Fault positioning and diagnosing method for hydraulic system of wave power generation device

Abstract

According to the fault positioning and diagnosis method for the hydraulic system of the wave energy power generation device, the fault of the hydraulic system is analyzed, the fault mechanism and the fault characteristics of the hydraulic system are analyzed according to the working form and the environmental characteristics, fault feature extraction is carried out, the chaos rough set theory is introduced into fault data processing and fault feature extraction, and the fault positioning and diagnosis accuracy of the hydraulic system is improved. The redundancy and complexity of data are reduced, and a theoretical basis is provided for selection of real machine monitoring points and improvement of monitoring efficiency through a TL-chaos rough set-based traits reduction method. According to the method, a support vector machine is adopted to carry out fault diagnosis on a system, an optimization genetic algorithm is adopted to optimize parameters of various kernel functions, and an effective method for improving the fault diagnosis precision is found through comprehensive comparison of fault feature extraction, kernel function selection, selection of different training sample numbers, selection of different sea condition training samples and the like. Through comprehensive comparison of diagnosis precision, effectiveness and pertinence of the method on fault diagnosis of the hydraulic system are proved.

Description

Technical Field

[0001] The present application relates to a wave energy power generation fault location diagnosis method, and in particular to a wave energy power generation device hydraulic system fault location diagnosis method, belonging to the technical field of clean power generation device fault diagnosis. Background Art

[0002] Wave energy comes from solar energy. It is the energy generated by the waves on the ocean surface. It is a kind of renewable energy. It is distributed in the vast ocean, with very rich reserves and high energy density. As a new energy source, the market prospects for the development and utilization of wave energy are very promising. Although wave energy has the advantages of high density, rich reserves, green environmental protection and low cost, it also has the disadvantages of poor stability, large periodic changes and low efficiency. Existing wave energy power generation devices generally have the disadvantages of low energy capture efficiency, poor stability of secondary conversion energy, poor survival ability in the marine environment, high failure rate, and difficulty in finding and troubleshooting faults. Due to these defects of existing wave energy power generation devices, wave energy is currently only developed on a small scale and it is difficult to commercialize it on a large scale. How to improve efficiency, enhance stability, improve reliability and fault detection efficiency is a technical problem that needs to be solved urgently in current wave energy power generation.

[0003] In view of the disadvantages of existing wave power generation devices, due to the poor stability and continuity of waves, the application of hydraulic systems to wave energy power generation devices can effectively solve the problem of poor stability of secondary conversion energy of the energy conversion system, give full play to the flexible and controllable characteristics of hydraulics, and achieve precise control of stiffness adjustment and effective balance of gravity.

[0004] The hydraulic system is a key system to ensure the effective operation of the wave energy power generation device. Once a fault occurs, the entire device will not work. The hydraulic system is a nonlinear system in which each component is in a closed operation state. It is difficult to extract fault information. After a fault occurs, it is difficult to find the cause of the fault and quickly and accurately locate the fault. In this power generation device composed of mechanical, hydraulic and electrical systems, the fault diagnosis difficulty of the hydraulic system is significantly higher than that of mechanical components, and the maintainability and detectability are also lower than those of the electrical system. The working characteristics also determine that the device works in the offshore surge area or around the island. The working environment is worse than that on land and indoors, and there is no obvious occupancy period in the working time. Moreover, once it is put into operation, it cannot be maintained and maintained in real time. Only regular inspections can ensure the reliable operation of the system. If a fault occurs, it is towed back to the shore and personnel go to sea to eliminate it. The cost of towing back to the shore for troubleshooting is high, so this method is generally not used. For some faults of various components of the hydraulic system, personnel go to sea for troubleshooting. The troubleshooting environment is harsh and the maintenance space is small, which puts forward high requirements for on-site judgment, analysis, and troubleshooting. How to improve the efficiency of fault detection and the accuracy of fault location, so as to effectively eliminate faults and ensure the reliable operation of the system, is a problem that needs to be considered in the early stage of the development of the power generation device, and is also a technical problem that must be solved in the current technical verification, trial operation and later investment and adoption process. Therefore, it is very necessary and practical to conduct fault analysis and diagnosis research on the hydraulic system of the device.

[0005] The problems that need to be solved by the wave power generation fault location diagnosis method in the prior art and the key technical difficulties of this application include:

[0006] (1) Once the hydraulic system fails, the wave energy power generation device will not work. The hydraulic system of the prior art is a nonlinear system in which each component is in a closed operating state. It is difficult to extract fault information. After a fault occurs, it is difficult to find the cause of the fault and quickly and accurately locate the fault. The fault diagnosis difficulty of the hydraulic system in the power generation device is significantly higher than that of mechanical components, and the maintainability and detection are also lower than those of the electrical system. The working characteristics also determine that the device works in the offshore surge area or around the island. The working environment is worse than that on land and indoors. There is no obvious occupancy period in the working time. Once it is put into operation, it cannot be maintained in real time. Only regular inspections can ensure the reliable operation of the system. If a fault occurs, it is towed back to the shore and personnel go to sea to eliminate it. The cost of towing back to the shore for troubleshooting is high. If personnel go to sea for troubleshooting, the troubleshooting environment is harsh and the maintenance space is small. It requires high requirements for on-site judgment, analysis, and troubleshooting. How to improve the efficiency of fault detection and the accuracy of fault location, so as to efficiently eliminate faults and ensure the reliable operation of the system, is a problem that needs to be considered in the early stage of the development of the power generation device, and it is also a technical problem that must be solved in the current technical verification, trial operation and later investment and adoption process.

[0007] (2) For hydraulic equipment that works for a long time in the special environment of the ocean, the requirements for discovering and eliminating faults are high. The existing technology lacks a good method for fault analysis and diagnosis of the hydraulic system of the wave energy power generation device. The actual machine is unable to fully collect fault data. It is impossible to simulate and analyze the fault conditions of each hydraulic component in the system, extract fault characteristics, find the correlation between data, and remove redundant data. By comparing the simplified conditions of conditions under different sea conditions, it is impossible to obtain the selection range of system monitoring points based on fault diagnosis. The accuracy of hydraulic system fault diagnosis is very low. The influence of different numbers of training samples, single sea condition and multiple sea condition training samples on the accuracy of fault diagnosis is unclear. There is a lack of diagnostic methods suitable for this hydraulic system. The existing technology cannot provide a theoretical basis and practical guidance for fault diagnosis and elimination.

[0008] (3) The existing technology is unable to analyze the faults of the hydraulic system of the wave energy power generation device, unable to improve the efficiency of fault diagnosis, and unable to ensure that the power generation device can operate stably and reliably in ground experiments and sea trials. It lacks the impact of the failure of each hydraulic component on the system based on the working characteristics of the system and each hydraulic component of the system. There is a lack of simulation models for the three subsystems of the hydraulic system. There is a lack of establishing conditional feature sets based on each monitoring data, and establishing decision feature sets based on each system fault and normal state. The method of extracting fault features based on chaos rough sets is not used. It is impossible to use Matlab as a calculation tool to obtain the discrimination matrix of each fault sample and synthesize it through normalization, similarity matrix, and lower approximation set, and it is impossible to obtain the range of system monitoring points based on fault diagnosis. There is a lack of analysis of fault feature extraction, kernel function selection, GA algorithm optimization penalty coefficient and kernel function parameters, as well as the impact of different numbers of training samples, single sea condition and multiple sea condition training samples on the accuracy of fault diagnosis, resulting in inaccurate fault location and diagnosis. Summary of the invention

[0009] This application analyzes the faults of the hydraulic system of the resonant wave energy power generation device, studies the causes and mechanisms of the faults, and uses the AMESim simulation software as a platform. According to the actual parameters of each hydraulic component, simulation models of the transducer system, stiffness adjustment system, and gravity balancer are built, and fault injection is performed based on the simulation model. The fault conditions of each hydraulic component in the system are simulated and analyzed. Each monitoring point is used as a conditional feature set, and the normal and fault states are used as decision feature sets. The chaos rough set theory is used to extract fault features, and the correlation between data is found through mathematical methods to remove redundant data. By comparing the conditional feature reduction under different sea conditions, the selection range of system monitoring points based on fault diagnosis is obtained, which provides guidance for the selection of real machine monitoring points. The genetic algorithm is used to optimize the parameters of each kernel function, and the influence of different kernel functions on the fault diagnosis accuracy of the hydraulic system is studied. Based on the influence of different numbers of training samples, single sea condition and multi-sea condition training samples on the fault diagnosis accuracy, the most suitable diagnosis method for the hydraulic system is found through comparison.

[0010] In order to achieve the above technical effects, the technical solutions adopted in this application are as follows:

[0011] The fault location and diagnosis method of the hydraulic system of the wave energy power generation device is to simulate the fault conditions and their impacts of each hydraulic component of the system through the AMESim simulation software, introduce chaos rough sets into fault data processing and fault feature extraction, find the correlation between data, and then use the T-based L - The characteristic reduction method of chaotic rough sets simplifies the conditional characteristics, uses support vector machines to diagnose system faults, optimizes genetic algorithms to optimize the parameters of various kernel functions, and finds effective methods to improve the accuracy of fault diagnosis through comprehensive comparison of fault feature extraction, kernel function selection, selection of different numbers of training samples, and selection of training samples under different sea conditions;

[0012] P1: According to the working characteristics of the hydraulic system and each hydraulic component of the system, analyze the impact of each hydraulic component failure on the system, and build simulation models of the three subsystems of the hydraulic system based on the physical parameters and failure mechanisms of the hydraulic components;

[0013] P2: Establish conditional characteristic sets based on each monitoring data, establish decision characteristic sets based on each system fault and normal state, adopt the fault feature extraction method based on chaos rough set, use Matlab as the calculation tool, through normalization, similarity matrix, lower approximation set, and then find the discrimination matrix of each fault sample, and synthesize it. Use the synthesized total identification to find the importance of each conditional characteristic and simplify it. By comparing the simplification of conditional characteristics under different sea conditions, the range of system monitoring points based on fault diagnosis is obtained;

[0014] P3: The fault diagnosis method of SVM is used to diagnose the fault of the hydraulic system, analyze the fault feature extraction, kernel function selection, GA algorithm optimization penalty coefficient and kernel function parameters, as well as the influence of different numbers of training samples, single sea condition and multiple sea condition training samples on the fault diagnosis accuracy. By comparison, the most suitable model, parameters and training sample selection method for the hydraulic system are selected, and Matlab software is used to build a visual fault diagnosis interface.

[0015] Preferably, the determination of conditional characteristics and decision characteristics: the hydraulic system adopts T L - Chaotic rough sets are used to extract fault features, select the data of each monitoring point as the conditional characteristics, and use the fault cause as the decision-making characteristic, and find the correlation between the fault phenomenon and the fault cause;

[0016] Adopt T L - Chaotic rough sets perform global trait reduction by using T L The residual implication and lower approximation operator are used for calculation. The specific calculation steps include:

[0017] (1) Standardization of original data: The trait value of the conditional trait variable is converted to the interval [0,1]. The conversion formula is:

[0018]

[0019] x ij is the element of the matrix obtained by normalizing the original matrix, d ij is the jth characteristic variable value of the i-th sample of the original data, d j min is the minimum value of the jth characteristic variable, d j max is the maximum value of the jth characteristic variable;

[0020] (2) Calculate the similarity matrix sim(R) and the lower approximation set: For each chaotic characteristic C k , define a T L - Chaotic Similarity Relationship R k For R k (x i , x j )=1-|C k (x i )-C k (x j )|, then the elements in the similarity matrix sim(R) are:

[0021] R k (x i , x j )=min{1-|C k (x i)-C k (x j )|}

[0022] Based on the calculation of the similarity matrix sim(R), the lower approximation set is calculated:

[0023] Sim(R) g (K)(x)=inf u∈U θ(Sim(R)(x,u),K(u))

[0024] in inf means taking the minimum value. When u is in K, K(u)=1, and when u is not in K, K(u)=0. The above formula is expressed as

[0025] (3) Calculate the identification matrix to simplify the attributes: Use the identification matrix to simplify the attributes. Let U = {x1, x2, ..., x n}, R={R1, R2, …, R m},M D (U, R) represents an n×n matrix {c ij}, is the identification matrix of (U, RUD), defined as:

[0026] c ij ={R∈R:T(R(x i , x j ),λ(x i ))=0}, if

[0027] If x j ∈[x i ] D ∈K

[0028] Where λ(x)= Yes(R) θ (K)(x),T L (x, y) = max{0, x+y-1},

[0029] In the identification matrix, element c ij In the calculation formula, if x i With x j In the same decision class, If x i With x j If they are not in the same decision class, then find the following c ij Conditional attribute c k That is c ij Elements:

[0030] c ij ={R∈R:max(0,R(xi , x j )+Sim(R) θ (K)(x i ) 1)=0}

[0031] Rk satisfies R(x i , x j )+Sim(R) θ (K)(x i )≤1 by calculating c k In identifying the matrix element c ij The importance is obtained by summing up the proportions in , removing the element corresponding to the attribute with the largest importance, and continuing to calculate the importance until the identification matrix is ​​empty, and then simplifying it according to the importance.

[0032] Preferably, the trait reduction method:

[0033] (1) Normalization: When the wave cycle is 6s, 200 samples are selected for each of the 24 decision characteristics in the hydraulic system decision set to establish a 4800×38 raw data matrix, which is then normalized to obtain a normalized matrix.

[0034] (2) Calculate the similarity matrix: According to the similarity matrix calculation method, obtain the similarity matrix. The similarity matrix is ​​a 4800 × 4800 symmetric matrix. Each element in the matrix represents the similarity between different samples. The elements on the diagonal are all 1, which represents the similarity between the sample and itself.

[0035] (3) Find the lower approximate set: Take 200 samples for each decision feature. Among the 24 decision features of the hydraulic system, the normal state decision is X1 = {x1, x2, … x 200}, the energy conversion system intake decision is X2 = {x 201 ,x 202 ,…x 400},…, the air intake decision of the rigid system is changed to X 24 ={x 2201 ,x 2202 ,…x 2400},but:

[0036]

[0037] (4) Calculate the recognition matrix and importance: There are 24 items in the decision set. According to the calculation, 24 4800×4800×38 recognition matrices M are obtained. {1} (U,R),M {2} (U,R),…,M {24} (U, R), and merge them into a 4800×4800×38 total recognition matrix M {SUM} (U,R),x i With xj In the same decision class, The identification matrix is ​​a sparse matrix;

[0038] The number of columns in the matrix represents 38 conditional traits. 1 means that the conditional trait is in the identification matrix, and 0 means that it is not in the identification matrix. The following partial identification matrix is ​​constructed based on the calculation results:

[0039]

[0040] The importance of each conditional trait is calculated through the identification matrix. 200 samples are selected for each decision trait under 4s and 8s sea conditions, and the trait simplification is performed to calculate the importance of each conditional trait.

[0041] Under the three sea conditions of 4s, 6s, and 8s, 200 samples were selected from each decision trait, totaling 14,400 samples, and trait reduction was performed to calculate the importance of each decision trait.

[0042] Preferably, the monitoring point selection is based on fault feature extraction: the operation of the hydraulic system is monitored from the perspective of fault diagnosis. When selecting monitoring points, every hydraulic component is monitored as much as possible, and the extracted parameters are used so that the fault condition of the component can be characterized by the parameters of the monitoring point and the fault is accurately located. The focus of fault diagnosis is to find reasonable monitoring points. Based on the working characteristics of the power generation device, most of the monitoring points are arranged in the hydraulic system, and are selected and arranged based on experience, experiments and debugging requirements.

[0043] The operation of the device is analyzed based on the data and the chaos rough set method is used to simplify the conditional characteristics through comprehensive comparison of similar situations.

[0044] The monitoring points are determined based on the simplification of the characteristics of fault sample conditions under different sea conditions and the actual situation during equipment operation.

[0045] 5. The method for locating and diagnosing faults in the hydraulic system of a wave energy power generation device according to claim 1 is characterized by introducing an improved support vector machine for hydraulic system fault diagnosis: introducing an improved support vector machine learning method to improve classification accuracy in the case of small samples, processing each sample through a chaotic rough set to extract data features, and using a genetic algorithm to optimize the kernel function parameters to further improve sample recognition and fault detection efficiency;

[0046] An improved SVM is introduced for fault diagnosis. By maximizing the interval, the classification problem is transformed into a convex quadratic programming problem. Finally, an optimal hyperplane that can correctly divide different samples is found, and the distance between this hyperplane and each classification is also the largest.

[0047] Preferably, the fault diagnosis of the hydraulic system of the wave energy power generation device: based on the fault analysis and simulation data of the hydraulic system of the power generation device, a fault diagnosis platform is built under the Matlab software, the LibSVM software package is used as a tool, and the GA algorithm toolkit is used to perform parameter optimization, and the fault diagnosis of the hydraulic system of the power generation device is performed to find the faulty parts of the power generation device including the hydraulic transducer system, the stiffness adjustment system, and the gravity balancer;

[0048] In the process of building the fault diagnosis model, the conditional characteristics of the hydraulic system are used as the input of the fault diagnosis model, and the decision characteristics are used as the output of the fault diagnosis model. Suppose the sample set of the hydraulic system is m represents the number of samples, x i Represents the input vector, corresponding to the conditional trait set of the i-th sample. For the original sample set, the input amount is x i =[x i1 , x i2 , …x i38 ], x i1 , x i2 , …x i38 ∈R respectively corresponds to the conditional feature set of the i-th sample as {C1, C2, ..., C 38}, for the sample set after chaos rough set reduction, the input is x i ={x i1 , x i2 , …x i15 ], the conditional feature set corresponding to the i-th sample is {C1, C2, …, C 15},y i represents the decision feature corresponding to the i-th sample, y∈{D1, D2, …, D 24}, a one-to-one classification method is used to establish a support vector machine classifier for the hydraulic system. There are 24 normal and fault states of the hydraulic system. In the one-to-one classification method, there are M types of problems that need to be classified. The number of classifiers should be (M-1)M / 2. For this hydraulic system, 276 two-class classifiers are required;

[0049] Under the three sea conditions of 4s, 6s and 8s, 200 samples are selected from the normal and each type of fault state data samples, totaling 14,400 samples, for support vector machine training. Under other sea conditions, 20 samples are selected from the normal and each type of fault state data samples, totaling 480 samples, for support vector machine detection.

[0050] Preferably, the chaotic rough set simplified fault diagnosis: adopt C-SVM model and Gaussian radial basis kernel function, adopt Libsvm toolkit to establish the fault diagnosis model, and the model structure generated by the unreduced data set is:

[0051]

[0052]

[0053] The model structure generated by the simplified data set:

[0054]

[0055] Chaotic rough sets are used to extract fault features, remove redundant data, reduce data dimensions, shorten training and detection time, and reduce the number of support vectors while keeping classification accuracy small.

[0056] Preferably, the diagnostic accuracy of different kernel functions and parameters is: the parameter of the polynomial kernel function has the order d, and the parameter of the RBF kernel function is the kernel width g (1 / σ 2 ), the parameters of the Sigmoid kernel function are v, c. The GA algorithm is used to optimize the penalty parameter C and different kernel function parameters. Under different kernel functions, 20 samples are randomly selected from the normal and each type of fault state data samples, a total of 480 samples, for support vector machine detection;

[0057] (1) Polynomial kernel function

[0058] The initial range of the penalty parameter C is set to [0,100], d is 1 to 9, the penalty parameter and the order are binary coded, and the binary coded parameters C and d are substituted into the SVM model to train the samples, perform the simulated biological evolution process of selection, crossover, and mutation, and determine whether the termination condition is met;

[0059] When it was inherited to the 89th generation, the best fitness tended to be stable and reached 89.2%. The optimization results of the polynomial kernel function were as follows: the order d was 2, the penalty parameter C was 26.2154, and the optimized kernel function model was used to classify the test samples with a classification accuracy of 83.3%.

[0060] (2) RBF kernel function

[0061] The penalty parameter C of the RBF kernel function ranges from [0,100], and the kernel function parameter g ranges from [0,1000]. When it is inherited to the 92nd generation, the best fitness tends to be stable, reaching 95.1%. The RBF kernel function optimization results are: C is 46.9587, g is 2.6541, and the optimized kernel function model is used to classify the test samples with a classification accuracy of 92.9%.

[0062] Preferably, the influence of the number of selected training samples on diagnosis is as follows: 50, 100, 200, 300 samples are randomly selected from the normal and each type of fault state data samples under the three sea conditions of 4s, 6s, and 8s, totaling 3600, 7200, 14400, and 21600 samples for support vector machine training; when the wave period is 7.5s, 20 samples are randomly selected from the normal and each type of fault state data samples, totaling 480 samples, for support vector machine detection;

[0063] When the number of training samples is small, it has a greater impact on the detection accuracy. As the number of training samples increases, the detection accuracy also increases accordingly and tends to be stable. When the number of training samples is small, it affects the detection accuracy. When there are too many training samples, it has little effect on improving the detection accuracy.

[0064] Preferably, the influence of different sea conditions on the diagnostic accuracy: considering the influence of sea conditions, three sea conditions of 4s, 6s and 8s are selected, 200 samples are randomly selected from the normal and each type of fault state data samples, a total of 14400 samples are used for the training of the support vector machine, and the sea conditions of 4s, 6s and 8s are considered separately. 600 samples are randomly selected from the normal and each type of fault state data samples, a total of 14400 samples are used for the training of the support vector machine. When the wave period is 7.5s, 20 samples are randomly selected from the normal and each type of fault state data samples, a total of 480 samples are used for the detection of the support vector machine;

[0065] It can be seen from the diagnosis results that when only one sea condition is considered, the greater the difference between the sea condition of the diagnosis sample and the detection sample, the lower the diagnosis accuracy. Considering the working sea condition range of the power generation device, the diagnosis accuracy of the training set selected when the sea conditions are 4s, 6s, and 8s meets the fault diagnosis requirements.

[0066] Compared with the prior art, the innovations and advantages of this application are:

[0067] (1) This application analyzes the faults of the hydraulic system of the resonant wave energy power generation device, studies the causes and mechanisms of the faults, and uses the AMESim simulation software as a platform. According to the actual parameters of each hydraulic component, simulation models of the transducer system, stiffness adjustment system, and gravity balancer are built, and fault injection is performed based on the simulation model. The fault conditions of each hydraulic component in the system are simulated and analyzed. Each monitoring point is used as a conditional feature set, and the normal and fault states are used as decision feature sets. The chaos rough set theory is used to extract fault features, and the correlation between data is found through mathematical methods to remove redundant data. By comparing the conditional feature reduction under different sea conditions, the selection range of system monitoring points based on fault diagnosis is obtained, which provides guidance for the selection of monitoring points for the actual machine. The genetic algorithm is used to optimize the parameters of each kernel function, and the influence of different kernel functions on the fault diagnosis accuracy of the hydraulic system is studied. Based on the influence of different numbers of training samples, single sea condition and multi-sea condition training samples on the fault diagnosis accuracy, the most suitable diagnosis method for the hydraulic system is found through comparison. Both experimental and actual troubleshooting results show that the proposed method can provide an effective theoretical basis and practical guidance for fault diagnosis and troubleshooting.

[0068] (2) Based on the experimental and debugging data of the hydraulic system of the resonant wave energy power generation device, this application is developed from four aspects: the fault mechanism and fault simulation of the device, fault feature extraction method, fault diagnosis method, experiment and diagnosis. The fault of the hydraulic system is analyzed, and the fault mechanism and fault characteristics of the hydraulic system are analyzed according to the working form and environmental characteristics. Fault feature extraction is performed, and chaos rough set theory is introduced into fault data processing and fault feature extraction to reduce the redundancy and complexity of the data. L -The characteristic reduction method of chaotic rough sets provides a theoretical basis for the selection of real machine monitoring points and improving monitoring efficiency. The system is diagnosed for faults, and it is proposed to use support vector machines to diagnose the system faults, optimize the genetic algorithm to optimize the parameters of various kernel functions, and find an effective method to improve the accuracy of fault diagnosis through comprehensive comparison of fault feature extraction, kernel function selection, selection of different numbers of training samples, and selection of training samples for different sea conditions. Through the comprehensive comparison of diagnostic accuracy and the elimination of faults found in the experiment, the effectiveness and pertinence of this method for hydraulic system fault diagnosis are proved.

[0069] (3) This application analyzes the failure principle of each hydraulic component in the system, and the fault simulation model based on AMESim can more realistically reflect the system failure characteristics. Based on the simulation data, the conditional characteristic set is simplified using the chaos rough set theory, and the range of monitoring point selection is determined, which can be used as a reference when diagnosing faults on the actual machine. After the monitoring points are optimized, it is beneficial to improve the accuracy of fault diagnosis. Through simulation data verification, the RBF kernel function and samples of more than three sea conditions are selected for training, and the penalty coefficient C and the kernel function width g are determined by GA optimization, and the detection accuracy obtained is relatively high. Under this model, the actual machine experimental data is used as a training sample, and the diagnostic accuracy also meets the requirements. It also has a great guiding role in the troubleshooting of the actual machine. It can improve the efficiency of fault detection and the accuracy of fault location, thereby efficiently troubleshooting and ensuring the reliable operation of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 It is a schematic diagram of the conditions and characteristics of the hydraulic system of the wave energy power generation device.

[0071] Figure 2 It is a schematic diagram of the decision characteristics of the hydraulic system of the wave energy power generation device.

[0072] Figure 3 This is a schematic diagram of the importance of characteristics of each condition of the hydraulic system under three sea conditions.

[0073] Figure 4 It is a schematic diagram of monitoring points determined by simplifying the conditional characteristics of fault samples under different sea conditions.

[0074] Figure 5 This is a flow chart of SVM parameter optimization based on genetic algorithm optimization.

[0075] Figure 6 It is a flow chart of fault diagnosis of hydraulic system of wave energy power generation device.

[0076] Figure 7 It is a schematic diagram of the classification of unreduced samples.

[0077] Figure 8 It is a schematic diagram of the classification of simplified samples.

[0078] Fig. 9 It is a schematic diagram of the impact of fuzzy rough set simplification of the v-sVM model on fault diagnosis.

[0079] Fig.10 It is a schematic diagram of the evolutionary generation and fitness change curve.

[0080] Fig.11 It is a schematic diagram of the evolutionary generation and fitness change curve.

[0081] Fig.12It is a schematic diagram of sensor distribution before and after optimization. DETAILED DESCRIPTION

[0082] The following further describes the technical solution of the fault location and diagnosis method of the hydraulic system of the wave energy power generation device provided by the present application in conjunction with the accompanying drawings, so that those skilled in the art can better understand the present application and implement it.

[0083] With the country's vigorous advocacy of clean energy utilization, wave energy as a clean energy with abundant reserves has attracted more and more attention. Research on wave energy power generation is increasingly popular, and various wave energy power generation devices are emerging. Hydraulic systems are widely used in various wave energy power generation devices due to mature technology, good transmission, control and other performance. For hydraulic equipment that works for a long time in the special environment of the ocean, the requirements for discovering and troubleshooting are high. Therefore, it is very necessary to conduct fault analysis and diagnosis research on the hydraulic system of wave energy power generation devices.

[0084] Taking the hydraulic system of the resonant wave energy power generation device as the research object, the fault of the system was analyzed, and the cause and mechanism of the fault were studied. In view of the limitation that the actual machine cannot fully collect fault data, AMESim simulation software is used as a platform. According to the actual parameters of each hydraulic component, simulation models of the energy conversion system, stiffness adjustment system, and gravity balancer are built, and fault injection is performed based on the simulation model to simulate and analyze the fault conditions of each hydraulic component in the system.

[0085] Taking the monitoring points of the simulation model as the conditional feature set, and the normal and fault states as the decision feature set, the chaos rough set theory is used to extract fault features. The correlation between data is found through mathematical methods, and redundant data is removed. By comparing the simplification of conditional features under different sea conditions, the selection range of system monitoring points based on fault diagnosis is obtained, which provides theoretical guidance for the selection of actual machine monitoring points.

[0086] The fault diagnosis method based on support vector machine is studied, and the parameters of each kernel function are optimized through genetic algorithm, and the influence of different kernel functions on the fault diagnosis accuracy of the hydraulic system is studied. The influence of different numbers of training samples, single sea condition and multiple sea condition training samples on the fault diagnosis accuracy is studied, and the most suitable diagnostic method for the hydraulic system is found through comparison.

[0087] By conducting fault simulation, training and diagnosis on the actual machine, it is verified that the use of chaos rough set theory to simplify simulation data can provide theoretical guidance for the selection of monitoring points for the actual machine. The feasibility of using support vector machines for fault diagnosis is verified. Both experimental and actual troubleshooting results show that this method can provide theoretical basis and practical guidance for fault diagnosis and troubleshooting.

[0088] 1. Fault feature extraction of hydraulic system of wave energy power generation device

[0089] 1. Determination of conditional characteristics and decision characteristics

[0090] The hydraulic system adopts T L - Chaotic rough set is used to extract fault features, select the data of each monitoring point as the conditional characteristics, and use the cause of the fault as the decision-making characteristic (such as Figure 1 , Figure 2 ), more clearly and intuitively describe each fault of the hydraulic system, find the correlation between the fault phenomenon and the fault cause, and also facilitate the corresponding fault diagnosis with the actual machine.

[0091] Adopt T L - Chaotic rough sets perform global trait reduction by using T L The residual implication and lower approximation operator are used for calculation. The specific calculation steps include:

[0092] (1) Standardization of original data: The trait value of the conditional trait variable is converted to the interval [0, 1]. The conversion formula is:

[0093]

[0094] x ij is the element of the matrix obtained by normalizing the original matrix, d ij is the jth characteristic variable value of the i-th sample of the original data, d j min is the minimum value of the jth characteristic variable, d j max is the maximum value of the jth characteristic variable;

[0095] (2) Calculate the similarity matrix sim(R) and the lower approximation set:

[0096] For each chaotic trait C k , define a T L - Chaotic Similarity Relationship R k For R k (x i , x j )=1-|C k (x i )-C k (x j )|, then the elements in the similarity matrix sim(R) are:

[0097] R k (x i , X j )=min{1-|( k (x i )-C k (x j )|}

[0098] Based on the calculation of the similarity matrix sim(R), the lower approximation set is calculated:

[0099] Yes(R) θ (K)(x)=inf u∈U θ(Sim(R)(x,u),K(u))

[0100] where θ L (x, y) = min{1, 1-x+y}, inf means taking the minimum value. When u is in K, K(u) = 1, and when u is not in K, K(u) = 0. The above formula is expressed as Sim(R) θ (K)(x)=min{min u∈U (1,1-Sim(R)(x,u)+K(u))}.

[0101] (3) Calculate the identification matrix to simplify attributes

[0102] Using the identification matrix calculation to simplify, let U = {x1, x2, ..., x n}, R={R1, R2, …, R m},M D (U, R) represents an n×n matrix {c ij}, is the identification matrix of (U, RUD), defined as:

[0103] c ij ={R∈R:T(R(x i , x j ),λ(x i ))=0}, if

[0104] If x j ∈[x i ] D ∈K

[0105] Where λ(x)= Yes(R) θ (K)(x),T L (x, y) = max{0, x+y-1},

[0106] In the identification matrix, element c ij In the calculation formula, if x i With x j In the same decision class, If x i With x j If they are not in the same decision class, then find the following c ij Conditional attribute c k That is c ij Elements:

[0107] c ij ={R∈R:max(0,R(x i , x j )+Sim(R) θ (K)(x i )-1)=0}

[0108] Rk satisfies R(x i , x j )+Sim(R) θ (K)(x i )≤1 by calculating c k In identifying the matrix element c ij The importance is obtained by summing up the proportions in , removing the element corresponding to the attribute with the largest importance, and continuing to calculate the importance until the identification matrix is ​​empty, and then simplifying it according to the importance.

[0109] (II) Trait Reduction Method

[0110] (1) Normalization

[0111] When the wave cycle is 6s, 200 samples are selected for each of the 24 decision characteristics in the hydraulic system decision set to establish a 4800×38 original data matrix, which is normalized to obtain a normalized matrix.

[0112] (2) Find the similarity matrix

[0113] According to the similarity matrix calculation method, the similarity matrix is ​​obtained. The similarity matrix is ​​a 4800×4800 symmetric matrix. Each element in the matrix represents the similarity between different samples. The elements on the diagonal are all 1, representing the similarity between the sample and itself.

[0114] (3) Find the lower approximate set

[0115] 200 samples are taken for each decision trait. Among the 24 decision traits of the hydraulic system, the normal state decision is X1 = {x1, x2, ... x 200}, the energy conversion system intake decision is X2 = {x 201 ,x 202 ,…x 400},…, the air intake decision of the rigid system is changed to X 24 ={x 2201 ,x 2202 ,…x 2400},but:

[0116]

[0117] (4) Calculate the recognition matrix and importance

[0118] The decision set has 24 items, and according to the calculation, 24 4800×4800×38 discriminative matrices M are obtained. {1} (U,R),M {2} (U,R),…,M {24} (U, R), and merge them into a 4800×4800×38 total recognition matrix M {SUM} (U,R),x i With x j In the same decision class, The identification matrix is ​​a sparse matrix.

[0119] The number of columns in the matrix represents 38 conditional traits. 1 means that the conditional trait is in the identification matrix, and 0 means that it is not in the identification matrix. The following partial identification matrix is ​​constructed based on the calculation results:

[0120]

[0121] The importance of each conditional trait is calculated through the identification matrix. 200 samples are selected for each decision trait under 4s and 8s sea conditions, and the trait simplification is performed to calculate the importance of each conditional trait.

[0122] Under the three sea conditions of 4s, 6s, and 8s, 200 samples are selected from each decision trait, totaling 14,400 samples for trait simplification calculation. The importance of each decision trait is as follows: Figure 3 shown.

[0123] 2. Monitoring point selection based on fault feature extraction

[0124] The operation of the hydraulic system is monitored from the perspective of fault diagnosis. When selecting monitoring points, every hydraulic component is monitored as much as possible. The extracted parameters can be used to characterize the fault condition of the component through the parameters of the monitoring points, thereby accurately locating the fault. The focus of fault diagnosis is to find reasonable monitoring points. Based on the working characteristics of the power generation device, most of the monitoring points are arranged in the hydraulic system, and are selected and arranged based on experience, experiments and debugging requirements.

[0125] The operation of the device is analyzed based on the data and the chaos rough set method is used to simplify the conditional characteristics through comprehensive comparison of similar situations.

[0126] Based on the simplification of the characteristics of fault samples under different sea conditions and the actual situation of equipment operation, the monitoring points are determined as follows: Figure 4 shown.

[0127] 3. Introduction of improved support vector machine for hydraulic system fault diagnosis

[0128] The available sample size for fault diagnosis of wave power generation devices is relatively limited. This is because the fault occurs suddenly and it is difficult to simulate the fault on the actual machine. Moreover, compared with the massive monitoring data of some mature large hydraulic equipment, the fault sample size of this device is still relatively small. The learning method of improved support vector machine is introduced to improve the classification accuracy in the case of small samples. The chaotic rough set is used to process each sample and extract data features. At the same time, the genetic algorithm is used to optimize the parameters of the kernel function to further improve the sample recognition and fault detection efficiency.

[0129] Fault diagnosis for locating fault samples in hydraulic systems is a classification problem. The hydraulic system is a closed flow system. Under the action of hydraulic force, the various components of the system influence and relate to each other. There are many nonlinear factors, and the information obtained is also chaotic and rough. The diagnosis is difficult, and it is very difficult to accurately locate the fault.

[0130] An improved SVM is introduced for fault diagnosis. By maximizing the interval, the classification problem is transformed into a convex quadratic programming problem. Finally, an optimal hyperplane that can correctly divide different samples is found, and the distance between this hyperplane and each classification is also the largest.

[0131] (I) SVM parameter optimization based on genetic algorithm optimization

[0132] When using SVM to diagnose hydraulic system faults, the classification accuracy is greatly affected by the kernel function parameters and penalty factors. GA is used to optimize the SVM parameters to improve the classification recognition rate. The parameter optimization process is as follows: Figure 5 shown.

[0133] (II) Fault diagnosis of hydraulic system of wave energy generator

[0134] Based on the data of fault analysis and simulation of hydraulic system of power generation device, a fault diagnosis platform is built under Matlab software, LibSVM software package is used as a tool, and GA algorithm toolkit is used to optimize parameters and perform fault diagnosis on hydraulic system of power generation device. The fault parts of power generation device including hydraulic transducer system, stiffness adjustment system and gravity balancer are found. The fault diagnosis process is as follows: Figure 6 shown.

[0135] In the process of building the fault diagnosis model, the conditional characteristics of the hydraulic system are used as the input of the fault diagnosis model, and the decision characteristics are used as the output of the fault diagnosis model. Suppose the sample set of the hydraulic system is , m represents the number of samples, x i Represents the input vector, corresponding to the conditional trait set of the i-th sample. For the original sample set, the input amount is x i =[x i1 , xi2 , …x i38 ], x i1 , x i2 , …x i38 ∈R respectively corresponds to the conditional feature set of the i-th sample as {C1, C2, ..., C 38}, for the sample set after chaos rough set reduction, the input is x i ={x i1 , x i2 , …x i15 ], the conditional feature set corresponding to the i-th sample is {C1, C2, …, C 15},y i represents the decision feature corresponding to the i-th sample, y∈{D1, D2, …, D 24}, a one-to-one classification method is used to establish a support vector machine classifier for the hydraulic system. There are 24 normal and fault states of the hydraulic system. In the one-to-one classification method, there are M types of problems to be classified, and the number of classifiers should be (M-1)M / 2. For this hydraulic system, 276 two-class classifiers are required. Although this classification method requires relatively more classifiers in solving the classification problem of the hydraulic system, the scale of each two-class classifier is relatively small, and the number of support vectors is relatively small, which is more convenient to learn.

[0136] Under the three sea conditions of 4s, 6s and 8s, 200 samples are selected from the normal and each type of fault state data samples, totaling 14,400 samples, for support vector machine training. Under other sea conditions, 20 samples are selected from the normal and each type of fault state data samples, totaling 480 samples, for support vector machine detection.

[0137] 1. Fault diagnosis based on simplified chaos rough set

[0138] The C-SVM model and Gaussian radial basis kernel function are used, and the Libsvm toolkit is used to establish a fault diagnosis model.

[0139] The model structure generated by the unreduced dataset is:

[0140]

[0141] The model structure generated by the simplified data set:

[0142]

[0143]

[0144] Select the v-SVM model and Gaussian radial basis kernel function, use the unreduced parameters for training and testing, and use the reduced parameters for training and testing to obtain the fault diagnosis and testing results as follows: Figure 7 , Figure 8 , Fig. 9 shown.

[0145] By comparing the two models, it can be clearly seen that the use of chaos rough sets to extract fault features removes redundant data, reduces data dimensions, shortens training and detection time, and reduces the number of support vectors when the classification accuracy changes slightly. In the processing of hydraulic system fault diagnosis classification, chaos rough sets can be used to extract fault features. In the diagnostic comparison, it was also found that the C-SVM model and the v-SVM model are not much different in the diagnostic classification of the hydraulic system. This application uses the C-SVM model.

[0146] 2. Diagnostic accuracy of different kernel functions and parameter selections

[0147] The parameter of the polynomial kernel function is the order d, and the parameter of the RBF kernel function is the kernel width g (1 / σ 2 ), the parameters of the Sigmoid kernel function are v, c. The GA algorithm is used to optimize the penalty parameter C and different kernel function parameters. Under different kernel functions, 20 samples are randomly selected from the normal and each type of fault state data samples, totaling 480 samples, for support vector machine detection.

[0148] (1) Polynomial kernel function

[0149] The initial range of the penalty parameter C is set to [0,100], d is 1 to 9, the penalty parameter and the order are binary coded, and the binary coded parameters C and d are substituted into the SVM model to train the samples, perform the selection, crossover, and mutation simulation biological evolution process, and determine whether the termination condition is met. Fig.10 shown.

[0150] It can be seen that in the iterative process, when it is inherited to the 89th generation, the best fitness tends to be stable and reaches 89.2%. The optimization result of the polynomial kernel function is: the order d is 2, the penalty parameter C is 26.2154, and the optimized kernel function model is used to classify the test samples with a classification accuracy of 83.3%.

[0151] (2) RBF kernel function

[0152] The penalty parameter C of the RBF kernel function ranges from [0,100], and the kernel function parameter g ranges from [0,1000]. During the iteration process, when it is inherited to the 92nd generation, the best fitness tends to be stable and reaches 95.1%. The RBF kernel function optimization result is: C is 46.9587, g is 2.6541, and the optimized kernel function model is used to classify the test samples with a classification accuracy of 92.9%.

[0153] (3) Sigmod kernel function

[0154] The penalty parameter C of the Sigmod kernel function ranges from [0,100], the kernel function parameter v ranges from [0,1000], and the c ranges from [-100,0]. The optimization process is as follows Fig.11 shown.

[0155] It can be seen that in the iterative process, when it is inherited to the 90th generation, the best fitness tends to be stable and reaches 92.8%. The Sigmod kernel function optimization results are: v is 25.21, C is 42.9685, c is -2.6911, and the optimized kernel function model is used to classify the test samples with a classification accuracy of 88.1%.

[0156] Three different kernel functions are selected for classification detection. By comparison, it can be seen that the Gaussian radial basis kernel function has the highest optimal fitness and relatively high detection accuracy. Compared with other kernel functions, it is more suitable for the fault classification of the hydraulic system of this type of power generation device.

[0157] 3. The impact of the number of training samples on diagnosis

[0158] The power generation device is designed to operate with a wave period of 4-9s. In order to improve the classification accuracy, 50, 100, 200, and 300 samples, totaling 3600, 7200, 14400, and 21600 samples, are randomly selected from the normal and each type of fault state data samples under three sea conditions of 4s, 6s, and 8s for support vector machine training. When the wave period is 7.5s, 20 samples, totaling 480 samples, are randomly selected from the normal and each type of fault state data samples for support vector machine detection.

[0159] When the number of training samples is small, it has a greater impact on the detection accuracy. As the number of training samples increases, the detection accuracy also increases accordingly and tends to be stable. When the number of training samples is small, it affects the detection accuracy. When there are too many training samples, it has little effect on improving the detection accuracy.

[0160] 4. Impact of different sea conditions on diagnostic accuracy

[0161] Considering the influence of sea conditions, three sea conditions of 4s, 6s and 8s are selected, and 200 samples are randomly selected from the normal and each type of fault state data samples, totaling 14400 samples, which are used for support vector machine training. Considering the 4s, 6s and 8s sea conditions separately, 600 samples are randomly selected from the normal and each type of fault state data samples, totaling 14400 samples, which are used for support vector machine training. When the wave period is 7.5s, 20 samples are randomly selected from the normal and each type of fault state data samples, totaling 480 samples, are used for support vector machine detection.

[0162] It can be seen from the diagnosis results that when only one sea condition is considered, the greater the difference between the sea condition of the diagnosis sample and the detection sample, the lower the diagnostic accuracy. Considering the working sea condition range of the power generation device, the diagnostic accuracy can meet the fault diagnosis requirements when the training set selects 4, 6, and 8s sea conditions.

[0163] 5. Build a visual fault diagnosis interface based on Matlab software

[0164] The work of Matlab's gui interface program is implemented through the callback function callback. The program clicks the corresponding button, and then the program background executes the corresponding pushbutton_callback module.

[0165] 4. Hydraulic system fault diagnosis experiment

[0166] In this power generation device, a monitoring system is provided to enable the energy conversion system to efficiently extract and convert energy, and the energy capture system to accurately control the stiffness of the pendulum body and balance the weight of the pendulum body at all times. A PLC-based lower computer is used to complete the flow of the energy conversion system and the displacement of the variable stiffness cylinder, and to monitor the hydraulic system. The upper computer control program developed based on the LabVIEW platform is used to complete the functions of transmission, storage and display of monitoring data.

[0167] 1. Purpose of the experiment

[0168] (1) To verify the influence of the fault extraction method based on chaos rough set and the selection of monitoring points based on this method on the fault diagnosis accuracy of the hydraulic system;

[0169] (2) Verify whether the kernel function selection, parameter optimization, and training sample selection of SVM can meet the requirements of actual machine fault diagnosis;

[0170] (3) The fault diagnosis method is used to diagnose abnormal data that appears during the debugging and experimental process of the actual machine, and the feasibility of the fault diagnosis method is further verified through on-site troubleshooting.

[0171] (II) Experimental content and results analysis

[0172] (1) Selection of training samples and fault simulation

[0173] Since it is difficult to obtain fault samples on the actual machine, especially leakage faults such as leakage of the reversing valve, oil pump, and oil cylinder, the overflow valve and accumulator faults of the transducer system are selected as diagnostic experimental objects without affecting the reliability of the hydraulic components.

[0174] The data of normal operation under 6s sea conditions were collected. Under experimental conditions, the accumulator oil replenishment flow was controlled by the throttle valve opening connected to the accumulator outlet to simulate the accumulator leakage and the failure of pressure and flow stabilization. As the throttle valve opening decreased, the system pressure fluctuation gradually increased. When the throttle valve was closed, the system pressure and flow fluctuation rates were 51.16% and 7.41% respectively.

[0175] By adjusting the working pressure of the relief valve to below the system working pressure, the fault conditions of the relief valve overflowing prematurely due to reduced spring force of the relief valve, large internal leakage, and valve core stagnation are simulated. When the relief pressure is 6MPa, which is the same as the initial pressure of the accumulator, the pressure and flow fluctuation rates of the accumulator exiting the working system reach 92.68% and 25.11%, and violent fluctuations begin to occur.

[0176] The normal operation of the system and the failure of the accumulator and the overflow valve under the sea conditions of 4s, 6s and 8s were simulated respectively. 200 normal samples, accumulator failure samples and overflow valve failure samples were selected from the three sea conditions, totaling 1800 samples for training. Under the sea condition of 7s, 7 samples were selected from the accumulator failure samples and the overflow valve failure samples for testing.

[0177] (2) Comparison of diagnostic accuracy before and after monitoring point optimization

[0178] The conditional characteristics of the hydraulic system simulation data were simplified by the chaos rough set principle. The conditional characteristics importance data obtained after multiple simplifications were integrated to optimize the monitoring points in the experiment. Sensor layout before and after optimization, "√" indicates selection, "" indicates abandonment. (e.g. Fig.12 )

[0179] In addition, two transducer cylinder displacement sensors are added to detect transducer cylinder faults. The diagnostic accuracy of accumulator and overflow valve faults before and after the monitoring point optimization is compared. The diagnostic accuracy is 71.4% before the monitoring point optimization, and 85.7% after the monitoring point optimization by chaos rough set theory. Analysis of the diagnostic results shows that the fault feature extraction and fault diagnosis methods based on simulation data can basically meet the diagnostic requirements when applied to real machine diagnosis, and the simplification of conditional characteristics helps to improve the diagnostic accuracy.

[0180] (3) Fault measurement

[0181] During the system debugging and operation, a fault state occurred in which the pressure and flow rate decreased and the fluctuation rate increased. The pressure and flow fluctuation rates were 82.35% and 27.94% respectively.

[0182] Check that the displacement of the transducer cylinder is normal, and check that the throttle valve at the outlet of the accumulator is fully open. According to the monitoring data, the pressure is 6MPa lower than the initial pressure of the accumulator, and the accumulator has stopped working. Due to the small number of actual machine fault samples, the diagnosis of the fault diagnosis model can only be used as a reference during the troubleshooting process.

[0183] The model trained based on simulation data diagnosed 36 samples as overflow valve failures, accounting for 72% of the number of samples tested. The model trained based on real machine data diagnosed 40 samples as overflow valve failures, accounting for 80% of the number of samples tested. Based on the comprehensive monitoring data and model diagnosis, it was determined that the overflow valve caused system pressure and flow fluctuations. After removing the overflow valve for inspection, the overflow pressure could not be accurately adjusted. After disassembly and inspection, it was found that the overflow valve sealing rubber ring was damaged, causing leakage, which reduced the system pressure.

[0184] This application performs fault diagnosis on an actual machine and verifies the fault diagnosis model. The diagnostic situation is verified by simulating accumulator and overflow valve faults. The verification results show that the model and parameters selected based on simulation data can meet the diagnostic requirements of the actual machine, and the optimization of monitoring points based on chaos rough sets can improve the diagnostic accuracy. Through the actual troubleshooting application during system operation and debugging, it is proved that the SVM fault diagnosis method helps to improve the fault diagnosis efficiency and fault detection rate of the hydraulic system of the power generation device.

Claims

1. A method for locating and diagnosing faults in a hydraulic system of a wave energy power generation device, characterized in that: AMESim simulation software is used to simulate the fault conditions and impacts of each hydraulic component in the system, and chaotic rough sets are introduced into fault data processing and fault feature extraction to find the correlation between data. L - The characteristic reduction method of chaotic rough sets simplifies the conditional characteristics, uses support vector machines to diagnose system faults, optimizes genetic algorithms to optimize the parameters of various kernel functions, and finds effective methods to improve the accuracy of fault diagnosis through comprehensive comparison of fault feature extraction, kernel function selection, selection of different numbers of training samples, and selection of training samples under different sea conditions; P1: According to the working characteristics of the hydraulic system and each hydraulic component of the system, analyze the impact of each hydraulic component failure on the system, and build simulation models of the three subsystems of the hydraulic system based on the physical parameters and failure mechanisms of the hydraulic components; P2: Establish conditional characteristic sets based on each monitoring data, establish decision characteristic sets based on each system fault and normal state, adopt the fault feature extraction method based on chaos rough set, use Matlab as the calculation tool, through normalization, similarity matrix, lower approximation set, and then find the discrimination matrix of each fault sample, and synthesize it. Use the synthesized total identification to find the importance of each conditional characteristic and simplify it. By comparing the simplification of conditional characteristics under different sea conditions, the range of system monitoring points based on fault diagnosis is obtained; P3: The fault diagnosis method of SVM is used to diagnose the fault of the hydraulic system, analyze the fault feature extraction, kernel function selection, GA algorithm optimization penalty coefficient and kernel function parameters, as well as the influence of different numbers of training samples, single sea condition and multiple sea condition training samples on the fault diagnosis accuracy. By comparison, the most suitable model, parameters and training sample selection method for the hydraulic system are selected, and Matlab software is used to build a visual fault diagnosis interface.

2. The method for locating and diagnosing faults in the hydraulic system of a wave energy power generation device according to claim 1, characterized in that: Determination of conditional characteristics and decision characteristics: Hydraulic system adopts T L - Chaotic rough sets are used to extract fault features, select the data of each monitoring point as the conditional characteristics, and use the fault cause as the decision-making characteristic, and find the correlation between the fault phenomenon and the fault cause; Adopt T L - Chaotic rough sets perform global trait reduction by using T L The residual implication and lower approximation operator are used for calculation. The specific calculation steps include: (1) Standardization of original data: The trait value of the conditional trait variable is converted to the interval [0,1]. The conversion formula is: x ij is the element of the matrix obtained by normalizing the original matrix, d ij is the jth characteristic variable value of the i-th sample of the original data, d j min is the minimum value of the jth characteristic variable, d j max is the maximum value of the jth characteristic variable; (2) Calculate the similarity matrix sim(R) and the lower approximation set: For each chaotic characteristic C k , define a T L - Chaotic Similarity Relationship R k For R k (x i , x j )=1-|C k (x i )-C k (x j )|, then the elements in the similarity matrix sim(R) are: R k (x i ,x j )=min{1-C k (x i )-C k (x j )|} Based on the calculation of the similarity matrix sim(R), the lower approximation set is calculated: Yes(R) θ (K)(x)=inf u∈U θ(Sim(R)(x,u),K(u)) where θ L (x, y) = min{1, 1-x+y}, inf means taking the minimum value. When u is in K, K(u) = 1, and when u is not in K, K(u) = 0. The above formula is expressed as Sim(R) θ (K)(x)=min{min u∈U (1,1-Sim(R)(x,u)+K(u))}; (3) Calculate the identification matrix to simplify the attributes: Use the identification matrix to simplify the attributes. Let U = {x1, x2, ..., x n }, R={R1, R2, …, R m },M D (U, R) represents an n×n matrix {c ij }, is the identification matrix of (U, RUD), defined as: c ij ={R∈R:T(R(x i , x j ),λ(x i ))=0}, if If x j ∈[x i ] D ∈K where λ(x) = Yes(R) θ (K)(x), T L (x, y) = max{0, x + y - 1}, In the identification matrix, element c ij In the calculation formula, if x i With x j In the same decision class, If x i With x j If they are not in the same decision class, then find the following c ij Conditional attribute c k That is c ij Elements: c ij ={R∈R:max(0,R(x i ,x j )+Sim(R) θ (K)(x i )-1)=0} R k Satisfy R(x i , x j )+Sim(R) θ (K)(x i )≤1 by calculating c k In identifying the matrix element c ij The importance is obtained by summing up the proportions in , removing the element corresponding to the attribute with the largest importance, and continuing to calculate the importance until the identification matrix is ​​empty, and then simplifying it according to the importance.

3. The method for locating and diagnosing faults in the hydraulic system of a wave energy power generation device according to claim 1, characterized in that: Trait reduction method: (1) Normalization: When the wave cycle is 6s, 200 samples are selected for each of the 24 decision characteristics in the hydraulic system decision set to establish a 4800×38 raw data matrix, which is then normalized to obtain a normalized matrix. (2) Calculate the similarity matrix: According to the similarity matrix calculation method, obtain the similarity matrix. The similarity matrix is ​​a 4800 × 4800 symmetric matrix. Each element in the matrix represents the similarity between different samples. The elements on the diagonal are all 1, which represents the similarity between the sample and itself. (3) Find the lower approximate set: Take 200 samples for each decision feature. Among the 24 decision features of the hydraulic system, the normal state decision is X1 = {x1, x2, ... x 200 }, the energy conversion system intake decision is X2 = {x 201 , x 202 , …x 400 }, ..., the air intake decision of the rigid system is changed to X 24 ={x 2201 , x 2202 , …x 2400 },but: (4) Calculate the recognition matrix and importance: There are 24 decision sets, and 24 4800×4800×38 recognition matrices M are obtained by calculation. {1} (U,R),M {2} (U,R),…,M {24} (U, R), and merge them into a 4800×4800×38 total recognition matrix M {SUM} (U,R),x i With x j In the same decision class, The identification matrix is ​​a sparse matrix; The number of columns in the matrix represents 38 conditional traits. 1 means that the conditional trait is in the identification matrix, and 0 means that it is not in the identification matrix. The following partial identification matrix is ​​constructed based on the calculation results: The importance of each conditional trait is calculated through the identification matrix. 200 samples are selected for each decision trait under 4s and 8s sea conditions, and the trait simplification is performed to calculate the importance of each conditional trait. Under the three sea conditions of 4s, 6s, and 8s, 200 samples were selected from each decision trait, totaling 14,400 samples, and trait reduction was performed to calculate the importance of each decision trait.

4. The method for locating and diagnosing faults in the hydraulic system of a wave energy power generation device according to claim 1, characterized in that: Selection of monitoring points based on fault feature extraction: Monitor the operation of the hydraulic system from the perspective of fault diagnosis. When selecting monitoring points, try to monitor every hydraulic component and extract parameters so that the fault condition of the component can be characterized by the parameters of the monitoring points and the fault can be accurately located. The focus of fault diagnosis is to find reasonable monitoring points. Based on the working characteristics of the power generation device, most of the monitoring points are arranged in the hydraulic system and are selected and arranged based on experience, experiments and debugging requirements. The operation of the device is analyzed, based on the data, the chaotic rough set method is used to simplify the conditional characteristics through comprehensive comparison of similar situations; The monitoring points are determined based on the simplification of the conditional characteristics of fault samples under different sea conditions and the actual situation during equipment operation.

5. The method for locating and diagnosing faults in the hydraulic system of a wave energy power generation device according to claim 1, characterized in that: Introducing improved support vector machine for hydraulic system fault diagnosis: Introducing improved support vector machine learning method to improve classification accuracy in small sample conditions, processing each sample through chaos rough set to extract data features, and using genetic algorithm to optimize kernel function parameters to further improve sample recognition and fault detection efficiency; An improved SVM is introduced for fault diagnosis. By maximizing the interval, the classification problem is transformed into a convex quadratic programming problem. Finally, an optimal hyperplane that can correctly divide different samples is found, and the distance between this hyperplane and each classification is also the largest.

6. The method for locating and diagnosing faults in the hydraulic system of a wave energy power generation device according to claim 1, characterized in that: Fault diagnosis of hydraulic system of wave power generation device: Based on the fault analysis and simulation data of hydraulic system of power generation device, a fault diagnosis platform is built under Matlab software, and LibSVM software package is used as a tool to optimize parameters through GA algorithm toolkit, and fault diagnosis of hydraulic system of power generation device is carried out to find the fault parts of power generation device including hydraulic transducer system, stiffness adjustment system and gravity balancer; In the process of building the fault diagnosis model, the conditional characteristics of the hydraulic system are used as the input of the fault diagnosis model, and the decision characteristics are used as the output of the fault diagnosis model. Suppose the sample set of the hydraulic system is m represents the number of samples, x i Represents the input vector, corresponding to the conditional trait set of the i-th sample. For the original sample set, the input amount is x i =[x i1 , x i2 , …x i38 ], x i1 , x i2 , …x i38 ∈R respectively corresponds to the conditional feature set of the i-th sample as {C1, C2, ..., C 38 }, for the sample set after chaos rough set reduction, the input is x i =[x i1 , x i2 , …x i15 ], the conditional feature set corresponding to the i-th sample is {C1, C2, …, C 15 },y i represents the decision feature corresponding to the i-th sample, y∈{D1, D2, …, D 24 }, a one-to-one classification method is used to establish a support vector machine classifier for the hydraulic system. There are 24 normal and fault states of the hydraulic system. In the one-to-one classification method, there are M types of problems that need to be classified. The number of classifiers should be (M-1)M / 2. For this hydraulic system, 276 two-class classifiers are required; Under the three sea conditions of 4s, 6s and 8s, 200 samples are selected from the normal and each type of fault state data samples, totaling 14,400 samples, for support vector machine training. Under other sea conditions, 20 samples are selected from the normal and each type of fault state data samples, totaling 480 samples, for support vector machine detection.

7. The method for locating and diagnosing faults in the hydraulic system of a wave energy power generation device according to claim 1, characterized in that: Fault diagnosis based on simplified chaotic rough sets: C-SVM model and Gaussian radial basis kernel function are used, and Libsvm toolkit is used to establish the fault diagnosis model. The model structure generated by the unreduced data set is as follows: The model structure generated by the simplified data set: Chaotic rough sets are used to extract fault features, remove redundant data, reduce data dimensions, shorten training and detection time, and reduce the number of support vectors while keeping classification accuracy small.

8. The method for locating and diagnosing faults in the hydraulic system of a wave energy power generation device according to claim 1, characterized in that: Diagnostic accuracy of different kernel functions and parameter selections: The parameter of the polynomial kernel function is the order d, and the parameter of the RBF kernel function is the kernel width g (1 / σ 2 ), the parameters of the Sigmoid kernel function are v, c; the GA algorithm is used to optimize the penalty parameter C and different kernel function parameters. Under different kernel functions, 20 samples are randomly selected from the normal and each type of fault state data samples, a total of 480 samples, for support vector machine detection; (1) Polynomial kernel function The initial range of the penalty parameter C is set to [0,100], d is 1 to 9, the penalty parameter and the order are binary coded, and the binary coded parameters C and d are substituted into the SVM model to train the samples, perform the simulated biological evolution process of selection, crossover, and mutation, and determine whether the termination condition is met; When the best fitness is inherited to the 89th generation, it tends to be stable and reaches 89.2%. The optimization result of the polynomial kernel function is: the order d is 2, the penalty parameter C is 26.2154, and the optimized kernel function model is used to classify the test samples, with a classification accuracy of 83.3%; (2) RBF kernel function The penalty parameter C of the RBF kernel function ranges from [0,100], and the kernel function parameter g ranges from [0,1000]. When it is inherited to the 92nd generation, the best fitness tends to be stable, reaching 95.1%. The RBF kernel function optimization results are: C is 46.9587, g is 2.6541, and the optimized kernel function model is used to classify the test samples with a classification accuracy of 92.9%.

9. The method for locating and diagnosing faults in the hydraulic system of a wave energy power generation device according to claim 1, characterized in that: The influence of the number of training sample selection on diagnosis: Under the three sea conditions of 4s, 6s and 8s, 50, 100, 200 and 300 samples were randomly selected from the normal and each type of fault state data samples, totaling 3600, 7200, 14400 and 21600 samples for support vector machine training. When the wave period is 7.5s, 20 samples were randomly selected from the normal and each type of fault state data samples, totaling 480 samples for support vector machine detection. When the number of training samples is small, it has a greater impact on the detection accuracy. As the number of training samples increases, the detection accuracy also increases accordingly and tends to be stable. When the number of training samples is small, it affects the detection accuracy. When there are too many training samples, it has little effect on improving the detection accuracy.

10. The method for locating and diagnosing faults in the hydraulic system of a wave energy power generation device according to claim 1, characterized in that: The impact of different sea conditions on diagnostic accuracy: Considering the impact of sea conditions, three sea conditions of 4s, 6s, and 8s are selected. 200 samples are randomly selected from the normal and each type of fault state data samples, totaling 14,400 samples, for support vector machine training. Considering the 4s, 6s, and 8s sea conditions separately, 600 samples are randomly selected from the normal and each type of fault state data samples, totaling 14,400 samples, for support vector machine training. When the wave period is 7.5s, 20 samples are randomly selected from the normal and each type of fault state data samples, totaling 480 samples, for support vector machine detection; It can be seen from the diagnosis results that when only one sea condition is considered, the greater the difference between the sea condition of the diagnosis sample and the detection sample, the lower the diagnosis accuracy. Considering the working sea condition range of the power generation device, the diagnosis accuracy of the training set selected when the sea conditions are 4s, 6s, and 8s meets the fault diagnosis requirements.

Images