A Fiber Optic State Detection Method Based on Multi-Source Information Fusion

Abstract

A fiber optic state detection method based on multi-source information fusion relates to the technical field of fiber optic line state detection, and solves problems such as easy occurrence of false alarms for faults, misjudgment of fault types, and difficulty in accurately identifying the line state in the fiber optic state recognition with a single parameter. The present invention is realized through steps such as data preprocessing, establishing an ELM recognition model, model optimization, and line state recognition. In the present invention, the white shark algorithm is used to optimize the extreme learning machine model, improving the recognition accuracy of the power fiber optic icing and galloping state and meeting the requirements of uninterrupted transmission of fiber optic communication. Compared with other traditional methods, the structure of the ELM recognition and classification model is simple and the learning speed is fast, which better realizes the detection of the fiber optic line state.

Description

Technical Field

[0001] The present invention relates to the technical field of optical fiber line state evaluation and analysis, and particularly relates to a method for detecting the state of an optical fiber by fusing multi-source information. Background Art

[0002] The power optical fiber communication network is becoming increasingly complex. The optical fiber composite overhead ground wire (OPGW) is constantly affected by various environmental factors such as wind dancing, sunlight exposure, and humidity all year round, and is prone to problems such as aging, cracks, and surface friction. These all affect the accuracy of optical fiber communication and even cause the optical fiber to break and interrupt the signal transmission, seriously threatening the safe and stable operation of the OPGW optical cable.

[0003] The scale of the power communication system is becoming increasingly large, which increases the difficulty of line maintenance work. As an important information transmission medium in the power system, if a failure occurs in the OPGW optical cable, it will not only affect the transmission of power communication information, but also cause incalculable economic losses to the power system. Relying solely on manual repair and maintenance is inefficient and costly. To solve this problem, it is necessary to accurately detect the state of the OPGW optical cable and promptly know the factors causing the failure, provide more basis for the maintenance and management of the line, and effectively improve the service life of the OPGW optical cable. Therefore, researching the method for detecting the state of an optical fiber by fusing multi-source information is of great significance for improving the disaster prevention ability and operation and maintenance level of the power grid. Summary of the Invention

[0004] The present invention provides a method for detecting the state of an optical fiber by fusing multi-source information to solve the problem of low accuracy in identifying the state of an optical fiber line in the prior art.

[0005] The method for detecting the state of an optical fiber by fusing multi-source information is realized by the following steps:

[0006] Step 1: Data preprocessing;

[0007] Denoise and normalize the temperature, stress, and vibration data collected by the Brillouin optical time domain reflectometer (B-OTDR) and the phase-sensitive optical time domain reflectometer respectively;

[0008] Step 2: Extract and fuse signal features;

[0009] Compare the data processed in Step 1 with the data when the optical fiber line is in the state of ice coating and wind dancing, extract the effective features in the data, and perform correlation fusion;

[0010] Step 3. Establish an Extreme Learning Machine (ELM) recognition model. Use the fused features in Step 2 as the input for training the ELM recognition model. Employ the White Shark Optimizer (WSO) to optimize the initial weights and bias values of the ELM recognition model to obtain an optimized ELM recognition model. The specific process is as follows:

[0011] Step 3-1. Use the randomly generated initial weights and bias values in the ELM recognition model as individuals of the white shark algorithm to form an initial population.

[0012] Step 3-2. Set the original position \(x\) of the white shark in the initialized population γ , the initial velocity \(v\) of the white shark γ , the maximum number of iterations \(T\) as the termination condition, and the spatial dimension \(D\).

[0013] Step 3-3. The white shark moves towards the prey to update its position.

[0014] Step 3-4. Calculate the fitness value according to the updated position of the white shark, update the historical optimal position and the global optimal position, and determine whether the termination condition is reached. If the termination condition is reached, output the optimal solution of the initial weights and bias to obtain an optimized ELM recognition model. Otherwise, re-determine the initial position and velocity of the white shark and then go back to Step 3-3.

[0015] Step 4. Fiber optic line status recognition: Input the real-time collected data into the optimized ELM model to achieve real-time detection of the fiber optic line status.

[0016] Advantages of the present invention: By studying the strain, temperature, and vibration data measured by two OTDRs, the present invention successfully correlates the data with the state characteristics during ice accretion and galloping of the fiber optic line, correlates and fuses the obtained characteristics to form a fusion vector, constructs an ELM recognition model, and the recognition rate of the trained ELM algorithm for the ice accretion and galloping state of the optical fiber is 94.44%. Use the white shark algorithm to optimize the initial weight \(\omega\) i and bias \(b\) i parameter values of the ELM recognition model, and the recognition accuracy is improved to 98.89%.

[0017] The method of the present invention realizes a fiber optic state detection method with multi-source information fusion, improves the recognition accuracy of the ice accretion and galloping state of power optical fibers, has good recognition effects, and meets the requirements of fiber optic communication reliability. Description of the Drawings

[0018] Figure 1 is a flowchart of the fiber optic state detection method with multi-source information fusion according to the present invention;

[0019] Figure 2 Flow chart of optimizing the ELM model by the white shark algorithm in the multi-source information fusion optical fiber state detection method of the present invention;

[0020] Figure 3 Flow chart of the ELM algorithm adopted in the multi-source information fusion optical fiber state detection method of the present invention. Specific implementation manners

[0021] Specific implementation manner 1. In combination with Figures 1 to 3 This implementation manner is described. A multi-source information fusion optical fiber state detection method is realized by the following steps:

[0022] Step 1. Data preprocessing;

[0023] Denoise and normalize the stress, temperature and vibration data measured by B-OTDR and ;

[0024] The collected data information contains various complex information, and the effective data and noise are mixed together. In order to reduce the influence of noise on the experimental results, it is necessary to filter the noise in the data. The data of signal feature vectors with different influence degrees are not in the same order of magnitude, so it is necessary to unify the data within a reasonable range. Standardizing the signal can prevent the influence of the differences in the data itself on the feature extraction process.

[0025] This implementation manner adopts the L 2 -norm normalization method. Suppose there is a data vector x = (x1, x2,..., x m ), and its normalized result vector is x' = (x1', x'2,..., x' m ), and its method is defined as:

[0026]

[0027] After data normalization, the Kalman filtering algorithm is used to denoise the data. The Kalman filter can estimate the state of a linear system and the estimation error variance is very small. The Kalman filter can estimate the state of a dynamic system from a series of data with measurement noise when the measurement variance is known, so as to remove the noise and restore the real data.

[0028] Step 2. Signal feature extraction and fusion;

[0029] Compare the data processed in Step 1 with the data when the optical fiber line is in the state of ice coating and wind dancing. When ice coating occurs, it will cause the line sag, and at this time, the temperature and stress data will change. When encountering wind dancing, the vibration data fluctuates greatly. Based on this, the effective features in the data can be extracted and associated and fused;

[0030] Using B-OTDR and Extract the short-time zero-crossing rate, average value, short-time energy, variance, and root mean square of the two signals respectively. Represent the extracted feature vectors as V1 and V2, and use a1 and a2 to represent the feature fusion coefficients of the two feature vectors. The feature fusion coefficient represents the weight of the two feature vectors. The two feature vectors are fused through the weight to obtain the fusion vector V = [a1V1, a2V2].

[0031] Step 3: Establish an ELM recognition model;

[0032] Use the fusion vector in Step 2 as the input for training the recognition model, combined with Figure 3 The specific steps are as follows:

[0033] In the model of the standard single-hidden layer feedforward neural network with L hidden layer nodes for N training samples (x i , t i ):

[0034]

[0035] Where y i is the output function of the hidden layer, the value of i is i = 1, 2,....L, a i = [a i1 , a i2 ,...a in is the input weight from the input layer node to the i-th hidden layer node. β i is the weight vector connecting the i-th hidden layer node and the output node. b i is the threshold on the i-th hidden layer node. g(x) represents an activation function that is differentiable within any range.

[0036] If the activation function is g(x) and the single-hidden layer feedforward neural network with T hidden nodes can approximate these N training samples (x i , t i ) without error, that is:

[0037]

[0038] It means that there exists a set of (a i , b i , β i ) such that:

[0039]

[0040] The above formula is represented in matrix form as:

[0041] Hβ = T (5)

[0042] Where H is the output matrix of the network hidden layer.

[0043]

[0044]

[0045] The weights β of the output layer are obtained by finding the least squares solution with the minimum norm of the linear model:

[0046]

[0047] where represents the Moore - Penrose generalized inverse of the hidden layer output matrix H.

[0048] Step 4: Optimization of the initial weights and bias values of the ELM recognition model:

[0049] The white shark algorithm is used to optimize the initial weights and bias values of the ELM so that the ELM recognition model can obtain the best classification results;

[0050] In this embodiment, the core concept and basic idea of the white shark algorithm are inspired by the behavior of white sharks when hunting, including their extraordinary hearing and smell when navigating and foraging in the sea. Mathematical modeling of the foraging behavior is carried out to achieve a sufficient balance between exploration and exploitation for white sharks. The main parameters of this algorithm are as follows:

[0051] The contraction factor μ of the white shark algorithm:

[0052] Controls the exploration behavior of white sharks, helps to avoid early convergence, and prevents the solution from degenerating into a local optimum. The value of μ depends on the parameter τ, and fine - tuning this parameter can improve the global search ability to reach the global optimal solution;

[0053]

[0054] Iterative parameters p1, p2:

[0055] p1 and p2, as functions of iteration, control the speed update of white sharks and achieve a stable balance between exploration and exploitation during the global and local searches of the WSO. It controls the influence of the best position known to the entire white shark group on the currently reached position.

[0056]

[0057] where T is the maximum number of iterations, t is the t - th iteration, p min and p max represent the initial and subordinate speeds to achieve good movement of white sharks;

[0058] Exploration parameter mυ:

[0059] This parameter is selected based on experience and should be less than 1. As shown in Equation (11), the higher the coefficient a1 in mυ, the better the exploration ability, and thus the lower the development accuracy. Similarly, the lower a0, the lower the development accuracy and the better the exploration ability. Therefore, this parameter can be fully adjusted to expand the exploration and development capabilities;

[0060]

[0061] Direction control parameter sgn(r2 - 0.5):

[0062] The parameter sgn(r2 - 0.5) is responsible for controlling the exploration direction, and since r2 is uniformly distributed between [0, 1], the positive and negative probabilities of the parameter are equal;

[0063]

[0064] where is the updated position of the γ-th white shark relative to the prey position, is the best position of the white shark relative to the prey at the t-th iteration, is the distance between the prey (i.e., the food source) and the white shark, defined as follows:

[0065]

[0066] where is the current position of the white shark relative to ;

[0067] Random parameter r1: Helps the solution run randomly in the given search space, and the change of the parameter r1 can better perform random exploration.

[0068] As Figure 1 and Figure 2 shown, the specific optimization process of the ELM recognition model; The output weights and bias values of the ELM model are optimized using the white shark algorithm;

[0069] (1) Initialize the population;

[0070] In the ELM network structure, the hidden layer is obtained by mapping the input samples under the combined influence of the weights and biases between the input layer and the hidden layer through a non-linear function. a i =[a i1 , a i2 ,...a in is the input weight from the input layer node to the i-th hidden layer node, and b i represents the bias of the i-th hidden node. Its value is randomly selected, so it is used as an individual of the white shark algorithm to form the initial population.

[0071] (2) Initialize the original position x of the white shark in the populationγ , the initial velocity v of the great white shark γ , the maximum number of iterations T, and the spatial dimension D.

[0072] (3) The great white shark uses its sensitive hearing, vision, and smell to track prey. When the great white shark senses the waves generated by the movement of the prey, it moves towards the prey in a undulating motion, which can be defined as the following formula:

[0073]

[0074] where γ = 1, 2,..., n, which is the number of great white sharks in the population of size n. represents the new velocity vector of the γ-th great white shark in the (t + 1)-th iteration, is the current velocity vector of the γ-th great white shark in the t-th iteration. is the global best position vector obtained by any great white shark so far in the t-th iteration. is the current position vector of the γ-th great white shark in the t-th iteration. is the known best position vector in the population, and c1 and c2 are random numbers uniformly distributed in the range [0, 1]. p1 and p2 are iteration parameters as shown in formula (10).

[0075] (4) The behavior of the great white shark when moving towards the prey can be described by the following formula:

[0076]

[0077] where is the new position vector of the γ-th great white shark in the (t + 1)-th iteration, is a negation operator, l and u represent the lower and upper bounds of the search space respectively, f represents the wave motion frequency of the great white shark, mυ is the exploration parameter of the great white shark, and the definitions of a, b, and ω0 are as follows:

[0078]

[0079]

[0080]

[0081] where a and b are one-dimensional vectors defined by equations (16) and (17), and ω0 is a logical vector defined by equation (18). is the bitwise exclusive OR operation.

[0082] (5) Formula (15) can be used to describe the behavior of the great white shark approaching the best position of the prey;

[0083] (6) Calculate the fitness value according to the position of the white shark, update the historical optimal position and the global optimal position, and determine whether the algorithm reaches the termination condition; if the end condition is reached, output the optimal solution of the initial weight and bias, and end the iterative process, and execute step five; otherwise, rearrange a fixed number of iterative times (about 10 times) each time in sequence to determine the position and initial velocity of the white shark, and then go to step (3).

[0084] Step Five, Optical Fiber Line State Identification: Input the newly collected real-time optical fiber state data into the ELM recognition model optimized by the white shark algorithm after step four, so as to achieve accurate identification of three situations of optical fiber line galloping, icing, and icing galloping.

[0085] The technical features of the above-described embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0086] The above-described embodiments only represent several implementation manners of the present invention. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several deformations and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention patent should be subject to the appended claims.

Claims

1. A fiber optic state detection method based on multi-source information fusion, characterized in that: This method is implemented by the following steps: Step 1: Data preprocessing; Denoise and normalize the fiber optic data collected by the Brillouin optical time domain reflectometer and the phase-sensitive optical time domain reflectometer; Step 2: Signal feature extraction and fusion; Extract the features of the OTDR signals obtained by the two optical time domain reflectometers in Step 1, and then correlate and fuse the extracted features; Step 3: Establish an ELM recognition model. Use the fused features in Step 2 as the input for training the ELM recognition model, and use the white shark algorithm to optimize the weights and bias values output by the ELM recognition model to obtain an optimized ELM recognition model. The specific process is as follows: Step 3-1: Use the randomly generated initial weights and bias values in the ELM recognition model as individuals of the white shark algorithm to form an initial population; Step 32: Set the original position x of the white sharks in the initial population γ , the initial velocity v of the white sharks γ , the maximum number of iterations T as the termination condition, and the spatial dimension D; Step 3-3: The white shark moves towards the prey to update its position; Step 3-4: Calculate the fitness value according to the updated position of the white shark, and update the historical optimal position and the global optimal position, and determine whether the termination condition is reached; If the termination condition is reached, output the optimal solution of the initial weights and biases to obtain an optimized ELM recognition model; otherwise, re-determine the initial position and speed of the white shark and then go back to Step 3-3; Step 4: Fiber optic line state recognition: Input the real-time collected data into the optimized ELM model to achieve real-time detection of the fiber optic line state.

2. The fiber optic state detection method for multi-source information fusion according to claim 1, characterized in that: The fiber optic data mentioned in Step 1 includes temperature, stress, and vibration data.

3. A fiber optic state detection method for multi-source information fusion according to claim 1, characterized in that: In Step 2, extract the features of the two OTDR signals, including the short-time zero-crossing rate, average value, short-time energy, variance, and root mean square of the signals. Let the feature vectors of the two OTDR signals be V1 and V2 respectively, and the feature fusion coefficients of the two feature vectors be a1 and a2 respectively. The feature fusion coefficients represent the weights of the two feature vectors respectively. The two feature vectors are fused through the weights to obtain the fusion vector V = [a1V1 a2V2].

Images