Radial gate flood discharge vibration state monitoring and diagnosis method based on improved permutation entropy algorithm

Abstract

The invention discloses a radial gate flood discharge vibration state monitoring and diagnosis method based on an improved permutation entropy algorithm, and belongs to the technical field of gate flood discharge vibration monitoring. According to the method, data-level dynamic fusion of multi-sensor vibration acceleration signals is realized based on a correlation variance contribution rate method, and flood discharge vibration characteristics reflecting the whole gate component are obtained; the defects of a coarse graining method in the prior art are improved, and dynamic calculation of IMMWPE is achieved in combination with an overlapped sliding window method. According to a nonparametric test method of kernel density estimation, entropy values of a large amount of normal and stable flood discharge vibration are collected so as to determine an entropy value alarm threshold value of abnormal vibration of the gate; and according to the fused support arm vibration data, calculating a composite multi-scale weighted permutation entropy to perform feature extraction, and introducing a penalty coefficient and a kernel parameter which are optimized by a Genetheshi shark optimization algorithm to construct a support vector machine recognition model to diagnose and recognize the gate vibration state. Compared with the prior art, the method improves the fault diagnosis recognition accuracy.

Description

Technical Field

[0001] The present invention belongs to the technical field of gate flood discharge vibration monitoring, and particularly relates to a method for monitoring and diagnosing the flood discharge vibration state of an arc gate by improving the permutation entropy algorithm. Background Technique

[0002] With the improvement of dam construction technology, high-head and large-discharge water discharge structures are continuously put into operation, and the problem of strong vibration of water discharge facilities induced by high-speed water flow is becoming increasingly prominent. Especially affected by the operation control water level during the flood season, the flood discharge facilities are frequently opened. Therefore, it is particularly important to monitor the vibration of working gates with high heads and partial opening flood discharge requirements. The flow-induced vibration of the gate structure, especially the support arm, is the main inducement for the instability of the gate. Therefore, it is very important to conduct real-time intelligent monitoring of the vibration parameters during the partial opening operation of the gate to ensure the safe operation of the entire water conservancy project and improve the level of water conservancy informatization and intelligence.

[0003] At present, different monitoring systems have been developed for the operation of gates. Although the existing developed systems have achieved real-time health monitoring during the operation of gates, such systems all conduct conventional analysis based on the vibration data picked up by sensors and cannot deeply mine the vibration characteristic information of the gates. In addition, at present, the vibration monitoring of gates mostly independently focuses on the vibration indexes (such as vibration acceleration and vibration stress) picked up by each sensor. The single-channel sensor cannot comprehensively reflect the vibration state of the overall components of the gate, and causes redundant observation data and poor monitoring feedback effect for operation and management personnel. At the same time, the flood discharge vibration of the gate is often characterized as a non-linear and non-stationary process. The traditional vibration monitoring method cannot conduct real-time fault diagnosis based on the measured vibration signal and cannot provide accurate and timely decisions for operation and management personnel in time. Therefore, how to efficiently fuse the vibration indexes picked up by multiple single sensors and accurately provide vibration monitoring early warning and diagnosis for operation personnel according to the fused vibration signal is of great significance for improving the safety and reliability of the flood discharge vibration of the gate.

[0004] Due to its high robustness and fast and simple algorithm, permutation entropy is widely used in non-linear processing methods. However, the permutation entropy algorithm only extracts the ordinal structure, loses the amplitude information, and is easily affected by noise. Both PE (permutation entropy) or WPE (weighted permutation entropy) only consider the single-scale dynamic behavior. Therefore, before calculating WPE, the time series is further coarsened at multiple scales to obtain the multi-scale weighted permutation entropy (MWPE). This method not only overcomes the deficiencies of a single scale but also fully retains the amplitude information. However, the way of coarsening the time series adopted by MWPE results in the length of the coarsened sequence decreasing with the increase of the scale factor, and at the same time, it is easy to lose the signal information of the omitted elements in the time series that cannot be evenly divided. Summary of the Invention

[0005] The purpose of the present invention is to solve the problem that the traditional vibration monitoring method cannot perform real-time fault diagnosis based on the measured vibration signal, which leads to inaccuracy and delay. A state prediction and diagnosis method for radial gate flood discharge vibration based on improved multi-scale weighted permutation entropy (IMMWPE) is proposed. The method can realize the dynamic fusion of vibration signals of multiple channel sensors on the arm and the calculation of multi-scale weighted permutation entropy. Firstly, the data-level dynamic fusion of vibration signals is realized based on the correlation variance contribution rate method, and the flood discharge vibration characteristics that can reflect the gate component as a whole (especially the radial gate arm) are obtained; secondly, the shortcomings of the coarse-grained method in the prior art are improved, and the dynamic calculation of IMWPE is realized by combining the overlapping sliding window method; according to the non-parametric test method of kernel density estimation, a large number of entropy values ​​of normal and stable flood discharge vibrations are collected to determine the entropy value alarm threshold of abnormal vibration of the gate; finally, the composite multi-scale weighted permutation entropy is calculated based on the fused arm vibration data for feature extraction, and the penalty coefficient and kernel parameter obtained by the Genghis Khan shark optimization algorithm are introduced to construct a support vector machine recognition model to diagnose and identify the gate vibration state.

[0006] The results show that the improved multivariate multiscale weighted permutation entropy combined with Genghis Khan shark optimization of support vector machine hyperparameters proposed in this invention can efficiently perform data-level fusion to achieve real-time monitoring of gate structures, and can timely diagnose the fault type of gate flood discharge vibration based on early warning information, and can greatly improve the diagnostic accuracy compared with the existing technology. This method provides a new idea for vibration monitoring during gate partial opening operation.

[0007] In order to solve the above technical problems, the present invention provides a method for monitoring and diagnosing the vibration state of an arc gate flood discharge using an improved permutation entropy algorithm, comprising the following steps:

[0008] S1. Collect vibration signals and conduct verification when the arc-shaped working gate is partially opened to discharge vibration;

[0009] S2. Classify the vibration signals, calculate the total correlation energy of the sensor based on the correlation between the vibration signals, further calculate the variance contribution rate, and calculate the dynamic fusion coefficient;

[0010] S3, according to the dynamically fused vibration signal obtained in S2, the vibration signal time series is improved by coarse-graining processing, sequence reconstruction, and probability calculation to obtain an improved multivariate multi-scale weighted permutation entropy value;

[0011] S4. Use the moving time domain sliding window technology to process the vibration signal after dynamic fusion, calculate the multi-scale weighted permutation entropy and determine whether the entropy value is lower than the alarm threshold. If it is lower, an alarm is issued and the gate operation status is adjusted in time;

[0012] S5. Use the entropy value of the alarm threshold trigger in S4 as a feature vector and input it into the classifier of the support vector machine (SVM). Optimize the penalty coefficient and kernel parameter using the Genghis Khan shark optimization algorithm and perform iterative optimization. Output the optimal solution to reverse-train the final SVM classifier.

[0013] Preferably, S2 specifically includes the following steps:

[0014] S21. Classify the vibration signals collected by the vibration sensors installed on each component of the radial gate according to the collection location. Among them, focus on the vibration signals of the radial gate's support arms.

[0015] S22. Based on the correlation between each signal, calculate the total correlation energy of the sensors through the data fusion principle of the correlation function weighting method.

[0016] S23. Combine the idea of variance contribution rate fusion, normalize the energy of the vibration signals, and further calculate the mean and variance of the normalized discrete vibration signal energy to obtain the variance contribution rate.

[0017] S24. Based on the variance contribution rate, calculate the dynamically allocated fusion coefficient at this moment.

[0018] S25. Calculate the dynamically allocated fusion coefficients of each sensor at each moment and perform traversal fusion to obtain the time history curve.

[0019] Preferably, S22 specifically includes the following steps:

[0020] S221. Assume that the vibration signals x 1,t , x 2,t , …, x r,t are measured by r sensors. Among them, the cross-correlation operation is performed on the signal pair of the i-th and j-th sensors to obtain:

[0021]

[0022] In the formula, represents the cross-correlation function of the vibration signals x i and x j . N represents the total number of signal data, m represents the time coordinate shift value, x i,t represents the vibration value of the i-th sensor at time t, and x j,t+m represents the vibration value of the j-th sensor at time t + m;

[0023] S222. Calculate the signal energy E ij through cross-correlation. The expression is:

[0024]

[0025] Wherein, i is the i-th sensor signal;

[0026] S223. Then, the total correlation energy E between the i-th sensor and all other vibration signals i is:

[0027]

[0028] Preferably, the improved coarse-graining process uses the basic idea of an overlapping sliding window. During the coarse-graining process, the overlapping sliding window method is used to calculate the multi-scale permutation entropy value. According to the selected coarse-graining scale factor s value, the vibration energy mean value within each scale factor is calculated based on the energy of the discrete signal for the elements within each scale factor, thereby obtaining an improved calculation method for the permutation entropy value.

[0029] Preferably, the specific expression of the improved coarse-graining is:

[0030]

[0031] Wherein, represents the coarse-grained sequence, s represents the scale factor, generally s takes a positive integer, and n represents the number of discrete vibration signal sample points.

[0032] Preferably, the specific expression of the improved multi-variable multi-scale weighted permutation entropy value is:

[0033]

[0034] Wherein, τ represents the time delay; WPE(·) represents the weighted permutation entropy operator, including three processing processes: sequence reconstruction, probability calculation, and entropy value calculation.

[0035] Preferably, the processing of the vibration signal after dynamic fusion using the moving time-domain sliding window technology is specifically expressed as:

[0036] X i→i+M-1 =[x i ,x i+1 ,x i+2 ,L,x i+M-1 ;

[0037] Wherein, X i→i+M-1 represents the window data within the time period from the i-th moment to the i+M-1-th moment of the vibration signal after dynamic fusion.

[0038] Preferably, the alarm threshold is statistically analyzed based on the entropy value of the normal and stable flood discharge vibration of the radial gate, and the vibration alarm and early warning are determined using the kernel density estimation method, which is specifically expressed as:

[0039]

[0040] Wherein, f(z) is the probability density function; l is the total number of samples; w is the width of the kernel function; K represents the kernel function; z is a random variable, which is the improved multi - variable multi - scale weighted permutation entropy value in this application; z i is the recording point of the random variable.

[0041] Preferably, the kernel density estimation method calculates the permutation entropy probability density distribution map reflecting the vibration state of the lower boom of the gate. Among them, the expression of the confidence level δ is:

[0042]

[0043] Among them, the confidence level δ is taken as 99%, that is, the probability that the permutation entropy value is not less than z lim during the normal and stable vibration of the gate is 99%. When the permutation entropy value is lower than z lim it is determined that the gate is in an abnormal vibration state.

[0044] Preferably, the S5 specifically includes: randomly generating Q shark individuals in the parameter space. Each individual includes a penalty coefficient C and a kernel parameter γ. The initial position of each individual represents the initial parameter combination of the SVM. Substitute the parameters C and γ of each shark individual into the SVM model for cross - validation, calculate the classification accuracy or error rate as the fitness value of this individual. If the optimal solution does not update during multiple iterations, terminate the algorithm, output the SVM penalty coefficient C and kernel parameter γ of the optimal shark individual, and finally use the optimal C and γ parameters to train the final SVM classifier.

[0045] Compared with the prior art, the beneficial effects of the present invention are:

[0046] 1. The method (IMMWPE - GKSO - SVM) of improving the multi - variable multi - scale weighted permutation entropy combined with the Genghis Khan shark optimization of SVM hyperparameters proposed in this scheme has the main advantages and improvement methods mainly divided into three directions: ① The method based on the correlation variance contribution rate realizes the dynamic data - level fusion of multi - sensors, realizes the overall vibration characteristics of the flood - discharging gate components, and combines the overlapping sliding window to realize dynamic real - time monitoring. ② The traditional calculation method is improved by using the moving coarse - graining method, and the mean value of the data points under each scale factor is calculated according to the vibration energy of the signal. ③ The GKSO method is used to optimize the penalty coefficient and kernel parameter to optimize the SVM recognition model, which can effectively ensure the accuracy of the flood - discharging vibration fault diagnosis and recognition of the gate.

[0047] 2. The method based on the correlation variance contribution rate proposed in the present invention realizes the data - level dynamic fusion of vibration signals, and obtains the overall flood - discharging vibration characteristics of the gate components; improves the deficiencies of the existing coarse - graining method, combines the moving coarse - graining and the method of calculating the mean value considering the signal vibration energy to realize the dynamic calculation of IMMWPE, and solves the deficiencies of the existing multi - scale permutation entropy.

[0048] 3. The present invention provides real-time early warning for management and operation personnel according to the entropy value alarm threshold of the abnormal vibration of the gate; after the early warning, fault diagnosis is carried out according to the early warning of the abnormal vibration state of the gate, providing an important new idea for the vibration monitoring of the local opening and flood discharge vibration of the radial working gate.

[0049] 4. The present invention introduces a method for jointly identifying faults in the flood discharge vibration state of the gate, namely IMMWPE-GKSO-SVM, which improves the accuracy of fault diagnosis and identification compared with the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a schematic diagram of the overall process of an embodiment of the present invention;

[0051] Figure 2 It is a flowchart of dynamic data-level fusion based on the contribution rate of correlation variance for an embodiment of the present invention;

[0052] Figure 3 It is a diagram of the data-level dynamic fusion result of the vibration acceleration signals of multiple sensors of the gate based on the contribution rate of correlation variance for an embodiment of the present invention;

[0053] Figure 4 It is a schematic diagram of the overlapping sliding window of the vibration acceleration signal after dynamic fusion for an embodiment of the present invention;

[0054] Figure 5 It is a diagram of the coarse-graining process of s = 2 and s = 3 in the prior art;

[0055] Figure 6 It is a diagram of the improved coarse-graining process of s = 2 and s = 3 for an embodiment of the present invention;

[0056] Figure 7 It is a diagram of the permutation entropy value of the prior art and the improved permutation entropy for an embodiment of the present invention;

[0057] Figure 8 It is a diagram of the change in entropy value under the normal and stable flood discharge vibration state of the radial gate for an embodiment of the present invention;

[0058] Figure 9 It is the probability density distribution of the entropy value under the normal and stable flood discharge vibration state of the radial gate for an embodiment of the present invention;

[0059] Figure 10 It is a diagram of the change in entropy value under the normal and stable flood discharge vibration state and abnormal vibration of the radial gate for an embodiment of the present invention;

[0060] Figure 11 It is a diagram of fault identification of the flood discharge vibration of the gate in an embodiment based on MMWPE-SVM in the prior art;

[0061] Figure 12The figure for identifying the vibration fault of the sluice gate during flood discharge based on IMMWPE-SVM in the embodiment of the present invention;

[0062] Figure 13 The figure for identifying the vibration fault of the sluice gate during flood discharge based on IMMWPE-GKSO-SVM in the embodiment of the present invention. Specific implementation manner

[0063] Embodiment 1: As Figure 1 - Figure 13 shown, the present invention discloses a method for monitoring and warning the vibration signal of the sluice gate based on the improved multi-source multi-scale weighted permutation entropy. Figure 1 The schematic diagram of the warning and diagnosis process for the vibration state of the radial gate based on IMMWPE proposed in this embodiment includes the following steps:

[0064] Step 1: Collect vibration acceleration signals

[0065] Collect vibration acceleration signals according to the vibration during the partial opening and discharge of the radial working gate, review whether the vibration signal indicators of the gate leaf and the support arm are normal, and judge whether there is a fault in the acceleration sensor.

[0066] Step 2: Calculate the dynamic fusion coefficient according to the correlation variance contribution rate to realize the magnitude fusion of multi-source sensor vibration signals

[0067] Classify the vibration signals picked up by the vibration acceleration sensors installed on each component of the sluice gate according to different positions, focus on the vibration acceleration signals of the support arm of the radial gate, calculate the total correlation energy of the sensors based on the correlation between each signal, and further calculate the mean and variance after normalizing the energy of the discrete vibration signals, so as to obtain the variance contribution rate.

[0068] Assume that the vibration signals x 1,t 、x 2,t , …, x r,t measured by r sensors, where the cross-correlation operation is performed on the signal pair of the i-th and j-th sensors to obtain

[0069]

[0070] On this basis, calculate the signal energy E through cross-correlation ij The expression is

[0071]

[0072] Then the total correlation energy of the i-th sensor and all other vibration signals is

[0073]

[0074] The above formulas (1) to (3) are the data fusion principle based on the correlation function weighting method. According to the weight p of each channel i it is possible to obtain p in direct proportion to the energy of the correlation function i , and then obtain the result of the fused signal.

[0075] The present invention further combines the idea of variance contribution rate fusion. First, the energy of the single-sensor signal x i,t is normalized to obtain

[0076]

[0077] When there are F homogeneous sensors collecting h discrete data at the same sampling frequency and sampling duration, the data signal sequence measured by the f-th sensor is xf,1 、x f,2 , …, x f,h . Let the p-th discrete data be x f,p , and its energy is normalized to obtain y f,p . The expression for calculating its variance contribution rate is

[0078]

[0079] where

[0080]

[0081]

[0082] In the formula: μ f represents the mean value after normalization of the energy of h discrete vibration signals collected by the sensor f, and σ f 2 is the variance.

[0083] The dynamic fusion coefficients allocated by different sensors at different times are obtained based on the correlation variance contribution rate, thereby realizing the data-level fusion of multi-sensor. The flowchart is shown in Figure 2 .

[0084] After calculating the variance contribution rate of the sensor f at the p-th discrete data, the dynamic fusion coefficient k allocated at this moment is further calculated f,p as

[0085]

[0086] Finally, the dynamic fusion coefficients of each sensor at each moment are calculated and traversed for fusion to obtain the time history curve, and its expression is

[0087]

[0088] The performance of the correlation variance contribution rate proposed by the present invention is as follows:

[0089] According to the actual engineering construction length scale of 1:25 of the ecological flood discharge hole of the actual gravity dam, a gate flow-induced vibration hydroelastic model is built. The material of the radial gate is an organic composite material filled with a weight-increasing agent, and by changing the formula and production process of the model material, the density and elastic modulus meet the requirements of elastic similarity. Finally, the selected material has an elastic modulus of 8.24 GPa and a unit weight of 7.8 kN / m 3 .

[0090] With the help of the gate flow-induced vibration test, three vibration acceleration signals on the lower boom during the flood discharge vibration of the radial working gate are obtained, and the vertical vibration signals of the three vibration acceleration sensors are dynamically fused by using the method proposed by the present invention. Figure 3 This is the result diagram of the data-level dynamic fusion of the multi-sensor gate vibration acceleration signal based on the correlation variance contribution rate in the embodiment of the present invention. It can be seen from the figure that the signal fusion coefficient of each channel is a fixed value relative to the prior art. The data-level fusion method proposed by the present invention dynamically calculates the fusion system at each channel and each sampling moment according to the correlation variance contribution rate.

[0091] In addition, for the purpose of realizing real-time dynamic monitoring, the vibration acceleration signal fused by the method based on the correlation variance contribution rate is preprocessed by an overlapping sliding window. The window length is 10,000 sampling points, that is, 20 s, and the sliding window step size is 500, that is, 1 s. The entropy value of the vibration acceleration signal under each window is calculated, and the entropy value in the stable vibration state of the gate is recorded in real time dynamically. The process is shown in Figure 4 as shown.

[0092] Step 3: According to the dynamically fused multi-sensor vibration acceleration signal obtained in Step 2, perform improved coarse-graining processing, sequence reconstruction, and probability calculation on its vibration signal time series, and finally obtain the IMMWPE reflecting the flood discharge vibration state of the gate.

[0093] When calculating the multi-scale weighted permutation entropy in the prior art, the time series is first coarsely grained, that is, the elements in the scale factor sequence are averaged. However, the unreasonable selection of the scale factor will have a greater impact on the calculation result. When the scale factor is large, the length of the time series will be greatly reduced, which is likely to lose effective time information; when the scale factor is small, the calculation of the entropy value result will be inaccurate. To address the above problems, the present invention proposes an improved time series coarse-graining method. This method uses the basic idea of the overlapping sliding window, adopts the overlapping sliding window method during the coarse-graining process, and considers the vibration energy of the time series to calculate the multi-scale permutation entropy value. The specific method is: according to the selected coarse-graining scale factor s value, calculate the average vibration energy of each scale factor according to the energy of the discrete signal for the elements within each scale factor, so as to obtain an improved calculation method for the permutation entropy value.

[0094] Let the one-dimensional time series be \(x(N)=x(1),x(2),x(3),\cdots,x(n)\). The prior art's coarse-graining processing of it can obtain:

[0095]

[0096] In the formula: represents the coarse-grained sequence, \(s\) represents the scale factor, and generally \(s\) takes a positive integer. When \(s = 1\), it is the original time series, and the corresponding calculated entropy value is the permutation entropy; when \(s\geq2\), the time series is coarse-grained into a new sequence of length \([n / s]\). When \(n / s\) is not a positive integer, it is rounded down. The schematic diagram of its coarse-graining process is shown in Figure 5 .

[0097]

[0098] Perform an overlapping sliding window process on the one-dimensional time series of the vibration acceleration signal after dynamic fusion. Let the time series within each window be \(x(N)=x(1),x(2),x(3),\cdots,x(n)\). The schematic diagrams of coarse-graining when \(s = 2\) and \(s = 3\) are shown in Figure 6 . Coarse-grain the time series within each window. Let the scale factor be \(s\), then the first sequence \(\{x 1 ,x 2 ,x 3 ,\cdots,x s \}\) is obtained, and the mean value considering the vibration energy of the first sequence is obtained Further move the sliding window forward by one time acquisition point to obtain the second sequence \(\{x 2 ,x 3 ,x 4 ,\cdots,x s+1 \}\), and the mean value considering the vibration energy is obtained And so on, the last sequence (i.e., the \((n - s + 1)\)-th sequence) is \(\{x n-s+1 ,x n-s+2 ,x n-s+3 ,\cdots,x n \}\), and the mean value considering the vibration energy after the last improved coarse-graining is obtained

[0099] Next, verify the performance of the improved coarse-graining method proposed by the present invention in calculating the permutation entropy value:

[0100] To verify the improved coarse-graining method proposed in the present invention, an ideal white noise time series is constructed, with a data length of 1000 selected, and its theoretical entropy value is 1. After performing moving coarse-graining processing on the data within the window, a new sequence is obtained. On this basis, phase space reconstruction is carried out. For the selection of the embedding dimension m and the time delay τ, the false nearest neighbor method and the mutual information method are respectively used to determine that the embedding dimension m = 5 and the time delay τ = 1 in this embodiment.

[0101] It can be Figure 7 seen that when the data length and the scale factor are the same, as the scale factor increases, the entropy value result of IMWPE (Improved Multiscale Permutation Entropy) is closer to 1 and the entropy value decreases more slowly compared with MPE (Multiscale Permutation Entropy), MWPE (Multiscale Weighted Permutation Entropy), CMPE (Composite Multiscale Permutation Entropy) and CMMWPE (Composite Multivariate Multiscale Weighted Permutation Entropy). The weighted entropy value is more stable; at the same time, IMWPE2 (the entropy value after improved coarse-graining considering the amplitude and energy information of the signal) has a higher entropy value and is closer to the theoretical entropy value than IMWPE1 (only sliding and taking the mean value in the prior art), indicating that the entropy value after weighting and improved coarse-graining contains the amplitude information and energy information of the signal, and the obtained entropy value is closer to the theoretical value; therefore, IMWPE is adopted for subsequent analysis of the gate vibration entropy value, which can effectively solve the deficiency in the calculation of permutation entropy in the prior art.

[0102] Step Four: Implement the calculation of IMMWPE based on the overlapping sliding window method, and determine whether the entropy value is lower than the alarm threshold. If it is lower, an alarm is issued and the gate operation state is adjusted in a timely manner.

[0103] The present invention statistically analyzes the entropy values of a large number of normal and stable flood discharge vibrations of the radial gate, and uses the kernel density estimation method to determine the vibration alarm and early warning, specifically as follows:

[0104]

[0105] In the formula: l is the total number of samples; w is the width of the kernel function; K represents the kernel function; z is a random variable, which is the improved multivariate multiscale weighted permutation entropy value in the present invention; z i is the recording point of the random variable.

[0106] The present invention calculates the permutation entropy probability density distribution map reflecting the vibration state of the lower boom of the gate by using the kernel density estimation method according to the permutation entropy values of a large number of stable and normal states during the flood discharge vibration of the gate. Its calculation expression is

[0107]

[0108] To minimize the probability of false alarms, the confidence level δ is taken as 99%, that is, the probability that the permutation entropy value is not lower than z lim during the normal and stable vibration of the gate is 99%. When the permutation entropy value is lower than zlim If it is, it is determined that the gate is vibrating abnormally, and an alarm is immediately sent to the management and operation personnel.

[0109] The following verifies the performance of the proposed IMMWPE in the vibration monitoring of the gate during flood discharge:

[0110] To verify the accuracy of the proposed method of the present invention, flow-induced vibration tests were carried out at relative gate openings of 50%, 30%, 25%, 20% and 10% respectively. Among them, for the relative gate opening of 25%, tests were carried out under the conditions of normal flood discharge vibration of the gate, abnormal vibration when the water stops leaking on both sides of the gate, and abnormal vibration of the submerged hydraulic jump behind the gate. To achieve the purpose of real-time dynamic monitoring, the vibration acceleration signals fused by the method of correlation variance contribution rate were preprocessed by overlapping sliding windows. The window length was 10,000 sampling points, that is, 20 s, and the sliding window step size was 500, that is, 1 s. The entropy value of the vibration acceleration signal under each window was calculated, and the entropy value in the stable vibration state of the gate was recorded in real time and dynamically.

[0111] According to the present invention, under the five fixed opening stable vibration states of the relative gate openings of 50%, 30%, 25%, 20% and 10%, the three vertical vibration acceleration signals of the lower boom of the radial gate are fused and the weighted permutation entropy value is calculated. The entropy value distribution diagram is shown in Figure 8 as follows.

[0112] The alarm threshold x for abnormal vibration of the gate is determined according to the non-parametric test method of kernel density estimation lim is 0.819, as shown in Figure 9 as follows.

[0113] As shown in Figure 10 as follows, when the gate vibrates normally, its entropy value is greater than 0.82. Under different openings, its entropy value changes slightly but there is no obvious difference and it will not trigger an early warning; when the gate vibrates abnormally, the permutation entropy decreases significantly, indicating that the permutation entropy change diagram can be used as an important index for monitoring the vibration state of the gate during flood discharge. When the water stops leaking on both sides of the gate during flood discharge, the entropy value decreases significantly. If a submerged hydraulic jump occurs, the water body behind the gate beats the back surface of the gate. This phenomenon is a typical problem of the gate vibrating induced by turbulence. At this time, its entropy value is lower than that when the water stops leaking on both sides, indicating that the vibration is more intense in this operating state, and its vibration randomness and complexity decrease, belonging to a typical non-stationary process. When the entropy value drops below the alarm threshold, it indicates that the gate has abnormal vibration, which has an adverse impact on the safe operation of the gate during flood discharge.

[0114] The above results show that there is an obvious difference in the entropy value between the normal flood discharge vibration and the abnormal flood discharge vibration of the gate calculated by the proposed IMMWPE method of the present invention. When the entropy value is lower than the alarm threshold, an alarm will be triggered, indicating that there is abnormal vibration in the gate during flood discharge, verifying that the method proposed by the present invention is effective in monitoring the operating state of the gate during flood discharge vibration.

[0115] Step 5: Use the entropy value in Step 4 as a feature vector and input it into the classifier of the support vector machine (SVM). Then, use the Genghis Khan shark optimizer (GKSO) to adaptively select the penalty coefficient and kernel parameter.

[0116] The Genghis Khan shark optimization algorithm (GKS) mimics the behaviors of sharks in nature, including hunting, moving, foraging, and self - protection mechanisms. The introduction of its algorithm and its corresponding mathematical expressions are as follows:

[0117] ① The hunting mechanism means that sharks update their positions by sensing the positions of prey to approach the optimal solution. This mechanism can be expressed as

[0118] X i (t + 1)=X i (t)+α·r 1 ·(X best (t)-X i (t)) (14)

[0119] where X i is the position of the i - th shark at the t - th iteration; X best (t) is the best position in the current iteration; α is the weight coefficient; r 1 is a random vector between 0 and 1.

[0120] ② The moving mechanism simulates the moving behavior of sharks in the search space, exploring new areas to find possible optimal solutions. This mechanism can be expressed by the following formula:

[0121] X i (t + 1)=X i (t)+β·r 2 ·(X mean (t)-X i (t)) (15)

[0122] X mean (t) is the average value of all sharks' positions; β is the weight coefficient; r 2 is a random vector between 0 and 1.

[0123] ③ The foraging mechanism simulates the behavior of sharks looking for food, adjusting their positions to obtain food resources and avoid falling into local optimal solutions. This mechanism can be expressed by the following formula:

[0124] X i (t + 1)=X i (t)+γ·r3 ·(X best (t)+X mean (t)-2·X i (t)) (16)

[0125] γ is the weight coefficient; r 3 is a random vector between 0 and 1.

[0126] ④ The self-protection mechanism simulates the behavior of sharks to avoid danger or predation. This mechanism ensures that the algorithm has a certain robustness and avoids premature convergence. The mechanism can be expressed by the following formula:

[0127] X i (t+1)=X i (t)-δ·r 4 ·(X worst (t)-X i (t) (17)

[0128] X worst (t) is the worst position in the current iteration; δ is the weight coefficient; r 4 is a random vector between 0 and 1.

[0129] ⑤ Combining the above four mechanisms, the update formula of Genghis Khan Shark Optimization Algorithm can be expressed as:

[0130]

[0131] This formula incorporates the shark's hunting, movement, foraging, and self-preservation behaviors, ensuring the algorithm conducts efficient search and optimization on a global scale.

[0132] Step 6: Based on the extracted entropy value, the penalty coefficient and kernel parameter obtained by GKSO optimization are used to build an SVM recognition model, and the gate operation status and gate maintenance after flood discharge are adjusted in time according to the identified fault status results.

[0133] The diagnostic accuracy of the IMMWPE-GKSO-SVM proposed by the invention is verified as follows:

[0134] In order to verify the diagnostic accuracy of the IMWPE-GKSO-SVM proposed in the present invention, 25 groups of vibration signals of each measuring point under three conditions were collected: normal flood discharge vibration of the radial gate, water leakage on both sides of the gate, and flooding water jump of the gate, accumulating 75 groups of vibration signal samples. Table 1 is a table of the division of vibration sample sets for gate flood discharge vibration faults. Table 2 is a comparison table of the diagnostic model of the IMWPE-GKSO-SVM proposed in the present invention and the accuracy of MWPE-SVM and IMWPE-SVM.

[0135] Table 1 Division of vibration sample sets for gate flood discharge vibration faults

[0136] Operating condition of flood discharge through the gate Number of training samples Number of test samples Number of labels Normal vibration during flood discharge 15 10 1 Vibration caused by water leakage at both sides of the water stop 15 10 2 Vibration of submerged hydraulic jump behind the gate 15 10 3

[0137] Table 2 Comparison table of vibration diagnosis accuracies for gate flood discharge vibration faults

[0138] Diagnostic model for gate flood discharge vibration MWPE-SVM IMMWPE-SVM IMMWPE-GKSO-SVM Accuracy rate 86.67％ 93.33％ 96.67％

[0139] Figures 11 to 13 They are respectively the classification result diagrams of the test sets of MWPE-SVM, IMMWPE-SVM, and IMMWPE-GKSO-SVM. Combining the charts, it can be seen that the classification accuracy rate of the MWPE-SVM diagnosis model of the existing technology is less than 90%. When IMMWPE of the present invention is used to replace MWPE, the accuracy rate is increased to 93.33%. To further improve the accuracy rate and reduce the probability of classification errors, the GKSO method is used to optimize the penalty coefficient and kernel parameters, and the accuracy rate is further increased to 96.67%, that is, only one sample is misclassified among 30 test classification samples, indicating that the GKSO method has higher performance than the template function templateSVM of the existing technology for the support vector machine model.

[0140] The above comparison results verify that the IMMWPE-GKSO-SVM joint diagnosis method proposed by the present invention has high reliability and effectively guarantees the fault recognition accuracy.

Claims

1. A method for monitoring and diagnosing the vibration state of a flood discharge arc gate using an improved permutation entropy algorithm, characterized in that: The steps include: S1. Collect vibration signals and conduct verification when the arc-shaped working gate is partially opened to discharge vibration; S2. Classify the vibration signals, calculate the total correlation energy of the sensor based on the correlation between the vibration signals, further calculate the variance contribution rate, and calculate the dynamic fusion coefficient; S3, according to the dynamically fused vibration signal obtained in S2, the vibration signal time series is improved by coarse-graining processing, sequence reconstruction, and probability calculation to obtain an improved multivariate multi-scale weighted permutation entropy value; S4. Use the moving time domain sliding window technology to process the vibration signal after dynamic fusion, calculate the multi-scale weighted permutation entropy and determine whether the entropy value is lower than the alarm threshold. If it is lower, an alarm is issued and the gate operation status is adjusted in time; S5. The entropy value of the alarm threshold triggered in S4 is used as a feature vector and input into the support vector machine SVM classifier. The penalty coefficient and the kernel parameter are optimized by the Genghis Khan shark optimization algorithm and iterative optimization is performed. The optimal solution is output to reversely train the final SVM classifier.

2. The method for monitoring and diagnosing the vibration state of a flood discharge arc gate using an improved permutation entropy algorithm according to claim 1 is characterized in that: The S2 specifically includes the following steps: S21, classifying the vibration signals collected by the vibration sensors installed on the components of the arc-shaped working gate according to the collection positions; among which, the vibration signals of the arc-shaped gate support arms are focused on; S22, based on the correlation between each signal, the total correlation energy of the sensor is calculated by the data fusion principle of the correlation function weighting method; S23, combining the idea of ​​variance contribution rate fusion, normalizing the energy of the vibration signal, further calculating the mean and variance of the normalized discrete vibration signal energy, and thus obtaining the variance contribution rate; S24. Calculate the dynamic fusion coefficient allocated at this moment based on the variance contribution rate; S25, calculating the dynamic fusion coefficient of each sensor at each moment and performing traversal fusion to obtain a time course.

3. The method for monitoring and diagnosing the vibration state of flood discharge of arc gates using an improved permutation entropy algorithm according to claim 1 is characterized in that: The S22 specifically includes the following steps: S221, assuming that the vibration signal x measured by r sensors 1,t 、x 2,t , …, x r,t , where the i-th and j-th sensor signal pairs are cross-correlated and the following is obtained: In the formula, Represents the vibration signal x i and x j The cross-correlation function, N represents the total number of signal data, m represents the time coordinate movement value, x i,t represents the vibration value of the i-th sensor at time t, x j,t+m represents the vibration value of the jth sensor at time t+m; S222, calculating signal energy E by cross-correlation ij The expression is: Where i is the i-th sensor signal; S223, then the total correlation energy E of the i-th sensor and all other vibration signals i for:

4. The method for monitoring and diagnosing the vibration state of a flood discharge arc gate using an improved permutation entropy algorithm according to claim 1 is characterized in that: The improved coarse-graining process uses the basic idea of ​​overlapping sliding windows. In the coarse-graining process, an overlapping sliding window method is used to calculate multi-scale permutation entropy values. According to the selected coarse-graining scale factor s value, the vibration energy mean within each scale factor is calculated for the elements within each scale factor according to the energy of the discrete signal, thereby obtaining an improved permutation entropy value calculation method.

5. The method for monitoring and diagnosing the vibration state of a flood discharge arc gate using an improved permutation entropy algorithm according to claim 4 is characterized in that: The improved coarse-grained specific expression is: In the formula, represents the coarse-grained sequence, s represents the scale factor, generally s is a positive integer, and n represents the number of discrete vibration signal sample points.

6. The method for monitoring and diagnosing the vibration state of a flood discharge arc gate using an improved permutation entropy algorithm according to claim 5 is characterized in that: The specific expression of the improved multivariate multi-scale weighted permutation entropy value is: Where τ represents the time delay; WPE(·) represents the weighted permutation entropy operator, which includes three processing steps: sequence reconstruction, probability calculation, and entropy value calculation.

7. The method for monitoring and diagnosing the vibration state of flood discharge of arc gates using an improved permutation entropy algorithm according to claim 1 is characterized in that: The vibration signal after dynamic fusion is processed by using the moving time domain sliding window technology is specifically expressed as follows: X i→i+M-1 =[x i ,x i+1 ,x i+2 ,L,x i+M-1 ]; Where, X i→i+M-1 It represents the window data of the vibration signal after dynamic fusion from time i to time i+M-1.

8. The method for monitoring and diagnosing the vibration state of flood discharge of arc gates using an improved permutation entropy algorithm according to claim 1 is characterized in that: The alarm threshold is statistically calculated based on the entropy value of the normal and stable flood discharge vibration of the radial gate, and the vibration alarm warning is determined by the kernel density estimation method, which is specifically expressed as: Where f(z) is the probability density function; l is the total number of samples; w is the width of the kernel function; K represents the kernel function; z is a random variable, which is the improved multivariate multiscale weighted permutation entropy value in this application; z i is the recording point of the random variable.

9. The method for monitoring and diagnosing the vibration state of the arc gate flood discharge using an improved permutation entropy algorithm according to claim 8 is characterized in that: The kernel density estimation method calculates the probability density distribution diagram of the permutation entropy reflecting the vibration state of the lower arm of the gate, where the confidence degree δ is expressed as: The confidence level δ is taken as 99%, that is, the value of the permutation entropy is not less than z when the gate vibrates normally and stably. lim The probability is 99% when the permutation entropy is lower than z lim It is determined to be an abnormal gate vibration state.

10. The method for monitoring and diagnosing the vibration state of a flood discharge arc gate using an improved permutation entropy algorithm according to claim 1 is characterized in that: The S5 specifically includes: randomly generating Q shark individuals in the parameter space, each individual includes a penalty coefficient C and a kernel parameter γ, the initial position of each individual represents the initial parameter combination of the SVM, substituting the parameters C and γ of each shark individual into the SVM model, performing cross-validation, calculating the classification accuracy or error rate as the fitness value of the individual, terminating the algorithm if the optimal solution is no longer updated in multiple iterations, outputting the SVM penalty coefficient C and kernel parameter γ of the optimal shark individual, and finally using the optimal C and γ parameters to train the final SVM classifier.

Images