CN115563490A - Super-large-scale low-orbit satellite fault diagnosis method based on improved locust optimization algorithm - Google Patents
Super-large-scale low-orbit satellite fault diagnosis method based on improved locust optimization algorithm Download PDFInfo
- Publication number
- CN115563490A CN115563490A CN202211222229.0A CN202211222229A CN115563490A CN 115563490 A CN115563490 A CN 115563490A CN 202211222229 A CN202211222229 A CN 202211222229A CN 115563490 A CN115563490 A CN 115563490A
- Authority
- CN
- China
- Prior art keywords
- satellite
- locust
- algorithm
- population
- individual
- 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
- 238000004422 calculation algorithm Methods 0.000 title claims abstract description 89
- 238000005457 optimization Methods 0.000 title claims abstract description 71
- 238000003745 diagnosis Methods 0.000 title claims abstract description 51
- 238000000034 method Methods 0.000 title claims abstract description 48
- 238000007781 pre-processing Methods 0.000 claims abstract description 8
- 230000006870 function Effects 0.000 claims description 51
- 241000238814 Orthoptera Species 0.000 claims description 13
- 230000008569 process Effects 0.000 claims description 13
- 238000012706 support-vector machine Methods 0.000 claims description 13
- 230000036541 health Effects 0.000 claims description 6
- 230000007246 mechanism Effects 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 5
- 238000011161 development Methods 0.000 claims description 5
- 238000009826 distribution Methods 0.000 claims description 5
- 238000012549 training Methods 0.000 claims description 5
- 230000006399 behavior Effects 0.000 claims description 4
- 238000012423 maintenance Methods 0.000 claims description 4
- 230000003997 social interaction Effects 0.000 claims description 4
- 238000007906 compression Methods 0.000 claims description 3
- 230000006835 compression Effects 0.000 claims description 3
- 230000005484 gravity Effects 0.000 claims description 3
- 230000009467 reduction Effects 0.000 claims description 3
- 241000254032 Acrididae Species 0.000 claims description 2
- 230000009286 beneficial effect Effects 0.000 claims description 2
- 230000007423 decrease Effects 0.000 claims description 2
- 230000007774 longterm Effects 0.000 claims description 2
- 238000012544 monitoring process Methods 0.000 claims description 2
- 238000012360 testing method Methods 0.000 claims description 2
- 238000012546 transfer Methods 0.000 claims description 2
- 238000012163 sequencing technique Methods 0.000 claims 1
- 230000005540 biological transmission Effects 0.000 abstract description 6
- 238000007726 management method Methods 0.000 description 18
- 238000004891 communication Methods 0.000 description 6
- 230000008901 benefit Effects 0.000 description 3
- 238000003860 storage Methods 0.000 description 3
- 238000007418 data mining Methods 0.000 description 2
- 238000005538 encapsulation Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 230000001418 larval effect Effects 0.000 description 2
- 239000002245 particle Substances 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 241000282461 Canis lupus Species 0.000 description 1
- DGAQECJNVWCQMB-PUAWFVPOSA-M Ilexoside XXIX Chemical compound C[C@@H]1CC[C@@]2(CC[C@@]3(C(=CC[C@H]4[C@]3(CC[C@@H]5[C@@]4(CC[C@@H](C5(C)C)OS(=O)(=O)[O-])C)C)[C@@H]2[C@]1(C)O)C)C(=O)O[C@H]6[C@@H]([C@H]([C@@H]([C@H](O6)CO)O)O)O.[Na+] DGAQECJNVWCQMB-PUAWFVPOSA-M 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000001174 ascending effect Effects 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000013144 data compression Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 230000007786 learning performance Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000004806 packaging method and process Methods 0.000 description 1
- 238000009304 pastoral farming Methods 0.000 description 1
- 238000012847 principal component analysis method Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000001172 regenerating effect Effects 0.000 description 1
- 229910052708 sodium Inorganic materials 0.000 description 1
- 239000011734 sodium Substances 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B7/00—Radio transmission systems, i.e. using radiation field
- H04B7/14—Relay systems
- H04B7/15—Active relay systems
- H04B7/185—Space-based or airborne stations; Stations for satellite systems
- H04B7/1851—Systems using a satellite or space-based relay
- H04B7/18519—Operations control, administration or maintenance
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Astronomy & Astrophysics (AREA)
- Aviation & Aerospace Engineering (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Radio Relay Systems (AREA)
Abstract
A super-large-scale low-orbit satellite fault diagnosis method based on an improved locust optimization algorithm adopts a double-layer satellite network management architecture, data preprocessing is carried out through an improved locust optimization feature selection algorithm LS-BGOA under the condition that the fault diagnosis accuracy rate is not influenced, and the algorithm combines a Levy flight strategy, a mixed complex evolution strategy and the locust optimization algorithm: 1) And performing data preprocessing work by using an improved locust optimization feature selection algorithm LS-BGOA, eliminating irrelevant or noisy data to reduce data dimension of satellite transmission, and selecting an optimal feature subset suitable for satellite fault diagnosis. 2) And an MEO/LEO double-layer hierarchical management architecture is adopted, an MEO network layer is used as a management layer, and sensing data corresponding to the optimal feature subset obtained by all LEO satellites are uploaded to the corresponding MEO management satellite. 3) And the MEO satellite carries out fault diagnosis work through the information uploaded by the LEO satellite, and judges whether the LEO satellite is in a fault state.
Description
Technical Field
The invention belongs to the technical field of satellite fault management, and mainly relates to a fault diagnosis method for a super-large-scale low-orbit satellite constellation.
Background
In recent years, due to the increasing emerging services resulting from the increasing popularity of smart devices, there is an increasing demand for information services and communication quality. It is obvious that the conventional terrestrial network communication alone cannot provide a high-rate, high-reliability communication service on a global scale. The satellite network leads new research trend due to the advantages of wide coverage range, high flexibility, long communication distance, no geographic environment limitation and the like. Among them, low earth orbit satellites (LEO) are an important component of air-ground integrated networks due to their advantages of low production cost, high capacity, high rate, low time delay, etc. In recent years, with thousands of LEO satellites covering the earth, a very large-scale low-earth satellite network can provide coverage and services which cannot be realized only by a ground communication system, and has become a research hotspot at present.
However, the ultra-large scale means that the number, design life and model types of on-orbit satellites are more and more, and with the development of the ultra-large scale low orbit satellite system, the satellites can face many different types of faults, and the wrong diagnosis of the health states of the satellites causes great economic loss, so that the satellite fault diagnosis technology also becomes important. Therefore, it is necessary to employ a reasonable failure diagnosis technique.
From the search of the existing literature, p.m. frank et al, in 1996, published in European Journal of control (European Journal of control), "Analytical and qualitative model-based fault diagnosis-a surface and sodium new results", classified fault diagnosis methods into three main types: analytical model based methods, signal processing based methods and data driven based methods.
With the rapid development of artificial intelligence technology, a data-driven fault diagnosis method has become a main development direction in the field with high efficiency. Because a super-large-scale low-orbit satellite network generates high-dimensional, huge and complex data volume and the storage and calculation resources on the satellite are limited, the data is preprocessed by using a data mining means in order to facilitate later data analysis, and feature selection is a preprocessing step in the data mining process. The main idea is to compress data by eliminating irrelevant, redundant or noisy features without sacrificing the accuracy of the fault diagnosis.
Common feature selection algorithms fall into two categories: filtration and encapsulation methods. The filtering method is irrelevant to a learning algorithm, and the feature correlation is mainly judged according to the internal attribute and data association of statistical analysis, so that the feature subset is selected. The method is simple and easy to implement, but has more redundant features and poor precision, such as a mutual information method and a principal component analysis method. The packaging method closely associates the machine learning performance with the feature subset evaluation, can effectively select the feature subset with higher relevance, eliminates redundant features and reduces feature dimensions. Since the goal of feature selection is to reduce the number of features on the basis of maximizing the classification accuracy, it can be regarded as an optimization problem. In recent years, many scholars select the optimal feature subset by using a group intelligence algorithm as a search mechanism of a feature selection encapsulation method, and obtain good results.
From the search of the existing literature, k.eberhartrc et al published in 1995 "a new optimization using particle swarm theory" in IEEE service center (IEEE international service association) "and s.saremi et al published in 2015" Neural Computing and Applications "and" Evolutionary position dynamics optimizer and grazing optimization "in Engineering Software (Engineering-computer)": in the article of the theory and application, the particle swarm optimization algorithm, the wolf optimization algorithm and the locust algorithm are respectively used as search mechanisms of the optimization problem, and good effects are achieved. The locust optimization algorithm can be used as a search mechanism in the feature selection process in fault diagnosis by virtue of the rapid convergence and the high optimizing efficiency.
However, the traditional locust optimization algorithm has the problems of lack of population diversity, lack of selection randomness, easiness in falling into local optimization and the like, so that the algorithm needs to be improved, the quality of a population in the optimization process is improved, the population diversity is increased, the population is helped to jump out of the local optimization, a better solution is obtained, and the purpose of effective data compression is achieved under the condition of guaranteeing the diagnosis accuracy.
Disclosure of Invention
The invention aims to: the method is oriented to the requirement of fault diagnosis in the super-large-scale low-orbit satellite constellation, and provides a super-large-scale low-orbit satellite fault diagnosis method based on an improved locust optimization algorithm.
In order to achieve the purpose, the invention adopts the technical scheme that: a very large-scale low-orbit satellite fault diagnosis method based on an improved locust Optimization Algorithm adopts a double-layer satellite network management architecture, firstly performs data dimension reduction and compression through an improved locust Optimization feature selection Algorithm LS-BGOA (Levy Flight and crushed Complex Evolution with Binary Grasshopper Optimization Algorithm) under the condition of not influencing the fault diagnosis accuracy, and then judges the health state of a satellite through compressed data information, and comprises the following steps:
step 1: and (3) performing data preprocessing work by using a feature selection algorithm LS-BGOA based on a modified locust optimization algorithm (GOA), eliminating irrelevant or noisy data to reduce data dimensionality of satellite transmission, and selecting an optimal feature subset suitable for satellite fault diagnosis. The feature selection algorithm LS-BGOA comprises the following steps:
step 1.1: taking historical data containing various sensing information and state information of the satellite as a training data set of an LS-BGOA algorithm, wherein 80% of the data are selected as training data, and 20% of the data are selected as test data;
step 1.2: defining input sensing data as characteristic information, satellite state information as fault judgment information, and a characteristic number N representing the dimensionality of the sensing data, defining the size of a selected characteristic subset as R and representing the characteristic number of the LEO satellite needing to upload data;
step 1.3: the fitness function defining the LS-BGOA algorithm:
the fitness function is taken as an optimization target of the algorithm, and the method comprises the following two aspects: wherein gamma is R (D) The error rate is the error rate of classifying data information corresponding to a certain feature subset, | R | is the number of features included in the feature subset, | N | is the number of all features, and parameters α and β respectively represent the classification precision and the weight of the subset size, and satisfy α + β =1. Thus, the fitness function takes into account two criteria for selecting a subset of features: error rate gamma of fault diagnosis by using data corresponding to the characteristic subset R (D) And the size of the feature subset | R |. When the error rate of fault diagnosis by using data corresponding to a certain feature subset is lower and the number of features is less, the fitness function value corresponding to the feature subset is lower, so that the performance of a solution (namely the feature subset) corresponding to a lower fitness function value is better;
step 1.4: the fitness function is taken as an optimization target, the feature selection problem is changed into an optimization problem, each feature subset is regarded as an individual, the quality of the solution set is measured by calculating the fitness function value of the individual, the individual with the lowest fitness function value in the population evolution process is regarded as the current optimal solution, and the individual with the lower fitness function value is taken as the new optimal solution. And generating a final optimal solution when population evolution is finished and the maximum iteration number is reached, so that an improved locust optimization feature selection algorithm LS-BGOA is utilized to select the optimal solution with high accuracy and small feature number, and the feature information corresponding to the optimal solution is the sensing data which needs to be uploaded by the LEO satellite in fault monitoring.
Step 2: and (3) adopting an MEO/LEO double-layer hierarchical management architecture, taking an MEO network layer as a management layer, and uploading the sensing data corresponding to the optimal feature subset obtained in the step (1) to the corresponding MEO management satellite by all LEO satellites. The two-tier hierarchical management architecture comprises:
step 2.1, each MEO satellite manages LEO satellites in the coverage range, and if a certain LEO satellite is covered by a plurality of MEO satellites at the same time, the MEO satellite with the shortest distance is selected as the management satellite;
2.2, randomly selecting one LEO Satellite as a Representative Satellite RS (reconstructed Satellite) in the management range of each MEO Satellite, regularly collecting sensing information of other LEOs by the Representative Satellite, and uploading the collected data to the corresponding MEO management Satellite;
and step 3: and the MEO satellite carries out fault diagnosis work through the information uploaded by the LEO satellite, and whether the LEO satellite is in a dangerous state or not is judged. The diagnostic work includes:
and 3.1, judging the health state of the satellite by the MEO satellite by utilizing a Support Vector Machine (SVM) classifier.
And 3.2, when the MEO diagnoses that a certain LEO satellite is in a dangerous state, transmitting the satellite information to the ground station, wherein the ground station is responsible for the maintenance work of the satellite.
Further, the fault diagnosis error rate γ is solved in the calculation of the fitness function in step 1.3 R (D) The classifier used in the process is an SVM (support vector machine) classifier, the classifier can realize rapid classification and high accuracy when judging whether the satellite state is healthy or faulty, for each feature subset, the corresponding data information is used as input, the classified state is used as a diagnosis result, and therefore the fault diagnosis error rate corresponding to each feature subset can be obtained, the fitness function value of the individual is calculated, and the quality of the solution set is measured.
Further, the specific steps of the feature selection algorithm LS-BGOA based on the improved locust optimization algorithm in step 1.4 when searching for the optimal feature subset are as follows:
step 1.4.1, initializing a certain amount of solutions generated randomly into a population;
step 1.4.2, randomly dividing the population into two sub-populations;
step 1.4.3, combining the traditional locust optimization algorithm with Levy flight, and improving the randomness and the population diversity of the traditional locust optimization algorithm;
step 1.4.4, converting the improved locust optimization algorithm from a continuous form into a binary form for feature selection;
step 1.4.5, respectively finding out a temporary optimal solution in the two sub-populations by using the feature selection algorithm in the step 1.4.4, and taking a solution with a lower fitness function value as a current optimal solution;
and step 1.4.6, forcibly mixing the population and randomly dividing the population into two sub-populations again, repeating the step 1.4.5 until an optimal solution is obtained after a certain number of iterations, and obtaining the characteristic information which is finally required to be transmitted by the LEO satellite corresponding to the characteristic subset.
Firstly, training and preprocessing satellite sensing data, and selecting an optimal feature subset through an improved locust optimization feature selection algorithm; secondly, a medium orbit satellite MEO/low orbit satellite LEO double-layer satellite management framework is adopted to realize the autonomous fault diagnosis of the satellite network, and the low orbit LEO satellite uploads the feature sensing information corresponding to the feature subset; finally, the MEO satellite judges the health state of the satellite according to the data information uploaded by the LEO satellite through a Support Vector Machine (SVM) classifier; when the satellite is determined to be in a fault state, the management satellite MEO sends information to the ground station responsible for the maintenance work. The method realizes the autonomous fault identification and response of the super-large scale low earth orbit satellite constellation system, can compress the information transmission amount of the low earth orbit satellite through the characteristic selection in the data preprocessing work, greatly reduces the consumption of on-satellite calculation and storage resources in the fault diagnosis process, and improves the overall transmission efficiency of the network.
Has the beneficial effects that: compared with the prior art, the invention has the following advantages:
the invention discloses a fault diagnosis method of a super-large-scale low-orbit satellite based on an improved locust optimization algorithm, which is oriented to the fault diagnosis problem in a super-large-scale low-orbit satellite constellation. The feature selection algorithm LS-BGOA adopted by the invention combines the traditional GOA algorithm with the Levy flight and mixed complex evolution strategy, improves the potential problems of low population diversity, easy falling into local optimization and the like of the traditional GOA algorithm, not only improves the solution set quality of the algorithm, but also enhances the optimization capability of the algorithm. The feature selection algorithm LS-BGOA can select the optimal feature subset under the condition that the fault diagnosis accuracy is not influenced, and data dimension reduction and compression are effectively carried out. And secondly, the invention is based on a double-layer satellite network management architecture, the MEO satellite network is used as a management layer of the LEO satellite network, the autonomous identification and response of satellite faults can be realized, and the compressed data information corresponding to the optimal characteristic subset is uploaded to the MEO satellite by the LEO satellite for fault diagnosis. The super-large-scale low-orbit satellite fault diagnosis method based on the improved locust optimization algorithm effectively selects the optimal feature subset under the condition of ensuring the accuracy of satellite fault diagnosis, relieves the transmission load of a satellite network, and reduces the loss of on-satellite storage and calculation resources.
Drawings
FIG. 1 is a diagram of a fault diagnosis method based on a two-layer network architecture;
FIG. 2 is a flow chart of the feature selection algorithm LS-BGOA.
Detailed Description
The present invention will now be described in further detail with reference to specific examples, which are intended to be illustrative, but not limiting, of the invention.
The present invention considers a fault diagnosis scenario based on a two-tier network management architecture, as shown in fig. 1. The invention only considers the physical fault of the satellite, and the communication function of the satellite is normal. And taking the MEO network layer as a management layer, managing the LEO satellites in the coverage range of each MEO satellite, and selecting the MEO satellite with the shortest distance as the management satellite if a certain LEO satellite is covered by a plurality of MEO satellites at the same time. In the management range of each MEO Satellite, randomly selecting one LEO Satellite as a Representative Satellite RS (regenerative Satellite) in the management range, regularly collecting sensing information of other LEOs by the Representative Satellite, and uploading the collected data to the corresponding MEO Satellite; the MEO satellite carries out fault diagnosis work through the information uploaded by the LEO satellite, and the MEO satellite judges whether the LEO satellite is in a dangerous state or not by utilizing an SVM (support vector machine) classifier. When the MEO diagnoses that a certain LEO satellite is in a dangerous state, the information of the satellite is sent to the ground station, and the ground station is responsible for the maintenance work of the satellite.
In the invention, I MEO satellites and L LEO satellites are assumed, and the MEO satellite set is expressed as M = { M = { (M) } 1 ,...,M I The set of LEO satellites managed by the MEO satellite is denoted K = { K = } 1 ,...,K I Is satisfied with
So for each MEO satellite M i E M, with the set of managed LEO satellites expressed as
Wherein the representative satellite is represented as
In the invention, the total number of generated features of each LEO satellite is assumed to be N, so that the feature set is F = { F = { (F) 1 ,...,F N And if the fault diagnosis is carried out by using all the characteristic information, the overall transmission efficiency of the satellite network is reduced, and irrelevant characteristics have negative influence on the classification precision, so that an optimal characteristic subset is selected by using an improved locust optimization characteristic selection algorithm LS-BGOA, and in the optimization process, two criteria are considered to judge the performance of the subset: classification accuracy and subset size, i.e., accuracy and number of features. The fitness function in the optimization is therefore expressed as:
wherein gamma is R (D) The error rate of classification is determined by using data information corresponding to a subset, and the classifier of the SVM is still usedAnd calculating to obtain the error rate of classification, wherein | R | is the number of the features contained in the feature subset, and | N | is the number of all the features. The parameters α and β represent the weight of the classification accuracy and the subset size, respectively, and satisfy α + β =1. Therefore, the lower the fitness function value is, the higher the quality of the solution set is, and the feature subset corresponding to the lowest value of the fitness function is the optimal solution.
The feature selection algorithm LS-BGOA is improved based on a locust optimization algorithm (GOA), aiming at the potential problems that the population diversity of the traditional GOA algorithm is low, the traditional GOA algorithm is easy to fall into local optimization and the like, on the basis of the traditional GOA algorithm, the improved feature algorithm LS-BGOA is obtained by combining mixed complex evolution and a Levy flight strategy, the solution set quality of the algorithm is improved, the optimization capability of the algorithm is enhanced, the flow of the algorithm is shown in figure 2, and the specific principle is as follows:
the GOA algorithm is a random method inspired by nature and proposed by Saremi et al, simulating the behavior of a locust population. Locusts mainly go through two stages: larval and adult, with different characteristics: the main features in the larval stage are slow movements and small steps. In contrast, large range and sudden movements are essential features of the adult population. And because the searching of the feed source is an important characteristic of locust swarm aggregation, the behavior of the locust swarm can be modeled into an optimization mechanism by the characteristics. In GOA, locust populations are initialized randomly, and each individual can be considered as a candidate solution, with the position of the individual locust being represented by the lower model:
X i ＝S i +G i +W i ，
wherein X i Indicates the location of the ith individual locust, which is socially affected by S i G, gravity G i Wind power W i Where social roles play a major role, it can be modeled as:
wherein d is ij Is the Euclidean distance between the i and j individuals of locusts, i.e. d ij ＝|x j -x i |，
A unit vector representing the distance between the ith and jth individual, i.e.
The social interaction function s (r) determines the magnitude and type of social interaction, which is defined as:
where f represents the intensity of the attraction and l represents the length scale of the attraction. Take f =0.5,l =1.5.
When the distance d is less than 2.079, the function value is less than 0, the locust is mainly subjected to repulsive force and is positioned in a repelled area; when the distance d is greater than 2.079, the function value is greater than 0, the locust is mainly attracted by the attraction force and is positioned in an attracted area; when the distance d is equal to 2.079, the social acting force on the locust is 0, the attraction force is equal to the repulsion force, and the locust is in a comfortable area. Since the value of the s (r) function is close to 0 when the distance is too large, the acting force is not obvious, so the distance between the locust individuals is normalized in the range of [1,4 ].
In conjunction with the above equations, the following equations are typically employed to solve the optimization problem:
the gravity component is not taken into account here and it is assumed that the wind direction is always oriented towards the optimal solution, ub d And lb d Respectively an upper bound and a lower bound for the target d-dimensional variable,
is the target in d-dimensional space, i.e. the best solution currently found. Since locusts reach the comfort zone at a distance of 2.079, the population is liable to converge to the comfort zone, which is not favorable for their local mining, i.e. they cannot converge quicklyTo a specific point, the parameter c is therefore introduced to adjust the behavior of the extensive exploration and local development in the evolution process. c as the number of iterations decreases, the expression is
L is the current iteration number, L is the maximum iteration number, c max Is a maximum value, c min Is the minimum value. The first c is used to balance the search range of locusts at different stages, and the second c narrows the attraction zone, comfort zone and repulsion zone between locusts. The locust individual will update the location based on the current location at the time of the search, the global best location, and the location of all other searching locust.
The traditional GOA algorithm initialization population is completely random, so that the diversity and the ergodicity of the population cannot be guaranteed, and in order to make the population jump away from the local optimal solution as far as possible and increase the diversity of the population, a Levy flight strategy is introduced to optimize the algorithm. Levy flight is a random non-gaussian walk with long-term small step search and short-term large step search pattern with a Levy distribution of steps:
Levy(β)～u＝t -1-β ,0＜β≤2,
wherein beta is a Levy index for adjusting the distribution stability, and the step length of Levy flight satisfies the following conditions:
where β is taken to be 1.5, u and v are both normally distributed, σ follows the following equation, and Γ is a standard gamma function:
combining the Levy strategy with the GOA algorithm, the location of each individual is updated as follows:
X i is the position of the locust individual without combining the Levy flight strategy variation,
the method is combined with the position of the Levy flying individual after variation, and the variation value of the position depends on the random value of the Levy and the current position. The updated step vector can be expressed as:
because the feature selection is a binary optimization problem, we need to convert continuous values into binary values, the function selected here is a sigmoid function, the update of the individual position is converted into a binary form, and the transfer function and the corresponding individual position update expression are as follows:
however, it cannot be guaranteed that the new individual after Levy flight variation is better than the original individual, so the evolution direction needs to be defined:
and for the current individual, if the updated fitness value of the individual is better than that of the current individual, replacing the original individual with the new individual, otherwise, keeping the fitness value unchanged. Therefore, the searching range of the population is expanded, a better value is easier to search, and the overall quality of the population is improved.
In order to increase the ergodicity and diversity of the population, a GOA algorithm is combined with a Levy flight strategy, and in addition, in order to increase the search range of the population and jump out of a comfort zone to explore a wider space, a mixed complex evolution strategy is introduced. The principle of the hybrid complex evolution strategy is to regard global search as a natural evolution process, equally divide the population into several sub-populations, each of which evolves independently. After a certain time of evolution, the sub-populations are remixed, new sub-populations are generated, and then the sub-populations are remixed after independent evolution until a certain number of iterations is reached or a target solution meeting the conditions is selected.
The method comprises the following specific steps:
1) Initializing a population, calculating fitness function values of individuals in the population, then sorting the fitness function values according to the ascending order of the fitness values, and randomly dividing the population into two sub-populations with the same number.
2) Each sub-population is independently evolved, and the evolution mechanism is an improved binary locust optimization algorithm combined with a Levy flight strategy. After a period of time of evolution, each sub-population generates an optimal solution, and the optimal solution generated by the two sub-populations with a lower fitness function value is used as the current optimal solution.
3) And forcibly remixing the two sub-populations into a new population, repeating the steps, splitting the population into the sub-populations again, independently evolving and remixing the sub-populations, and replacing the original optimal solution with the solution with better performance if the solution with lower fitness function value exists. After a certain number of iterations or a certain optimization condition is met, a final optimal solution is obtained.
It is worth noting that the algorithm can realize independent evolution in parallel by dividing the population into sub-populations, so that the running time of the algorithm is greatly shortened, and the optimizing efficiency is improved. Therefore, the LS-BGOA algorithm combining the Levy flight and mixed complex evolution strategy with the traditional GOA algorithm not only improves the optimization capability and the optimization efficiency of the population, but also effectively improves the quality of the solution set, improves the performance of the feature selection algorithm, and helps us to effectively compress data while ensuring the diagnosis accuracy in the large-scale low-orbit satellite constellation fault diagnosis.
The above description is only a preferred embodiment of the present invention, and it should be noted that: it will be apparent to those skilled in the art that various modifications and adaptations can be made without departing from the principles of the invention and these are intended to be within the scope of the invention.
Claims (5)
1. A very large-scale low-orbit satellite fault diagnosis method based on an improved locust Optimization Algorithm is characterized in that a double-layer satellite network management architecture is adopted, data preprocessing is performed through an improved locust Optimization feature selection Algorithm LS-BGOA (Levy Flight and short Complex Evolution with Binary Grasshopper Optimization Algorithm) under the condition that the fault diagnosis accuracy is not influenced, a Levy Flight strategy, a mixed Complex Evolution strategy and a locust Optimization Algorithm (GOA) are combined through the Algorithm to perform data dimension reduction and compression, and then the health state of a satellite is judged through compressed data information, and the method comprises the following steps:
step 1: and performing data preprocessing work by using an improved locust optimization feature selection algorithm LS-BGOA, eliminating irrelevant or noisy data to reduce data dimensionality transmitted by the satellite, and selecting an optimal feature subset suitable for satellite fault diagnosis. The feature selection algorithm LS-BGOA comprises the following steps:
step 1.1: taking historical data containing various sensing information and state information of the satellite as a training data set of an LS-BGOA algorithm, wherein 80% of the data are selected as training data, and 20% of the data are selected as test data;
step 1.2: defining input sensing data as characteristic information, satellite state information as fault judgment information, and a characteristic number N representing the dimensionality of the sensing data, defining the size of a selected characteristic subset as R and representing the characteristic number of the LEO satellite needing to upload data;
step 1.3: the fitness function defining the LS-BGOA algorithm:
the fitness function is taken as an optimization target of the algorithm, and the method comprises two aspects: wherein gamma is R (D) Is to use a certain subset of featuresThe error rate of classifying corresponding data information, | R | is the number of features included in the feature subset, | N | is the number of all the features, and parameters α and β respectively represent the classification precision and the weight of the subset size, and satisfy α + β =1; thus, the fitness function takes into account two criteria for selecting a subset of features: error rate gamma of fault diagnosis by using data corresponding to the characteristic subset R (D) And the size of the feature subset | R |. When the error rate of fault diagnosis by using data corresponding to a certain feature subset is lower and the number of features is smaller, the fitness function value corresponding to the feature subset is lower, so that the performance of a solution (namely the feature subset) corresponding to the lower fitness function value is better;
step 1.4: the fitness function is taken as an optimization target, the feature selection problem is changed into an optimization problem, each feature subset is regarded as an individual, the quality of the solution set is measured by calculating the fitness function value of the individual, the individual with the lowest fitness function value in the population evolution process is regarded as the current optimal solution, and the individual with the lower fitness function value is taken as the new optimal solution. When population evolution is finished, namely the maximum iteration number is reached, a final optimal solution is generated, an improved locust optimization feature selection algorithm LS-BGOA is utilized to select the optimal solution with high accuracy and small feature number, and feature information corresponding to the optimal solution is sensing data which needs to be uploaded by an LEO satellite in fault monitoring;
step 2: adopting an MEO/LEO double-layer hierarchical management architecture, taking an MEO network layer as a management layer, and uploading sensing data corresponding to the optimal feature subset obtained in the step 1 to a corresponding MEO management satellite by all LEO satellites; the two-tier hierarchical management architecture comprises:
step 2.1, each MEO satellite manages LEO satellites in the coverage range, and if a certain LEO satellite is covered by a plurality of MEO satellites at the same time, the MEO satellite with the shortest distance is selected as the management satellite;
2.2, randomly selecting one LEO Satellite as a Representative Satellite RS (reconstructed Satellite) in the management range of each MEO Satellite, regularly collecting sensing information of other LEOs by the Representative Satellite, and uploading the collected data to the corresponding MEO management Satellite;
and step 3: the MEO satellite carries out fault diagnosis work through the information uploaded by the LEO, and whether the LEO satellite is in a fault state or not is judged; the diagnostic work includes:
step 3.1, the MEO satellite judges the health state of the satellite by utilizing a Support Vector Machine (SVM) classifier;
and 3.2, when the MEO diagnoses that a certain LEO satellite is in a fault state, the information of the satellite is sent to the ground station, and the ground station is responsible for the maintenance work of the satellite.
2. The method for diagnosing the fault of the ultra-large-scale low-orbit satellite based on the improved locust optimization algorithm as claimed in claim 1, wherein the fault diagnosis error rate γ is obtained in the calculation of the fitness function in the step 1.3 R (D) The classifier used in the process is an SVM (support vector machine) classifier, the classifier can realize rapid classification and high accuracy when judging whether the satellite state is healthy or faulty, for each feature subset, the corresponding data information is used as input, the classified state is used as a diagnosis result, and therefore the fault diagnosis error rate corresponding to each feature subset can be obtained, the fitness function value of the individual is calculated, and the quality of the solution set is measured.
3. The method for diagnosing the fault of the ultra-large-scale low-orbit satellite based on the improved locust optimization algorithm, as claimed in claim 1, wherein the improved locust optimization feature selection algorithm LS-BGOA in step 1.4 is specifically performed by the following steps:
step 1.4.1, initializing a certain amount of solutions generated randomly into a population;
step 1.4.2, randomly dividing the population into two sub-populations;
step 1.4.3, combining the traditional locust optimization algorithm with Levy flight, and improving the randomness and the population diversity of the traditional locust optimization algorithm;
step 1.4.4, converting the improved locust optimization algorithm from a continuous form into a binary form for feature selection;
step 1.4.5, respectively finding out a temporary optimal solution in the two sub-populations by using the feature selection algorithm in the step 1.4.4, and taking a solution with a lower fitness function value as a current optimal solution;
and step 1.4.6, forcibly mixing the population and randomly dividing the population into two sub-populations again, and repeating the step 1.4.5 until an optimal solution is obtained after a certain number of iterations, so that the characteristic information which is finally required to be transmitted by the LEO satellite corresponding to the characteristic subset is obtained.
4. The ultra-large-scale low-orbit satellite fault diagnosis method based on the improved locust optimization algorithm as claimed in claim 1, wherein the improved locust optimization feature selection algorithm LS-BGOA: in GOA, the location of individual locusts is socially affected S i Is modeled as:
wherein d is ij Is the Euclidean distance between the i and j individuals of locusts, i.e. d ij ＝|x j -x i |，
A unit vector representing the distance between the ith and jth individual, i.e.
The social interaction function s (r) determines the magnitude and type of social interaction, which is defined as:
where f represents the intensity of the attraction and l represents the length scale of the attraction; take f =0.5,l =1.5.
When the distance d is less than 2.079, the function value is less than 0, the locust is mainly subjected to repulsive force and is positioned in a repelled area; when the distance d is more than 2.079, the function value is more than 0, the locust is mainly attracted by the attraction force and is positioned in an attracted area; when the distance d is equal to 2.079, the social acting force borne by the locust is 0, the attraction force is equal to the repulsion force, and the locust is in a comfortable area; normalizing the distance between individual locusts to be in the range of [1,4 ]; the following equations are used to solve the optimization problem:
the gravity component is not taken into account here and it is assumed that the wind direction is always oriented towards the optimal solution, ub d And lb d Respectively an upper bound and a lower bound for the target d-dimensional variable,
is a target in d-dimensional space, i.e. the best solution currently found; because the locusts reach a comfort zone when the distance is 2.079, the population is easy to converge to the comfort zone, which is not beneficial to the local exploitation of the locusts, namely, the locusts cannot quickly converge to a specific point, the parameter c is introduced to adjust the behaviors of large-scale exploration and local development in the evolution process; c as the number of iterations decreases, the expression is
L is the current iteration number, L is the maximum iteration number, c max Is a maximum value, c min Is the minimum value; the first c is used for balancing the searching range of locusts in different periods, and the second c narrows the attraction zone, the comfort zone and the repulsion zone among the locusts; the locust individual can update the position according to the current position, the global optimal position and the positions of all other locust searches;
a Levy flight strategy optimization algorithm is introduced, levy flight is random non-Gaussian walk, and has a long-term small step search mode and a short-term large step search mode, wherein the step length has Levy distribution:
Levy(β)～u＝t -1-β ,0＜β≤2,
wherein beta is a Levy index for adjusting the distribution stability, and the step length of Levy flight satisfies the following conditions:
where β is taken to be 1.5, u and v both follow a standard normal distribution σ, which follows the following equation, where Γ is a standard gamma function:
combining the Levy strategy with the GOA algorithm, the location of each individual is updated as follows:
X i is the position of the locust individual without combining the Levy flight strategy variation,
the position of the variant of the Levy flying individual is combined, and the variation value of the position depends on the Levy random value and the current position; the updated step vector can be expressed as:
because the feature selection is a binary optimization problem, continuous values need to be converted into binary values, the function selected here is a sigmoid function, the updating of the individual positions is converted into a binary form, and the transfer function and the corresponding individual position updating expression are as follows:
however, it cannot be guaranteed that the new individual after Levy flight variation is better than the original individual, so the evolution direction needs to be defined:
and for the current individual, if the updated fitness value of the individual is better than that of the current individual, replacing the original individual with the new individual, otherwise, keeping the fitness value unchanged.
5. The super-large scale low earth orbit satellite fault diagnosis method based on the improved locust optimization algorithm as claimed in claim 1,
in order to increase the ergodicity and diversity of a population, a GOA algorithm and a Levy flight strategy are combined, and a mixed complex evolution strategy is introduced; the principle of the hybrid complex evolution strategy is to regard global search as a natural evolution process, equally divide the population into several sub-populations, each of which evolves independently. After a certain time of evolution, the sub-populations are mixed again to generate a new sub-population, and then the new sub-populations are mixed again after independent evolution until a certain number of iterations is reached or a target solution meeting the conditions is selected;
the method comprises the following specific steps:
1) Initializing a population, calculating fitness function values of individuals in the population, then sequencing the fitness function values in an increasing order, and randomly dividing the population into two sub-populations with the same number;
2) Each sub-population is independently evolved, and the evolution mechanism is an improved binary locust optimization algorithm combined with a Levy flight strategy. After a period of time of evolution, each sub population generates an optimal solution, and the optimal solution generated by the two sub populations with a lower fitness function value is taken as the current optimal solution;
3) Forcibly mixing the two sub-populations again to form a new population, repeating the steps, splitting the population into the sub-populations again, independently evolving and then mixing again, and replacing the original optimal solution with the better solution if the solution with better performance, namely the solution with lower fitness function value, exists; and obtaining the final optimal solution after a certain number of iterations or a certain optimization condition is met.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211222229.0A CN115563490A (en) | 2022-10-08 | 2022-10-08 | Super-large-scale low-orbit satellite fault diagnosis method based on improved locust optimization algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211222229.0A CN115563490A (en) | 2022-10-08 | 2022-10-08 | Super-large-scale low-orbit satellite fault diagnosis method based on improved locust optimization algorithm |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115563490A true CN115563490A (en) | 2023-01-03 |
Family
ID=84745789
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211222229.0A Pending CN115563490A (en) | 2022-10-08 | 2022-10-08 | Super-large-scale low-orbit satellite fault diagnosis method based on improved locust optimization algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115563490A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117370877A (en) * | 2023-12-06 | 2024-01-09 | 长春理工大学 | Agricultural machinery fault prediction method based on multiple sensors and IPSO-GPR |
| CN117744465A (en) * | 2023-10-20 | 2024-03-22 | 佛山科学技术学院 | Structural damage identification method, system, equipment and medium |
-
2022
- 2022-10-08 CN CN202211222229.0A patent/CN115563490A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117744465A (en) * | 2023-10-20 | 2024-03-22 | 佛山科学技术学院 | Structural damage identification method, system, equipment and medium |
| CN117370877A (en) * | 2023-12-06 | 2024-01-09 | 长春理工大学 | Agricultural machinery fault prediction method based on multiple sensors and IPSO-GPR |
| CN117370877B (en) * | 2023-12-06 | 2024-04-26 | 长春理工大学 | Agricultural machinery fault prediction method based on multiple sensors and IPSO-GPR |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115563490A (en) | Super-large-scale low-orbit satellite fault diagnosis method based on improved locust optimization algorithm | |
| Jiang et al. | Deep learning-based edge caching in fog radio access networks | |
| CN112039687A (en) | Small sample feature-oriented fault diagnosis method based on improved generation countermeasure network | |
| CN113259469B (en) | Edge server deployment method, system and storage medium in intelligent manufacturing | |
| CN109193756B (en) | Scene decoupling dynamic economic dispatching model solving method of wind power access system | |
| CN110442143A (en) | A kind of unmanned plane situation data clustering method based on combination multiple target dove group's optimization | |
| CN114091667A (en) | Federal mutual learning model training method oriented to non-independent same distribution data | |
| CN107679484A (en) | A kind of Remote Sensing Target automatic detection and recognition methods based on cloud computing storage | |
| CN113485409A (en) | Unmanned aerial vehicle path planning and distribution method and system for geographic fairness | |
| CN115912502A (en) | Intelligent power grid operation optimization method and device | |
| Du et al. | The Internet of Things as a deep neural network | |
| CN113676357B (en) | Decision method for edge data processing in power internet of things and application thereof | |
| Hitomi et al. | Extracting and applying knowledge with adaptive knowledge-driven optimization to architect an earth observing satellite system | |
| CN114936708A (en) | Fault diagnosis optimization method based on edge cloud collaborative task unloading and electronic equipment | |
| CN108012345A (en) | A kind of more mobile energy supplement method of single pair in wireless sensor network | |
| CN108876144B (en) | Transformer substation pre-addressing method based on deep learning algorithm | |
| Gad et al. | Communication-efficient federated learning in drone-assisted IoT networks: Path planning and enhanced knowledge distillation techniques | |
| CN114629769B (en) | Traffic map generation method of self-organizing network | |
| CN113807005B (en) | Bearing residual life prediction method based on improved FPA-DBN | |
| CN115906959A (en) | Parameter training method of neural network model based on DE-BP algorithm | |
| CN115034608A (en) | Distribution network risk assessment method based on distribution network element and neural network | |
| CN115906153A (en) | Federal learning optimization method, device and system under sample imbalance scene | |
| CN115345320A (en) | Method for realizing personalized model under layered federal learning framework | |
| CN114595816A (en) | Neural network model training method based on edge calculation | |
| CN112613227A (en) | Model for predicting remaining service life of aero-engine based on hybrid machine learning |
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 |