A method for diagnosing vibration faults of hydropower units

Abstract

The present invention provides a method for diagnosing vibration faults of hydropower units, and belongs to the technical field related to the method for diagnosing faults of hydropower units. Firstly, an improved adaptive noise complete set empirical mode decomposition ICEEMDAN is used to perform modal decomposition on the vibration signal of the hydropower unit, calculate the correlation coefficients of each mode, remove the redundant components, and reconstruct the vibration signal; secondly, the multi-scale weighted permutation entropy MWPE is used to characterize the faults of the vibration signal; finally, an improved dung beetle algorithm MSIDBO is used to optimize the support vector machine and establish the MSIDBO-SVM classifier model to identify different faults of the hydropower unit. Based on this method, the accuracy of signal processing of the hydropower unit is improved, the disadvantage of poor noise resistance of the current feature extraction is overcome, and the stability of vibration fault identification of the hydropower unit is improved.

Description

Technical Field

[0001] The invention relates to the technical field of hydroelectric generator set fault diagnosis, and in particular to a hydroelectric generator set vibration fault diagnosis method. Background Art

[0002] As a key equipment for clean energy conversion and supply, hydropower units play a key role in ensuring power supply and promoting sustainable energy development. However, with the continuous expansion of the scale of units and the improvement of their intelligence level, their operating environment has become more complex, and the interaction of internal mechanisms has become closer, which has led to a corresponding increase in the possibility of failure and the difficulty of diagnosis. During operation, hydropower units may experience vibration, noise, temperature abnormalities and other problems due to various factors such as hydraulic imbalance, mechanical wear, and electrical failure. If these problems are not handled in a timely and effective manner, they will not only affect the stable operation and power generation efficiency of the units, but may also pose a potential threat to the safety of the entire hydropower station. According to relevant statistics, among the various faults that occur in hydropower units, nearly 80% of the faults can be reflected in the vibration signal. Therefore, the study of the characteristic information of the vibration signal of the hydropower unit can effectively realize the situation analysis of the unit's operating status, so as to help the staff of the hydropower station discover the faults in the unit as early as possible, thereby preventing the faults from further deteriorating.

[0003] In summary, the research on fault diagnosis technology of hydropower units is particularly important. However, due to the insufficiency of signal processing technology, the immaturity of feature extraction methods and the lack of fault diagnosis mechanism, the operation of hydropower units faces many challenges. To solve these problems, it is urgent to introduce and deeply study new intelligent fault diagnosis methods. By accurately identifying fault characteristics and building reliable diagnostic models, real-time monitoring of the operating status of the unit and fault warning can be achieved, which is of great significance to improving the operation and maintenance management level of hydropower stations and ensuring the safe and stable operation of the power grid. Summary of the invention

[0004] In view of this, in view of the problems raised in the background technology, the present invention provides a method for fault diagnosis of hydropower units. First, the improved adaptive noise complete set empirical mode decomposition is used to perform modal decomposition on the vibration signal of the hydropower unit, calculate the correlation coefficients of each mode, remove the redundant components, and reconstruct the vibration signal; secondly, the multi-scale weighted permutation entropy is used to characterize the fault of the vibration signal; finally, the improved dung beetle algorithm is used to optimize the support vector machine and establish the MSIDBO-SVM classifier model to identify different faults of the hydropower unit. This method effectively retains the effective components of the vibration signal, solves the problem of sensitive support vector machine parameters, improves the computational stability of the classifier, avoids the model from falling into the local optimum, and ensures the accuracy of signal analysis and fault diagnosis of the hydropower unit.

[0005] The technical solution adopted by the present invention is as follows:

[0006] A method for diagnosing vibration faults of a hydropower unit, the method comprising the following steps:

[0007] Step 1: Use vibration sensors to obtain acoustic vibration signals of hydropower units under different operating conditions;

[0008] Step 2: The acoustic vibration signal is modally decomposed using the improved adaptive noise complete ensemble empirical mode decomposition ICEEMDAN, the correlation coefficient of each modal component is calculated, the redundant components are removed, and a new acoustic vibration signal is reconstructed;

[0009] Step 3: Calculate the multi-scale weighted permutation entropy of the reconstructed acoustic vibration signal as a feature vector to characterize the fault characteristics of the hydropower unit;

[0010] Step 4: Divide the feature vector into a training set and a test set, input the training set into the fault diagnosis model based on the improved dung beetle algorithm to optimize the support vector machine, perform model training, and obtain an optimized classifier;

[0011] Step 5: Input the test set into the trained classifier to identify the faults of the hydropower unit.

[0012] Furthermore, in step 2, the acoustic vibration signal is modally decomposed using the improved adaptive noise complete set empirical mode decomposition ICEEMDAN, the correlation coefficient of each modal component is calculated, the redundant components are removed, and a new acoustic vibration signal is reconstructed. The specific ICEEMDAN decomposition is as follows:

[0013] Add i groups of Gaussian white noise to the original signal to form a new signal. Use EMD to calculate the average value of the upper and lower envelopes to obtain the first-order residual and modal components. The decomposition process is as follows:

[0014] x i t=x(t)+β0E1(ξ i (t)) (27)

[0015] r1=M(x i (t)) (28)

[0016] β0=δ0std(x) / std(E1(ξ i (t))) (29)

[0017]

[0018] Among them, ξ i(t) represents Gaussian white noise with mean and unit variance of 0 and 1 respectively, which is used to adjust the required signal-to-noise ratio (SNR) between the added noise and the residual after adding the noise. When k ≥ 1, β k =δ0std(r k ), δ0 represents the inverse of the required SNR between the input signal sequence and the first added noise, std() is the noise standard deviation operator, and the operator E k () represents the kth modal component obtained by EMD decomposition, the operator M() represents the local mean of the signal envelope, and < > represents the average calculation;

[0019] Construct a new decomposition signal r1(t)+ω1E2(ξ i (t)), the second-order residual is estimated by calculating the local mean through EMD decomposition, and the second modal component is defined as:

[0020]

[0021] Calculate the remaining order modal components k=3,4…,K and obtain the k-order residual:

[0022] r k (t)= <M(r k-1 (t)+ω k-1 E k (ξ i (t)))> (32)

[0023] Calculate the k-order modal component:

[0024]

[0025] Repeat the above steps to get the next modal component. When the residual r k When it is a monotonic function, terminate the calculation;

[0026] The correlation coefficient is a parameter indicator to measure the similarity between two signals. The larger the absolute value of the correlation coefficient between the modal component and the original signal, the higher the correlation and the more effective components it contains.

[0027]

[0028] Where N is the sample length, For x i The average value of the signal, for y i The average value of the signal is calculated by calculating the correlation coefficient of each mode, removing the redundant components and retaining the effective components of the vibration signal to reconstruct the signal.

[0029] Furthermore, in step 3, the multi-scale weighted permutation entropy of the reconstructed acoustic vibration signal is calculated as a feature vector to characterize the fault characteristics of the hydropower unit. The calculation process of the multi-scale weighted permutation entropy is as follows:

[0030] The basic theory of weighted permutation entropy algorithm is as follows:

[0031] For a given signal sequence Using phase space reconstruction theory, multiple subsequences are reconstructed as follows:

[0032]

[0033] Where m is the embedding dimension, τ is the delay, and N is the signal length;

[0034] In order to introduce the amplitude characteristics of the signal, each subsequence is calculated The weight value ω i as follows:

[0035]

[0036]

[0037] For each signal subsequence Use the weight value ω i and permutation and combination patterns Represents characteristic information. There are K kinds of signal sequence arrangement modes in total, and each mode The probability weighted value of is:

[0038]

[0039] Finally, the weighted permutation entropy value WPE(X,m,τ) of the signal sequence is calculated as follows:

[0040]

[0041] ②The calculation process of the multi-scale weighted permutation entropy algorithm is as follows:

[0042] WPE is a single scale to measure the complexity and dynamics of data, while MWPE is to coarse-grain the time series and then calculate the WPE value at each scale. For a time series x(i) of length N, coarse-grain processing is performed to obtain the coarse-grained series y (s) (j) as follows:

[0043]

[0044] Where s is the scale factor, Z + is a positive integer;

[0045] Calculate the WPE values ​​at multiple scales for the coarse-grained sequence:

[0046] E MWPE (X,τ,s,m)=WPE(y (s) (j),m,τ) (41).

[0048] Further, in step 4, the feature vector is divided into a training set and a test set, and the training set is input into the fault diagnosis model of the support vector machine optimized based on the improved dung beetle algorithm, and the model is trained to obtain an optimized classifier. Specifically, the support vector machine is very sensitive to the penalty factor c and the kernel function parameter g, and it is input into the dung beetle algorithm improved based on PWLCM chaotic mapping, nonlinear boundary convergence factor, rolling ball behavior fused with dynamic golden sine, sparrow vigilant mechanism and parameter adaptive adjustment strategies to iteratively search for optimization and output the optimal parameters. The improved dung beetle algorithm process is as follows:

[0049] Initialization population of PWLCM chaotic map

[0050] The PWLCM chaotic mapping is described as follows:

[0051]

[0052] Where: The control parameter p∈(0,0.5) is 0.4; the chaotic random value x(t)∈(0,1) is used to divide the piecewise function;

[0053] Nonlinear boundary convergence factor

[0054] The reproduction and foraging behaviors of the dung beetle algorithm are restricted to a certain area, that is, the spawning area and foraging area can be defined

[0055]

[0056] Among them, R = 1-t / T, T is the maximum number of iterations, Lb * and Ub * They define the lower and upper bounds of the spawning area, respectively. * is the optimal position of the current population; X b represents the global optimal position, Lb b and Ub b Respectively represent the optimal lower and upper limits of the foraging area; the spawning and foraging areas are dynamically adjusted with the number of iterations, and the reset boundary convergence factor R is expressed as follows:

[0057]

[0058] Among them, t is the number of iterations of the current population, and T is the maximum number of iterations of the population;

[0059] Rolling ball behavior with dynamic golden sine

[0060] The golden sine strategy is introduced to use the sine function to comprehensively search for potential areas, improve the convergence speed, and get rid of the limitation of local optimality. The update formula of the improved dung beetle rolling behavior is as follows:

[0061]

[0062] In the formula, R1 determines the moving distance of the current individual, which is a random number between [0, 2π]; R2 determines the update direction of the current individual position in the next iteration, which is a random number between [0, π]; X a is the position of the current optimal individual, X c is the position in the i-th dimension, x i (t) represents the position information of the ith dung beetle in the tth iteration; k∈(0,0.2) is the perturbation coefficient, α is -1 or 1, b is a random number between (0,1), δ is a random number between [0,1], x1 and x2 are the golden section coefficients; τ is the golden section ratio, which is 0.618033;

[0063] In addition to improving the movement path of the rolling dung beetle, in order to further balance the exploration and development capabilities of the algorithm in iteration, the inertia weight w is introduced into the golden sine formula:

[0064]

[0065] Among them, w min is the minimum inertia weight, w max is the maximum inertia weight, t is the current iteration number, and T is the maximum iteration number;

[0066] Sparrow Vigilante Mechanism

[0067] The vigilant strategy of the sparrow search algorithm has the function of reminding the population to take anti-predation actions. It is introduced into the dung beetle algorithm. At the same time, the shrinkage factor is introduced. In order to balance the exploration and development capabilities and enable the algorithm to have flexible search capabilities at different stages, the strategy update formula is as follows:

[0068]

[0069] Where: is the current global optimal position, T is the maximum number of iterations, and t is the current number of iterations; β is a normally distributed random number with mean 0 and variance 1, which is used to control the step length; K is the step length control parameter of the sparrow's moving direction; ε is the minimum constant to avoid the denominator being 0; f i is the fitness value of the i-th sparrow; f w 、f g are the worst fitness value and the current best fitness value respectively;

[0070] Parameter adaptive adjustment

[0071] The small step length and long jump mechanism of Levy flight and the wide trajectory movement of Brownian motion, this combined strategy effectively prevents the thieving dung beetles from quickly approaching the optimal individual in the early stage of iteration, and keeps the thieving dung beetles active in the later stage of iteration.

[0072] The position of the thief dung beetle after the introduction of Brownian motion is updated as follows:

[0073] x i (t+1)=X b +S·g·(|x i (t)-X * |+|x i (t)-X b |)+R BM (51)

[0074] Where g is a 1×D random vector, S is a constant, and R BM is the Brownian motion strategy,

[0075] The position of the thief dung beetle after the introduction of Levy's flight has been updated to:

[0076] x i (t+1)=X b +S·g·(|x i (t)-X * |+|x i (t)-X b |)+R LF (52)

[0077] Among them, R LF Flying strategy for Levy.

[0078] Compared with the prior art, the present invention has the following beneficial effects:

[0079] In the dung beetle algorithm, the PWLCM chaotic map is used to initialize the population, which enriches the diversity of the population, balances the search space of the population, and strengthens the global exploration ability. At the same time, the improvement of the boundary convergence factor and the introduction of the dynamic golden sine strategy, on the one hand, strengthen the global search ability and accelerate the convergence speed of the algorithm, on the other hand, enhance the information exchange of dung beetle individuals, balance the early and late exploration, avoid the algorithm from falling into the optimal solution area, and improve the optimization accuracy of the algorithm; the introduction of Levy flight and Brownian motion expands the search range of the early population, and enables the algorithm to search near the optimal solution in the later stage; the introduction of the sparrow vigilance mechanism reminds the population to perform anti-predation behavior, so that the algorithm has flexible search capabilities at different stages. Based on the improved dung beetle algorithm, it is used to optimize the penalty factor c and kernel function parameter g of the support vector machine, which enhances the robustness of the classifier and improves the accuracy of fault identification. Multi-scale weighted permutation entropy is used to extract the characteristics of vibration signals of hydropower units. The algorithm performs weighted processing on the permutation entropy of the original vibration signal and then analyzes it from multiple scales, which can effectively extract the state characteristic information of the hydropower unit.

[0080] The method proposed in the present invention has demonstrated excellent performance in the efficiency and stability of hydropower unit fault diagnosis. Experimental verification shows that the method has achieved extremely high accuracy, becoming a beneficial supplement to existing hydropower unit fault detection technology and demonstrating its high value in engineering applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only preferred embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0082] Figure 1 This is a flow chart of the present invention.

[0083] Figure 2 It is a time domain waveform diagram of the acoustic vibration data of the hydropower unit under different operating conditions described in the present invention.

[0084] Figure 3 This is the ICEEMDAN modal decomposition diagram of the normal operating condition of the hydropower unit described in the present invention.

[0085] Figure 4 This is the entropy value diagram of the four operating conditions of MWPE described in the present invention.

[0086] Figure 5 This is a visual classification diagram of the four operating conditions of the hydropower unit described in the present invention.

[0087] Figure 6This is a confusion matrix diagram of four operating conditions of the hydropower unit described in the present invention. DETAILED DESCRIPTION

[0088] The principles and features of the present invention are described below in conjunction with the accompanying drawings. The enumerated embodiments are only used to explain the present invention and are not used to limit the scope of the present invention.

[0089] Reference Figure 1 The present invention provides 1. a method for diagnosing vibration faults of a hydropower unit, the method comprising the following steps:

[0090] Step 1: Use vibration sensors to obtain acoustic vibration signals of hydropower units under different operating conditions;

[0091] The signal acquisition sensor equipment uses an industrial-grade noise sensor (CRY2301) as a spectrum analyzer. The sensor is used to detect slight noise. It is equipped with an advanced DSP processor, preamplifier, data acquisition card and microphone. The entire device is integrated into one, and the structure is compact, which also ensures the accuracy of the collected signal. The specific parameters of the noise sensor are shown in Table 1. The performance parameters of the hydro-turbine generator are shown in Table 2. The acoustic vibration signals of the hydro-turbine unit under normal working conditions, water hammer working conditions, sediment wear working conditions, and impact working conditions are collected. The time domain signals are shown in Table 2. Figure 2 As shown. The collected signals are processed, and the four working condition signals are divided into 120 samples, 30 samples for each working condition, and 10240 data for each sample. In each hydropower unit working condition data, 20 samples are selected as training sets, and the remaining 10 samples are used as test sets, with a total of 80 training sets and 40 test sets.

[0092] Table 1 Noise sensor parameters

[0093]

[0094] Table 2 Hydrogenerator performance parameters

[0095]

[0096] Step 2: The acoustic vibration signal is modally decomposed using the improved adaptive noise complete ensemble empirical mode decomposition ICEEMDAN, the correlation coefficient of each modal component is calculated, the redundant components are removed, and a new acoustic vibration signal is reconstructed as follows:

[0097] ICEEMDAN decomposition: Add i groups of Gaussian white noise to the original signal to form a new signal, and use EMD to calculate the average value of the upper and lower envelopes to obtain the first-order residual and modal components. The decomposition process is as follows:

[0098] x i t=x(t)+β0E1(ξi (t)) (53)

[0099] r1=M(x i (t)) (54)

[0100] β0=δ0std(x) / std(E1(ξ i (t))) (55)

[0101]

[0102] Among them, ξ i (t) represents Gaussian white noise with mean and unit variance of 0 and 1 respectively, which is used to adjust the required signal-to-noise ratio (SNR) between the added noise and the residual after adding the noise. When k ≥ 1, β k =δ0std(r k ), δ0 represents the inverse of the required SNR between the input signal sequence and the first added noise, and std() is the noise standard deviation operator. Operator E k () represents the kth modal component obtained by EMD decomposition, the operator M() represents the local mean of the signal envelope, and <> represents the average calculation.

[0103] Construct a new decomposition signal r1(t)+ω1E2(ξ i (t)), the second-order residual is estimated by calculating the local mean through EMD decomposition, and the second modal component is defined as:

[0104]

[0105] Calculate the remaining order modal components k=3,4…,K and obtain the k-order residual:

[0106] r k (t)= <M(r k-1 (t)+ω k-1 E k (ξ i (t)))> (58)

[0107] Calculate the k-order modal component:

[0108]

[0109] Repeat the above steps to get the next modal component. When the residual r k When it is a monotonic function, the calculation is terminated. Figure 3 This is the modal decomposition diagram of ICEEMDAN under normal operating conditions of the hydropower unit.

[0110] The correlation coefficient is a parameter indicator to measure the similarity between two signals. The larger the absolute value of the correlation coefficient between the modal component and the original signal, the higher the correlation and the more effective components it contains.

[0111]

[0112] Where N is the sample length, For x i The average value of the signal, for y i The average value of the signal. The signal is reconstructed by calculating the correlation coefficient of each mode, taking 0.1 as the boundary, retaining those greater than it and removing those less than it. The correlation coefficients of the modal components under normal working conditions are shown in Table 3.

[0113] Table 3 Correlation coefficients of modal components under normal working conditions

[0114] Modal component number Correlation coefficient Keep 1 0.39740 yes 2 0.58172 yes 3 0.65265 yes 4 0.53711 yes 5 0.29095 yes 6 0.21215 yes 7 0.15944 yes 8 0.11768 yes 9 0.03872 no 10 0.02242 no 11 0.02237 no 12 0.01703 no

[0115] Step 3: Calculate the multi-scale weighted permutation entropy of the reconstructed acoustic vibration signal as the feature vector to characterize the fault characteristics of the hydropower unit; extract the multi-scale permutation entropy of acoustic vibration under different working conditions of the hydropower unit as follows: Figure 4 shown.

[0116] The calculation process of multi-scale weighted permutation entropy is as follows:

[0117] ①The basic theory of weighted permutation entropy algorithm is as follows:

[0118] (1) For a given signal sequence Using phase space reconstruction theory, multiple subsequences are reconstructed as follows:

[0119]

[0120] Where m is the embedding dimension, τ is the delay, and N is the signal length.

[0121] (2) In order to introduce the amplitude characteristics of the signal, calculate each subsequence The weight value ω i as follows:

[0122]

[0123] (3) For each signal subsequence Use the weight value ω i and permutation and combination patterns Represents characteristic information. There are K kinds of signal sequence arrangement modes in total, and each mode The probability weighted value of is:

[0124]

[0125] (4) The weighted permutation entropy value WPE(X,m,τ) of the signal sequence is finally calculated as follows:

[0126]

[0127] ②The basic theory of multi-scale weighted permutation entropy algorithm is as follows:

[0128] WPE is a single scale to measure the complexity and dynamics of data, while MWPE is to coarse-grain the time series and then calculate the WPE value at each scale, which is a quantitative description of the time series at different scales. For a time series x(i) of length N, coarse-grain processing is performed to obtain the coarse-grained series y (s) (j) as follows:

[0129]

[0130] Where s is the scale factor, Z + Is a positive integer.

[0131] Calculate the WPE values ​​at multiple scales for the coarse-grained sequence:

[0132] E MWPE (X,τ,s,m)=WPE(y (s) (j),m,τ) (67)

[0133] Step 4: Divide the feature vector into a training set and a test set. The training set is input into the fault diagnosis model based on the improved dung beetle algorithm to optimize the support vector machine for model training to obtain an optimized classifier. The support vector machine is very sensitive to the penalty factor c and the kernel function parameter g. It is input into the improved dung beetle algorithm based on PWLCM chaotic mapping, nonlinear boundary convergence factor, rolling ball behavior fused with dynamic golden sine, sparrow vigilant mechanism and parameter adaptive adjustment strategies to iteratively search for the best parameters and output the optimal parameters. The improved dung beetle algorithm process is as follows:

[0134] (1) PWLCM chaotic mapping initialization population

[0135] The PWLCM chaotic mapping, with its randomness, non-repetitiveness and ergodicity, makes the population distribution more uniform, which is better than the random method. The PWLCM chaotic mapping is described as follows:

[0136]

[0137] In the formula: The control parameter p∈(0,0.5) is generally taken as 0.4; the chaotic random value x(t)∈(0,1) is used to divide the piecewise function.

[0138] (2) Nonlinear boundary convergence factor

[0139] The reproduction and foraging behaviors of the dung beetle algorithm are restricted to a certain area, that is, the egg-laying area and the foraging area can be defined.

[0140]

[0141] Among them, R = 1-t / T, T is the maximum number of iterations, Lb * and Ub * They define the lower and upper bounds of the spawning area, respectively. * is the optimal position of the current population; X b represents the global optimal position, Lb b and Ub b Respectively represent the optimal lower and upper limits of the foraging area; the spawning and foraging areas are dynamically adjusted with the number of iterations, and the reset boundary convergence factor R is expressed as follows:

[0142]

[0143] Among them, t is the number of iterations of the current population, and T is the maximum number of iterations of the population.

[0144] (3) Rolling behavior integrating dynamic golden sine

[0145] The golden sine strategy is introduced to use the sine function to comprehensively search for potential areas, improve the convergence speed, and get rid of the limitation of local optimality. The update formula of the improved dung beetle rolling behavior is as follows:

[0146]

[0147]

[0148] In the formula, R1 determines the moving distance of the current individual, which is a random number between [0, 2π]; R2 determines the update direction of the current individual position in the next iteration, which is a random number between [0, π]; X a is the position of the current optimal individual, X c is the position in the i-th dimension, x i (t) represents the position information of the ith dung beetle in the tth iteration; k∈(0,0.2) is the perturbation coefficient, α is -1 or 1, b is a random number between (0,1), δ is a random number between [0,1], x1 and x2 are the golden section coefficients; τ is the golden section ratio, which is 0.618033;

[0149] In addition to improving the movement path of the rolling dung beetle, in order to further balance the exploration and development capabilities of the algorithm in iteration, the inertia weight w is introduced into the golden sine formula:

[0150]

[0151] Among them, w min is the minimum inertia weight, w max is the maximum inertia weight, t is the current iteration number, and T is the maximum iteration number.

[0152] (4) Sparrow Sentinel Mechanism

[0153] The vigilant strategy of the sparrow search algorithm has the function of reminding the population to take anti-predation actions, which is introduced into the dung beetle algorithm. At the same time, the shrinkage factor is introduced In order to balance the exploration and development capabilities, the algorithm has flexible search capabilities at different stages. The update formula of this strategy is as follows:

[0154]

[0155] Where: is the current global optimal position, T is the maximum number of iterations, and t is the current number of iterations; β is a normally distributed random number with mean 0 and variance 1, which is used to control the step length; K is the step length control parameter of the sparrow's moving direction; ε is the minimum constant to avoid the denominator being 0; f i is the fitness value of the i-th sparrow; f w 、f g are the worst fitness value and the current best fitness value respectively.

[0156] (5) Parameter Adaptive Adjustment

[0157] The small step length and long jump mechanism of Levy flight and the wide trajectory movement of Brownian motion, this combined strategy can effectively prevent the thieving dung beetle from quickly approaching the optimal individual in the early stage of iteration, and enable the thieving dung beetle to maintain a certain activity in the later stage of iteration.

[0158] The position of the thief dung beetle after the introduction of Brownian motion is updated as follows:

[0159] x i (t+1)=X b +S·g·(|x i (t)-X * |+|x i (t)-X b |)+R BM (77)

[0160] Where g is a 1×D random vector, S is a constant, and R BM is the Brownian motion strategy,

[0161] The position of the thief dung beetle after the introduction of Levy's flight has been updated to:

[0162] x i (t+1)=X b +S·g·(|xi (t)-X * |+|x i (t)-X b |)+R LF (78)

[0163] Among them, R LF Flying strategy for Levy.

[0164] Step 5: Input the test set into the trained classifier to identify the faults of the hydropower unit. Figure 5 and Figure 6 They are respectively the visualization classification diagram and confusion matrix diagram of the hydropower unit. It can be seen from the figure that the average precision rate of the fault diagnosis classifier of MSIDBO-SVM is 100%, the average recall rate is 100%, and the average F1 score value is 1. This method has good fault identification performance and provides useful guidance value for the vibration fault identification of hydropower units.

[0165] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for diagnosing vibration faults of a hydropower unit, characterized in that: The method comprises the following steps: Step 1: Use vibration sensors to obtain acoustic vibration signals of hydropower units under different operating conditions; Step 2: The acoustic vibration signal is modally decomposed using the improved adaptive noise complete ensemble empirical mode decomposition ICEEMDAN, the correlation coefficient of each modal component is calculated, the redundant components are removed, and a new acoustic vibration signal is reconstructed; Step 3: Calculate the multi-scale weighted permutation entropy of the reconstructed acoustic vibration signal as a feature vector to characterize the fault characteristics of the hydropower unit; Step 4: Divide the feature vector into a training set and a test set, input the training set into the fault diagnosis model based on the improved dung beetle algorithm to optimize the support vector machine, perform model training, and obtain an optimized classifier; Step 5: Input the test set into the trained classifier to identify the faults of the hydropower unit.

2. A method for diagnosing vibration faults of a hydroelectric unit according to claim 1, characterized in that: In step 2, the acoustic vibration signal is modally decomposed using the improved adaptive noise complete set empirical mode decomposition ICEEMDAN, the correlation coefficient of each modal component is calculated, the redundant components are removed, and a new acoustic vibration signal is reconstructed. The specific ICEEMDAN decomposition is as follows: Add i groups of Gaussian white noise to the original signal to form a new signal. Use EMD to calculate the average value of the upper and lower envelopes to obtain the first-order residual and modal components. The decomposition process is as follows: x i t=x(t)+β0E1(ξ i (t)) (1) r1=M(x i (t)) (2) β0=δ0std(x) / std(E1(ξ i (t))) (3) Among them, ξ i (t) represents Gaussian white noise with mean and unit variance of 0 and 1 respectively, which is used to adjust the required signal-to-noise ratio (SNR) between the added noise and the residual after adding the noise. When k ≥ 1, β k =δ0std(r k ), δ0 represents the inverse of the required SNR between the input signal sequence and the first added noise, std() is the noise standard deviation operator, and the operator E k () represents the kth modal component obtained by EMD decomposition, the operator M() represents the local mean of the signal envelope, and <> represents the average calculation; Construct a new decomposition signal r1(t)+ω1E2(ξ i (t)), the second-order residual is estimated by calculating the local mean through EMD decomposition, and the second modal component is defined as: Calculate the remaining order modal components k=3,4…,K and obtain the k-order residual: r k (t)= <M(r k-1 (t)+ω k-1 E k (x) i (t)))> (6) Calculate the k-order modal component: Repeat the above steps to get the next modal component. When the residual r k When it is a monotonic function, terminate the calculation; The correlation coefficient is a parameter indicator to measure the similarity between two signals. The larger the absolute value of the correlation coefficient between the modal component and the original signal, the higher the correlation and the more effective components it contains. Where N is the sample length, For x i The average value of the signal, for y i The average value of the signal is calculated by calculating the correlation coefficient of each mode, removing the redundant components and retaining the effective components of the vibration signal to reconstruct the signal.

3. A method for diagnosing vibration faults of a hydroelectric unit according to claim 1, characterized in that: In step 3, the multi-scale weighted permutation entropy of the reconstructed acoustic vibration signal is calculated as a feature vector to characterize the fault characteristics of the hydropower unit. The calculation process of the multi-scale weighted permutation entropy is as follows: The basic theory of weighted permutation entropy algorithm is as follows: For a given signal sequence Using phase space reconstruction theory, multiple subsequences are reconstructed as follows: Where m is the embedding dimension, τ is the delay, and N is the signal length; In order to introduce the amplitude characteristics of the signal, each subsequence is calculated The weight value ω i as follows: For each signal subsequence Use the weight value ω i and permutation and combination patterns Represents characteristic information. There are K kinds of signal sequence arrangement modes in total, and each mode The probability weighted value of is: Finally, the weighted permutation entropy value WPE(X,m,τ) of the signal sequence is calculated as follows: ②The calculation process of the multi-scale weighted permutation entropy algorithm is as follows: WPE is a single scale to measure the complexity and dynamics of data, while MWPE is to coarse-grain the time series and then calculate the WPE value at each scale. For a time series x(i) of length N, coarse-grain processing is performed to obtain the coarse-grained series y (s) (j) as follows: Where s is the scale factor, Z + is a positive integer; Calculate the WPE values ​​at multiple scales for the coarse-grained sequence: E MWPE (X,τ,s,m)=WPE(y (s) (j),m,τ) (15)。 4. A method for diagnosing vibration faults of a hydroelectric unit according to claim 1, characterized in that: In step 4, the feature vectors are divided into a training set and a test set. The training set is input into the fault diagnosis model of the support vector machine optimized based on the improved dung beetle algorithm for model training to obtain an optimized classifier. Specifically, the support vector machine is very sensitive to the penalty factor c and the kernel function parameter g. The training set is input into the dung beetle algorithm improved based on strategies such as PWLCM chaotic mapping, nonlinear boundary convergence factor, rolling ball behavior fused with dynamic golden sine, sparrow vigilant mechanism and parameter adaptive adjustment to iteratively search for the best parameters and output the optimal parameters. The improved dung beetle algorithm process is as follows: Initialization population of PWLCM chaotic map The PWLCM chaotic mapping is described as follows: Where: The control parameter p∈(0,0.5) is 0.4; the chaotic random value x(t)∈(0,1) is used to divide the piecewise function; Nonlinear boundary convergence factor The reproduction and foraging behaviors of the dung beetle algorithm are restricted to a certain area, that is, the spawning area and foraging area can be defined Among them, R = 1-t / T, T is the maximum number of iterations, Lb * and Ub * They define the lower and upper bounds of the spawning area, respectively. * is the optimal position of the current population; X b represents the global optimal position, Lb b and Ub b Respectively represent the optimal lower and upper limits of the foraging area; the spawning and foraging areas are dynamically adjusted with the number of iterations, and the reset boundary convergence factor R is expressed as follows: Among them, t is the number of iterations of the current population, and T is the maximum number of iterations of the population; Rolling ball behavior with dynamic golden sine The golden sine strategy is introduced to use the sine function to comprehensively search for potential areas, improve the convergence speed, and get rid of the limitation of local optimality. The update formula of the improved dung beetle rolling behavior is as follows: In the formula, R1 determines the moving distance of the current individual, which is a random number between [0, 2π]; R2 determines the update direction of the current individual position in the next iteration, which is a random number between [0, π]; X a is the position of the current optimal individual, X c is the position in the i-th dimension, x i (t) represents the position information of the ith dung beetle in the tth iteration; k∈(0,0.2) is the perturbation coefficient, α is -1 or 1, b is a random number between (0,1), δ is a random number between [0,1], x1 and x2 are the golden section coefficients; τ is the golden section ratio, which is 0.618033; In addition to improving the movement path of the rolling dung beetle, in order to further balance the exploration and development capabilities of the algorithm in iteration, the inertia weight w is introduced into the golden sine formula: Among them, w min is the minimum inertia weight, w max is the maximum inertia weight, t is the current iteration number, and T is the maximum iteration number; Sparrow Vigilante Mechanism The vigilant strategy of the sparrow search algorithm has the function of reminding the population to take anti-predation actions. It is introduced into the dung beetle algorithm. At the same time, the shrinkage factor is introduced. In order to balance the exploration and development capabilities and enable the algorithm to have flexible search capabilities at different stages, the strategy update formula is as follows: Where: is the current global optimal position, T is the maximum number of iterations, and t is the current number of iterations; β is a normally distributed random number with mean 0 and variance 1, which is used to control the step length; K is the step length control parameter of the sparrow's moving direction; ε is the minimum constant to avoid the denominator being 0; f i is the fitness value of the i-th sparrow; f w 、f g are the worst fitness value and the current best fitness value respectively; Parameter adaptive adjustment The small step length and long jump mechanism of Levy flight and the wide trajectory movement of Brownian motion, this combined strategy effectively prevents the thieving dung beetle from quickly approaching the optimal individual in the early stage of iteration, and keeps the thieving dung beetle active in the later stage of iteration. The position update of the thieving dung beetle after the introduction of Brownian motion is: x i (t+1)=X b +S·g·(|x i (t)-X * |+|x i (t)-X b |)+R BM (25) Where g is a 1×D random vector, S is a constant, and R BM is the Brownian motion strategy, The position of the thief dung beetle after the introduction of Levy's flight has been updated to: x i (t+1)=X b +S·g·(|x i (t)-X * |+|x i (t)-X b |)+R LF (26) Among them, R LF Flying strategy for Levy.

Images