or speed of this particle should be replaced by the boundary position or boundary speed. The update formula of the position and speed of the i-th particle is:

[0088]

[0089] Where c1 and c2 are acceleration constants (learning factors) that adjust the step size of the calculation; r1 and r2 are random numbers to increase the randomness of the search; w is the inertia weight. Different inertia weight values ​​determine the different local convergence capabilities of the algorithm. The larger the inertia weight, the better the global convergence ability, and the smaller the inertia weight, the better the local convergence ability.

[0090] The right side of the equal sign in the above formula is divided into three parts. The first part corresponds to the diversified characteristics, indicating that the particle trusts its current state of motion and provides a certain momentum for the next state, so that it moves at its own speed. The second and third parts correspond to the characteristics of centralized search. Among them, the second part is the individual cognitive part, which represents the thinking behavior of the particle, so that the particle constantly approaches the best position that the particle has experienced; the third part is the social cognitive part, that is, the particles exchange information and the solidarity and cooperation among members to make the particles approach the best position of the entire particle group. The main performance of PSO is determined by the balance and limitations of these three parts.

[0091] Random disturbances are added to the limited data to increase the amount of data. Combined with the occurrence of line galloping at each voltage level, wind speed, wind direction, temperature and humidity are used as characteristic quantities for galloping warning, and five levels of warning are issued for the galloping risk of overhead lines. The five-level warning includes: zero-level warning (no galloping risk), first-level warning (with no ice galloping risk), second-level warning (slight ice galloping risk), third-level warning (moderate ice galloping risk) and fourth-level warning (severe ice galloping risk). The specific rules for the division of warning levels are shown in Table 2.

[0092] Table 2

[0093]

[0094] The GRNN network is trained using historical meteorological data to construct a meteorological prediction model based on GRNN. The real-time monitored meteorological data is used as input to predict the meteorological information for the next moment. Similarly, the PSO-SVM classification model is trained using the accumulated meteorological and dancing historical data to construct a PSO-SVM-based overhead line dancing graded warning model. The predicted meteorological data is used as input to obtain the risk level for the next moment; the real-time monitored meteorological data is used as input to obtain the real-time risk level. In this process, you can prepare in advance according to the predicted risk level, and then make adjustments according to the real-time risk level to decide whether to take action. The process of the distribution network overhead line dancing graded warning method based on meteorological forecast is as follows: Figure 4 shown.

[0095] It should be emphasized that the embodiments described in the present invention are illustrative rather than restrictive. Therefore, the present invention includes but is not limited to the embodiments described in the specific implementation manner. Any other implementation manners derived by those skilled in the art based on the technical solution of the present invention also fall within the scope of protection of the present invention.

Claims

1. A prediction and early warning system based on the galloping characteristics of power transmission lines under complex meteorological conditions, characterized in that: The specific steps are: Step 1: Use micro-meteorological sensors to monitor meteorological parameters such as wind speed, wind direction, temperature, humidity, and ice thickness around the transmission line in real time; Step 2: Analyze the collected meteorological data based on deep learning algorithms to identify potential factors that trigger dancing, including sudden changes in wind speed and increased ice cover, and establish a correlation model between meteorological factors and dancing amplitude; Step 3: Use a neural network model combined with a support vector machine to make real-time predictions of the amplitude, frequency and dancing trend of the wire dancing to generate a dancing risk level; Step 4: Send the warning results to the control center through the data transmission module to achieve real-time warning and monitoring of transmission line vibration.

2. The power transmission line galloping monitoring and early warning system according to claim 1 is characterized in that: Step 2.1: Obtain 10 characteristic parameters including wind speed, wind direction, ice thickness, etc. through meteorological data, including peak wind speed, average wind speed, temperature change rate, ice thickness, etc.; Step 2.2: Use generalized regression neural network (GRNN) to reduce the dimension of feature parameters and generate three main feature components under each meteorological state for dancing excitation identification; Step 2.3: Train the dancing amplitude prediction model and the dancing fault classification model respectively to form a dancing early warning system.

3. The power transmission line dancing monitoring system according to claim 1, characterized in that: The system includes micro-meteorological sensors, data acquisition modules, data analysis modules and early warning modules; the micro-meteorological sensors collect meteorological data around the transmission lines in real time and transmit it to the back-end server through the data acquisition module; the data analysis module calculates the risk of dancing in real time and generates early warning information; The early warning module provides the operator with dancing level and early warning information through a visual interface.

4. The method for monitoring the galloping of a power transmission line according to claim 1, characterized in that: Dancing monitoring is based on multiple influencing factors under complex meteorological conditions, such as changes in wind speed and ice thickness. When the ice cover thickens, changes in wind speed cause the dancing amplitude to increase. Under certain conditions, the dancing amplitude may reach more than 10 meters. It is necessary to analyze the interaction of micro-meteorological factors in real time to predict the risk of dancing.

5. The power transmission line galloping warning system according to claim 1 is characterized by: The system can automatically adjust the warning threshold and dynamically adjust the dancing excitation threshold based on historical data and real-time monitoring results to adapt to changes in dancing risks under different meteorological conditions and improve the accuracy and response speed of the warning.

Images