CN108268710B - Genetic algorithm-based IMA system dynamic reconstruction strategy optimization method - Google Patents

Abstract

The invention discloses an IMA system dynamic reconstruction strategy optimization method based on a genetic algorithm, and belongs to the field of design of integrated electronic systems. The invention solves the problem of explosion of the state space of the configuration scheme when the IMA system carries out dynamic reconstruction by using a genetic algorithm technology, and obtains an effective IMA system dynamic reconstruction strategy model by adding an IMA system reliability, real-time performance and static load balance analysis method. The method can be converted into an IMA system configuration scheme AADL model by carrying out reconstruction configuration coding on the software, hardware and time configuration of the IMA system, and can obtain a lethal gene and a fitness function in genetic algorithm analysis and optimization by carrying out reliability, real-time constraint and multidimensional load balance analysis on the IMA system through the AADL model. Through the operation of a genetic algorithm, a better IMA system dynamic reconfiguration configuration scheme can be effectively found.

Description

Genetic algorithm-based IMA system dynamic reconstruction strategy optimization method

Technical Field

The invention belongs to the field of integrated electronic system design, relates to a method for configuring an integrated avionics system in the design of the avionics system, and particularly relates to a method for optimizing software and hardware configuration strategies of the integrated avionics system (IMA) through a genetic algorithm during dynamic reconstruction of the integrated avionics system.

Background

The avionics system relates to functions of communication, navigation, identification, detection, flight management, display control and the like, provides comprehensive and accurate battlefield situation information for a pilot, enhances the capability of the pilot in identifying, monitoring, analyzing and responding to potential threats in a battlefield environment, and is the central nerve of the whole warplane. The development of avionics goes through several stages of discrete type, combined type, integrated type and advanced integrated type, and aims to enhance the functions and performances of an avionics system, reduce the life cycle cost of an airplane and solve the problem of upgrading and transplanting software and hardware of the system.

Due to the limitations of the modern aircraft on structure and function, changes of fault positions and components, operation modes, operation environments, structures and other problems, engineers cannot define all configurations to deal with all possible key conditions after the aircraft takes off, and therefore the system needs to generate various configurations in real time aiming at planned or unplanned events, namely what we often say is system reconfiguration. The dynamic reconfiguration of the system is a main means for improving the reliability of the system, and the dynamic reconfiguration failure and the error of the system are difficult to model when the embedded system is designed.

The new generation of avionics systems needs to have fault-tolerant, reconfigurable functions, i.e. when a hardware failure occurs, the system needs to be reconfigured, i.e. the functions are subdivided into safe modules. Due to the universality of IMA system modules, the mode of system configuration is various and selectable, and the mode of system configuration is also various when the system is reconstructed according to needs caused by the change of task requirements (such as mode switching of cruise, search, attack and the like) or the uncertainty of faults (module faults, partition faults, application software faults and the like), different configuration changes form different configuration schemes and configuration paths, and the comprehensive performance of the various different configuration schemes is different. The existing analysis method for IMA reconstruction is limited to model-based computational analysis on the existing IMA system configuration scheme, the advantages and disadvantages of the configuration scheme are evaluated through comparison of multiple characteristics among different configuration schemes, and a strategy generation method for integrating reliability, instantaneity and load balance characteristics based on a model aiming at an IMA reconstruction process is lacked. Aiming at the design problem of the dynamic reconstruction strategy under the conditions of a multitask mode and multiple constraints, a comprehensive design method of the dynamic reconstruction strategy of the IMA system is urgently needed to be researched, a design support means is constructed, a foundation is laid for the comprehensive design of the IMA system, and the development of the next generation of IMA dynamic reconstruction technology is further supported.

The IMA system configuration relates to whether the key characteristics of the IMA system meet the requirements of system design after application software is deployed on an IMA platform, and domestic and foreign related research institutions develop special research on each single characteristic. At present, a plurality of perfect evaluation methods are provided for IMA system characteristic analysis based on models, reliability research aiming at IMA mainly focuses on reliability design of a system architecture, a plurality of methods for designing a system to improve reliability based on ARINC653 and ASSAC specifications are provided, and a real-time analysis method for dynamic reconfiguration and configuration of the system based on models is provided aiming at real-time requirements of the system. In addition, a load balance analysis method for system reconfiguration configuration is still very lacking, and the advantages and disadvantages of different configuration schemes cannot be compared under the condition that indexes such as system reliability real-time performance are met. At the same time, these methods are only suitable for evaluating existing IMA system configuration schemes. Although theoretically, all possible configuration schemes can be exhausted and evaluated to screen the optimal configuration scheme, in the actual IMA system dynamic reconfiguration strategy analysis, because the possible configuration schemes have the problem of state space explosion, dynamic reconfiguration through exhaustive optimization is difficult to realize. For complex software and hardware binding relations of an IMA system, an optimal IMA system dynamic reconfiguration strategy cannot be found out through exhaustively exhausting possible configuration schemes, and an optimization method capable of evaluating and screening the IMA system dynamic reconfiguration strategy through synthesizing various characteristics under a high-efficiency search algorithm to obtain the optimal IMA system configuration scheme is needed.

Disclosure of Invention

The invention solves the problem of explosion of the state space of the configuration scheme when the IMA system carries out dynamic reconstruction by using a genetic algorithm technology, and obtains an effective IMA system dynamic reconstruction strategy model by adding an IMA system reliability, real-time performance and static load balance analysis method. The method can be converted into an IMA system configuration scheme AADL model by carrying out reconstruction configuration coding on the software, hardware and time configuration of the IMA system, and can obtain a lethal gene and a fitness function in genetic algorithm analysis and optimization by carrying out reliability, real-time constraint and multidimensional load balance analysis on the IMA system through the AADL model. Through the operation of a genetic algorithm, a better IMA system dynamic reconfiguration configuration scheme can be effectively found.

Specifically, the dynamic reconfiguration strategy optimization method of the IMA system based on the genetic algorithm comprises the following steps:

establishing an IMA system dynamic reconstruction strategy analysis model based on AADL;

modeling an IMA system structure and a dynamic reconstruction process by using an AADL language, establishing an IMA system dynamic reconstruction strategy analysis model, and adding system reliability, real-time performance and load balance analysis attributes required by the IMA system strategy optimization analysis into the IMA system dynamic reconstruction strategy analysis model. The reliability analysis mainly comprises the basic reliability and the task transfer probability of each software and hardware of the system. The real-time analysis mainly comprises the response and running time of system software, task transition probability, starting identification and the like. For load balance analysis, a time scheduling mechanism of an IMA module needs to be provided, the length of a main time frame and a time slice unit allowed by each IMA module are determined, and in addition, the top-level configuration of communication interfaces and channels between IMA system partitions needs to be determined, namely the configuration under different software and hardware matching relations.

Step two, setting an IMA system genetic algorithm analysis algorithm;

the genetic algorithm is adopted to analyze the IMA system, and firstly, the software and hardware matching and the time configuration of the IMA system are coded. Under the condition that various parameters of software and hardware of the IMA system and a top-level task chain are determined, and only one piece of software works in each partition of the IMA system, the parameters of the dynamic reconfiguration structure problem of the IMA system can be represented by time parameters and space parameters, wherein the time parameters comprise the number and sequence of time slices occupied by the software, and the space parameters refer to the hardware position of the software. The code can adopt binary code, and comprises software number, hardware number, time slice number and sequence information of system configuration.

When different characteristics of an IMA system are used as individual judgment conditions to generate and screen populations, the reliability and real-time analysis are analyzed by adopting the existing stochastic Petri network model and analysis method. In the load balance analysis, aiming at the task load and the communication load of the system, a plurality of evaluation indexes are provided, and the task load of the system is described by using a running time ratio and a memory occupation ratio in time and space dimensions respectively; for communication load, it is proposed that bus bandwidth occupancy describes communication load in the system space dimension. And (4) synthesizing the characteristic analysis to obtain a judgment basis for optimizing the dynamic reconfiguration strategy of the IMA system.

When a genetic algorithm is adopted to search and optimize an IMA dynamic reconstruction scheme, a method for iteratively generating a better initial population on the basis of the optimization of a random configuration scheme is selected for the initial population. The genetic algorithm operation selects a roulette selection method, a single-point crossover method or a binary variation algorithm. When the search is terminated, the algebraic upper limit of the search is determined as a termination condition.

Thirdly, calculating the reliability and the real-time performance of system configuration as individual screening conditions;

the AADL configuration model of the IMA system can be obtained through system software and hardware matching and time configuration coding, meanwhile, the reliability and real-time property corresponding to system software and hardware are already contained in the AADL multi-characteristic analysis basic model of the IMA system, the AADL model can be converted into a Petri network model, the reliability and real-time property indexes of the IMA system under the current configuration are calculated on the basis of the Petri network model, and the reliability and real-time property indexes are compared with design allowable values to be used as death causing genes in a genetic algorithm to judge whether an individual survives.

Step four, carrying out load balance analysis on the configuration scheme to obtain individual fitness;

when the software and hardware of the IMA system are configured and dynamically reconstructed, the configuration scheme firstly needs to meet the performance requirements of the system, such as the requirements on reliability and real-time performance. On the basis of meeting the necessary conditions, the IMA system configuration scheme is still expected to better adapt to the current task requirements, so that the working efficiency of the IMA system is maximized, the communication requirements among IMA modules are reduced, and the possibility of faults is reduced. By using an analysis method for load balance problems in a distributed system for reference, static load balance analysis is carried out on an IMA system, the task processing capacity of the IMA system is described through a running time ratio and a memory occupation ratio, and the communication capacity of the IMA system is described through bus bandwidth occupied by communication among IMA modules, so that the static load balance analysis is used as an evaluation index of an IMA system configuration scheme. And integrating the three evaluation indexes as the fitness function of the individual.

Decoding the optimal gene obtained by searching the genetic algorithm to obtain an IMA system dynamic reconfiguration configuration scheme;

after the genetic algorithm is subjected to search convergence or specified algebra is completed, the individual genes with the optimal algorithm can be obtained, wherein the individual genes comprise software and hardware matching and time scheduling information of the IMA system dynamic reconfiguration configuration scheme, corresponding software and hardware matching matrixes and time scheduling information on each module are obtained through decoding, the information is corresponding to an AADL (architecture analysis and design language) model of a system architecture, and the corresponding IMA system dynamic reconfiguration configuration scheme can be obtained.

The invention has the advantages that:

1. the genetic algorithm is adopted to carry out dynamic reconstruction strategy optimization on the IMA system, and the problem of space explosion of the dynamic reconstruction configuration scheme of the IMA system is solved.

2. When the genetic algorithm describes the dynamic reconfiguration configuration of the IMA system, a mode of coding the software and hardware matching and the time configuration of the system is specifically adopted, and coding information is added into an AADL model of the IMA system during decoding to obtain a complete configuration scheme AADL model. The reliability and the real-time performance of the IMA system configuration are used as lethal genes to be added into genetic operation, so that the reliability and the real-time performance of the configuration are ensured to meet the requirements of the IMA system.

3. In the fourth step, the system load balance analysis method used for judging the adaptability of the configuration scheme integrates the task load and the communication load of the IMA system configuration scheme. Wherein the task load includes a task load in a time dimension described by a runtime ratio and a task load in a space dimension described by a memory occupancy ratio. In the description of the communication capacity, the communication delay of the IMA system is far less than the scheduling interval of the tasks rotating by time slices, and only the communication load on the space dimension described by the bus bandwidth occupied by the IMA module communication is included.

Drawings

FIG. 1 is a flow chart of a dynamic reconfiguration strategy optimization method of an IMA system based on a genetic algorithm, which is provided by the invention;

FIG. 2 is a diagram of an IMA system partition invocation process;

FIG. 3 is a schematic diagram of memory usage of different modules of an IMA system;

FIG. 4 is a multi-layer structure diagram of the inter-partition communication mode of the IMA system;

FIG. 5 is a diagram of an example IMA system basic application AADL model;

FIG. 6 is a diagram of an example IMA system reliability, real-time analysis software architecture Petri Net model;

fig. 7 is a schematic diagram of an AADL model decoded by an IMA system configuration scheme.

Detailed Description

The present invention will be described in detail below with reference to the accompanying drawings and examples.

The existing analysis method for dynamic reconfiguration of an IMA system is only limited to simply generating a configuration file by using a model containing configuration information, directly analyzing schedulability and resource constraint under different configurations through configuration information simulation, only judging whether the dynamic reconfiguration of the system meets single characteristic indexes such as functional requirements, reliability and instantaneity, and cannot automatically generate a proper dynamic reconfiguration configuration scheme according to the overall requirements of the system. Aiming at the requirements of high reliability, complete performance and the like which are required to be achieved after an IMA system is dynamically reconstructed, on the basis of establishing an IMA system dynamic reconstruction strategy analysis model containing characteristic analysis attributes, the reliability, real-time performance and load balance characteristics of the IMA system are respectively analyzed, and a plurality of IMA system single analysis models are established. On the basis, the configuration scheme of the system is coded through the genetic algorithm, the requirements of system reliability, real-time performance and load balance are added into the screening search condition of the genetic algorithm, so that a satisfactory dynamic reconfiguration configuration scheme of the IMA system is obtained, and the genetic algorithm-based dynamic reconfiguration strategy optimization method of the IMA system is completed.

As shown in the flowchart of fig. 1, the genetic algorithm based IMA system dynamic reconstruction strategy optimization method provided by the present invention includes the following steps:

step one, establishing an IMA system dynamic reconstruction strategy analysis model based on AADL.

Using AADL language as modeling language to model IMA system architecture, in the system architecture design model, firstly modeling the connection relation among software, hardware and software and hardware of the system according to the summary design document of the IMA system, including the number of modules of the system, the partition configuration and the bus connection mode on the modules, etc.; and then adding functional attributes of each part, such as bus bandwidth, port channel configuration, running time length and storage space required by software, maximum main time frame length allowed by a system module, time slice granularity, storage space and the like, into the model according to the system detailed design document. A dynamic reconstruction model is established on the basis of a system structure design model, an AADL error attachment and the like are used for describing an IMA system reconstruction triggering condition, and a behavior attachment and a mode are used for describing a reconstruction process and different states before and after reconstruction. And adding the system reliability, real-time performance and load balance analysis attributes required by the IMA system strategy optimization analysis into the dynamic reconstruction model to obtain the IMA system dynamic reconstruction strategy analysis model containing the characteristic analysis attributes.

The reliability analysis mainly comprises the basic reliability and the task transfer probability of each software and hardware of the system. The real-time analysis mainly comprises the response and running time of system software, task transition probability, starting identification and the like. For load balance analysis, a time scheduling mechanism of the IMA module needs to be given, the main time frame length and the time slice unit allowed by each module are determined, and in addition, the top-level configuration of communication interfaces and channels between the IMA system partitions needs to be determined, namely the configuration under different software and hardware matching relations. The added characteristic analysis attributes are shown in table 1 and table 2.

TABLE 1 reliability analysis input Attribute

TABLE 2 analysis of input attributes in real-time

Secondly, dynamically reconstructing configuration codes by the IMA system, and setting a genetic algorithm analysis algorithm of the IMA system;

the IMA system has more tasks in actual aircraft operation, and a two-stage configuration mode of distributing the tasks on the IMA module of the IMA system and distributing corresponding time slices for the IMA module is flexible and changeable, so that the problem of explosion of a configuration scheme state space exists during configuration and reconstruction of the IMA system. When the AADL language is used for modeling the IMA system, an ARINC653 accessory is adopted to model the partition structure inside the IMA module, and the component processors, processes and threads in the AADL structure model are respectively corresponding to the modules, partitions and processes in the ARINC653 accessory. In the ARINC653 attachment, 255 partition processes can be bound for a subsystem corresponding to an IMA module, corresponding to the actual general IMA system standard. After assigning tasks to partitions, the runtime of the tasks also needs to be configured, time-slicing the tasks over an uncertain master time frame, which will result in a further increase of the state space of the configuration scheme.

Setting an IMA system configuration space model:

F＝(d,s,n,T,t₀)

wherein d is the number of modules of the IMA system, and s is the number of software of the IMA system; n is the maximum number of partitions that a module of the IMA system can accommodate; t is the maximum main time frame period allowed by the IMA system, T₀Granularity of time-slices divided for modules of IMA systems, T being T₀Multiples of (a).

For a medium-sized IMA system, assume:

F＝(d,s,n,T,t₀)＝(4,20,225,120,10)

because each module in the IMA system is equivalent, the number of dynamic reconfiguration schemes of the IMA system can be calculated by a second Stirling number method as follows:

it can be seen that the number of possible reconfiguration schemes of the IMA system has already reached billions before the second-stage time dimension configuration of the IMA system is performed, and if an optimal reconfiguration scheme is obtained by using a simple arrangement and screening method, since the reliability, real-time performance and other analyses of the system need to be performed after the system configuration scheme is determined, failure to generate all reconfiguration schemes will result in failure to perform subsequent work.

It can be known through calculation that it is not practical to obtain the optimal reconfiguration scheme of the IMA system through a violent enumeration and screening method. Therefore, it is considered to find a better solution of the reconstruction scheme using a suitable heuristic algorithm. The genetic algorithm is a randomized search method evolved by using the evolution law of biology as a reference, can directly operate a structural object, does not have the limitation of derivation and function continuity, and has the inherent implicit parallelism and better global optimization capability. The genetic algorithm adopts a probabilistic optimization method, can automatically acquire and guide an optimized search space, adaptively adjusts the search direction and does not need a determined rule. When the optimal reconstruction scheme is searched for in the IMA system, the problems are that besides the overlarge search space, the comprehensive reliability, real-time performance and load balance characteristics are needed to evaluate the scheme, and the judgment standard is very complex. To overcome these problems, genetic algorithms were chosen for analytical optimization of the IMA system reconstruction scheme.

The genetic algorithm is adopted to analyze the IMA system, and a genetic algorithm framework needs to be designed firstly. The search process of genetic algorithm is based on biology, and needs to establish an initial population of an IMA reconfiguration scheme, wherein individuals in the initial population contain binary coded IMA system dynamic reconfiguration scheme information. Then, the fitness of the 'natural environment' of the individuals in the initial population needs to be evaluated, and the fitness is determined by the characteristics needing to be evaluated in the reconfiguration scheme. The initial population is selected and propagated to generate the next generation, forming a new population. At this point, the new population of individuals represents a new reconfiguration scheme resulting from gene crossover and mutation. The genetic algorithm can continuously repeat the process to find the most suitable individual, and the setting of the termination condition can obtain a better reconfiguration scheme within a limited time. Thus, the basic algorithm of genetic algorithms actually includes a) encoding, b) fitness function, c) initial population design, d) genetic manipulation and termination conditions, which will be described in detail separately below.

a) Coding;

under the condition that the software and hardware parameters of the IMA system and a top-level task chain are determined and only one piece of software works in each partition is set, the parameters of the IMA system reconstruction structure problem can be represented by time parameters and space parameters, wherein the time parameters comprise the number and the sequence of time slices occupied by the software, and the space parameters refer to the hardware position of the software.

The software number, the hardware number, the time slice number and the sequence are discrete variables, and the binary coding has direct semantics. The length of the idle time slice that the module has should likewise respect the time slice granularity t₀And may be represented by binary coding as well.

Let a function f (x)₁) And f (x)₂,x₃,x₄) Wherein, in the step (A),

x₁＝{1,2,…,s},x₂＝{1,2,…,T/t₀},x₃＝{1,2,…，n},x₄＝{1,2,…,d}

wherein s is the software quantity of the IMA system; t is the maximum main time frame period allowed by the IMA system, T₀Partitioning the granularity of the time slice for the module; n is the maximum number of partitions that the module can accommodate; d is the number of modules of the IMA system.

And performing fixed-length encoding by using a binary character string with the length L being sl + dl', wherein,

l＝l₁+l₂+l₃＝[log₂(T/t₀)]+[log₂n]+[log₂d]

l'＝[log₂(T/t₀)]

l₁,l₂,l₃respectively representing the unit length of a time slice occupied by software, the scheduling order of the module and the module, and l' representing the length of the main time frame of the module. Each value is counted from 0.

All configuration schemes constitute a state space:

S^L＝{a₁,a₂,…,a_k}

wherein, the coding for any reconstruction configuration scheme k is:

a_k＝(a_k1,a_k2,…,a_ks,a_k'1,…,a_k'd)

comprising s software-specific coding factors (a)_k1,a_k2,…,a_ks) And d areCoding factor (a) for hardware_k'1,…,a_k'd) Respectively is as follows:

a_ks＝(a_ks1,a_ks2,…,a_ksl),a_k'd＝(a_k'd1,a_k'd2,…,a_k'dl')

the reconstruction configuration scheme, namely the character string form of the individual codes, is as follows:

s_k＝a_k11a_k12…a_k1la_k21…a_ksla_k'11a_k'12…a_k'1l'a_k'21…a_k'dl'(indicating from lower to higher in order from right to left).

For example, for F ═ d, s, n, T₀) For an IMA system of (4,20,225,120,10), the code length for each configuration is 296 bits of binary code, where l₁＝l'＝4，l₂＝8，l₃＝2。

For a certain configuration scheme, for the 1 st software, the code is a_ks100110000001000, the code length is 14. Since the values are all counted from 0, a binary reduction to decimal requires + 1. So for this software, where the first l₁Bit 0011, indicating that the software occupies a time slice length of 4 units, after which₂The bit 00000010 indicates that the software is the 4 th software in the module to which it belongs, and finally l₃Bit 00 indicates that the software is on module 1;

for the 1 st module, the code is a_k'd11000, the length of the main time frame of the module is 9, which is equal to the sum of the length of the time slice occupied by all software in the module and the length of all idle time slices.

b) A fitness function;

and (3) reconstructing a Petri network model of the configuration scheme by the IMA system obtained by coding and decoding, and further calculating the reliability and real-time index of the IMA system. Reliability and real-time performance only need to meet system requirements, are not used as fitness function evaluation indexes, and are only considered as lethal genes when a subgroup is constructed.

In the load balance characteristic analysis, the time dimension task load of the system can be directly calculated through the partition information contained in the code, and the space dimension task load needs to read the memory allocation condition of the top layer, so that the calculation is performed through the restored AADL model. Because the genetic algorithm code does not contain information interaction mode information of the system, the system should preset intra-partition and inter-partition information interaction modes among tasks in a top-level software task chain. By reading the information interaction mode of the top layer, the AADL model containing the port channel information can be obtained by reduction, and then the corresponding communication load balance index of the system is calculated. After index synthesis, the load balancing index may evaluate the fitness of the individual as part of a fitness function.

The better the reconfiguration scheme when system load balancing takes less.

Then the fitness function f (x), when cost_max-cost(x)>0，f(x)＝cost_max-cost (x); otherwise, f (x) is 0.

Wherein, cost_maxRepresenting the maximum load cost that the system can reach, namely the load cost of the system when the configuration scheme with the maximum load cost is selected; cost (x) represents the load cost of the system when selecting configuration scheme x.

c) And (4) selecting an initial population.

After the codes are written, the initial population selection method of the genetic algorithm is actually verified and determined, and because the IMA system has a large state space and each task has different influences on characteristics, it is difficult to predict which type of distribution method is easy to achieve higher fitness, when the initial population is selected, a certain number of individuals are randomly generated, and then the best individual is selected from the individuals and added to the initial population. This process is iterated until the number of individuals in the initial population reaches a predetermined size.

For example, for an IMA system, F ═ d, s, n, T₀) The number of possible reconfiguration schemes is on the order of more than one billion (4,20,225,120, 10). When the initial population is obtained, 100 reconstruction configuration schemes are randomly generated each time, then the best individual is selected from the reconstruction configuration schemes, and iteration is continuously carried out until 100 individuals are obtained as the initial population.

d) Population evolution;

genetic manipulation (population evolution) involves the following three basic genetic operators: selection, crossover, and mutation. The operation of the individual genetic operators is performed under random perturbation conditions. Therefore, the rule of individual migration to the optimal solution in the population is random, but the randomization operation is different from the traditional random search method, and the method is an efficient and directional random search method. When the genetic algorithm is adopted to search and optimize an IMA reconstruction configuration scheme, a relatively common genetic operation method is selected.

In the selection operation, roulette wheel selection (roulette wheel selection) is used, in which the selection probability of each individual is proportional to its fitness value. The greater the fitness of an individual, the higher the probability that it is selected and vice versa. Individuals who do not meet the requirements of reliability and real-time performance appear to be non-viable and will not be selected into mating pairs. After the selection probability of each individual in the initial population is calculated, multiple rounds of selection are required in order to select mating individuals. Each round generates a random number which is uniform between [0, 1], and the random number is used as a selection pointer to determine the selected individuals. After the individuals are selected, mating pairs can be randomly formed for later crossover operations.

The genes of the mating pair are swapped in a crossover operation to create a new combination of genes, in hopes that beneficial genes can be combined together. The most common one-point cross (one-point cross) method is used. The specific operation is as follows: a cross point is randomly set in the individual string, and when the cross is executed, the partial structures of two individuals before or after the cross point are interchanged to generate two new individuals.

In mutation operations, changes are made to gene values at certain loci of individual strings in the starting population to avoid the algorithm entering local convergence. A binary mutation algorithm is adopted, and the mutation probability is 0.01.

e) A termination condition;

setting the algebraic upper limit of iteration as a termination condition to ensure that the system can complete the search process in a proper time, and for the scale of F ═ d, s, n, T, T₀)＝(4,20,225,120,10), the generation number of the genetic algorithm is set to 200 generations.

And thirdly, calculating the reliability and the real-time performance of the system configuration as individual screening conditions.

An AADL configuration model of the IMA system can be obtained through system software and hardware and time configuration codes, meanwhile, reliability and real-time property corresponding to system software and hardware are already contained in an AADL multi-characteristic analysis basic model of the IMA system, the AADL model can be converted into a Petri network model, reliability and real-time property indexes of the system under the current configuration are calculated on the basis of the Petri network model, and the reliability and real-time property indexes are compared with design allowable values to be used as lethal genes in a genetic algorithm to judge whether an individual survives.

For reliability analysis of an IMA system, respectively corresponding a member and a connecting piece in an AADL model to a library site and a transition in a Petri network model according to a conversion rule, corresponding a connection relation between the member and the connecting piece to a directed arc, converting the AADL model into a reliability analysis Petri network model, and generating a test path PW according to a migration probability of a weighted SAPetri network (Software Architecture Petri network), wherein the migration probability of the test path PW can be calculated as follows:

wherein, P_tiAnd representing the migration probability of the arc i in the software architecture Petri net.

Test path PW of C₁→L₁→C₂→L₂→……→L_n→C_n+1Comprising n +1 members C_iCorresponding depot and n connecting pieces L_iAnd (3) corresponding transitions, wherein 2n arcs are shared between the library and the transitions, and the reliability of the test path PW is calculated as follows:

wherein R is_ciIs that C is constructed in the test path PW of SAPN_iReliability of the corresponding library, R_LiIs a connecting piece L in the test path PW of the SAPN_iReliability of the corresponding transition, R_TiIs directed arc T between depot and transition_iReliability of (2);

reliability R of the system_SAThat is, the reliability of the whole SAPN is calculated by the reliability and the migration probability of all m test paths PW, where each test path considers the reliability of n +1 libraries, n transitions and the migration process therebetween, as follows:

wherein, P_PWiRepresenting the probability of migration, R, along the ith test path PW_PWiThe reliability of the ith test path PW is indicated.

Reliability R when reconfiguring a configuration scheme_SAWhen the system requirements are not met, the individuals corresponding to the codes of the reconfiguration scheme are excluded from the corresponding populations.

When real-time analysis is carried out, firstly, converting an AADL system architecture model into an SAPetri network model according to a conversion rule from an IMA system AADL model to the Petri network model; in addition to converting the time attribute describing task execution and information interaction into time characteristics in a position element and a migration element, the attribute describing task response time needs to be converted, test paths are generated according to a time SAPetri model and given start and end positions of a task, test paths from an initial point to an end point can be obtained through a Breadth First Search (BFS) algorithm, each test path is analyzed in real time, and calculation can be performed:

and accumulating the time attributes of the elements on the test path to obtain a real-time analysis result. The total number of elements O on the test path PW is 2n +1 (including the transitions corresponding to the libraries corresponding to the n +1 members and the n connecting members), wherein

Representing the time attribute, T, of the ith element O_PWRepresenting the total time taken by the test path PW. For all test paths, the IMA reconfiguration policy should be satisfied

Wherein

Represents the maximum time limit, T, of the test path j allowed by the system_PWjRepresenting the actual elapsed time for test path j in the current configuration. When the reconfiguration method does not satisfy the above conditions, the individual should be excluded from the population in which it is located.

Fourthly, carrying out load balance analysis on the reconfiguration scheme to obtain individual fitness;

when the software and hardware of the IMA system are configured and dynamically reconstructed, the reconstruction configuration scheme firstly needs to meet the performance requirements of the IMA system, such as the requirements on reliability and real-time performance. On the basis of meeting the necessary conditions, the IMA system configuration scheme is still expected to better adapt to the current task requirement, so that the working efficiency of the system is maximized, the communication requirement among modules is reduced, and the possibility of failure is reduced. By using an analysis method for load balance problems in a distributed system for reference, static load balance analysis is carried out on an IMA system, the task processing capacity of the IMA system is described through a running time ratio and a memory occupation ratio, and the communication capacity of the IMA system is described through bus bandwidth occupied by communication among modules and communication step length, so that the static load balance analysis is used as an evaluation index of a reconfiguration configuration scheme of the IMA system.

(1) An operating time ratio;

in an IMA system, unlike a preemptive scheduling method for tasks on processing units of a distributed system, the tasks are first assigned to different partitions, and each partition is assigned a different time window within a main time frame. The partition is based on the scheduling strategy of time slice rotation, and can only acquire the processing resource in the allocated time slot and be called to execute. As shown in fig. 2, 1 to k partitions and partially free partitions may be included in one master time frame, and a free partition may be located between any two working partitions. The partitions are sequentially executed from left to right in time sequence, and after the current main time frame is executed, the sequence is periodically and circularly operated.

For a general IMA system, the system contains d modules. Wherein the l-th module has k_lAnd (4) partitioning. The task load of the ith module in the time dimension can be expressed as:

therein, ζ₁The maximum operation load value of the ith module represents the calculation processing capacity of the module; omega_f1The task load of the f-th partition on the l-th module.

The tasks of the partitions on the same module of the IMA system are carried out in a time-sharing manner, so in the task load formula of the first module on the time dimension, the task load of the first module on the time dimension is expressed as the ratio of the sum of the task loads of all the partitions on the first module to the maximum operation load value of the first module. We can further simplify the formula for the task load of the ith module in the time dimension for the architectural and scheduling characteristics of IMA systems.

In the IMA system scheduling strategy based on the time slice round robin, even if a partition fails in the calling execution process, if the partition cannot finish running in the allocated time slot, the system can forcibly finish the partition and move the time window to the next time slice. Accordingly, even if the partition application has completed before the end of the assigned time slot, the system will not allocate this portion of the free time to other tasks. That is, the time dimension that a task occupies is independent of its own computational requirements, and is only dependent on the length of the assigned time slot. When the allocated time slices are too few to complete the task, the scheduling requirements of the system are not met, the reconfiguration configuration scheme is directly rejected, and no further load balancing analysis is needed. So the task load of the f-th partition on the l-th module

Comprises the following steps:

wherein, T₁The length of the main time frame of the first module;

the length of the time slice allocated to the f-th partition on the l-th module.

The formula for expressing the task load of the ith module in the time dimension can be converted into:

note the book

The total working time of the ith module in a main time frame is represented as follows, the task load of the ith module in the time dimension is further simplified as follows:

in an IMA system, d modules are operated simultaneously, and the task load cost of the IMA system in the time dimension_timescaleExpressed as the maximum value of the load of each module, i.e.:

from the above equation, for a reconfiguration scheme that meets the scheduling requirements of the system, the task load of the system in the time dimension is independent of the actual computation requirements of the tasks of the partitions of the module, and is only related to the allocation of the time slices. When cost_timescaleMinimum sizeAnd then, a reconfiguration scheme with optimal computing capacity load of the system in the time dimension can be obtained.

(2) Memory occupancy ratio;

each module of the IMA system is provided with an independent processing unit, each partition occupies processing unit resources according to a time slice round-robin scheduling method, and when each partition task runs, except that the lengths of allocated time slices are different, the sizes of occupied module memories are also different. In order to make better use of system resources and improve the robustness of the system, the memory occupation ratio of each module needs to be balanced as much as possible.

The minimum unit of the change of the configuration scheme of the system is a partition, and the fluctuation of the memory occupation amount of the operation of a single partition can be ignored relative to the difference of the memory occupation amount between different partition tasks. Because the system needs to ensure the operation function, the memory resources occupied by a single partition cannot exceed the upper limit of the module memory, and the space dimension processing capacity load of the partition can be represented by the average value of the occupied amount of the task operation memory.

As shown in FIG. 3, the main time frame of module 1 has a length T₁When the operation reaches the idle time slice, the memory occupation amount of the module is 0. The length of the main time frame of module 2 is T₂The length of the main time frame of the module r is T_rThe main time frames of the modules are different in length, the time slice division of the partitions is different except that the same minimum granularity is followed, and further, the change rule of the memory occupied by the partitions on each module is different. The quantity of partitions of each module of the IMA system, the length of a time slice occupied by the partitions and the length of a main time frame are different, and when the running time of two tasks which occupy the same amount of memory is different, the influence on the load balance of the system is also different. So when considering the load capacity of the IMA system in the spatial dimension, the memory footprint still needs to be weighted according to the temporal attributes.

The task load on the space can be expressed as:

wherein M is₁The total memory size of the first module; m is_f1And the average memory amount occupied by the operation of the f-th partition on the l-th module is calculated.

In order to meet the hardware replaceability of the IMA system, the total memory amount of each module is the same, and for any module l, M_l＝M₁Task load cost in system space dimension as M_spacescaleExpressed as the maximum value of the space load of each module, namely:

when the load capacity of the system in the space dimension is described by using the memory occupation of the task operation, the size and time of the memory occupation of the task need to be considered, so that the memory occupation ratio also describes the load capacity of the system in the space-time coupling dimension. When cost_spacescaleAnd when the minimum time is reached, the optimal reconfiguration scheme of the system load capacity can be obtained.

(3) Bus bandwidth occupancy;

in an IMA system, inter-partition communication is performed through a channel cache between ports according to a sampling or queue mode, as shown in fig. 4, a channel is connected to a partition port of an application layer upward and a hardware port of a hardware layer downward, and may be connected to a source partition and one to a plurality of destination partition ports. In terms of time resources, the communication delay between the partitions is far smaller than the time deviation caused by the rotation of the partition tasks according to the time slices, the communication time between the partitions is determined by the time slice configuration of the module where the partitions are located, and extra time resource consumption can not be caused under the condition that the real-time requirement of the system is met.

On the aspect of space resources, in the same module, the module is pre-configured with enough cache for information interaction between partitions, and the space resources occupied by task operation are not influenced. The partitioned communication among the modules needs to be carried out through a system bus, and the communication capacity of the system can be described by using the occupation amount of the bandwidth of the system bus. In IMA systemsDifferent from dynamic bandwidth allocation of information transfer in a distributed system network, ports used when two partitions communicate through a system bus and bus bandwidth capable of being occupied are configured in advance, so that cost of communication resources of an IMA system is cost_spacecommOnly the number of configured information interaction channels and the preset bandwidth value of the channels are related, and can be represented as follows:

wherein m is_ckBus bandwidth occupied for channel k, M_cIs the total amount of bandwidth of the system bus.

Because the bandwidth allocation of the system bus is preset, the use efficiency of the channel does not affect the communication overhead of the system, and in the IMA system, the communication overhead of the system is mainly reduced by reducing the number of channels on the system bus, particularly the channels occupying larger bus bandwidth.

(4) Integrating evaluation indexes;

as can be seen from the IMA system load balance evaluation index, the IMA system can perform static load balance analysis according to the system structure parameters, and the influence generated by task operation is not required to be considered by means of other models. That is, the load balance analysis of the IMA system can be performed by means of its AADL structural model without establishing a separate load balance analysis model.

The pre-configuration of the IMA system ensures the system function and reduces the coupling between the system module and the partition, so that the evaluation indexes are relatively independent, and index utility change caused by the system operation state is not required to be avoided when index integration is carried out. For more flexible description of the load balance characteristic of the IMA system, the three indexes are weighted and summed:

wherein the coefficients

Can be set through experience and test analysis.

When the total load cost of the system is smaller, the better the load balance characteristic of the system configuration scheme is.

The fitness function f (x) of the genetic algorithm is:

when cost_max-cost(x)＞0，f(x)＝cost_max-cost (x), otherwise, f (x) is 0.

And fifthly, decoding the optimal gene obtained by searching the genetic algorithm to obtain the dynamic reconfiguration configuration scheme of the IMA system.

After the genetic algorithm is subjected to search convergence or specified algebra is completed, the individual genes with the optimal algorithm can be obtained, wherein the individual genes comprise software and hardware matching and time scheduling information of the IMA system dynamic reconfiguration configuration scheme, corresponding software and hardware matching matrixes and time scheduling information on each module are obtained through decoding, the information is corresponding to an AADL (architecture analysis and design language) model of a system architecture, and the corresponding IMA system dynamic reconfiguration configuration scheme can be obtained.

The specific implementation process of the IMA system dynamic reconfiguration strategy optimization method is described below by a simplified flight control system example.

In a flight control system, a "flight State selector" (State _ selector) and a "flight control console" (Operating) are human-machine interfaces of the flight control system. The system has the function of setting the working state of the system by a driver and the function of indicating the working state of the system. The system consists of stability augmentation/control augmentation (pcs, lcs), automatic pilot control/mode (ap _ manage) and automatic balancing (pap, lap); wherein, the stability increasing/automatic stability increasing and the automatic pilot control/mode are respectively provided with a stability sensor and a GPS as information input interfaces. Meanwhile, the system also has a control line subsystem which mainly comprises a flight state selector (CDS), starting logic, redundancy management, an operating system and a system BIT. The method comprises the following specific implementation steps.

Step one, establishing an IMA system dynamic reconstruction strategy analysis model based on AADL.

The schematic diagram of the AADL model of the main software program of the system is shown in FIG. 5, the operating system, the starting logic, the BIT, the redundancy management, the stability/control stability augmentation, the automatic balancing, the automatic pilot control/mode and the flight state selection are used as software applications, the flight control console, the flight state selector, the stability sensor and the GPS are used as peripheral equipment, and the connection relationship between the software and the applications is shown in the figure. The reliability and real-time analysis attribute data given by the system are added into the AADL model as shown in tables 3 and 4.

TABLE 3 reliability analysis data

TABLE 4 real-time analytical data (unit: ms)

The dynamic reconfiguration strategy of the IMA system is configured, after the system is reconfigured, the 10 basic software applications are allocated to 5 equivalent IMA system modules, and the time configuration of the modules is given in the following configuration space model.

The system configuration space model is as follows:

F＝(d,s,n,T,t₀)

wherein d is the number of modules of the system, and s is the number of software of the system; n is the maximum number of partitions that the module can accommodate; t is the maximum main time frame period allowed by the system, T₀The granularity of the time slice is divided for the modules.

For this IMA system there are:

F＝(d,s,n,T,t₀)＝(5,10,10,10,1)

and step two, setting an IMA system genetic algorithm analysis algorithm.

a) Encoding

Let a function f (x)₁),f(x₂,x₃,x₄)，Wherein the content of the first and second substances,

x₁＝{1,2,…,10},x₂＝{1,2,…,10},x₃＝{1,2,…,10},x₄＝{1,2,…,5}

and performing fixed-length encoding by using a binary character string with the length L being sl + dl' being 130, wherein,

l＝l₁+l₂+l₃＝[log₂(T/t₀)]+[log₂n]+[log₂d]＝11

l'＝[log₂(T/t₀)]＝4

establishing a bit string space:

S^L＝{a₁,a₂,…,a_k}

the coding of configuration scheme k is:

a_k＝(a_k1,a_k2,…,a_ks,a_k'1,…,a_k'd)

the method comprises s coding factors aiming at software and d coding factors aiming at hardware:

a_ks＝(a_ks1,a_ks2,…,a_ksl),a_k'd＝(a_k'd1,a_k'd2,…,a_k'dl')

arrangement, i.e. vector a of individuals_k＝(a_k1,a_k2,…,a_ks,a_k'1,…,a_k'd) In the form of a string of characters

s_k＝a_k11a_k12…a_k1la_k21…a_ksla_k'11a_k'12…a_k'1l'a_k'21…a_k'dl'(indicating from lower to higher in order from right to left). b) Fitness function

The fitness function f (x),

when cost_max-cost(x)>0，f(x)＝cost_max-cost (x); otherwise, f (x) is 0.

Wherein cost_maxAnd cost (x) are given in step 4.

c) Initial population selection

For example,for IMA systems F ═ d, s, n, T, T₀) (5,10,10,10,1), the software and hardware matching state space number is calculated according to the second type Stirling number method

And each software and hardware matching configuration has different number of time slice scheduling configuration schemes. Selecting 100 state schemes randomly generated each time when obtaining the initial population, then picking out the best individual from the state schemes, and continuously iterating until obtaining 100 individuals as the initial population.

d) Genetic manipulation

In the selection operation, a roulette selection method is used, and the selection probability of each individual is proportional to the fitness value thereof. Individuals who do not meet the requirements of reliability and real-time performance appear to be non-viable and will not be selected into mating pairs. After the selection probability of each individual in the population is calculated, it is necessary to perform multiple rounds of selection in order to select mating individuals. Each round generates a random number which is uniform between [0, 1], and the random number is used as a selection pointer to determine the selected individuals. The genes of the mating pair are swapped in a crossover operation to create a new combination of genes, in hopes that beneficial genes can be combined together. The most common single point crossing method is used. In the mutation operation, a binary mutation algorithm is adopted, and the mutation probability is 0.01.

e) Termination conditions

An iterative algebra upper limit is set as a termination condition to ensure that the system can complete the search process within a proper time, and through practical verification, for the IMA system, the genetic algorithm algebra is set to be 50 generations to obtain a better search result.

Step three, calculating the reliability and the real-time performance of system configuration as individual screening conditions

Establishing a Petri network model for analyzing the reliability and real-time performance of an IMA system reconstruction strategy, converting the model into a Petri network graph according to an AADL system structure model as shown in FIG. 6, wherein C1 to C10 represent libraries obtained by converting components C1 to C10, L1 to L10 represent transitions obtained by converting connecting pieces L1 to L10, and searching to obtain a test path as shown in Table 5.

TABLE 5 test paths

According to the path transfer weight averaging principle, the average transfer probability of each path is obtained, and then the path reliability of each test path is calculated as shown in table 6:

TABLE 6 test Path reliability

Numbering	Transition probability	Reliability of path
			P1	0.031250	0.790
P2	0.078125	0.900
			P3	0.062500	0.922
P4	0.125000	0.922
			P5	0.078125	0.890
P6	0.125000	0.900
			P7	0.125000	0.922
P8	0.125000	0.922
			P9	0.031250	0.860

Substituting the calculation result into a total calculation formula of the system structure, wherein the reliability of the system structure is calculated as follows:

R_SA＝0.7074085/0.78125＝0.9055

reliability index R of system_o＝0.9000，R_SA＞R_oAnd the reliability requirement of the system is met.

The running time of each test path is calculated according to the execution time and the trigger time of the test path passing through the member as shown in table 7:

TABLE 7 test Path runtime

For all paths, the runtime of the IMA reconfiguration configuration policy should be less than the longest runtime limit.

Individuals who do not meet the requirements for reliability and real-time should be excluded from the progeny population. The calculation can show that the reliability and the real-time performance in the configuration scheme are both satisfied, and the individual fitness analysis can be continuously carried out.

Step four, carrying out load balance analysis on the configuration scheme to obtain individual fitness

The following tables 8 and 9 are used for extracting system configuration information and communication information from the system reconfiguration strategy AADL model:

TABLE 8 System hardware and software configuration (time unit: s)

TABLE 9 System communication configuration

And the total amount of memory of each module of the system is the same, and has M_l1000, total system bus bandwidth M_c＝3000。

From the above information, three load balancing characteristics of the system configuration scheme are calculated:

(1) run time ratio

The task load of the system module l in the time dimension can be expressed as:

wherein

The total working time of the module in a main time frame.

The task load of each module is as in table 10:

TABLE 10 module task time load

The task load of the system in the time dimension is represented as the maximum value of the load of each module, and the maximum value comprises the following components:

cost_timescale＝max(l)_tcost＝0.9

(2) memory footprint

Total amount of memory of each module is the same, M_lThe task load of each module is as in table 11:

TABLE 11 Module task space load

The task load in the system space dimension is expressed as the maximum value of the space load of each module, namely:

cost_spacescale＝max(l)_scost＝0.85

(3) bus bandwidth occupancy

The communication resource cost of the IMA system is only related to the number of the configured information interaction channels and the preset bandwidth value of the channels, the communication in the same module in the table is omitted, and the bandwidth of the rest channels is added to obtain:

wherein m is_ckThe bus bandwidth occupied by channel k.

(4) Integration of evaluation indexes

The three indices are summed:

cost＝cost_timescale+cost_spacescale+cost_spacecomm＝0.9+0.85+0.8＝2.55

when the total load cost of the system is smaller, the better the load balance characteristic of the system configuration scheme is.

The fitness function of the genetic algorithm is f (x) cost_max-cost(x)＝3-2.55＝0.45。

Step five, decoding the optimal gene obtained by genetic algorithm search to obtain an IMA system dynamic reconfiguration configuration scheme

After the genetic algorithm is subjected to search convergence or specified algebra is completed, the individual genes with the optimal algorithm can be obtained, wherein the individual genes comprise software and hardware matching and time scheduling information of the IMA system dynamic reconfiguration configuration scheme, corresponding software and hardware matching matrixes and time scheduling information on each hardware module are obtained through decoding, the information is corresponding to an AADL (architecture analysis and design language) model of a system architecture, and the corresponding IMA system dynamic reconfiguration configuration scheme can be obtained.

Assuming that the current scheme is the optimal scheme obtained by genetic algorithm, the information available from the individual codes of the scheme is shown in table 12 and table 13:

TABLE 12 software configuration information

Component numbering	Position of the component	Time slice length	Memory size
				C1	M3	4	500
C2	M1	4	500
				C3	M2		2	300
C4	M2		2						200
				C5	M3	3	200
C6	M4		2					300
				C7	M4		2		200
C8	M2	5	350
				C9	M5	4	400
C10	M5	3	350

Table 13 module configuration information

Module numbering	Length of main time frame
		M1	6
M2	10
		M3	8
M4	5
		M5	10

Corresponding these information to the AADL model of the system architecture, a corresponding dynamic reconfiguration AADL model of the IMA system can be obtained as shown in fig. 7. As in fig. 7, the components C1 to C10 are located in partitions on five modules M1 to M5, respectively, and the connection relationship between the components is determined by the system task requirements, as in fig. 5.

Claims (3)

1. A dynamic reconstruction strategy optimization method of an IMA system based on a genetic algorithm is characterized in that: the method comprises the following steps:

establishing an IMA system dynamic reconstruction strategy analysis model based on AADL;

modeling an IMA system structure and a dynamic reconstruction process by using an AADL language, adding system reliability, real-time performance and load balance analysis attributes in the dynamic reconstruction of the IMA system, and establishing an IMA system dynamic reconstruction strategy analysis model based on the AADL; aiming at reliability analysis, the reliability analysis comprises the basic reliability and the task transfer probability of each software and hardware of the system; aiming at real-time analysis, including response and running time of system software, task transition probability and initial identification; for load balance analysis, a time scheduling mechanism of an IMA module needs to be given, the length of a main time frame and a time slice unit allowed by each IMA module are determined, and in addition, the top-level configuration of communication interfaces and channels among IMA system partitions needs to be determined, namely the configuration under different software and hardware matching relations;

step two, setting an IMA system genetic algorithm analysis algorithm;

analyzing the IMA system by adopting a genetic algorithm, and firstly, coding software and hardware matching and time configuration of the IMA system; the code adopts binary code and comprises a software number, a hardware number, a time slice number and sequence information configured by the system;

adopting a genetic algorithm, and adopting the existing stochastic Petri net model and analysis method for analyzing reliability and real-time performance when generating and screening populations by taking different characteristics of an IMA system as individual judgment conditions; in the load balance analysis, aiming at the task load and the communication load of the system, the task load of the system is described by a running time ratio and a memory occupation ratio on time and space dimensions respectively; aiming at communication load, bus bandwidth occupation quantity is provided to describe the communication load on the system space dimension; the characteristic analysis is integrated to obtain a judgment basis for optimizing the dynamic reconfiguration strategy of the IMA system;

selecting a roulette selection method, a single-point crossing method or a binary variation algorithm in the genetic algorithm operation; when the search is terminated, determining an algebraic upper limit of the search as a termination condition;

thirdly, calculating the reliability and the real-time performance of system configuration as individual screening conditions;

obtaining an AADL configuration model of the IMA system through system software and hardware matching and time configuration coding, converting the AADL configuration model into a Petri network model and calculating reliability and real-time indexes of the IMA system under the current configuration on the basis of the Petri network model, and comparing the reliability and real-time indexes with design allowable values to be used as death causing genes in a genetic algorithm to judge whether an individual survives, wherein the reliability and real-time attributes corresponding to system software and hardware are already contained in the AADL multi-characteristic analysis basic model of the IMA system;

step four, carrying out load balance analysis on the configuration scheme to obtain individual fitness;

when the software and hardware of the IMA system are configured and dynamically reconstructed, the configuration scheme firstly needs to meet the requirements of the reliability and the real-time performance of the system; on the basis of meeting the necessary conditions, static load balance analysis is carried out on the IMA system, the task processing capacity of the IMA system is described through a running time ratio and a memory occupation ratio, and the communication capacity of the IMA system is described through bus bandwidth occupied by communication among IMA modules to serve as an evaluation index of the IMA system configuration scheme; the evaluation indexes are three, and are respectively as follows: the running time ratio, the memory occupation ratio and the bus bandwidth occupation amount; integrating the three evaluation indexes as the fitness function of the individual;

decoding the optimal gene obtained by searching the genetic algorithm to obtain an IMA system dynamic reconfiguration configuration scheme;

after the genetic algorithm is subjected to search convergence or specified algebra is completed, an individual gene with the optimal algorithm is obtained, wherein the individual gene comprises software and hardware matching and time scheduling information of an IMA system dynamic reconfiguration scheme, corresponding software and hardware matching matrixes and time scheduling information on each module are obtained through decoding, and the information is corresponding to an AADL (architecture analysis and design language) model of a system architecture to obtain a corresponding IMA system dynamic reconfiguration scheme.

2. The method for optimizing the dynamic reconfiguration strategy of an IMA system based on genetic algorithm as claimed in claim 1, wherein: the genetic algorithm comprises a) encoding, b) fitness function, c) initial population design, d) genetic manipulation and termination conditions;

a) coding;

under the conditions that various parameters of software and hardware of the IMA system and a top-level task chain are determined and only one piece of software works in each partition is set, the parameters of the IMA system reconstruction structure problem are expressed by time parameters and space parameters, the time parameters comprise the number and the sequence of time slices occupied by the software, and the space parameters refer to the hardware position of the software;

the software number, the hardware number, the time slice number and the sequence are discrete variables, and binary coding is adopted to have direct semantics; the length of the idle time slice that the module has should likewise respect the time slice granularity t₀Also represented in binary code;

b) a fitness function;

the Petri network model of the IMA system reconfiguration configuration scheme is obtained through coding and decoding, and then the reliability and real-time performance indexes of the IMA system are calculated; reliability and real-time performance only need to meet system requirements, are not used as fitness function evaluation indexes, and are only considered as lethal genes when subgroups are constructed;

in the load balance characteristic analysis, the time dimension task load of the system is directly calculated through partition information contained in codes, and the space dimension task load needs to read the memory allocation condition of the top layer, so that the calculation is carried out through a restored AADL model; when the system load balancing cost is lower, the reconfiguration configuration scheme is better; then the fitness function f (x), when cost_max-cost(x)>0，f(x)＝cost_max-cost (x); otherwise, f (x) is 0;

wherein, cost_maxRepresenting the maximum load cost that the system may achieve, i.e. selecting the party with the highest load costThe load cost of the time plan system; cost (x) represents the load cost of the system when selecting configuration scheme x;

c) selecting an initial group;

when an initial population is selected, a certain number of individuals are randomly generated, and then the best individual is selected from the individuals and added into the initial population; the process is iterated continuously until the number of individuals in the initial population reaches a predetermined scale;

the best individuals refer to: individuals containing information about the dynamically reconfigurable configuration scheme of an IMA system encoded with binary;

d) population evolution;

in the selection operation, a roulette selection method is adopted, in which the selection probability of each individual is proportional to the fitness value thereof; the larger the individual fitness, the higher the probability that it is selected, and vice versa; individuals with unsatisfactory reliability and instantaneity appear to be non-viable and will not be selected into mating pairs; after the selection probability of each individual in the initial population is calculated, in order to select mating individuals, multiple rounds of selection are needed, each round generates a uniform random number between [0, 1], and the random number is used as a selection pointer to determine the selected individual; after the individuals are selected, randomly forming a mating pair for later crossing operation; in the mutation operation, a binary mutation algorithm is adopted, and the mutation probability is 0.01;

e) and (4) terminating the conditions.

3. The method for optimizing the dynamic reconfiguration strategy of an IMA system based on genetic algorithm as claimed in claim 1, wherein: in the fourth step, static load balance analysis is performed on the IMA system, the task processing capacity of the IMA system is described by a running time ratio and a memory occupation ratio, and the communication capacity of the IMA system is described by a bus bandwidth and a communication step length occupied by inter-module communication, so as to serve as an evaluation index of a reconfiguration configuration scheme of the IMA system, which is specifically as follows:

(1) an operating time ratio;

in an IMA system, tasks are firstly distributed to different partitions, and each partition is distributed with different time windows in a main time frame; the partition is based on a scheduling strategy of time slice rotation, and can only acquire processing resources in an allocated time slot and be called and executed;

for a generic IMA system, the system comprises d modules, the first of which has k_lAnd the task load of the first module in the time dimension is represented as:

therein, ζ₁Representing the computing processing capacity of the first module for the maximum operational load value of the module; omega_f1The task load of the f-th partition on the first module is;

task load of f-th partition on first module

Comprises the following steps:

wherein, T₁The length of the main time frame of the first module;

the length of the time slice allocated to the f-th partition on the first module;

the formula for the first module's task load in the time dimension translates to:

note the book

For the total working time of the first module in a main time frame, thenThe task load of the first module in the time dimension is further simplified as follows:

t_1wtotal operating time in a master time frame for the first module;

in an IMA system, d modules are operated simultaneously, and the task load cost of the IMA system in the time dimension_timescaleExpressed as the maximum value of the load of each module, i.e.:

t_lwthe total working time of the module in a main time frame;

from the above formula, for the reconfiguration configuration scheme meeting the scheduling requirement of the system, the task load of the system in the time dimension is not related to the actual calculation requirement of each partition task of the module, but only related to the distribution of the time slice, when cost is used_timescaleWhen the minimum time is reached, obtaining a reconfiguration scheme with the optimal computing capacity load of the system in the time dimension;

(2) memory occupancy ratio;

the task load on the space is expressed as:

wherein M is₁The total memory size of the first module; m is_f1Running the average memory occupied for the f-th partition on the first module;

in order to meet the hardware replaceability of the IMA system, the total memory amount of each module is the same, and for any module l, M_l＝M₁Task load cost in system space dimension as M_spacescaleExpressed as the maximum value of the space load of each module, namely:

when cost_spacescaleWhen the minimum time is reached, obtaining the optimal reconfiguration scheme of the system load capacity;

(3) bus bandwidth occupancy;

cost of communication resources of IMA system_spacecommOnly the number of configured information interaction channels and the preset bandwidth value of the channels are related, and the information interaction channels are represented as follows:

wherein m is_ckBus bandwidth occupied for channel k, M_cIs the total bandwidth of the system bus;

(4) integrating evaluation indexes;

the three indexes are weighted and summed:

wherein the coefficients

Setting through experience and test analysis;

when the total load cost of the system is smaller, the load balance characteristic of the system configuration scheme is better; the fitness function f (x) of the genetic algorithm is:

when cost_max-cost(x)＞0，f(x)＝cost_max-cost (x), otherwise, f (x) is 0.

Images