Optical fiber composite submarine cable vibration signal classification method based on energy entropy and ELM

Abstract

The invention discloses an optical fiber composite submarine cable vibration signal classification method based on energy entropy and ELM, and the method comprises the steps: firstly building an optical fiber composite submarine cable working model under each working condition through finite element simulation software, and extracting the vibration acceleration data of an optical unit in a submarine cable; performing multilayer decomposition on the vibration acceleration by using an empirical mode algorithm to obtain a plurality of IMF components and a residual term; taking the energy entropy ratio of the first m IMF components as a characteristic quantity to form a characteristic vector T = [p1, p2,..., pm], and integrating the characteristic vectors under all working conditions to form an integral sample; and finally, randomizing the samples into a training set and a test set, inputting the training set into an ELM classifier to train the training set, and inputting the test set into the trained extreme learning machine classifier to complete vibration signal identification. According to the method, the energy entropy characteristic of the signal and the advantages of high calculation accuracy and short calculation time of the extreme learning machine are combined, the real-time working state of the optical fiber composite submarine cable is rapidly judged, and a new method is provided for online monitoring of the optical fiber composite submarine cable in actual engineering.

Description

Technical Field

[0001] The present invention relates to the field of optical fiber composite submarine cable signal detection, and in particular to an optical fiber composite submarine cable vibration signal classification method based on energy entropy and ELM. Background Art

[0002] The vibration signal of the submarine cable can accurately reflect the real-time working status of the optical fiber composite submarine cable. Complex sea conditions and anchor damage are the most common factors causing submarine cable failures. At the least, it will cause pulling on the submarine cable, causing certain deformation and damage to each structural layer, and may evolve into a fault over time, resulting in power outages of the submarine cable line; at the worst, it will directly lead to the breakage of the submarine cable. There are large differences in the vibration signals of optical fiber composite submarine cables under different sea conditions. By analyzing the vibration characteristics of the submarine cable and judging the working status of the submarine cable based on the vibration data, it can provide great guidance and reference value for the daily monitoring, fault analysis and engineering construction of the submarine cable. In the context of the popularization of intelligent technology, the amount of data required for submarine cable status monitoring has increased dramatically, and the difficulty of monitoring multiple fault signals of submarine cables under different working conditions has increased. Traditional fault monitoring methods can no longer meet new needs.

[0003] At present, the information detection method of submarine cables is mostly based on the traveling wave method. However, in actual use, this method has a long transmission distance, which causes a large attenuation of the traveling wave information during the transmission process, and there is the effect of electromagnetic radiation, which makes the cable information impossible to identify and detect. Therefore, in actual work, it is necessary to develop a more effective working status information monitoring solution based on the characteristics of the long energy transmission distance of the optical cable composite submarine cable.

[0004] The Empirical Mode Decomposition (EMD) algorithm is highly adaptive and can decompose vibration signals into multiple modal functions (IMF) with inherent characteristics. At present, fault diagnosis technology combined with machine learning and deep learning has been widely used in various practical occasions. Compared with traditional neural network algorithms, the Extreme Learning Machine (ELM) has the advantages of faster training speed, better generalization performance and applicability to large-scale data. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides a fiber optic composite submarine cable vibration signal classification method based on energy entropy and ELM. The method combines the energy entropy characteristics of the signal and the advantages of high calculation accuracy and calculation time period of the extreme learning machine ELM, and realizes the rapid judgment of the real-time working status of the fiber optic composite submarine cable, which provides an effective means for online monitoring of the fiber optic composite submarine cable in actual engineering.

[0006] The technical solution adopted by the present invention is:

[0007] The optical fiber composite submarine cable vibration signal classification method based on energy entropy and ELM includes the following steps:

[0008] Step 1: Establish a three-dimensional simulation model of the optical fiber composite submarine cable under complex sea conditions, and extract the vibration acceleration data of a single node in the optical unit structure layer under normal state, suspended state, anchoring and hooking conditions;

[0009] Step 2: Import the vibration acceleration data extracted in step 1 into the MATLAB workspace, and use the empirical mode decomposition (EMD) algorithm to decompose the extracted vibration acceleration data into multiple mode functions (IMFs) with inherent characteristics and a residual term;

[0010] Step 3: Calculate the energy entropy proportion of each modal function IMF component, and construct the first m energy entropy proportions into a feature vector T = [p1, p2, ..., p m ]; p1, p2, …, p m They represent the energy entropy proportions of the 1st to mth mode functions IMF respectively;

[0011] Step 4: Input a part of the feature vectors in step 3 into the extreme learning machine ELM for training to obtain a trained extreme learning machine classifier, and then input the remaining data of the other part of the feature vectors into the trained extreme learning machine classifier to obtain the output value, determine the category of the vibration signal according to the output result, and realize the classification of the vibration signal of the optical fiber composite submarine cable.

[0012] In step 1, a three-dimensional simulation model of the optical fiber composite submarine cable in the suspended state and when anchor damage occurs is first established. Figures 10(a), 10(b), and 10(c) are the three-dimensional simulation models of the optical fiber composite submarine cable under various working conditions. In the suspended state, the submarine cable is only affected by gravity acceleration and wave speed, while in the anchor-smashing and hooking states, the anchor is applied downward and obliquely upward respectively, so the three cases are modeled and discussed separately.

[0013] The simulation time is set to 2.5×10 -4 In order to expand the number of samples, 4 groups of data were extracted under normal state, suspended state, anchoring and hooking conditions, and a total of 40 groups of samples were analyzed.

[0014] In the step 1, a three-dimensional simulation model of the optical fiber composite submarine cable is built using finite element simulation software ANSYS WORKBENCH to obtain vibration acceleration data required for calculating energy entropy;

[0015] Fiber-optic composite submarine cables are normally laid on the seabed and are only affected by waves and their own gravity.

[0016] The suspended state is similar to the normal state. The suspended part of the submarine cable shakes irregularly in the seawater under the scouring of waves and the action of its own gravity. The submarine cable is prone to local stress concentration under long-term shaking.

[0017] The geometric model of anchoring and hooking of optical fiber composite submarine cable consists of two parts: optical fiber composite submarine cable and anchor model, as shown in Figure 10(a), Figure 10(b), and Figure 10(c). A node in the optical unit of the submarine cable that is closer to the anchor position is selected to calculate the vibration acceleration data of the optical fiber composite submarine cable under various working conditions.

[0018] like Fig.11 As shown in the figure, a single node in the optical unit structure layer is closer to the anchor position, and the vibration acceleration data of this node is extracted as a sample. The number of sample points of vibration acceleration data of each working condition is large, which has been presented in the form of a drawing. Figure 7(a) to Figure 7(d) Place.

[0019] In step 2, the vibration acceleration data of the nodes of the optical unit structure layer are extracted through the simulation results; the data is calculated by finite element software, see Figure 7(a) to Figure 7(d) .

[0020] The Empirical Mode Decomposition (EMD) algorithm can decompose complex signals into a finite number of Intrinsic Mode Functions (IMFs). It is an intuitive, direct, a posteriori, and adaptive signal decomposition method. It can decompose complex signals into a finite number of Intrinsic Mode Functions (IMFs). Each IMF component after decomposition contains the local characteristics of the original signal at different time scales.

[0021] In step 2, the modal function IMF decomposition process includes the following steps:

[0022] 2.1: Import the vibration acceleration data extracted in step 1 into the MATLAB workspace; read the simulation data into the workspace using the MATLAB built-in function xlsread;

[0023] 2.2: Find the maximum and minimum points of the original vibration signal X(t). This function is a built-in function of the EMD algorithm. When using the function, it can automatically find the function extreme points and then mark them;

[0024] Then, these extreme points are fitted by the curve interpolation method to obtain the upper envelope X of the signal. max (t) and the lower envelope X min (t); Curve interpolation is a mathematical method that constructs a continuous function on given discrete data points so that the function passes through all given discrete data points.

[0025] 2.3: Calculate the average value of the uplink and downlink routes:

[0026]

[0027] 2.4: Subtract the original vibration signal X(t) from the average value m1(t) of the upper and lower packet routes to obtain the residual signal d1(t);

[0028] If the original vibration signal X(t) is a stationary signal, the residual signal d1(t) is the first modal function of the original vibration signal X(t);

[0029] If the original vibration signal X(t) is a non-stationary signal, the monotonicity of the original vibration signal X(t) will continue to change, resulting in multiple inflection points. If these inflection points containing the specific characteristics of the original vibration signal X(t) are not selected, the residual signal d1(t) obtained in this case does not meet the conditions, so the following steps are required;

[0030] 2.5: Process the remaining signal d1(t) again in steps 2.2 to 2.4, and stop when SD is less than the threshold value. SD is the screening threshold value, which is generally 0.2-0.3. The first-order modal component c1(t), i.e. the first IMF component, obtained by this method meets the requirements of step 2.6. The screening threshold value SD is calculated as follows:

[0031]

[0032] In the above formula, d k-1 (t) is the k-1th order residual signal; d k (t) is the k-th order residual signal; T is the signal cycle time; t is the calculation time.

[0033] 2.6: Subtract the original vibration signal X(t) from the first-order modal component c1(t) to obtain the first-order residual r1(t), and then use the first-order residual r1(t) to replace the original vibration signal X(t);

[0034] 2.7: After repeating steps 2.3 to 2.6 for n times, the nth-order mode function c can be obtained. n (t) and the final residual r that meets the requirements n (t);

[0035] The expression of the original vibration signal X(t) decomposed by EMD is:

[0036]

[0037] In step 3, the energy entropy of each modal function IMF component is calculated as follows:

[0038] Assuming that the original vibration signal is decomposed into a total of n IMF components through the EMD algorithm, the energy entropy value E of the jth IMF is j for:

[0039]

[0040] Where x(i) represents the signal sequence, i = 1, 2, 3, ..., n;

[0041] The proportion of the energy of the jth IMF component in the total energy of the vibration signal p(j) is:

[0042]

[0043] Where x(i) is the vibration signal sequence; i = 1, 2, 3, ..., n, n is the number of sampling points of the vibration signal; l is the number of IMF components decomposed from the vibration signal; the first m energy entropy proportions are constructed into a feature vector T = [p1, p2, ..., p m ].

[0044] In step 4, the Extreme Learning Machine (ELM) is essentially a single-hidden layer feedforward neural network (SLFN). The biggest feature of ELM is that it can maintain high learning efficiency while maintaining high training accuracy;

[0045] For a typical single hidden layer neural network, such as Figure 1 As shown, assuming that there are any number of samples (X i ,t i ), where X i =[x i1 ,x i2 ,...,x in ] T ∈R n ,t i=[t i1 ,t i2 ,...,t im ] T ∈R m . For a single hidden layer neural network with L hidden layer nodes, it can be expressed as:

[0046]

[0047] Among them, g(x) is the activation function, W i =[w i,1 ,w i,2 ,...,w i,n ] T is the input weight, β i is the output weight, b i is the bias of the i-th hidden layer unit. W i ·X j W i and X j The inner product of .

[0048] The learning purpose of a single hidden layer neural network is to minimize the output error, which can be expressed as:

[0049]

[0050] That is, there is β i , W i and b i , so that

[0051]

[0052] The matrix can be represented as

[0053] Hβ=T

[0054] Among them, H is the output of the hidden layer node, β is the output weight, and T is the expected output.

[0055]

[0056] In order to achieve the best effect of single hidden layer neural network, it is necessary to and Meet the following requirements:

[0057]

[0058] Among them, i = 1,...,L, which is equivalent to minimizing the loss function

[0059]

[0060] For such problems, several commonly used existing gradient descent methods can solve such problems well, and the most fundamental point is to adjust each parameter at each iteration. In the ELM method, the output matrix of the hidden layer is unique by randomly determining the input weights and hidden layer biases. The training problem of a single hidden layer neural network is converted into solving a linear system. The output weight is thus determined as:

[0061]

[0062] Among them, H T is the Moore–Penrose generalized inverse of the matrix.

[0063] In step 4, in order to verify the feasibility of the invention, the vibration signals in normal state, suspended state, anchoring and hooking state are sampled, and a part of the data is randomly selected as a training set and input into the extreme learning machine classifier for training, the purpose of which is to grasp the hidden characteristics contained in each signal, and the remaining data is input into the trained extreme learning machine classifier as a test set for signal classification, and the digital labels "1", "2", "3" and "4" represent the four working states of the optical fiber composite submarine cable respectively.

[0064] The specific steps are as follows:

[0065] 4.1: Using the finite element simulation software ANSYS WORKBENCH, the vibration signal of the optical unit of the optical fiber composite submarine cable under different working conditions is calculated;

[0066] 4.2: The measured optical unit vibration acceleration signal is decomposed into multiple layers using the empirical mode algorithm, and the energy entropy of each IMF component and its proportion in the total energy are calculated respectively. The characteristic vector T = [p1, p2, ..., p m ];

[0067] 4.3: The extracted energy entropy ratio features under different working conditions are randomly divided into training samples and test samples, and the training samples are input into the extreme learning machine ELM for training;

[0068] 4.4: Using the trained extreme learning machine ELM classifier, the remaining fiber-optic composite submarine cable optical unit vibration signal test samples under different conditions are identified to complete the classification and identification of vibration signals.

[0069] The present invention provides a method for classifying vibration signals of optical fiber composite submarine cables based on energy entropy and ELM, and the technical effects are as follows:

[0070] 1) The present invention uses finite element simulation software to calculate the vibration acceleration data of the optical fiber composite submarine cable under different sea conditions. When the optical fiber composite submarine cable is in different states, the frequency distribution and energy distribution of the collected structural layer vibration signal are different, and the corresponding energy is also significantly different.

[0071] 2) The method of the present invention decomposes the signal into several IMF components with strong characteristics through the EMD algorithm, and constructs the signal feature vector based on this, which helps to improve the accuracy of signal classification and identification

[0072] 3) The method of the present invention directly uses the vibration signal feature vectors under different sea conditions as the input of the extreme learning machine ELM, achieving high-precision state recognition.

[0073] 4) The method of the present invention inputs the extracted feature vector into the extreme learning machine ELM. The calculation results show that the accuracy of ELM is 96.7% and the running time is 0.9685s. It can be well used for the recognition of submarine cable vibration signals and has certain practical application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 Schematic diagram of a typical single hidden layer neural network structure.

[0075] Figure 2 This is a flow chart of the steps of the optical fiber composite submarine cable vibration signal classification method based on energy entropy and ELM.

[0076] Figure 3 This is a simplified schematic diagram of the submarine cable model structure.

[0077] Figure 4 This is a schematic diagram of the anchor model structure.

[0078] FIG5( a ) is a schematic diagram of the structure of a three-dimensional simulation model under a suspended state;

[0079] Figure 5(b) is a schematic diagram of the structure of the three-dimensional simulation model under the anchoring condition;

[0080] Figure 5(c) is a schematic diagram of the three-dimensional simulation model structure under the hook condition.

[0081] Figure 6 It is the interface for extracting vibration acceleration data.

[0082] Figure 7(a) is a graph of vibration acceleration under normal working conditions;

[0083] Figure 7(b) is a graph of vibration acceleration under suspended state;

[0084] Figure 7(c) is a graph of the vibration acceleration under the anchoring condition;

[0085] Figure 7(d) is a graph of the vibration acceleration under the hooked state.

[0086] Figure 8(a) is the EMD decomposition diagram under normal working conditions;

[0087] Figure 8(b) is the EMD decomposition diagram under the suspended state condition;

[0088] Figure 8(c) is the EMD decomposition diagram under the anchor smashing condition;

[0089] Figure 8(d) is the EMD decomposition diagram under the hook state.

[0090] Fig. 9 This is the classification result of the extreme learning machine.

[0091] FIG10( a ) is a three-dimensional simulation model of a fiber-optic composite submarine cable in a suspended state;

[0092] FIG10( b ) is a three-dimensional simulation model of the optical fiber composite submarine cable in the anchor state;

[0093] FIG10( c ) is a three-dimensional simulation model of the optical fiber composite submarine cable in the hooked state.

[0094] Fig.11 Solve the schematic for the node. DETAILED DESCRIPTION

[0095] The optical fiber composite submarine cable vibration signal classification method based on energy entropy and ELM includes the following steps:

[0096] Step 1: Establish a three-dimensional simulation model of the optical fiber composite submarine cable under various working conditions, and extract the vibration acceleration data of a single node in each structural layer. After consulting the relevant literature: "The Impact of Anchor Damage on the Structural Layer of 500kV Submarine Cables" Document No.: 10.13336 / j.1003-6520.hve.20200131021., it can be seen that when anchor damage occurs, since the contact point between the anchor and the submarine cable is very small, the size of the anchor damage is almost unaffected by the shape of the anchor, but only depends on the mass and speed of the anchor, so the anchor is simplified into a trapezoid. The internal components of the optical fiber composite submarine cable adopt a bonded contact method to prevent the submarine cable from penetrating when hooked, resulting in inaccurate vibration waveform results. The contact method between the submarine cable and the anchor is set to friction (Frictional), which fully considers the Coulomb friction between the anchor and the submarine cable in actual conditions. After setting the contact type, it is also necessary to set the necessary constraints for this model. Since the hooking process only occurs in part of the submarine cable, the length of the submarine cable affected by the fault is much shorter than the actual total length of the submarine cable. It can be assumed that the displacement caused by the fault at both ends of the submarine cable is negligible. Therefore, displacement constraints are applied to all structural surfaces at both ends of the submarine cable to completely fix the cross-sections at both ends. The healthy fiber-optic composite submarine cable model only needs to suppress the velocity load in the above model and let the model only be under the action of its own gravity acceleration to simulate the suspended section. After the settings are completed, perform simulation and extract vibration acceleration data.

[0097] Step 2: Read the vibration acceleration data obtained by simulation calculation into the MATLAB workspace, and use the empirical mode decomposition (EMD) algorithm in MATLAB to decompose the vibration signal into multiple IMF components;

[0098] Step 3: Assume that the vibration signal is decomposed into n IMF components through the EMD algorithm, derive the data value of each IMF component in the figure, substitute it into the formula to calculate the energy value of each IMF component, and the jth IMF energy value E j The calculation expression is:

[0099]

[0100] The proportion of the jth IMF energy in the total energy of the vibration signal p(j) is

[0101]

[0102] Where x(i) (i=1, 2, 3, …, n) is the sample value of the vibration signal; n is the number of sampling points of the vibration signal; l is the number of IMF components decomposed from the vibration signal.

[0103] Finally, the characteristic vector T is constructed based on the energy proportion p(j) of each component, and its expression is:

[0104] T=[p1,p2,…,p m ]

[0105] Step 4: The feature vectors of the three conditions of suspended state, anchoring and hooking calculated in step 3 are used as training samples, and the remaining ones are input into the extreme learning machine ELM as test samples. The category of the vibration signal is determined according to the output results to realize the classification of the vibration signal of the optical fiber composite submarine cable.

[0106] Example:

[0107] In this embodiment, a three-core optical fiber composite submarine cable is selected as the finite element simulation object.

[0108] First of all, the structure of the three-core submarine cable is relatively complex, and it is difficult and time-consuming to model it. In order to reduce the time-consuming finite element simulation, the submarine cable modeling was optimized accordingly while ensuring the accuracy of the dynamic structure. The conductor shielding layer, the insulating shielding layer and the semiconductor waterproof layer are classified as one part because of their small thickness and the same mechanical properties as XLPE insulation. According to the mechanical properties and thickness of the internal components, it can be assumed that the anti-corrosion layer is part of the polypropylene filling layer, and the winding tape and inner cushion layer of the armor layer are part of the outer armor. At the same time, the number of steel wire armor is reduced. The simplified submarine cable model is as follows: Figure 3 The original structural parameters of the submarine cable and the simplified structural parameters are shown in Table 1. The present invention selects the anchor of a small ship for hooking simulation. The shape of the anchor is as follows: Figure 4 shown.

[0109] Table 1 Comparison of submarine cable structural parameters before and after simplification

[0110]

[0111]

[0112] According to the current relevant research results, when anchor damage occurs, the contact point between the anchor and the cable body is small, so the anchor damage has little to do with the shape of the anchor, but is related to the mass and speed of the anchor. The anchor can be simplified into a trapezoidal body. The simplified geometric models of the suspended state, anchor smashing and hooking are shown in Figure 5(a), Figure 5(b), and Figure 5(c), respectively. The normal state model only needs to add a fixed support force to the suspended state model, so the model diagram is omitted here.

[0113] Considering the complexity of the submarine cable working conditions of the present invention, the model contains regular and irregular objects, so two methods, sweeping method and free grid, are selected when classifying the grid. In order to ensure the convergence of the results during the calculation process, the three-core optical fiber composite submarine cable selects the free grid division method, which fully guarantees the regularity and uniformity of the grid, thereby effectively improving the accuracy of the three-dimensional simulation of each working condition of the submarine cable and appropriately reducing the calculation workload. After completing the grid division, it is necessary to apply loads to all models. First, gravity acceleration is applied to all models, and then a velocity load is applied to the anchor model alone to simulate the anchor damage process. In addition, since the submarine cable is laid in the silt under the seabed, the changes in its own weight and wave force act on the entire section of the submarine cable, while the changes caused by anchor damage only act on the local part of the submarine cable, so that the influence of factors such as the submarine cable's own weight and ocean currents can be ignored in the simulation. The internal components of the three-core optical fiber composite submarine cable adopt a bonded contact method, the purpose of which is to prevent the submarine cable from penetrating when hooked, resulting in inaccurate vibration waveform results. The contact mode between the submarine cable and the anchor is set to friction (Frictional), which fully considers the Coulomb friction between the anchor and the submarine cable in actual situations. After setting the contact type, it is also necessary to set the necessary constraints for this model. Since the hooking process only occurs locally in the submarine cable, and the length affected by the fault is much smaller than the actual total length of the submarine cable, it can be assumed that both ends of the submarine cable are completely fixed, that is, displacement constraints are added to all degrees of freedom at both ends of the submarine cable to completely fix the cross-sections at both ends. Finally, in order to realize the effect of anchor hooking, a velocity load in the oblique upward direction is applied to the anchor model separately, so that the anchor drives the submarine cable to displace. The vibration acceleration data extraction interface and vibration acceleration curve of the optical unit of the fiber-optic composite submarine cable are shown in the figure below. Figure 6 shown.

[0114] In the present invention, finite element simulation software is used to calculate the vibration acceleration data of the optical unit structure layer of the optical fiber composite submarine cable with different damage degrees. After multi-layer decomposition using the empirical mode algorithm, the IMF component images and data under each working condition are obtained. Each sample contains 100 original data points. The energy entropy value, total energy and proportion of each IMF component are calculated according to the above theoretical formula, and the characteristic vector T is constructed. The IMF component decomposition diagram under each working condition is shown in the figure below: Figure 8(a) to Figure 8(d) shown.

[0115] Sufficient samples were collected to train the extreme learning machine. In this example, 10 groups of samples were selected for normal state, suspended state, anchoring state and hooking state, totaling 40 groups of samples for analysis.

[0116] Table 2 Feature vector table

[0117]

[0118] After shuffling all the samples, 10 groups of samples are randomly selected as training sets, and the remaining 30 groups are input into the trained extreme learning machine classifier as test sets for classification. The results are as follows Fig. 9 shown.

[0119] The digital labels "1", "2", "3" and "4" represent the optical fiber composite submarine cable in normal state, suspended state, anchored state and hooked state respectively. As can be seen from the figure, the accuracy of the results after identification by the extreme learning machine classifier reaches 96.7%.

[0120] In order to verify the advantages of the extreme learning machine over other algorithms, this embodiment also compares the algorithm with other typical machine learning algorithms such as k-Means algorithm, k nearest neighbor algorithm (KNN), classification and regression tree (CART), Bayesian network (BBN), and support vector machine (SVM). The above six methods are run ten times respectively by averaging multiple times to find their respective average accuracy and average running time. The comparison results are shown in Table 3 below.

[0121] Table 3 Comparison of test results

[0122]

[0123] As can be seen from Table 3, the accuracy of the extreme learning machine ELM and k-Means algorithm are relatively high, 98.3% and 97.6% respectively, followed by Bayesian network BBN (96.3%), k nearest neighbor algorithm KNN (95.6%), support vector machine SVM (92.7%), and the classification and regression tree CART has the lowest accuracy (91.1%). In terms of computing time, the classification and regression tree CART has the shortest running time (0.8245s), followed by the extreme learning machine ELM (0.9685s), k nearest neighbor algorithm KNN (0.9856s), support vector machine SVM (0.9934s), k-Means algorithm (1.1344s), and the Bayesian network BBN has the longest running time (1.7648s). Given the accuracy first and the computing time comprehensively considered, the extreme learning machine ELM method can obtain higher accuracy and less running time, and this method has certain advantages.

[0124] The specific steps are as follows:

[0125] 1. Establish a three-dimensional simulation model of the optical fiber composite submarine cable under the conditions of suspension, anchoring and hooking, and sample the vibration acceleration data of a single node in the optical unit of the optical fiber composite submarine cable under the three working conditions;

[0126] 2. The vibration acceleration data obtained by simulation calculation is read into MATLAB, and the EMD algorithm is used to perform multi-layer decomposition on each vibration signal to extract 4 IMF components with obvious characteristics and 1 residual term;

[0127] 3. Calculate the energy of each IMF component and the total energy, and construct the feature vector T = [p1, p2, ..., p m ], used to identify the category of vibration signals of optical fiber composite submarine cables;

[0128] 4. The vibration acceleration data of the optical unit in four conditions, namely, normal state, suspended state, anchor hit and hook hung, are sampled. A part of each type of vibration signal samples is randomly selected as training sample data, and the other part is input into the trained extreme learning machine classifier as test samples to obtain the signal classification results.

[0129] The optical fiber composite submarine cable vibration signal classification method based on energy entropy and ELM has the following functional features:

[0130] 1) In terms of the evaluation characteristics of the fiber-optic composite submarine cable signal classification, the vibration acceleration data obtained by simulation is processed by a related algorithm to convert several modal function components with inherent characteristics and a residual term. When the fiber-optic composite submarine cable is subjected to different external forces, its internal energy distribution is also different, and the energy entropy in each IMF component will change. The feature vector constructed using this characteristic can effectively distinguish different working conditions of the fiber-optic composite submarine cable, thereby improving the accuracy of the classification results.

[0131] 2) In terms of vibration signal analysis technology, the empirical mode decomposition (EMD) algorithm is used to analyze the vibration acceleration signals of the optical fiber composite submarine cable model under different working conditions, and several modal function (IMF) components with inherent characteristics and a residual term are extracted. The EMD algorithm has applications in many fields, such as signal processing, image processing, vibration analysis, meteorology, financial analysis, etc. In addition, since EMD is a data-driven method, it can also be used in combination with other technologies to improve the analysis and understanding of complex signals. This algorithm can be adaptively applied to the analysis of nonlinear and non-stationary signals without pre-setting basis functions or filters; this calculation method can adaptively analyze nonlinear and non-smooth signals without pre-setting basis functions or filters; the algorithm can randomly generate the connection weights between the input layer and the hidden layer and the thresholds for the hidden layer neurons. Only the number of hidden layer neurons needs to be adjusted to obtain the only optimal solution, and no adjustment is required during the training process.

[0132] 3) In terms of the accuracy of classification of vibration signals of optical fiber composite submarine cables, the present invention establishes a three-dimensional simulation model of four different working conditions: normal, suspended, anchored and hooked, and obtains the vibration acceleration data of the optical fiber composite submarine cable under each working condition. A large number of samples are classified by a trained extreme learning machine classifier. The analysis results show that the classification accuracy of the vibration signals of optical fiber composite submarine cables is 96.7%, and the calculation time is 0.9685s. The results show that the method using the extreme learning machine ELM can obtain higher accuracy and less running time, and this method has certain advantages.

Claims

1. A method for classifying vibration signals of optical fiber composite submarine cables based on energy entropy and ELM, characterized in that The following steps are involved: Step 1: Establish a three-dimensional simulation model of the optical fiber composite submarine cable under complex sea conditions, and extract the vibration acceleration data of the nodes in the optical unit structure layer under normal state, suspended state, anchoring and hooking conditions; Step 2: Use the Empirical Mode Decomposition (EMD) algorithm to decompose the extracted vibration acceleration data into multiple intrinsic mode functions (IMFs) with inherent characteristics and a residual term; Step 3: Calculate the energy entropy proportion of each modal function IMF component, and construct the first m energy entropy proportions into a feature vector T = [p1, p2, ..., p m ]; p1, p2, …, p m They represent the energy entropy proportions of the 1st to mth mode functions IMF respectively; Step 4: Input a part of the feature vectors in step 3 into the extreme learning machine ELM for training to obtain a trained extreme learning machine classifier, and then input another part of the feature vectors into the trained extreme learning machine classifier to obtain an output value, determine the category of the vibration signal according to the output result, and realize the classification of the vibration signal of the optical fiber composite submarine cable.

2. The optical fiber composite submarine cable vibration signal classification method based on energy entropy and ELM according to claim 1 is characterized in that: In the step 1, a three-dimensional simulation model of the optical fiber composite submarine cable in the suspended state and under the condition of anchor damage is first established. In the suspended state, the submarine cable is only affected by the acceleration of gravity and the speed of the waves, while the anchor-smashed and hooked states are respectively affected by the downward and oblique upward forces of the anchor on this basis; in order to expand the number of samples, data under normal state, suspended state, anchor-smashed and hooked conditions are extracted for analysis.

3. The optical fiber composite submarine cable vibration signal classification method based on energy entropy and ELM according to claim 2 is characterized in that: In the step 1, a three-dimensional simulation model of the optical fiber composite submarine cable is built using finite element simulation software ANSYS WORKBENCH to obtain vibration acceleration data required for calculating energy entropy; Fiber-optic composite submarine cables are normally laid on the seabed and are only affected by waves and their own gravity. The suspended state is similar to the normal state. The suspended part of the submarine cable shakes irregularly in the seawater under the scouring of waves and the action of its own gravity. The submarine cable is prone to local stress concentration under long-term shaking. The geometric model of anchoring and hooking of fiber-optic composite submarine cable consists of two parts: fiber-optic composite submarine cable and anchor model. A node in the optical unit of the submarine cable that is relatively close to the anchor position is selected to calculate the vibration acceleration data of the fiber-optic composite submarine cable under various working conditions.

4. The optical fiber composite submarine cable vibration signal classification method based on energy entropy and ELM according to claim 1, characterized in that: In step 2, the modal function IMF decomposition process includes the following steps: 2.1: Import the vibration acceleration data extracted in step 1 into the MATLAB workspace; 2.2: Find the maximum and minimum points of the original vibration signal X(t); Then, these extreme points are fitted by the curve interpolation method to obtain the upper envelope X of the signal. max (t) and the lower envelope X min (t); 2.3: Calculate the average value of the uplink and downlink routes: 2.4: Subtract the original vibration signal X(t) from the average value m1(t) of the upper and lower packet routes to obtain the residual signal d1(t); If the original vibration signal X(t) is a stationary signal, the residual signal d1(t) is the first modal function of the original vibration signal X(t); If the original vibration signal X(t) is a non-stationary signal, the monotonicity of the original vibration signal X(t) will continue to change, resulting in multiple inflection points. If these inflection points containing the specific characteristics of the original vibration signal X(t) are not selected, the residual signal d1(t) obtained in this case does not meet the conditions, so the following steps are required. 2.5: Process the remaining signal d1(t) again in steps 2.2 to 2.4, and stop when SD is less than the threshold value. SD is the screening threshold value. The first-order modal component c1(t), i.e. the first IMF component, obtained by this method meets the requirements of step 2.

6. The screening threshold value SD is calculated as follows: In the above formula, d k-1 (t) is the k-1th order residual signal; d k (t) is the k-th order residual signal; t is the signal cycle time; t is the calculation time; 2.6: Subtract the original vibration signal X(t) from the first-order modal component c1(t) to obtain the first-order residual r1(t), and then use the first-order residual r1(t) to replace the original vibration signal X(t); 2.7: After repeating steps 2.3 to 2.6 for n times, the nth-order mode function c can be obtained. n (t) and the final residual r that meets the requirements n (t); The expression of the original vibration signal X(t) decomposed by EMD is:

5. The optical fiber composite submarine cable vibration signal classification method based on energy entropy and ELM according to claim 1 is characterized in that: In step 3, the energy entropy of each modal function IMF component is calculated as follows: Assume that the original vibration signal is decomposed into n IMF components by EMD algorithm, then the energy entropy value of the jth IMF is E j for: Where x(i) represents the signal sequence, i = 1, 2, 3, ..., n; The proportion of the energy of the jth IMF component in the total energy of the vibration signal p(j) is: Where x(i) is the vibration signal sequence; i = 1, 2, 3, ..., n, n is the number of sampling points of the vibration signal; l is the number of IMF components decomposed from the vibration signal; the first m energy entropy proportions are constructed into a feature vector T = [p1, p2, ..., p m ].

6. The optical fiber composite submarine cable vibration signal classification method based on energy entropy and ELM according to claim 1, characterized in that: In step 4, the vibration signals in the normal state, suspended state, anchoring state and hooking state are sampled, a part of the data is randomly selected as a training set and input into the extreme learning machine classifier for training, and the remaining data is input into the trained extreme learning machine classifier as a test set for signal classification. The labels "3" and "4" represent the four working states of the optical fiber composite submarine cable.

Images