CN113610179A - Equipment fault detection classifier training method, computing equipment and storage medium - Google Patents
Equipment fault detection classifier training method, computing equipment and storage medium Download PDFInfo
- Publication number
- CN113610179A CN113610179A CN202110942911.6A CN202110942911A CN113610179A CN 113610179 A CN113610179 A CN 113610179A CN 202110942911 A CN202110942911 A CN 202110942911A CN 113610179 A CN113610179 A CN 113610179A
- Authority
- CN
- China
- Prior art keywords
- population
- original sample
- original
- training
- individuals
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 88
- 238000012549 training Methods 0.000 title claims abstract description 76
- 238000001514 detection method Methods 0.000 title claims abstract description 19
- 238000003860 storage Methods 0.000 title claims abstract description 15
- 230000002068 genetic effect Effects 0.000 claims abstract description 73
- 210000000349 chromosome Anatomy 0.000 claims abstract description 30
- 238000012545 processing Methods 0.000 claims abstract description 18
- 230000006870 function Effects 0.000 claims description 34
- 230000035772 mutation Effects 0.000 claims description 26
- 238000009396 hybridization Methods 0.000 claims description 22
- 238000004891 communication Methods 0.000 claims description 19
- 239000011159 matrix material Substances 0.000 claims description 15
- 238000009402 cross-breeding Methods 0.000 claims description 2
- 238000004422 calculation algorithm Methods 0.000 description 27
- 238000001228 spectrum Methods 0.000 description 21
- 238000005457 optimization Methods 0.000 description 19
- 108090000623 proteins and genes Proteins 0.000 description 19
- 230000008569 process Effects 0.000 description 16
- 238000009826 distribution Methods 0.000 description 15
- 238000010586 diagram Methods 0.000 description 12
- 238000003745 diagnosis Methods 0.000 description 7
- 230000002028 premature Effects 0.000 description 7
- 230000003044 adaptive effect Effects 0.000 description 6
- 238000004088 simulation Methods 0.000 description 6
- 230000008859 change Effects 0.000 description 4
- 230000007423 decrease Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 3
- 230000006872 improvement Effects 0.000 description 3
- 230000002093 peripheral effect Effects 0.000 description 3
- 230000003068 static effect Effects 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 230000000052 comparative effect Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 230000002829 reductive effect Effects 0.000 description 2
- 230000000717 retained effect Effects 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000002759 chromosomal effect Effects 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000007667 floating Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 230000000670 limiting effect Effects 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 230000036961 partial effect Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000000746 purification Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000004904 shortening Methods 0.000 description 1
- 230000001360 synchronised effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 230000007723 transport mechanism Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biomedical Technology (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Physiology (AREA)
- Genetics & Genomics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a training method of an equipment fault detection classifier, which comprises the following steps: the method comprises the steps that a slave device obtains a device operation data set, each operation data record of the data set is processed to generate an original sample, and an original sample set comprising a plurality of original samples is obtained; encoding each original sample in the original sample set to generate an individual to obtain an initial population comprising a plurality of individuals; constructing a fitness function according to the fault types included in the original sample set; performing genetic iteration on the population according to the fitness function to obtain a target population; determining the optimal individual in the target population; determining an optimized feature set of the original sample set according to chromosomes of the optimal individuals; processing each original sample in the original sample set according to the optimized feature set to generate a training sample, and obtaining a training sample set comprising a plurality of training samples; and training the classifier according to the training sample set. The invention also discloses a computing device and a computer readable storage medium.
Description
Technical Field
The invention relates to the field of equipment fault diagnosis, in particular to an equipment fault detection classifier training method, computing equipment and a storage medium.
Background
With the development of computer and sensor technologies, computers monitor the operation conditions of equipment and automatically judge the operation problems of the equipment. The operation data of the equipment is collected by deploying sensors in the equipment, and the operation data is analyzed by a computer to obtain an analysis result. However, the operation data acquisition items are more, the operation condition of the equipment is more complex, and the corresponding relation between the operation data and the fault type of the equipment is not clear enough. Therefore, a large number of original features have to be provided for fault identification, but due to the restriction of a plurality of factors such as the scale of a classifier, the complexity of a training process, the capacity of a computer and the like, a good diagnosis effect cannot be achieved.
In the prior art, principal components are adopted for analysis, and a feature vector corresponding to the maximum feature value is used as a classification feature, so that some irrelevant features in an original feature set are removed. However, the feature set obtained after the features are removed may miss some features, so that the obtained classifier cannot obtain a good diagnosis effect.
For this reason, a new training method of the device fault detection classifier is required.
Disclosure of Invention
To this end, the present invention provides a method of training a device fault detection classifier in an attempt to solve, or at least alleviate, the problems identified above.
According to one aspect of the present invention, there is provided a device fault detection classifier training method, adapted to be executed in a server, the server being in communication connection with one or more devices, the method comprising the steps of: the method comprises the steps that a device operation data set is obtained from a device, each operation data record of the operation data set is processed to generate an original sample, an original sample set comprising a plurality of original samples is obtained, and the original sample set comprises original samples of various fault types; encoding each original sample in the original sample set to generate an individual to obtain an initial population comprising a plurality of individuals; constructing a fitness function according to the fault types included in the original sample set; performing genetic iteration on the initial population according to the fitness function to obtain a target population; determining the optimal individual in the target population; determining an optimized feature set of an original sample set according to the chromosome of the optimal individual; processing each original sample in the original sample set according to the optimized feature set to generate a training sample, and obtaining a training sample set comprising a plurality of training samples; and training the classifier according to the training sample set.
Optionally, in the method according to the present invention, the original samples include a plurality of feature dimensions, and encoding each original sample in the original sample set to generate an individual includes: binary coding the feature value of each feature dimension in the original sample generates individual chromosome codes.
Optionally, in the method according to the present invention, constructing the fitness function from the fault types included in the original sample set comprises the steps of: constructing an intra-class divergence matrix and an inter-class divergence matrix according to the number of fault types included in an original sample set, the number of corresponding characteristic dimensions held by the original samples, the number of the original samples included in each class of fault types and the ratio in the original sample set; and constructing a fitness function according to the trace of the intra-class divergence matrix and the trace of the inter-class divergence matrix.
Optionally, in the method according to the present invention, performing genetic iteration on the initial population according to the fitness function to obtain the target population comprises the steps of: setting iteration times, mutation operators, selection operators, hybridization operators and scale factors for genetic iteration of the initial population; and performing genetic iteration on the initial population according to the iteration times according to the fitness function, the mutation operator, the selection operator, the hybridization operator and the scale factor.
Optionally, in the method according to the present invention, when performing genetic iteration on the initial population according to iteration times according to the fitness function, the mutation operator, the selection operator, the hybridization operator, and the adulteration expression, each genetic iteration execution step includes: calculating a fitness score for each individual in the initial population according to a fitness function; selecting a plurality of individuals from the initial population according to the selection operator and the fitness score; a selection operation is performed on each of the selected plurality of individuals, resulting in a first intermediate population.
Optionally, in the method according to the present invention, further comprising the step of: selecting a plurality of individuals from the first intermediate population according to a crossover operator; performing a cross-breeding operation on each of the selected plurality of individuals to obtain a second intermediate population.
Optionally, in the method according to the present invention, further comprising the step of: selecting a plurality of individuals from the second intermediate population according to the mutation operator; and performing mutation operation on each individual in the selected plurality of individuals to obtain a third intermediate population.
Optionally, in the method according to the present invention, further comprising the step of: calculating population entropy for the third intermediate population based on the chromosomes of each individual in the third intermediate population; calculating a doping operator according to the population entropy and the scale factor of the third intermediate population; and generating a new individual according to the doping operator, and replacing the third intermediate population to obtain a fourth intermediate population.
Optionally, in the method according to the present invention, further comprising the step of: judging whether the genetic iteration is the last genetic iteration, and if not, taking the fourth intermediate population as the initial population for next iteration; and if the genetic iteration is the last genetic iteration, taking the population obtained after the doping operation as a target population.
Optionally, in the method according to the present invention, determining the optimal individual in the target quasi-group comprises the steps of: and calculating the fitness score of each individual in the target population according to the fitness function, and taking the individual with the highest fitness score of the target population as the optimal individual.
Optionally, in the method according to the present invention, the chromosome of the individual includes a plurality of genes, each gene corresponding to a feature dimension of the original sample, and the determining the optimized feature set of the original sample set according to the chromosome of the optimal individual includes the steps of: determining one or more genes retained by the chromosome of the optimal individual; and taking the feature dimension corresponding to each gene reserved by the optimal individual as an optimization feature to obtain an optimization feature set.
Optionally, in the method according to the present invention, processing each original sample in the original sample set according to the optimized feature set to generate the training sample includes: and (4) retaining the optimized features belonging to the optimized feature set in the original sample, and deleting the feature dimensions of the non-optimized features in the original sample to obtain the original sample only containing the optimized features as a training sample.
Optionally, in the method according to the present invention, training the classifier according to the training sample set comprises the steps of: and inputting the training samples in the optimized feature set into a classifier, and adjusting parameters of the classifier to enable the classifier to output fault types corresponding to the input training samples when the classifier inputs the training samples.
Optionally, in the method according to the present invention, each device connected to the service is deployed with a plurality of sensors, and acquiring the device data set from the device comprises the steps of: acquiring the acquired operation data from each sensor deployed in the equipment to obtain a plurality of items of operation data of the equipment; and recording a plurality of items of operation data of the equipment as one piece of operation data in the equipment data set.
Optionally, in the method according to the present invention, processing each operation data record of the operation data set to generate a raw sample includes the steps of: performing fast Fourier transform on each operating data in the operating data records to generate frequency spectrum data to obtain a plurality of frequency spectrum data in different directions; drawing a two-dimensional holographic spectrogram according to the frequency spectrum data in a plurality of different directions; and generating an original sample according to the two-dimensional holographic spectrogram.
According to another aspect of the present invention, there is provided a computing device comprising: one or more processors; a memory; an image display system; and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including instructions for performing any of the methods of the device fault detection classifier training method according to the present invention.
According to a further aspect of the invention, there is provided a computer readable storage medium storing one or more programs, the one or more programs comprising instructions, which when executed by a computing device, cause the computing device to perform any of a device failure detection classifier training method according to the invention.
The training method of the equipment fault detection classifier is suitable for being executed in a server, and the server is in communication connection with one or more pieces of equipment. Firstly, a server acquires a device operation data set from a device, and processes each operation data record of the operation data set to generate an original sample. Because the number of items recorded by the running data of the equipment is large, the feature dimension of the original sample is large, and the feature dimension of the original sample is difficult to correspond to the fault type corresponding to the original sample set. Therefore, the invention adopts the genetic algorithm to determine the optimized feature set from the original sample set, the optimized feature set selects the training sample, and trains the classifier, thereby improving the classification accuracy of the classifier, simultaneously reducing the data amount of the classifier for training, eliminating the redundant feature dimension and shortening the training time.
Furthermore, in order to avoid too fast convergence of the genetic algorithm when the genetic algorithm is adopted to determine the optimized feature set from the original sample set, a variable doping operator is added into the genetic algorithm to perform doping operation on the population, so that the convergence speed of the genetic algorithm is adjusted and the calculation accuracy of the obtained value is improved. The doping operator is obtained by calculating the population entropy of the population according to the population entropy and the scale factor, the size of the doping operator is dynamically adjusted in a self-adaptive mode, and the amplitude of each doping operation is controlled, so that the genetic algorithm obtains a more accurate optimization feature set.
Drawings
To the accomplishment of the foregoing and related ends, certain illustrative aspects are described herein in connection with the following description and the annexed drawings, which are indicative of various ways in which the principles disclosed herein may be practiced, and all aspects and equivalents thereof are intended to be within the scope of the claimed subject matter. The above and other objects, features and advantages of the present disclosure will become more apparent from the following detailed description read in conjunction with the accompanying drawings. Throughout this disclosure, like reference numerals generally refer to like parts or elements.
FIG. 1 shows a schematic diagram of a server and device communication connection according to an example embodiment of the present invention;
FIG. 2 illustrates a block diagram of a computing device 200, according to an exemplary embodiment of the invention;
FIG. 3 shows a flowchart of a device fault detection classifier training method 300 according to an exemplary embodiment of the invention
FIG. 4 shows a schematic flow chart for performing a genetic algorithm according to yet another exemplary embodiment of the present invention;
FIG. 5a shows a two-dimensional hologram of a rotor misalignment according to an exemplary embodiment of the present invention;
FIG. 5b shows a two-dimensional hologram of a rotor dynamic and static rub according to an exemplary embodiment of the present invention;
FIG. 5c shows a two-dimensional hologram of a rotor imbalance according to an exemplary embodiment of the invention;
FIG. 6 shows a schematic flow diagram of each genetic iteration performed on an initial population according to an exemplary embodiment of the present invention;
FIG. 7a shows a fitness score change curve for an optimal individual during an iterative process of inheritance according to an exemplary embodiment of the present invention;
FIG. 7b shows a schematic diagram of the variation of chromosome codes during genetic iterations for an optimal individual according to an exemplary embodiment of the invention;
FIG. 8a shows a schematic distribution of original samples in a feature space according to an exemplary embodiment of the present invention;
FIG. 8b shows a graph of different doping rates as an iteration progresses according to an exemplary embodiment of the present invention;
FIG. 8c shows a plot of population entropy for different doping rates in accordance with an exemplary embodiment of the present invention;
FIG. 8d shows a graph of the average fitness for different doping levels according to an exemplary embodiment of the present invention;
FIG. 8e shows the average evolutionary algebra resulting in an optimized feature set at different doping rates according to an exemplary embodiment of the invention. And
fig. 8f shows a schematic diagram of the change of the chromosome code of the optimal individual during the genetic iteration under the adaptive doping rate according to an exemplary embodiment of the invention.
Detailed Description
Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art. Like reference numerals generally refer to like parts or elements.
FIG. 1 shows a schematic diagram of a server and device communication connection according to an exemplary embodiment of the invention. As shown in FIG. 1, a server 140 is communicatively coupled to the devices 110-130. The devices 110-130 may be implemented as a mechanical device, and the specific type of the devices 110-130 is not limited in the present invention. The equipment 110-130 are all provided with a plurality of sensors, the equipment 110 is provided with the sensors 111-113, the equipment 120 is provided with the sensors 121-123, and the equipment 130 is provided with the sensors 131-133. Each sensor of the devices 110-130 is used to collect an operational data of the device. The invention does not limit the type and number of sensors deployed in the devices 110-130 and the type of operational data collected. When the scheme of the invention is specifically realized, different numbers and types of sensors can be flexibly deployed in the equipment according to actual needs to collect required operation data, for example, the sensors can be deployed according to operation data such as vibration, noise, temperature and the like in any equipment as required.
The server 140 acquires device operation data from sensors of the devices 110 to 130 communicatively connected thereto, and when a plurality of devices and respective sets of operation data thereof are acquired, the acquired operation data constitute an operation data set. The operational data set may include a plurality of operational data records, each operational data record including a plurality of operational data collected by a plurality of sensors of a piece of equipment. The server 140 performs multi-step processing on the operation data set to obtain a training sample set, and trains the classifier by using the training sample set, so as to analyze and process the subsequently received operation data by using the trained classifier.
The server 140 may be embodied as a computing device. FIG. 2 illustrates a block diagram of a computing device 200, according to an exemplary embodiment of the invention. As shown in FIG. 2, in a basic configuration 202, a computing device 200 typically includes a system memory 206 and one or more processors 204. A memory bus 208 may be used for communication between the processor 204 and the system memory 206.
Depending on the desired configuration, the processor 204 may be any type of processing, including but not limited to: a microprocessor (μ P), a microcontroller (μ C), a Digital Signal Processor (DSP), or any combination thereof. The processor 204 may include one or more levels of cache, such as a level one cache 210 and a level two cache 212, a processor core 214, and registers 216. Example processor cores 214 may include Arithmetic Logic Units (ALUs), Floating Point Units (FPUs), digital signal processing cores (DSP cores), or any combination thereof. The example memory controller 218 may be used with the processor 204, or in some implementations the memory controller 218 may be an internal part of the processor 204.
Depending on the desired configuration, system memory 206 may be any type of memory, including but not limited to: volatile memory (such as RAM), non-volatile memory (such as ROM, flash memory, etc.), or any combination thereof. System memory 206 may include an operating system 220, one or more programs 222, and program data 228. In some embodiments, the program 222 may be arranged to execute the instructions 223 of the method 300 according to the invention on an operating system by one or more processors 204 using the program data 228.
A network communication link may be one example of a communication medium. Communication media may typically be embodied by computer readable instructions, data structures, program modules, and may include any information delivery media, such as carrier waves or other transport mechanisms, in a modulated data signal. A "modulated data signal" may be a signal that has one or more of its data set or its changes made in such a manner as to encode information in the signal. By way of non-limiting example, communication media may include wired media such as a wired network or private-wired network, and various wireless media such as acoustic, Radio Frequency (RF), microwave, Infrared (IR), or other wireless media. The term computer readable media as used herein may include both storage media and communication media.
In a computing device 200 according to the present invention, the program 222 includes a plurality of program instructions of a device fault detection classifier training method 300 that may instruct the processor 204 to perform some of the steps of the device fault detection classifier training method 300 operating in the computing device 200 of the present invention such that the various portions of the computing device 200 implement training of a classifier for device fault detection by performing the device fault detection classifier training method 300 of the present invention.
FIG. 3 shows a flowchart of a device fault detection classifier training method 300 according to an exemplary embodiment of the invention. The method 300 is suitable for execution in the server 140 and may be embodied as being executed in the computing device 200. As shown in fig. 3, the training method 300 for device fault detection classifier starts with step S310, obtaining a device operation data set from a device, and processing each operation data record of the operation data set to generate a raw sample, so as to obtain a raw sample set including a plurality of raw samples, where the raw sample set includes raw samples of a plurality of fault types. The method 300 is described below using the device 110 and the server 140 as examples.
When server 140 obtains the device operational data set from device 110, the collected operational data is obtained from each sensor deployed in device 110, resulting in a plurality of operational data for device 110. Sensors 111-113 in equipment 110 each collect an item of operating data of equipment 110, and server 140 records a plurality of items of operating data of the equipment as an item of operating data in an equipment data set. Sensors 111-113 in equipment 110 may collect operational data of equipment 110 multiple times, with each operational data record representing multiple operational data of the equipment at a time.
Processing each operation data record of the operation data set to generate an original sample, firstly performing fast Fourier transform on each operation data record to generate frequency spectrum data to obtain frequency spectrum data in a plurality of different directions, secondly drawing a two-dimensional holographic spectrogram according to the frequency spectrum data in the plurality of different directions, and then generating the original sample according to the two-dimensional holographic spectrogram.
In step S310, the original sample also needs to be labeled, and the original sample corresponds to the equipment failure. When the original sample is labeled, the original sample can be judged manually, and the fault type of the sample is judged. The method for marking the original sample is not limited, and the fault type corresponding to the original sample can be obtained through different marking modes.
According to one embodiment of the present invention, the device 110 is a large rotating mechanical device, and the sensors 111 and 112 collect vibration data of two sections (section 1 and section 2) of the device 110 respectively. Fig. 4 shows a flow diagram for processing a running data record according to an exemplary embodiment of the present invention. As shown in fig. 4, the server 140 acquires vibration data of the section 1 and the section 2 of the device 110, and records the two vibration data as one piece of operation data. Firstly, data preprocessing is carried out on the operation data record, vibration data of the section 1 are decomposed into X-direction vibration and Y-direction vibration, and vibration data of the interface 2 are also decomposed into X-direction vibration and Y-direction vibration.
And then performing fast Fourier transform on the vibration data (X-direction vibration and Y-direction vibration) of the section 1 and the vibration data (X-direction vibration and Y-direction vibration) of the section 2 in the operation data record to obtain frequency spectrum data, wherein the frequency spectrum data comprises an X-direction frequency spectrum and a Y-direction frequency spectrum of the section 1 and an X-direction frequency spectrum and a Y-direction frequency spectrum of the section 2.
When a two-dimensional holographic spectrogram is drawn according to a plurality of frequency spectrum data in different directions, power frequency components are calculated according to the frequency spectrum data, the X-direction frequency spectrum and the Y-direction frequency spectrum of the section 1 and the X-direction frequency spectrum and the Y-direction frequency spectrum of the section 2 to obtain the X-direction power frequency component and the Y-direction power frequency component of the section 1 and the X-direction power frequency component and the Y-direction power frequency component of the section 2. And integrating the X-direction power frequency component and the Y-direction power frequency component of the cross section 1 to obtain a power frequency ellipse of the cross section 1, and integrating the X-direction power frequency component and the Y-direction power frequency component of the cross section 2 to obtain a power frequency ellipse of the cross section 2 in the same way. And finally, integrating the power frequency ellipse of the section 1 and the power frequency ellipse of the section 2 to obtain the two-dimensional hologram of the vibration data of the equipment 110.
In the fault diagnosis practice of rotary machines, there are three common fault types that cause synchronous vibration, namely, misalignment of the rotor, dynamic and static rubbing of the rotor, and unbalance of the rotor. Fig. 5a shows a two-dimensional hologram in a rotor misalignment according to an exemplary embodiment of the present invention. FIG. 5b shows a two-dimensional hologram of the dynamic and static rub of a rotor according to an exemplary embodiment of the present invention. Fig. 5c shows a two-dimensional hologram of a rotor imbalance according to an exemplary embodiment of the present invention. The two-dimensional hologram drawn by the vibration data of the device 110 is drawn according to different values of the data to obtain any one of the images shown in fig. 5a to 5 c.
The original sample is then generated from the two-dimensional holographic spectrogram. And when the original sample is generated according to the two-dimensional holographic spectrogram, extracting and calculating a characteristic value of a characteristic dimension corresponding to the original sample from the two-dimensional holographic spectrogram. Geometric parameters of 1X-5X frequency ellipses in the two-dimensional holographic spectrogram are as follows: the long axis 2a, the short axis 2b, the eccentricity e, the inclination angle phi of the long axis and the initial phase angle theta are normalized. Wherein, the major axis and the minor axis are normalized according to the power frequency ellipse to obtain characteristic values corresponding to 23 characteristic dimensions.
And after generating an original sample for each operation data record of the operation data set, marking the original samples in the original sample set after obtaining the original sample set, and marking the fault types corresponding to the original samples. When the original sample is marked, the two-dimensional holographic spectrogram of the original sample can be manually checked, and the fault type of the original sample is manually marked according to the two-dimensional holographic spectrogram.
Mechanical failure diagnosis is a typical mode classification problem, and an original sample set is obtained by processing an operation data set after the operation data set of equipment is obtained through a sensor. The original samples in the original sample set have more feature dimensions, and if the original samples are classified directly according to the feature dimensions, a good classification effect is difficult to obtain due to high redundancy of the original samples. Therefore, in order to improve the recognition accuracy, two types of redundant feature quantities must be removed before designing the classifier: 1. a feature quantity unrelated to the classification target; 2. selecting a group of optimal features with the number D (D < D) from a group of D features to minimize the classification error rate; therefore, the purification of the diagnosis characteristics is realized through the optimized selection of the characteristics, and finally, the classifier is learned and trained to effectively classify.
In order to reduce the number of feature dimensions of the training classifier, the original sample set needs to be processed to obtain a training sample set. And when the original sample set is processed to obtain a training sample set, determining an optimized feature set from the original sample set by adopting a genetic algorithm, and selecting feature dimensions of fault types from the original samples according to the optimized feature set to construct the training sample set.
Genetic algorithms solve the optimization problem by simulating hybridization, variation and natural selection during the evolution of organisms. Due to good parallelism, universality and robustness, the method has become a focus of attention of many subjects such as information science, computer science and artificial intelligence. In the evolution process of the biological world, the origin and development of life are gradually transited from a disordered state to an ordered state, and the evolution process is a typical self-organizing process, and genetic algorithms also have the characteristic.
To apply the genetic algorithm, step S320 is first performed to encode each original sample in the original sample set to generate an individual, and an initial population including a plurality of individuals is obtained. And when each original sample is coded, binary coding is carried out on the characteristic value of each characteristic dimension in the original sample to generate an individual chromosome code. When binary coding is carried out on the characteristic dimension, if chromosomes share x genes, the ith position is 1, the characteristic is marked to be selected, and if the ith position is 0, the characteristic is not selected.
According to one embodiment of the invention, the major axis and the minor axis of the geometric parameters of each frequency ellipse of 1X-5X in the two-dimensional holographic spectrogram are normalized according to the power frequency ellipse to obtain 23 characteristic values corresponding to the characteristic dimensions, and the 23 characteristic dimensions are subjected to binary coding to obtain 23 chromosomes in total. Specifically, when the binary encoding is performed, if the normalized feature value is less than 0.5, the normalized feature value may be encoded as 0, and if the normalized feature value is not less than 0.5, the normalized feature value may be encoded as 1.
In genetic algorithms, it is necessary to perform operations such as screening of a population to be inherited using a fitness function. The method adopts the intra-class inter-class distance as the fitness function, can fully utilize the hidden parallelism of the genetic algorithm, has strong parallelism and optimizing capability, efficiently eliminates the redundant features of the original sample set, and searches the optimal feature set for classifier design. The generated optimized feature set is used for training the neural network classifier, and the fault identification precision can be improved.
Subsequently, step S330 is performed to construct a fitness function according to the fault types included in the original sample set. When the redundant features in the original sample set are removed, different feature dimensions of the original samples are located in different areas in the feature space. When the fitness function is constructed, an intra-class divergence matrix and an inter-class divergence matrix are constructed according to the number of fault types included in an original sample set, the number of corresponding characteristic dimensions of the original samples, the number of original samples included in each class of fault types and the ratio in the original sample set, and then the fitness function is constructed according to the trace of the intra-class divergence matrix and the trace of the inter-class divergence matrix.
The expression of the intra-class divergence matrix is:
the expression of the inter-class divergence matrix is:
wherein:
c is the number of fault types included in the original sample set, when calculating the fitness score, namely the number of fault types in the corresponding initial population,
when calculating the fitness score for the kth original sample in the ith type of fault of the original sample set, the kth original sample is the kth individual in the ith type of fault of the corresponding initial population, niCalculating the fitness score for the number of original samples included in the ith fault type, i.e. the number of individuals included in the ith fault type of the corresponding initial population, PiCalculating the fitness score for the ratio of the ith fault type in all samples of the original sample set, namely the ratio of the ith fault type in all individuals of the initial population.
In order to maximize the inter-class divergence of the extracted feature dimensions, and minimize the intra-class divergence, a corresponding fitness function is constructed, the expression of which is:
where tr represents the trace operation, the return of which is the sum of the diagonal elements of the matrix.
Subsequently, step S330 is executed to perform genetic iteration on the initial population according to the fitness function to obtain the target population. When genetic iteration is performed on the initial population according to the fitness function, the iteration number, the mutation operator, the selection operator, the hybridization operator and the scale factor for genetic iteration on the initial population are set, and then genetic iteration is performed on the initial population according to the iteration number according to the fitness function, the mutation operator, the selection operator, the hybridization operator and the scale factor.
The invention does not limit the specific numerical values of the iteration times, mutation operators, selection operators, hybridization operators and scale factors, and when the method 300 of the invention is specifically implemented, the parameters can be flexibly set according to the requirements of the genetic algorithm. According to a preferred embodiment of the invention, the number of iterations is set to 200, the mutation operator is 0.06, the selection operator is 30%, the hybridization operator is 0.6, and the scale factor is 2.
Fig. 6 shows a schematic flow diagram of each genetic iteration performed on an initial population according to an exemplary embodiment of the present invention. As shown in fig. 6, when genetic iteration is performed on the initial population according to the iteration number according to the fitness function, the mutation operator, the selection operator, the hybridization operator, and the expression of the adulteration rate, the following steps are performed for each genetic iteration:
firstly, calculating a fitness score for each individual in the initial population according to the fitness function, and substituting the related parameters of each individual into the fitness function to obtain the fitness score of the individual.
A plurality of individuals is then selected from the initial population according to the selection operator and the fitness score. When selecting individuals from the initial population, sorting the individuals according to the fitness scores of all the individuals in the initial population, and selecting the individuals with the highest fitness scores in the initial population according to the proportion specified by the selection operator. According to one embodiment of the present invention, the selection operator is set to 30%, and then the individuals with the initial population number of 30% with the highest fitness score are selected.
A selection operation is then performed on each of the selected plurality of individuals to obtain a first intermediate population. When the selection operation is performed on the individuals, new individuals are added into the population by randomly crossing paired chromosomes of every two individuals from the selected individuals, and corresponding numbers of individuals with lower fitness in the initial population are replaced, so that a first intermediate population is obtained.
And then selecting a plurality of individuals from the first intermediate population according to a hybridization operator, and carrying out a hybridization operation on each individual of the selected plurality of individuals to obtain a second intermediate population. And when the hybridization operation is carried out according to the hybridization operator, randomly selecting a plurality of individuals in the second intermediate population according to the proportion specified by the hybridization operator, and randomly pairing chromosomes of every two individuals from the selected plurality of individuals to obtain new two individuals, putting the new two individuals back into the population, thereby obtaining a third intermediate population. The manner of chromosomal crossing can be by way of interchanging random pairs of genes, and the invention is not limited to the number of loci that are interchanged in the crossing. According to one embodiment of the present invention, the hybridization operator is set to 0.6, and then 60% of the number of individuals from the first intermediate population are randomly selected for the hybridization operation.
And then selecting a plurality of individuals from the second intermediate population according to the mutation operator, and performing mutation operation on each individual of the selected plurality of individuals to obtain a third intermediate population. And when mutation operation is carried out according to the mutation operator, a plurality of individuals in the second intermediate population are randomly selected according to the proportion specified by the mutation operator, the value of one gene is randomly changed on the chromosome of each individual in the plurality of individuals, and the chromosome is put back into the population to obtain a third intermediate population. For example, if the value of a certain gene in the individual is 1, the value is changed to 0, and if the value of the gene is 0, the value is changed to 1. The present invention does not limit the number of genes altered per chromosome. According to an embodiment of the present invention, the mutation operator is set to 0.06, and 6% of the number of individuals in the second intermediate population are randomly selected for mutation.
The execution process of the genetic algorithm is a typical dual target optimization process. On the one hand, it needs to ensure that the optimal individual can evolve generation by generation in the population and finally can stand out; on the other hand, the diversity of population individuals is ensured, and the distribution divergence is certain. The former is the final goal of optimization and the latter is material assurance to avoid premature convergence. The hybridization, mutation and selection of the above steps play different roles in the self-organizing evolution of populations. The hybridization operation is the mutual exchange of genetic information among individuals in a population, and the total amount of genetic information in a system is kept unchanged. The hybridization operator can generate more excellent individuals, but cannot ensure the diversity of population individuals, so that the population entropy is kept constant.
The mutation operation is equivalent to the function of sequence parameters in the self-organization process, and the total amount of genetic information in the population is increased by performing the mutation operation on the population individuals, so that the diversity of the population individuals is increased to a certain extent, the population entropy is increased to a certain extent, and the improvement of the fitness of the individuals cannot be ensured.
The selection operation selects a part of individuals with high fitness to be copied by a principle of high-out and low-out to form a new population, which ensures that the individuals with high fitness can be dredged, but reduces the total amount of population genetic information to a certain extent and reduces the diversity of population individuals.
The traditional genetic operation can ensure the improvement of the average fitness of the population, so that the optimal individual can stand out, but the divergence of population distribution cannot be ensured, and therefore, the population is easy to fall into local minimum, thereby causing premature convergence. Therefore, in order to avoid premature convergence, a new operator must be introduced to ensure that the optimal individual can grow continuously on the premise of population individual diversity. The doping operator (also called a catastrophe operator) is a common means for improving the population distribution divergence, the doping operation process corresponding to the doping operator is very simple, and after each iteration, some new individuals are randomly generated to replace the original partial individuals so as to form a new population. The choice of doping rate will directly influence the optimization effect of the genetic algorithm. On one hand, if the doping rate is too high, the overall process of population evolution is slowed down to a certain extent; on the other hand, the doping ratio is too small, and the distribution divergence of the population cannot be effectively improved, thereby causing premature convergence.
The invention adopts the self-adaptive principle to determine the doping rate, and selects the smaller doping rate when the population distribution divergence is larger so as to accelerate the convergence speed of the algorithm; when the population distribution divergence is small, a large doping rate is selected to avoid premature convergence of the algorithm. In order to know the divergence program of the population distribution, parameters are needed to be used for describing the population distribution, and the information entropy of the system is the most effective tool for describing the self-organization behavior.
In order to analyze the transformation situation of the total amount of genetic information in a population in the evolution process, the concept of population entropy E is introduced. In the evolution process, individuals of the initial population are randomly selected, so that the distribution divergence is maximum, the distribution divergence of the population is gradually reduced along with the evolution, and the population entropy quantitatively and visually depicts the attenuation process.
Population entropy of the third intermediate population is then calculated based on the chromosomes of each individual in the third intermediate population. The expression of population entropy is as follows:
wherein p isiThe frequency of the ith gene of all individuals in the population is 1, and D is the number of genes included in chromosomes of the individuals in the population.
Calculating a doping operator according to the population entropy and the scale factor of the third intermediate population, wherein the expression of the doping operator is as follows:
the adoption of the genetic algorithm based on the self-adaptive doping rate to remove the redundant features can improve the classification progress of the trained classifier. And when the third intermediate population is doped according to the doping operator, generating a new individual according to the doping operator, and replacing the third intermediate population to obtain a fourth intermediate population. Firstly, the doping operator is multiplied by the number of individuals included in the third intermediate population to obtain the doping number, then new individuals with the doping number are randomly generated, and the individuals of the third intermediate population are randomly replaced to complete the doping operation.
Then, judging whether the genetic iteration is the last genetic iteration, if not, taking the fourth intermediate population as the initial population for the next iteration until the set iteration times are finished, performing the next iteration on the population, and repeatedly executing the step in the step S340; and if the genetic iteration is the last genetic iteration, taking the population obtained after the doping operation as a target population.
Subsequently, step S350 is executed to determine the optimal individual in the target population, perform fitness score calculation for each individual in the target population according to the fitness function, and take the individual with the highest fitness score of the target population as the optimal individual. The manner of calculating the fitness score for each individual in the target population is the same as the manner of calculating the fitness score in step S340, and is not repeated here.
Subsequently, step S360 is performed to determine an optimized feature set of the original sample set according to the chromosome of the optimal individual. And when the optimization feature set is determined, determining one or more genes reserved in the chromosome of the optimal individual, and taking the feature dimension corresponding to each gene reserved in the optimal individual as the optimization feature to obtain the optimization feature set.
FIG. 7a shows a fitness score variation curve during genetic iteration for an optimal individual according to an exemplary embodiment of the present invention. FIG. 7b shows a schematic diagram of the change of chromosome code during genetic iteration for an optimal individual according to an exemplary embodiment of the invention. As shown in fig. 7a, the number of genetic iterations is set to 200, the fitness of the optimal individual gradually increases and decreases as the number of iterations increases, and the fitness score converges to 1.08 at 135 genetic iterations. As shown in fig. 7b, as the number of iterations increases, the chromosomes of the optimal individual gradually discard other irrelevant features that classify the fault features, and finally retain genes corresponding to feature dimensions of 2X to 5X long axis and eccentricity of 1X, 2X, 3X, 5X. From this, the major axis of 2X-5X, eccentricity of 1X, 2X, 3X, 5X, which are the optimization features included in the optimization feature set, can be derived.
Subsequently, step S370 is executed to process each original sample in the original sample set according to the optimized feature set to generate a training sample, resulting in a training sample set including a plurality of training samples. When the training sample is generated, the optimized features belonging to the optimized feature set in the original sample are reserved, the feature dimensions of the non-optimized features in the original sample are deleted, and the original sample only containing the optimized features is obtained and used as the training sample. The optimized features included in the optimized feature set are feature dimensions which need to be reserved in the original sample, and other feature dimensions which are not selected as the optimized features in the original sample are removed without participating in the subsequent training of the classifier.
According to one embodiment of the invention, when generating the training samples, the original samples are divided by the optimization features: and (3) removing the long axis of 2X-5X, the eccentricity of 1X, 2X, 3X and 5X, and the characteristic dimension and the characteristic value thereof except the eccentricity, and only keeping the characteristic dimension which becomes the optimized characteristic in the optimized characteristic set. And processing each original sample to obtain a training sample, and further obtaining a training sample set.
And finally, executing step S380, training the classifier according to the training sample set, inputting the training samples in the optimized feature set into the classifier, and adjusting parameters of the classifier to enable the classifier to output fault types corresponding to the input training samples when inputting the training samples. The invention does not limit the training mode adopted when the classifier is used for training the neural network, and the classifier trained by using the optimized feature set generated by the execution method 300 can improve the identification progress of the classifier on the fault type.
To test the ability of the method 300 of the present invention to determine optimized features from a set of raw features using a genetic algorithm, a simulation experiment was performed to set the method 300. The simulation experiment comprises the following steps: an original sample set is set, wherein the original sample set comprises two types of original samples, and each type of original sample comprises 100 original samples. The number of feature dimensions of the original sample is 23, wherein the first 3 features are optimized features, and the last 20 features are subjected to random number dereferencing between 0 and 1, and do not contain feature dimensions related to fault types.
Fig. 8a shows a schematic distribution diagram of original samples in a feature space according to an exemplary embodiment of the present invention. As shown in fig. 8a, feature 1, feature 2 and feature 3 are respectively taken as three coordinate axes of the feature space. The two types of original samples of the original sample set are respectively a type 0 original sample and a type 1 original sample.
In the simulation experiment, the population size of the initial population is set to be 50, and the initial population comprises 50 individuals. The crossover operator was 0.6, the mutation operator was 0.06, the selection operator was 30%, and the scale factor was 2. To examine the optimized performance of the genetic algorithm based on adaptive doping rates, comparative experiments were performed using three different doping schemes. The doping rates of the comparative experiments are pd ═ 0 respectively; pd is 0.2; pd is 0.4.
Fig. 8b shows a graph of the variation of different doping rates as an iteration progresses according to an exemplary embodiment of the invention. Fig. 8c shows a graph of the variation of population entropy at different doping rates according to an exemplary embodiment of the present invention. As shown in fig. 8c, at a fixed doping rate, population entropy decreases monotonically as the genetic iteration progresses. And the larger the doping ratio, the larger the decreasing amplitude thereof. Comparing fig. 8b with fig. 8c, it is found that there is an inverse interaction relationship between population entropy and doping rate under the action of the adaptive doping operator. Population entropy decreases, meaning that the population distribution divergence decreases, then the doping rate rises adaptively. In the evolution process, in order to maintain the requirement of population distribution divergence, the doping rate shows wave-like rise along with the evolution of the population.
Fig. 8d shows a graph of the average fitness for different doping rates according to an exemplary embodiment of the present invention. As shown in fig. 8d, the increase in average fitness is slowed down as the doping ratio is increased. FIG. 8e shows the average evolutionary algebra resulting in an optimized feature set at different doping rates according to an exemplary embodiment of the invention. As shown in fig. 8e, the larger the doping ratio is, the more times of iteration is required, and for the adaptive hybrid operator, since the improvement of the average fitness and the diversity of population distribution are considered at the same time, the premature convergence of the genetic algorithm is avoided, the optimization performance of the genetic algorithm can be effectively improved, and the average evolution algebra is reduced.
Fig. 8f shows a schematic diagram of the change of the chromosome code of the optimal individual during the genetic iteration under the adaptive doping rate according to an exemplary embodiment of the invention. As shown in fig. 8f, the optimal individuals gradually discarded the last 20-dimensional redundant features during the genetic iteration of the population. The optimal feature set containing a single feature is [10000000000000000000000 ]; the optimal feature set comprising two features is [11000000000000000000000 ]; the most preferred feature comprising three features is [11100000000000000000000 ]; this is consistent with the set-up of the simulation experiment. Therefore, the genetic algorithm based on the self-adaptive doping rate has excellent parallelism and strong optimizing capability in feature selection.
According to the method 300 and the simulation experiment, the method 300 of the present invention can be obtained to solve various defects of the traditional combinatorial optimization method for feature selection, and on the basis of the adaptive doping rate, the intra-class-inter-class distance criterion J is provided as a fitness function, so that the method has strong parallelism and optimization capability, and can efficiently remove redundant features of the original feature set. The generated optimized feature set is used for training a neural network classifier, so that the identification precision of the fault can be improved based on a feature selection strategy of a genetic algorithm. Simulation and rotary machine fault diagnosis example analysis prove that the method has stronger parallelism and optimizing capability. The diagnostic accuracy and efficiency can be improved by automatically selecting the feature combination, and redundant information can be effectively eliminated; and the genetic optimization algorithm based on the self-adaptive doping rate avoids premature convergence in feature selection, and can fully utilize the implicit parallelism of the genetic algorithm to search an optimal feature set for classifier design.
A9, the method of A8, wherein the method further comprises the steps of: judging whether the genetic iteration is the last genetic iteration, and if not, taking the fourth intermediate population as the initial population for next iteration; and if the genetic iteration is the last genetic iteration, taking the population obtained after the doping operation as a target population.
A10, the method of A9, wherein said determining the optimal individual in said target quasi-group comprises the steps of: and calculating the fitness score of each individual in the target population according to the fitness function, and taking the individual with the highest fitness score of the target population as the optimal individual.
A11, the method as claimed in a10, wherein the chromosome of the individual includes a plurality of genes, each gene corresponding to a characteristic dimension of the original sample, the determining the set of optimized features of the original sample set from the chromosome of the optimal individual includes the steps of: determining one or more genes retained by the chromosome of the optimal individual; and taking the feature dimension corresponding to each gene reserved by the optimal individual as an optimization feature to obtain an optimization feature set.
A12, the method as in A11, wherein the processing each original sample in the original sample set according to the optimized feature set to generate the training sample comprises the steps of: and reserving the optimized features belonging to the optimized feature set in the original sample, and deleting the feature dimensions of the non-optimized features in the original sample to obtain the original sample only containing the optimized features as a training sample.
A13, the method of A14, wherein the training a classifier according to the training sample set comprises the steps of: and inputting the training samples in the optimized feature set into a classifier, and adjusting parameters of the classifier to enable the classifier to output fault types corresponding to the input training samples when the classifier inputs the training samples.
A14, the method of a13, wherein each device connected to the service has deployed a plurality of sensors, the obtaining of device data sets from the devices comprises the steps of: acquiring the acquired operation data from each sensor deployed in the equipment to obtain a plurality of items of operation data of the equipment; and recording a plurality of items of operation data of the equipment as one operation data in the equipment data set.
A15, the method as in a14, wherein the processing each operational data record of the operational data set to generate a raw sample comprises the steps of: performing fast Fourier transform on each operating data in the operating data records to generate frequency spectrum data, and obtaining a plurality of frequency spectrum data in different directions; drawing a two-dimensional holographic spectrogram according to the frequency spectrum data in a plurality of different directions; and generating an original sample according to the two-dimensional holographic spectrogram.
Claims (10)
1. An equipment fault detection classifier training method, suitable for being executed in a server, wherein the server is in communication connection with one or more pieces of equipment, and the method comprises the following steps:
acquiring an equipment operation data set from the equipment, and processing each operation data record of the operation data set to generate an original sample to obtain an original sample set comprising a plurality of original samples, wherein the original sample set comprises original samples of a plurality of fault types;
encoding each original sample in the original sample set to generate an individual to obtain an initial population comprising a plurality of individuals;
constructing a fitness function according to the fault types included in the original sample set;
performing genetic iteration on the initial population according to the fitness function to obtain a target population;
determining the optimal individual in the target population;
determining an optimized feature set of the original sample set according to the chromosome of the optimal individual;
processing each original sample in the original sample set according to the optimized feature set to generate a training sample, and obtaining a training sample set comprising a plurality of training samples;
and training a classifier according to the training sample set.
2. The method of claim 1, wherein the original samples comprise a plurality of feature dimensions, and wherein encoding each original sample in the set of original samples to generate an individual comprises:
binary coding the feature values of each feature dimension in the original sample generates a chromosome code of the individual.
3. The method of claim 2, wherein said constructing a fitness function from the fault types included in the original sample set comprises the steps of:
constructing an intra-class divergence matrix and an inter-class divergence matrix according to the number of fault types included in the original sample set, the number of corresponding feature dimensions held by the original samples, the number of the original samples included in each class of fault types and the ratio in the original sample set;
and constructing a fitness function according to the trace of the intra-class divergence matrix and the trace of the inter-class divergence matrix.
4. The method according to any one of claims 1-3, wherein said performing genetic iterations on said initial population according to said fitness function to obtain a target population comprises the steps of:
setting iteration times, mutation operators, selection operators, hybridization operators and scale factors for genetic iteration of the initial population;
and performing genetic iteration on the initial population according to the iteration times according to the fitness function, the mutation operator, the selection operator, the hybridization operator and the scale factor.
5. The method of claim 4, wherein when performing genetic iterations on the initial population according to the fitness function, mutation operator, selection operator, hybridization operator and adulteration expression according to iteration number, each genetic iteration execution step comprises:
calculating a fitness score for each individual in the initial population according to a fitness function;
selecting a plurality of individuals from the initial population according to the selection operator and the fitness score;
a selection operation is performed on each of the selected plurality of individuals, resulting in a first intermediate population.
6. The method of claim 5, wherein the method further comprises the steps of:
selecting a plurality of individuals from the first intermediate population according to the crossover operator;
performing a cross-breeding operation on each of the selected plurality of individuals to obtain a second intermediate population.
7. The method of claim 6, wherein the method further comprises the steps of:
selecting a plurality of individuals from the second intermediate population according to the mutation operator;
and performing mutation operation on each individual in the selected plurality of individuals to obtain a third intermediate population.
8. The method of claim 7, wherein the method further comprises the steps of:
calculating population entropy for the third intermediate population from the chromosomes of each individual in the third intermediate population;
calculating a doping operator according to the population entropy and the scale factor of the third intermediate population;
and generating a new individual according to the doping operator, and replacing the third intermediate population to obtain a fourth intermediate population.
9. A computing device, comprising:
one or more processors;
a memory; and
one or more apparatuses comprising instructions for performing any of the methods of claims 1-8.
10. A computer readable storage medium storing one or more programs, the one or more programs comprising instructions, which when executed by a computing device, cause the computing device to perform any of the methods of claims 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110942911.6A CN113610179A (en) | 2021-08-17 | 2021-08-17 | Equipment fault detection classifier training method, computing equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110942911.6A CN113610179A (en) | 2021-08-17 | 2021-08-17 | Equipment fault detection classifier training method, computing equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113610179A true CN113610179A (en) | 2021-11-05 |
Family
ID=78340918
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110942911.6A Pending CN113610179A (en) | 2021-08-17 | 2021-08-17 | Equipment fault detection classifier training method, computing equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113610179A (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019090878A1 (en) * | 2017-11-09 | 2019-05-16 | 合肥工业大学 | Analog circuit fault diagnosis method based on vector-valued regularized kernel function approximation |
| CN109884419A (en) * | 2018-12-26 | 2019-06-14 | 中南大学 | A kind of wisdom grid power quality on-line fault diagnosis method |
| CN111027697A (en) * | 2019-12-08 | 2020-04-17 | 贵州大学 | Genetic algorithm packaged feature selection power grid intrusion detection method |
| CN112101227A (en) * | 2020-09-16 | 2020-12-18 | 江苏师范大学 | Mechanical state monitoring method based on CELMDAN and SSKFDA |
-
2021
- 2021-08-17 CN CN202110942911.6A patent/CN113610179A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019090878A1 (en) * | 2017-11-09 | 2019-05-16 | 合肥工业大学 | Analog circuit fault diagnosis method based on vector-valued regularized kernel function approximation |
| CN109884419A (en) * | 2018-12-26 | 2019-06-14 | 中南大学 | A kind of wisdom grid power quality on-line fault diagnosis method |
| CN111027697A (en) * | 2019-12-08 | 2020-04-17 | 贵州大学 | Genetic algorithm packaged feature selection power grid intrusion detection method |
| CN112101227A (en) * | 2020-09-16 | 2020-12-18 | 江苏师范大学 | Mechanical state monitoring method based on CELMDAN and SSKFDA |
Non-Patent Citations (4)
| Title |
|---|
| 刘运 等: "基于遗传算法的故障特征优化研究", 《计算机仿真》, vol. 27, no. 4, pages 192 - 196 * |
| 史东锋 等: "遗传算法在故障特征选择中的应用研究", 《振动、测试与诊断》, vol. 20, no. 3, pages 171 - 176 * |
| 徐姣 等: "改进遗传算法优选分形插值参数的短期负荷预测", 《广西电力》, no. 3, pages 1 - 5 * |
| 王正初 等: "水库优化调度的改进蚂蚁算法研究", 《PROCEEDINGS OF THE 8TH WORLD CONGRESS ON INTELLIGENT CONTROL AND AUTOMATION》, pages 5337 - 5342 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Pirgazi et al. | An Efficient hybrid filter-wrapper metaheuristic-based gene selection method for high dimensional datasets | |
| Sun et al. | Local-learning-based feature selection for high-dimensional data analysis | |
| Yusta | Different metaheuristic strategies to solve the feature selection problem | |
| Zhu et al. | Wrapper–filter feature selection algorithm using a memetic framework | |
| Abusamra | A comparative study of feature selection and classification methods for gene expression data of glioma | |
| Molina et al. | Memetic algorithms based on local search chains for large scale continuous optimisation problems: MA-SSW-Chains | |
| Grahn et al. | CudaRF: a CUDA-based implementation of random forests | |
| CN111240995B (en) | Multi-path coverage method and system combining key point probability and path similarity | |
| Astorino et al. | DC models for spherical separation | |
| Damaševičius | Optimization of SVM parameters for recognition of regulatory DNA sequences | |
| Zheng et al. | An automatic sampling ratio detection method based on genetic algorithm for imbalanced data classification | |
| Castellani | Evolutionary generation of neural network classifiers—An empirical comparison | |
| Saranya et al. | Feature selection techniques for disease diagnosis system: A survey | |
| Hu et al. | Combined gene selection methods for microarray data analysis | |
| WO2023148733A1 (en) | Method and system for predicting drug-drug interactions | |
| Al-Sawwa et al. | Performance evaluation of a cost-sensitive differential evolution classifier using spark–Imbalanced binary classification | |
| Yang et al. | Feature selection using memetic algorithms | |
| Sharma et al. | A semi-supervised generalized vae framework for abnormality detection using one-class classification | |
| CN112488188B (en) | Feature selection method based on deep reinforcement learning | |
| Brown et al. | LSHPlace: fast phylogenetic placement using locality-sensitive hashing | |
| Escobar et al. | Assessing parallel heterogeneous computer architectures for multiobjective feature selection on EEG classification | |
| CN107292320A (en) | System and its index optimization method and device | |
| Ding et al. | Going faster and hence further with lexicase selection | |
| CN113610179A (en) | Equipment fault detection classifier training method, computing equipment and storage medium | |
| Guney et al. | A robust ensemble feature selection technique for high‐dimensional datasets based on minimum weight threshold method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |