A bearing fault diagnosis method based on improved Empire competition algorithm to optimize SVC

Abstract

The present invention relates to a bearing fault diagnosis method based on an improved imperial competition algorithm to optimize SVC, wherein the penalty factor C and the kernel parameter Gamma of the SVC are obtained after being optimized by the improved imperial competition algorithm; the improved imperial competition algorithm refers to an imperial competition algorithm including an imperial reinforcement strategy; the imperial reinforcement strategy refers to taking each empire and four randomly selected empires as inputs of a differential evolution algorithm in each round of iteration of the imperial competition algorithm, and sequentially undergoing mutation, crossover and selection in the differential evolution algorithm, outputting the strongest empire as the empire of the colony group in the next iteration. The imperial reinforcement strategy provided by the improved imperial competition algorithm of the present invention can increase diversity, prevent the empire and the colony from falling into the local optimum, and further use the improved imperial competition algorithm to optimize the SVC, which can well help the SVC find the best hyperparameters, thereby helping the model improve the accuracy of fault diagnosis.

Description

Technical Field

[0001] The invention belongs to the technical field of bearing fault diagnosis, and relates to a bearing fault diagnosis method based on an improved Empire competition algorithm to optimize SVC. Background Art

[0002] With the development of science and technology and the accumulation of time, heuristic algorithms have been well developed. Nowadays, various heuristic algorithms have been proposed, such as the cuckoo algorithm, particle swarm algorithm, genetic algorithm and imperial competition algorithm, which are all various optimization algorithms created based on inspiration from various fields such as biology and sociology. These algorithms can be well combined with today's classifiers to help classifiers find more suitable hyperparameters and thus obtain better accuracy.

[0003] Current applications of the imperial competition algorithm: In the medical field, E.Naseri proposed an ICA-based clustering algorithm for the diagnosis of arrhythmias, and used ICA to find appropriate weighting factors to assist the clustering algorithm to obtain better clustering results; Korhan Güünel proposed the use of the imperial competition algorithm for the classification of epileptic seizures and psychological non-epileptic seizures (PNES); In scheduling problems, Li, Ming proposed an imperial competition algorithm with imperial cooperation (ECICA) to minimize the fuzzy manufacturing period, thereby solving the fuzzy distribution assembly flow shop scheduling; Jabari proposed an optimal chiller load problem based on the imperial competition algorithm to find the partial load rate, cooling flux and electricity demand of the chiller, and minimize the energy consumption of the entire system. In the engineering field, You, Ting introduced the wavelet mutation operator into the original ICA to improve its detection ability and adjust the parameters of the fractional-order PID controller of the automatic voltage regulator (AVR) system.

[0004] In the fault diagnosis of bearings, the Empire Competition Algorithm can help SVC find more appropriate hyperparameter penalty factors (C) and kernel parameters (Gamma), thereby improving the diagnostic accuracy. If you choose to adjust the parameters manually, it will take a lot of time and you will not be able to get satisfactory accuracy. In the current traditional optimization algorithms, there is a common problem that they are prone to fall into the local optimum, thus failing to maximize the performance of the classifier.

[0005] Therefore, it is necessary to improve the imperial competition algorithm to address this problem. Summary of the invention

[0006] In order to solve the problems existing in the prior art, the present invention provides a bearing fault diagnosis method based on optimizing SVC by using an improved Empire competition algorithm.

[0007] To achieve the above object, the scheme adopted by the present invention is as follows:

[0008] A bearing fault diagnosis method based on an improved imperial competition algorithm to optimize SVC, wherein the penalty factor C and the kernel parameter Gamma of the SVC are obtained after being optimized by the improved imperial competition algorithm;

[0009] The improved empire competition algorithm refers to an empire competition algorithm that includes an empire strengthening strategy;

[0010] The empire strengthening strategy refers to taking each empire and four randomly selected empires as inputs of the differential evolution algorithm in each round of iteration of the empire competition algorithm, and sequentially undergoing mutation, crossover and selection in the differential evolution algorithm, outputting the strongest empire (i.e., the one with the smallest fitness value) as the empire of the colony group in the next iteration.

[0011] In each round of iteration of the empire competition algorithm, the present invention uses each empire and four randomly selected empires as inputs of the differential evolution algorithm, so that each empire and the four randomly selected empires undergo mutation behavior, and the generated new empire moves toward a comprehensive direction containing five empires, and then the original empire and the new empire 1 are crossed to generate a new empire, which can ensure the diversity of the algorithm, and finally select the strongest empire from the original empire, the new empire 1 and the new empire 2 as the empire of the colony in the next round of iteration. This can not only increase the diversity of the population, but also help the algorithm break through the local optimal point to find the global optimal point.

[0012] In the existing strategy for strengthening the empire, the levy flight in the cuckoo algorithm is integrated into each round of iteration. Levy flight is a random movement, and the direction of the next flight cannot be determined, and it will not move in a certain good direction. In the present invention, the three processes of mutation, crossover, and selection in differential evolution are introduced into the strategy for strengthening the empire, so that the direction of movement of the empire in each round of iteration can be determined, and it must move in a right direction. In the present invention, four empires will be randomly selected for mutation for each empire to generate a new individual, which contains part of the information of the five empires. Then, the mutated country is crossed with the original country to generate a new crossover individual to reflect the diversity of the algorithm. Finally, the crossover-generated country, the mutated country, and the original country are selected, and the three strongest empires are retained for the next round of iteration. Doing so can help the algorithm avoid the empire falling into the local optimal value, which causes the colony to fall into the local optimal value together. Especially in the later stage of iteration, the algorithm is very easy to fall into it. The method of the present invention can explore a better development area together with the information of other empires, thereby breaking through the local optimality.

[0013] As the preferred technical solution:

[0014] As described above, in the bearing fault diagnosis method based on the improved Empire competition algorithm to optimize SVC, the mutation process is:

[0015] H imp =Imp best +F×(Random imp1 -Random imp2 )+F×(Random imp3 -Random imp4 );

[0016] Among them, Imp best For the strongest imperialist country, Random imp1 , Random imp2 , Random imp3 , Random imp4 are four randomly selected imperialist countries, and Random imp1 For the first randomly selected country, Random imp2 For the second randomly selected country, Random imp3 For the third randomly selected country, Random imp4 is the fourth randomly selected country, F is the scaling factor, H imp For the mutant country;

[0017] The scaling factor F is calculated as:

[0018]

[0019] Among them, index represents the current number of iterations, epoch represents the maximum number of iterations, π represents pi in mathematics, cos(·) is the cosine function; a is a constant with a value of 0.4.

[0020] As described above, a bearing fault diagnosis method based on the improved Empire competition algorithm to optimize SVC, the crossover process is:

[0021]

[0022] Among them, V imp is the new empire obtained by crossing the mutant empire with the empire sent into each iteration, Imp[i] is the i-th dimension sent into the empire, H imp [i] is the i-th dimension of the mutation empire, CR is the crossover probability

[0023] The calculation method of crossover probability CR is:

[0024]

[0025] Among them, index represents the current number of iterations, epoch represents the maximum number of iterations, π represents pi in mathematics, cos(·) is the cosine function; a is a constant with a value of 0.4.

[0026] As described above, a bearing fault diagnosis method based on the improved Empire competition algorithm to optimize SVC, the selection process is:

[0027] Imp new =min{fitness(H imp ), fitness(V imp ), fitness(Imp)};

[0028] That is, calculate the fitness values ​​of the three countries and select the smallest empire country (i.e. the strongest empire) to update the existing countries;

[0029] Among them, Imp new The new empire after strengthening the empire's strategy,

[0030] fitness(H imp ), fitness(v imp ), fitness(Imp) are the fitness values ​​of the mutant empire, the cross-generated empire, and the original empire respectively.

[0031] A bearing fault diagnosis method based on the improved imperial competition algorithm to optimize SVC as described above, wherein the improved imperial competition algorithm further includes an imperial learning strategy; the imperial learning strategy refers to sending colonies from the weakest empire to learn around the strongest empire;

[0032] The weakest empire mentioned here refers to the empire with the smallest total energy among all existing empires;

[0033] Sending colonies from the weakest empire means sorting all colonies in the weakest empire from largest to smallest according to fitness value, and sending out the top 3 colonies; if the total number of colonies is ≤ 3, all colonies are sent out;

[0034] The strongest empire mentioned here refers to the empire with the greatest total energy among all existing empires;

[0035] The learning strategy is as follows:

[0036] bounds[i]=[x best [i]×U(0,0.5),x best [i]×(1+(0,0.5))];

[0037]

[0038] Where bounds[i] is the range of the i-th dimension coordinates in x, where x represents the colony coordinates; best [i] represents the i-th dimension in x, U(0, 0.5) represents the generation of a random number uniformly distributed in the interval (0, 0.5), and x new Represents the coordinates of the newly generated colony, LHSample generates Latin hypercube sampling samples, dim represents the dimension of x, bounds represents the output boundary, and colonies represents the number of colonies.

[0039] In the imperial competition algorithm before the improvement, the communication between populations has two modes: empire and colony, and empire and empire. However, the communication between empires is carried out through a competitive relationship, which can only help the algorithm converge, but does not help the empire jump out of the original local optimal value. Therefore, the present invention adds a population communication strategy, namely, an imperial learning strategy, on the basis of the empire reinforcement strategy. In the iterative process, it helps the entire empire jump out of the local optimal point and better find the global optimal point. Specifically: the present invention improves the competitiveness of the weakest empire, prevents the algorithm from converging to the local optimal point too early, and helps the weakest empire break through the limitations of the local optimal point, so as to help the algorithm better find the global optimal point. In this regard, the present invention communicates with the colonies of the weakest empire to the strongest empire. If the number of colonies of the weakest empire is less than or equal to three, all colonies are sent to learn around the strongest empire. If the number of colonies is greater than three, three colonies are sent to learn around the strongest empire.

[0040] The imperial learning strategy is used to improve the classification accuracy of the classification model.

[0041] The biggest difference between the Empire Learning Strategy and the Empire Reinforcement Strategy is that they target different subjects. The Empire Learning Strategy targets colonies while the Empire Reinforcement Strategy targets empires. An empire is the best individual of a population, while a colony is just an individual of a population. The reason for combining the two optimization strategies is that the Empire Reinforcement Strategy is only to help the empire escape from the local optimum, while the optimization goal of the Empire Learning Strategy is not to optimize but to prevent the algorithm from converging to a local point too early. The Empire Learning Strategy not only has the goal of optimizing but also helps the weakest empire to improve the ability of its own population, thereby preventing the weakest empire from being annexed, which can prevent the algorithm from converging too early.

[0042] The combination of the empire reinforcement strategy and the empire learning strategy can help an optimization algorithm that only has one combination to have stronger breakthrough capabilities. In each iteration process, the optimization breakthrough in the empire is often local, which is to select the best among the best, while the optimization between colonies is large-scale, and batch searches are carried out in the area where they are located to help the population to open up better areas. The combination of the two allows the algorithm to take into account both the local and the global, which greatly improves the algorithm in both global and local optimization.

[0043] A bearing fault diagnosis method based on an improved imperial competition algorithm to optimize SVC as described above, wherein a reverse learning mechanism is added to the assimilation process of the initialization population and the colony in the improved imperial competition algorithm;

[0044] The assimilation process of the colony is:

[0045] x as =x+(β×random(dim)-1)×(x imp -x);

[0046] x invert =boundary[2×i]+boundary[2×i+1]-x[i]i=0, 1, 2…m;

[0047] x new =min{x as , x invert};

[0048] Among them, x as is the coordinate of the colony after assimilation, x is the coordinate of the colony, β is the assimilation coefficient, random(dim) is the random number with dimension dim, x imp is the coordinate of the empire, x invert is the coordinate of the colony after assimilation through reverse learning, i is the i-th dimension, boundary[2×i] is the 2i-th dimension in the boundary list, boundary[2×i+1] is the 2i+1-th dimension in the boundary list, x[i] is the i-th coordinate of the colony, x new The coordinates for the newly generated colony.

[0049] Since the prior art adopts the reverse learning strategy when the initial population is generated, a better initial population is obtained. However, only adopting the reverse learning strategy at the initial stage cannot significantly reduce the number of iterations in this method, and the calculation time of the algorithm is related to the distance between the individual and the optimal individual in the initial population. If the individual is born near the optimal value, then in this calculation, all individuals in the population will converge quickly. Since the individuals generated purely randomly have not been estimated, the convergence speed is unpredictable. However, if the reverse individual of each individual is considered at the same time, then the probability that the individual and the reverse individual are closer to the optimal individual is 50%, and the closer individual is selected as the initial individual of the population, that is, each individual is one step closer to the optimal solution. Therefore, the present invention not only adds a reverse learning mechanism when initializing the population, so that a better initial population can be obtained in the initial iteration, and the convergence speed of the algorithm can be accelerated, but also adds a reverse learning mechanism in the communication process of the colony, and the search range can be reduced in the search process, thereby further accelerating the convergence speed of the algorithm. Specifically, an initial population is generated at the beginning of the algorithm through Latin hypercube sampling, and then a reverse population is generated based on the initial population. The fitness functions of the individuals in the two populations are compared in turn, and the better individuals are selected to form a new population to serve as the initial population for the next operation of the algorithm. In the process of colony assimilation, a reverse learning mechanism is also added.

[0050] The existing technology improves assimilation by modifying its assimilation mechanism, and does not consider that the algorithm can be promoted to be closer to the optimal value by comparing the reverse coordinates after assimilation. The existing reverse learning is only used in the initialization of the population and is not reused in the algorithm. The use of the reverse mechanism in the initial population can obtain a better initial population, while the present invention adopts secondary utilization and also adopts this method in the assimilation process, which can help us obtain better colonies after assimilation.

[0051] The combination of the two stages of reverse learning not only greatly reduces the iterative parameters, but also allows the colony to move closer to the empire from two directions during each assimilation process, thus expanding the scope of colony search in each iteration, thereby accelerating the convergence of the algorithm. If there is only reverse learning at the initial moment, the direction of the colony moving toward the empire during the assimilation process is single, and the position that may become the optimal solution may be missed. The method of the present invention can reduce the probability of missing the optimal solution.

[0052] As described above, a bearing fault diagnosis method based on the improved Empire competition algorithm to optimize SVC, the reverse learning formula in the reverse learning mechanism is:

[0053] P i =a i +bi -p i ;

[0054] Among them, P i is the reverse individual, a i and b i are the upper and lower bounds respectively, p i For the initial individual.

[0055] As described above, a bearing fault diagnosis method based on the improved Empire competition algorithm to optimize SVC refers to extracting features from the vibration signal of the bearing to be tested, then reducing the dimension through a feature selector and inputting it into the SVC to output the diagnosis result.

[0056] As described above, a bearing fault diagnosis method based on the improved Empire competition algorithm to optimize SVC, the optimization process is:

[0057] (1) Input bearing vibration signal;

[0058] (2) Extracting features from bearing vibration signals;

[0059] (3) Initializing the initial parameters of the improved imperial competition algorithm (RE-ICA); the initial parameters are the population number, the number of empires, the number of iterations, the assimilation factor and the learning rate;

[0060] (4) Use feature selector to reduce the dimension of the extracted features;

[0061] (5) An improved imperial competition algorithm is used to generate an initial population; and the fitness value in the improved imperial competition algorithm is the inverse of the SVC classification accuracy; and both the empire and the colony contain the penalty factor C and the kernel parameter Gamma of the SVC;

[0062] (6) The reduced-dimensional features are used as the input of SVC. After input, the penalty factor C and the kernel parameter Gamma of SVC are optimized through the improved imperial competition algorithm. After iteration, when there is only one empire left in the algorithm or the maximum number of iterations is reached, the optimization is stopped, and the optimal parameters contained in the strongest empire are output to SVC for training. The trained SVC is used as the output model (i.e., the optimized SVC).

[0063] As described above, a bearing fault diagnosis method based on an improved Empire competition algorithm to optimize SVC, the process of the improved Empire competition algorithm is:

[0064] (1) Generate an initial population. Each initial population generates a reverse initial population through reverse learning, and retains individuals with smaller fitness values ​​as the initial population for the next iteration. The empire and colonies are then divided according to the fitness values.

[0065] (2) Start iteration;

[0066] (2.1) The Empire assimilates the colonies: In each process of the colonies approaching the Empire, the new colony positions are generated through reverse learning, and the best of the two colony positions is selected as the new position of the colony after the approach;

[0067] (2.2) Compare the fitness values ​​of the empire and the colony, and select the one with the smallest fitness value as the empire; that is, if there is a colony with a smaller fitness value than the empire, then the colony and the empire will swap identities; if not, then they will not swap identities;

[0068] (2.3) The empires compete with each other and divide up the weakest empire; the weakest empire refers to: comparing the total energy of each empire and its colonies, the empire with the lowest total energy is regarded as the weakest empire;

[0069] (2.4) Generate a random number and compare it with the adaptive value of the current iteration number. If it is less than the adaptive value, perform the empire learning strategy and then enter the empire splitting process to generate a new empire. If it is greater than the adaptive value, directly enter the empire splitting process to generate a new empire.

[0070] (2.5) Carry out empire strengthening strategies;

[0071] (2.6) Determine whether there is only one empire left. If so, stop iterating. Otherwise, determine whether the maximum number of iterations has been reached. If so, stop iterating. Otherwise, return to (2.1) to continue iterating.

[0072] Beneficial Effects

[0073] (1) The improved empire competition algorithm of the present invention provides an empire reinforcement strategy and a learning strategy that can increase the diversity of the algorithm, prevent the empire from falling into a local optimum and the colony from falling into a local optimum, thereby reducing the risk of the algorithm falling into a local optimum;

[0074] (2) The reverse learning strategy introduced in the improved imperial competition algorithm of the present invention during the initial population generation and assimilation process can improve the convergence speed of the algorithm and achieve higher accuracy;

[0075] (3) The present invention uses an improved imperial competition algorithm to optimize SVC, which can help SVC find the best penalty factor C and Gamma coefficient, thereby helping the model improve the accuracy of fault diagnosis, thereby improving the benefits for the enterprise. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 A flowchart of the improved imperial competition algorithm of the present invention;

[0077] Figure 2is the function graph of Ackely function;

[0078] Figure 3 is the function graph of Schaffer function;

[0079] Figure 4 A performance comparison chart of different algorithms in terms of their ability to escape from local optimality when encountering local optimality;

[0080] Figure 5 This is a comparison chart of the convergence speed of different algorithms;

[0081] Figure 6 Flow chart of optimizing SVC of the present invention;

[0082] Among them, RE-ICA-Improved Imperial Competition Algorithm, PSO-Particle Swarm Optimization, ICA-Imperial Competition Algorithm, CS-Cuckoo Algorithm. DETAILED DESCRIPTION

[0083] The present invention will be further described below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms fall within the scope limited by the appended claims of the application equally.

[0084] An improved imperial competition algorithm, such as Figure 1 As shown, the specific process is:

[0085] (1) Generate an initial population. Each initial population generates a reverse initial population through reverse learning. By comparing the fitness values, the individuals with smaller fitness values ​​are retained as the initial population for the next iteration. Finally, the empire and colonies are divided according to the fitness values.

[0086] The reverse learning formula in the reverse learning mechanism is:

[0087] P i =a i +b i -p;

[0088] Among them, P i is the reverse individual, a i and b i are the upper and lower bounds respectively, p i For the initial individual.

[0089] (2) Start iteration;

[0090] (2.1) The empire assimilates its colonies. The assimilation process is as follows:

[0091] xas =x+(β×random(dim)-1)×(x imp -x);

[0092] x invert =boundary[2×i]+boundary[2×i+1]-x[i]i=0, 1, 2…m;

[0093] x new =min{x as , x invert};

[0094] Among them, x as is the coordinate of the colony after assimilation, x is the coordinate of the colony, β is the assimilation coefficient, random(dim) is the random number with dimension dim, x imp is the coordinate of the empire, x invert is the coordinate of the colony after assimilation through reverse learning, i is the i-th dimension, boundary[2×i] is the 2i-th dimension in the boundary list, boundary[2×i+1] is the 2i+1-th dimension in the boundary list, x[i] is the i-th coordinate of the colony, x new The coordinates of the newly generated colony;

[0095] In each process of the colony approaching the empire, a new colony position is generated through reverse learning, and the best of the two colony positions is selected as the new position of the colony after the approach; the reverse learning formula in the reverse learning mechanism is:

[0096] P i =a i +b i -p i ;

[0097] Among them, P i is the reverse individual, a i and b i are the upper and lower bounds respectively, p i For the initial individual.

[0098] (2.2) Compare the fitness values ​​of the empire and the colony, and select the one with the smallest fitness value as the empire;

[0099] (2.3) Compare the total energy of each empire and its colonies, and take the empire with the lowest total energy as the weakest empire; the empires compete with each other and divide up the weakest empire;

[0100] (2.4) Generate a random number and compare it with the adaptive value of the current iteration number. If the random number is less than the adaptive value, perform the empire learning strategy and then enter the empire split to generate a new empire. If the random number is greater than or equal to the adaptive value, directly enter the empire split to generate a new empire.

[0101] The imperial learning strategy mentioned above refers to sending colonies from the weakest empire to learn around the strongest empire;

[0102] The weakest empire mentioned here refers to the empire with the smallest total energy among all existing empires;

[0103] Sending colonies from the weakest empire means sorting all colonies in the weakest empire from largest to smallest according to fitness value, and sending out the top 3 colonies; if the total number of colonies is ≤ 3, all colonies are sent out;

[0104] The strongest empire mentioned here refers to the empire with the greatest total energy among all existing empires;

[0105] The learning strategy is as follows:

[0106] bounds[i]=[x best [i]×U(0,0.5),x best [i]×(1+(0,0.5))];

[0107]

[0108] Where bounds[i] is the range of the i-th dimension coordinates in x, where x represents the colony coordinates; best [i] represents the i-th dimension in x, U(0, 0.5) represents the generation of a random number uniformly distributed in the interval (0, 0.5), and x new Represents the coordinates of the newly generated colony, LHSample generates Latin hypercube sampling samples, dim represents the dimension of x, bounds represents the output boundary, and colonies represents the number of colonies.

[0109] (2.5) Implement empire strengthening strategy: In each round of iteration, each empire and four randomly selected empires are used as inputs of the differential evolution algorithm. After mutation, crossover and selection in the differential evolution algorithm, the strongest empire is output as the empire of the colony group in the next iteration.

[0110] The mutation process is:

[0111] H imp =Imp best +F×(Random imp1 -Random imp2)+F×(Random imp3 -Random imp4 );

[0112] Among them, Imp best For the strongest imperialist country, Random imp1 , Random imp2 , Random imp3 , Random imp4 are four randomly selected imperialist countries, and Random imp1 For the first randomly selected country, Random imp2 For the second randomly selected country, Random imp3 For the third randomly selected country, Random imp4 is the fourth randomly selected country, F is the scaling factor, H imp For the mutant country;

[0113] The scaling factor F is calculated as:

[0114]

[0115] Among them, index represents the current number of iterations, epoch represents the maximum number of iterations, π represents pi in mathematics, cos(·) is the cosine function; a is a constant with a value of 0.4.

[0116] The crossover process is:

[0117]

[0118] Among them, V imp is the new empire obtained by crossing the mutant empire with the empire sent into each iteration, Imp[i] is the i-th dimension sent into the empire, H imp [i] is the i-th dimension of the mutation empire, CR is the crossover probability

[0119] The calculation method of crossover probability CR is:

[0120]

[0121] Among them, index represents the current number of iterations, epoch represents the maximum number of iterations, π represents pi in mathematics, cos(·) is the cosine function; a is a constant with a value of 0.4.

[0122] The selection process is:

[0123] Imp new =min{fitness(H imp ), fitness(vimp ), fitness(Imp)};

[0124] Among them, Imp new The new empire after strengthening the empire's strategy,

[0125] fitness(H imp ), fitness(v imp ), fitness(Imp) are the fitness values ​​of the mutant empire, the cross-generated empire, and the original empire respectively.

[0126] (2.6) Determine whether there is only one empire left. If so, stop iterating. Otherwise, determine whether the maximum number of iterations has been reached. If so, stop iterating. Otherwise, return to (2.1) to continue iterating.

[0127] In order to verify the effectiveness of our improved imperial competition algorithm, we use Ackely and Schaffer (the function graphs of the two are as follows: Figure 2 and 3 The two test functions are used to verify the performance of the improved imperial competition algorithm (RE-ICA), the unimproved imperial competition algorithm (ICA), the cuckoo algorithm (CS) and the particle swarm algorithm (PSO). Figure 2 It can be found that Ackely has many local extreme points, but its minimum point is hidden in the center of the function graph, and there are many local extreme points around it that are not much different, which are very difficult to find. It can be used to evaluate the ability of the algorithm to jump out of the local optimum when encountering a local optimum; Figure 3 It can be seen that the extreme points of the Schaffer function are relatively clear, and the algorithm can find the general trend of the extreme points very well, but the convergence speed of the algorithm is different. This test function can be used to compare the convergence speed of the algorithm very well; specifically, the minimum value of each generation of the test function obtained in these two test functions is compared (such as Figure 4 and Figure 5 As shown, Figure 4 correspond Figure 2 The function in Figure 5 correspond Figure 3 In comparison, the improved imperial competition algorithm of the present invention has a better optimization effect, and both the convergence speed and the convergence accuracy are greatly improved.

[0128] The improved imperial competition algorithm is used to optimize the penalty factor C and kernel parameter Gamma of SVC, as follows: Figure 6 As shown, the specific process is as follows:

[0129] (1) Input bearing vibration signal;

[0130] (2) Extracting features from bearing vibration signals;

[0131] (3) Initializing the initial parameters of the improved imperial competition algorithm (RE-ICA); the initial parameters are the population number, the number of empires, the number of iterations, the assimilation factor and the learning rate;

[0132] (4) Use feature selector to reduce the dimension of the extracted features;

[0133] (5) An improved imperial competition algorithm is used to generate an initial population; and the fitness value in the improved imperial competition algorithm is the inverse of the SVC classification accuracy; and both the empire and the colony contain the penalty factor C and the kernel parameter Gamma of the SVC;

[0134] (6) The reduced-dimensional features are used as the input of SVC. After input, the penalty factor C and the kernel parameter Gamma of SVC are optimized through the improved imperial competition algorithm. After iteration, when there is only one empire left in the algorithm or the maximum number of iterations is reached, the optimization is stopped, and the optimal parameters contained in the strongest empire are output to SVC for training. The trained SVC is used as the output model, that is, the optimized SVC.

[0135] The above optimized SVC is used for bearing fault diagnosis, which means extracting features from the vibration signal of the bearing to be tested, and then reducing the dimension through a feature selector and inputting it into the optimized SVC to output the diagnosis result; specifically: the present invention uses the bearing data of the driving end of the Western Reserve University of the United States as the experimental data of the present invention, and extracts the time domain and frequency domain features of the first 6 modal functions after empirical mode decomposition of the bearing data as the features of the bearing vibration signal at that moment, and selects 10 features through a feature selector and sends them to the classifier for training. 70% of the data is used as training data, and 30% of the data is used as test data. The classification accuracy of the test data is used as the measurement indicator of this experiment. Table 1 is a comparison of the performance of the improved imperial competition algorithm, the unimproved imperial competition algorithm and other algorithms to optimize the classifier. It can be seen from Table 1 that the improved imperial competition algorithm has been greatly improved in accuracy, and the classification accuracy has reached the highest among all optimization algorithms.

[0136] Table 1 Application effect of the algorithm in bearing failure

[0137]

Claims

1. A bearing fault diagnosis method based on an improved imperial competition algorithm to optimize SVC, characterized by: The penalty factor C and the kernel parameter Gamma of SVC in the method are obtained after optimization by the improved imperial competition algorithm; The improved empire competition algorithm refers to an empire competition algorithm that includes an empire reinforcement strategy and an empire learning strategy; The empire strengthening strategy refers to that in each round of iteration of the empire competition algorithm, each empire and four randomly selected empires are used as the input of the differential evolution algorithm, and the strongest empire is output as the empire of the colony group in the next iteration after mutation, crossover and selection in the differential evolution algorithm. The mutation process is: H imp =Imp best +F×(Random imp1 -Random imp2 )+F×(Random imp3 -Random imp4 ); Among them, Imp best For the strongest imperialist country, Random imp1 ,Random imp2 ,Random imp3 ,Random imp4 are four randomly selected imperialist countries, and Random imp1 For the first randomly selected country, Random imp2 For the second randomly selected country, Random imp3 For the third randomly selected country, Random imp4 is the fourth randomly selected country, F is the scaling factor, H imp For the mutant country; The scaling factor F is calculated as: Where index represents the current iteration number, epoch represents the maximum iteration number, π represents pi in mathematics, cos(·) represents the cosine function; a is a constant with a value of 0.4; The crossover process is: Among them, V imp is the new empire obtained by crossing the mutant empire with the empire sent into each iteration, Imp[i] is the i-th dimension sent into the empire, H imp [i] is the i-th dimension of the mutation empire, CR is the crossover probability; The calculation method of crossover probability CR is: Where index represents the current iteration number, epoch represents the maximum iteration number, π represents pi in mathematics, cos(·) represents the cosine function; a is a constant with a value of 0.4; The selection process is: Imp new =min{fitness(H imp ),fitness(V imp ),fitness(Imp)}; Among them, Imp new The new empire after strengthening the strategy for the empire, fitness (H imp ),fitness(V imp ), fitness(Imp) are the fitness values ​​of the mutant empire, the cross-generated empire, and the original empire respectively; The imperial learning strategy mentioned above refers to sending colonies from the weakest empire to learn around the strongest empire; The weakest empire mentioned here refers to the empire with the smallest total energy among all existing empires; Sending colonies from the weakest empire means sorting all colonies in the weakest empire from largest to smallest according to fitness value, and sending out the top 3 colonies; if the total number of colonies is ≤ 3, all colonies are sent out; The strongest empire mentioned here refers to the empire with the greatest total energy among all existing empires; The learning strategy is as follows: bounds[i]=[x best [i]×U(0,0.5),x best [i]×(1+(0,0.5))]; Where bounds[i] is the range of the i-th dimension coordinates in x, where x represents the colony coordinates; best [i] represents the i-th dimension in x, U(0,0.5) represents the generation of a random number uniformly distributed in the interval (0,0.5), and x new Represents the coordinates of the newly generated colony, LHSample generates Latin hypercube sampling samples, dim represents the dimension of x, bounds represents the output boundary, and colonies represents the number of colonies.

2. A bearing fault diagnosis method based on improved Empire competition algorithm to optimize SVC according to claim 1, characterized in that: In the improved imperial competition algorithm, a reverse learning mechanism is added to the initialization population and the assimilation process of the colony; The assimilation process of the colony is: x as =x+(β×random(dim)-1)×(x imp -x); x invert =boundary[2×i]+boundary[2×i+1]-x[i] i=0,1,2…m; x new =min{x as ,x invert }; Among them, x as is the coordinate of the colony after assimilation, x is the coordinate of the colony, β is the assimilation coefficient, random(dim) is the random number with dimension dim, x imp is the coordinate of the empire, x invert is the coordinate of the colony after assimilation through reverse learning, i is the i-th dimension, boundary[2×i] is the 2i-th dimension in the boundary list, boundary[2×i+1] is the 2i+1-th dimension in the boundary list, x[i] is the i-th coordinate of the colony, x new The coordinates for the newly generated colony.

3. The bearing fault diagnosis method based on the improved Empire competition algorithm to optimize SVC according to claim 2 is characterized in that: The reverse learning formula in the reverse learning mechanism is: P i =a i +b i -p; Among them, P i is the reverse individual, a i and b i are the upper and lower bounds respectively, p i For the initial individual.

4. The bearing fault diagnosis method based on the improved Empire competition algorithm to optimize SVC according to claim 1 is characterized in that: The bearing fault diagnosis method refers to extracting features from the vibration signal of the bearing to be tested, and then reducing the dimension through a feature selector and inputting the signal into the SVC to output the diagnosis result.

5. The bearing fault diagnosis method based on the improved Empire competition algorithm to optimize SVC according to claim 1 is characterized in that: The optimization process is as follows: (1) Input bearing vibration signal; (2) Extracting features from bearing vibration signals; (3) Initializing the initial parameters of the improved imperial competition algorithm; the initial parameters are the population number, the empire number, the number of iterations, the assimilation factor and the learning rate; (4) Use feature selector to reduce the dimension of the extracted features; (5) An improved imperial competition algorithm is used to generate an initial population; and the fitness value in the improved imperial competition algorithm is the inverse of the SVC classification accuracy; and both the empire and the colony contain the penalty factor C and the kernel parameter Gamma of the SVC; (6) The reduced-dimensional features are used as the input of SVC. After the input, the penalty factor C and the kernel parameter Gamma of SVC are optimized through the improved imperial competition algorithm. After iteration, when there is only one empire left in the algorithm or the maximum number of iterations is reached, the optimization is stopped. The optimal parameters contained in the strongest empire are output to SVC for training, and the trained SVC is used as the output model.

6. The bearing fault diagnosis method based on the improved Empire competition algorithm to optimize SVC according to claim 1 is characterized in that: The process of improving the imperial competition algorithm is as follows: (1) Generate an initial population. Each initial population generates a reverse initial population through reverse learning, and retains individuals with smaller fitness values ​​as the initial population for the next iteration. The empire and colonies are then divided according to the fitness values. (2) Start iteration; (2.1) The Empire assimilates the colonies: In each process of the colonies approaching the Empire, the new colony positions are generated through reverse learning, and the best of the two colony positions is selected as the new position of the colony after the approach; (2.2) Compare the fitness values ​​of the empire and the colony, and select the one with the smallest fitness value as the empire; (2.3) The empires compete with each other and divide up the weakest empire; the weakest empire refers to: comparing the total energy of each empire and its colonies, the empire with the lowest total energy is regarded as the weakest empire; (2.4) Generate a random number and compare it with the adaptive value of the current iteration number. If it is less than the adaptive value, perform the empire learning strategy and then enter the empire splitting process to generate a new empire. If it is greater than the adaptive value, directly enter the empire splitting process to generate a new empire. (2.5) Carry out empire strengthening strategies; (2.6) Determine whether there is only one empire left. If so, stop iterating. Otherwise, determine whether the maximum number of iterations has been reached. If so, stop iterating. Otherwise, return to (2.1) to continue iterating.

Images