High-speed data acquisition system mapping method based on improved whale algorithm

By improving the whale algorithm, combining genetic algorithm and Levy flight function, the thermal effect and local optimization problems in high-speed data acquisition systems are solved, the thermal balance and performance of the system are improved, power consumption is reduced, and the mapping quality of the data acquisition system is significantly improved.

CN119065837AInactive Publication Date: 2024-12-03GUILIN UNIV OF ELECTRONIC TECH
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202411094723.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-10
Publication Date
2024-12-03
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

High-speed data acquisition systems are prone to thermal effects problems in high-integration chips, resulting in performance degradation and increased power consumption. The existing whale algorithms are prone to problems of local optimization and low convergence accuracy in large-scale complex optimization.

Method used

Improved whale algorithms by introducing adaptive random sinusoidal functions and Levy flight functions, combining genetic algorithms, enhance the algorithm's local and global search capabilities, establish a new power consumption model, and optimize the thermal balance problem of three-dimensional on-chip networks.

Benefits of technology

It improves the thermal balance level and overall performance of the system, reduces power consumption, and significantly improves the quality and convergence speed of the data acquisition system mapping.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119065837A_ABST
    Figure CN119065837A_ABST
Patent Text Reader

Abstract

The invention discloses a high-speed data acquisition system mapping method based on an improved whale algorithm. The mapping method is composed of an evaluator, a calorific value weight matrix, a task graph to be mapped and an optimization algorithm. The evaluator includes an evaluation function and an NoC model. The NoC model records the processing data volume of each node in the NoC network. And the evaluation function calculates the fitness of the mapping solution according to the NoC model. And the processing data volume of each node is obtained after the mapping solution is mapped to the 3D NoC according to the task matrix. In this method, each individual of the whale population can be regarded as a one-dimensional sequence mapping solution. Elements in the sequence respectively represent an acquisition node, a storage node, a transmission node and a main control node according to different positions of the elements, and numerical values of the elements represent mapping positions of the nodes in the NoC. In the iteration process, the whale individual with the highest fitness at present is selected as a leader. And optimizing the local search capability of the whale algorithm by using a self-adaptive random sine function. The global search capability of the whale algorithm is enhanced through combination of a level flight function whale algorithm and a genetic algorithm, so that the whale algorithm jumps out of the local maximum. The method not only can effectively optimize the performance of the data acquisition system, but also can improve the efficiency and reliability of the system, and is suitable for various application scenes needing high-speed data processing.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of network-on-chip high-speed data acquisition systems, and in particular to a high-speed data acquisition system mapping method based on an improved whale algorithm. Technical Background

[0002] The development of high-speed data acquisition systems is limited and challenged by many factors such as data acquisition speed, accuracy and system complexity. With the advancement of technology and the increase in application requirements, traditional monolithic ADCs have been unable to meet the needs of high-speed data acquisition. Therefore, the development of multi-chip ADCs combined with time interleaving technology (TIADC) has become an important solution. Time interleaved ADC technology achieves high-speed sampling by using multiple low-speed ADC chips at the same time, thus breaking through the limitations of the speed limit of a single ADC. Its working principle is to slightly stagger the sampling clocks of multiple ADC chips so that they work alternately in different time segments, and each ADC is responsible for sampling a part of the entire frequency range. Through reasonable design and control of clock synchronization, the speed of data acquisition can be greatly improved while maintaining system accuracy and performance.

[0003] Three-dimensional Network on Chip (3D NoC) improves system integration and packaging density, and reduces chip area. Topology architecture and thermal effects such as Figure 2 As shown. However, due to the increase in the complexity of chip integration, a large thermal effect may be generated, which in turn affects the reliability of the chip. In actual applications, excessive local temperature will seriously affect the performance of the entire chip. Studies have shown that, especially in highly integrated chips, the delay increases by about 5% for every 10°C increase in temperature, and the line leakage current increases exponentially with the increase in temperature. Heat accumulation has become an important factor in energy consumption and has hindered the development of NoC systems. Through a reasonable mapping strategy, the overall performance of the system can be significantly improved, especially in terms of energy consumption. Specifically, the mapping optimization problem refers to arranging the position of each IP core in the on-chip network topology architecture and making the network have the best possible performance based on a given communication task graph, design constraints, IP core library, and network topology. Taking the on-chip network Mesh structure as an example, the on-chip network mapping process is as follows Figure 3 shown.

[0004] The mathematical description of the high-speed data acquisition system is as follows: Figure 4As shown in the figure, the connecting edges between task nodes represent the communication relationship between different functional modules, and the numbers on the connecting lines represent the data communication volume between task nodes. In the high-speed acquisition system based on time interleaving technology, all acquisition nodes have the same sampling rate, but there is a phase difference in the time when each node starts sampling. During the data acquisition process, the communication volume, data transmission bandwidth requirements and delay limits between each acquisition node and storage node are the same. When the controller issues a data transmission instruction, the corresponding storage node transmits the stored data to the specified transmission interface to complete the function of the data acquisition system. The more acquisition nodes there are in the high-speed data acquisition system, the higher the acquisition rate of the data acquisition system. Faced with the problem of a large number of nodes in the data acquisition system, exhaustive traversal and traditional heuristic methods will become very time-consuming.

[0005] The whale algorithm (WOA) simulates the hunting behavior of humpback whales. The algorithm's optimization process can be divided into three optimization mechanisms: surrounding prey, random walk to search for prey, and bubble net spiral attack on prey. These three strategies can cleverly solve many optimization problems. However, as a swarm intelligence algorithm, the whale algorithm itself has population characteristics, and the optimization process is inseparable from the information exchange and collaboration between populations. As the whale algorithm iterates, the global search capability gradually weakens, and the optimization mechanism is gradually replaced by two methods: surrounding prey and bubble net attack. Both search methods are based on the optimal individual, resulting in a relatively single diversity of the entire whale population. When faced with large-scale complex optimization such as high-speed data acquisition system mapping problems, there are problems such as easy to fall into local optimality, low convergence accuracy, and poor convergence speed.

[0006] The random sine function uses the relationship between the sine function and the unit circle to allow the population to traverse all values ​​on the sine function, that is, to search for all points within the target unit circle, so that the algorithm has better exploration capabilities. However, if the exploration capability is too prominent, the population will be too divergent and difficult to converge. In order to avoid this situation, the adaptive function is introduced to gradually weaken the algorithm's exploration capability as the algorithm iterates, thereby enhancing the convergence of the population.

[0007] Genetic Algorithm (GA) is a randomized search method with universal influence. It is evolved from the evolutionary laws of the biological world. It is a kind of heuristic algorithm. It uses a probabilistic search method to search effectively in a large space. It has good global optimization ability and simple operation process. The basic operations of genetic algorithm include: algorithm initialization, selection operation, crossover operation and mutation operation. Among them, the selection operation refers to the process of selecting superior individuals from the population and eliminating inferior individuals. The purpose of selection is mainly to directly inherit the optimized individuals or solutions to the next generation or to inherit the new individuals generated by other genetic operations to the next generation. The crossover operation and mutation operation are performed with a certain probability. The crossover operation is mainly to replace and recombine the partial structures of the two parent individuals to generate new individuals, so as to improve the fitness of the individuals. The mutation operation is mainly to change the values ​​of certain genes on the coding string of a specific individual in the population, and spontaneously generate random changes on each chromosome to replace one or more genes. The introduction of genetic algorithm into the whale algorithm is mainly to improve the global search ability of the algorithm and the diversity of solutions. Summary of the invention

[0008] The purpose of the present invention is to propose a high-speed data acquisition system mapping method based on an improved whale algorithm to improve the thermal balance level of the system and reduce overall power consumption in view of the shortcomings of the existing high-speed data acquisition system mapping method.

[0009] The technical solution of the present invention to solve the shortcomings of the high-speed data acquisition system mapping method is:

[0010] By starting from the perspective of thermal balance at the algorithm application level, the Access Noxim simulation software is used to conduct an in-depth study of the three-dimensional network on chip (3D NoC) architecture. The research method includes obtaining and normalizing the thermal effect matrix of the NoC as shown below Figure 5 As shown in the figure, a new power consumption model is established based on the existing NoC model. A three-dimensional on-chip network mapping algorithm based on the improved whale algorithm is proposed. The algorithm analyzes the problems of the whale optimization algorithm, such as easy to fall into local optimality and low convergence accuracy, optimizes the local search ability of the whale algorithm (WOA) by using an adaptive random sine function, and combines WOA with the genetic algorithm (GA) through the levy flight function to enhance the global search ability of WOA.

[0011] 1. Make the following improvements to the whale algorithm

[0012] The adaptive random sine formula is used to improve the prey search strategy of the whale algorithm. The adaptive parameter w' and the sine random numbers R1 and R2 are substituted into the differential expression to obtain the random adaptive sine differential expression. Improved random adaptive sine differential expression:

[0013] X(t+1)=X i (t).|sin(R1)|+R2.|sinR1|.|α.X rand -βX i (t)|

[0014] Where X rand To randomly select individuals, R1 is a random number in [0,2π], and R2 is a random number in [0,π]. The expressions of α and β are

[0015]

[0016] w' is an adaptive parameter that changes with the number of iterations. The adaptive function formula is as follows:

[0017]

[0018] w min and w max are the minimum and maximum values ​​of the weight coefficient, l is the curvature coefficient, which is generally a constant, and w' refers to the weight size at the tth iteration. The search step length of the algorithm at different times is regulated by adaptive weights, and the random adaptive sine formula is used for perturbation to enhance the randomness of the search strategy for prey, strengthen the local search ability of the algorithm, and help the group reduce the probability of falling into the local optimum, thereby ensuring the optimization effect of the algorithm.

[0019] The evolution mechanism of the GA algorithm is introduced and combined with the basic strategies of the whale optimization algorithm (encirclement, search, bubble attack), and the mathematical characteristics of the Levy function are used to coordinate the cooperation between WOA and GA to improve the overall performance of the optimization algorithm. Among them, Levy flight is a statistical description of random motion. The Levy flight formula is as follows:

[0020]

[0021] The parameter β is a constant with a range of [0,2], μ is a shift parameter, and v>0 controls the size of the distribution. Both obey the normal distribution. The evolutionary factor G is used to determine whether levy flight is a "long step" and thus make decisions on population behavior.

[0022] 2. Mapping of high-speed data acquisition system based on improved whale algorithm

[0023] By starting from the perspective of thermal balance at the algorithm application level, the Access Noxim simulation software is used to conduct an in-depth study of the three-dimensional network on chip (3D NoC) architecture. The research method includes obtaining and normalizing the thermal effect matrix of the NoC as shown below Figure 5As shown in the figure, a new power consumption model is established based on the existing NoC model. At the same time, a three-dimensional on-chip network mapping algorithm based on the improved whale algorithm is proposed. The algorithm analyzes the problems of the whale optimization algorithm, such as easy to fall into local optimality and low convergence accuracy, optimizes the local search ability of the whale algorithm (WOA) by using an adaptive random sine function, and combines WOA with the genetic algorithm (GA) through the levy flight function to enhance the global search ability of WOA.

[0024] Step 1: Establish a thermal balance model to calculate the power consumption of each node in the 3DNoC network. The formula is as follows:

[0025] In the formula, represents the total consumption at the node xyz(i, j, l). Ehorizontal is the energy consumed by the physical link transmission of adjacent resource nodes per unit bit in the horizontal direction. vertical is the energy consumed by the physical link transmission of a unit bit in the vertical direction between adjacent resource nodes; b ith With b itv They represent the data transmission of the routing node in the horizontal and vertical directions, E switch It is the energy consumed by each bit of data in the switching structure.

[0026] Step 2: Establish the optimization objective function. Multiply the total power consumption of the node by the heat value coefficient of the node, and the formula is as follows:

[0027]

[0028] The coefficient matrix in the above formula is It is obtained by adjusting the scale and normalizing the heat value matrix, which can reflect the thermal effect of 3D NoC and is obtained by Access Noxim simulation. The fitness of the mapping solution is obtained by adding the heat mean value and the heat mean square error in a certain proportion of the proportional coefficient. The calculation formula is as follows:

[0029]

[0030] In the above formula, the NoC network size is n×m×z. The first term is It is used to evaluate the balance of NoC heat. The second item is the average node heat value x avg To evaluate the total heat generation of NoC, the two items are multiplied by the weight parameter α and added to finally obtain the evaluation function.

[0031] Step 3: Use the improved whale algorithm to implement on-chip network mapping. The specific process of the algorithm is as follows:

[0032] Each whale individual represents a feasible solution. Each dimension of the solution corresponds to the acquisition node, storage node, transmission node and control node in the high-speed data acquisition system. The specific solution is in the form of x i =[α1,α2,α3,…,α n ], where α n It means that the task node numbered n in the high-speed data acquisition system is mapped to the node numbered α on the 3Dmesh architecture NoC. The algorithm design process is as follows:

[0033] S1, initialization. Determine the number of iterations as T, set the population size as M, and generate M random sequences x i =[α1,α2,α3,…,α n ], i∈(0,n] and calculate the fitness of each individual;

[0034] S2, update parameters and mark the prey position. Mark the individual with the best fitness as prey, and record the fitness and position X of the individual. *t =[β1,β2,β3,…,β n ]. Update the decision parameters P and A, where P∈[0.1] is a random number in the interval, and the update formula of A is as follows:

[0035] A=2×a×ra

[0036] Where the parameter r is a random value in the range [0,1], and the coefficient a decreases linearly from 2 to 0 according to the number of iterations.

[0037] S3, when the decision parameter P<0.5, |A|<1, the whale population implements the strategy of surrounding prey. i =[α1,α2,α3,…,α n ],i∈(0,n] to the prey position X *t =[β1,β2,β3,…,β n ] close, that is, each dimension α of the current individual i According to the strategy of surrounding the prey, the prey corresponds to the dimension β i The formula is as follows:

[0038]

[0039] Where C is the bracketing step length, which takes a random value in the interval [0,2]. After executing this strategy, jump to S7.

[0040] S4, when the decision parameter P < 0.5, |A| ≥ 1, the whale population implements the prey search strategy. Randomly select any individual in the population as the search target, and the randomly selected individual is marked as X rand =[θ1,θ2,θ3,…,θn , each whale individual moves towards X rand Update the position according to the adaptive random differential formula to explore the solution space and find unvisited regions. The formula is as follows:

[0041] X(t + 1) = X i (t).|sin(R1)| + R2.|sinR1|.|α.X rand - β.X i (t)|

[0042] where R1 is a random number in [0, 2π], R2 is a random number in [0, π], and α and β are adaptive parameters, which are multiplied by X rand and X i (t) respectively. After executing this strategy, jump to S7.

[0043] S5, when the decision parameter P ≥ 0.5, the whale population executes the bubble net attack strategy. Each whale individual x i = [α1, α2, α3, …, α n , i ∈ (0, n] updates its position by randomly selecting one of the encircling or spiral shrinking strategies with a 50% probability. The formulas for shrinking or encircling the prey are as follows:

[0044]

[0045] where e is the base of the natural logarithm, the spiral equation constant is taken as 1, and l is a random number in the [1, 1] interval. After executing this strategy, jump to S7.

[0046] S6, when the random number generated by the levy formula is greater than the evolution factor G, the whale population executes the evolution strategy. According to the process of the genetic algorithm, selection, crossover, mutation, and repair are performed in sequence according to the fitness of the whale individuals, and screening is carried out under the condition of ensuring that the population size remains unchanged before and after iteration for global search. After executing this strategy, jump to S7.

[0047] S7, repair the individuals and calculate the individual fitness. According to the principle that each dimension cannot be the same and is a real number not exceeding the number of NoC network nodes. Take all elements in the individual x i = [α1, α2, α3, …, α n , i ∈ (0, n] as integers, and then replace the repeated numbers with the unoccupied NoC node numbers. After the repair is completed, calculate the fitness of each individual, and mark the new prey position X *t+1 . Increment the iteration count times by one. If times < T, then jump to S2 to continue execution; otherwise, output the currently recorded optimal individual X *t = [β1, β2, β3, …, β n] and the optimal score Fitness(X *t ).

[0048] The present invention analyzes the physical characteristics of the 3Dmesh architecture NoC through the Access Noxim simulation software, and reconstructs the objective function with thermal balance as the goal in combination with the traditional model. Aiming at the problem that the whale optimization algorithm is prone to fall into the local optimum and has low convergence accuracy, an improved evolutionary whale optimization algorithm is proposed. The new algorithm improves the search strategy of the whale algorithm by introducing a random adaptive sine function, and enhances the local search capability of the whale optimization algorithm. At the same time, the genetic algorithm is integrated with the whale optimization algorithm by combining the levy function. The encoding and crossover mechanism in the GA provides diversity for the whale population, which helps to jump out of the local optimal solution. The application of the levy function effectively balances the global and local search capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 This is a system block diagram of a high-speed data acquisition system mapping method based on an improved whale algorithm of the present invention.

[0050] Figure 2 This is a diagram of the NoC network and thermal effect of a three-dimensional mesh architecture.

[0051] Figure 3 This is a schematic diagram of the on-chip network mapping of the present invention

[0052] Figure 4 It is a task graph model of on-chip network high-speed data acquisition system

[0053] Figure 5 This is the heat coefficient matrix processing flow chart of the present invention

[0054] Figure 6 This is the flowchart of the improved whale algorithm of the present invention

[0055] Figure 7 This is a comparison chart of the iteration trend of the improved whale algorithm of the present invention, the whale algorithm and the genetic algorithm DETAILED DESCRIPTION

[0056] The high-speed data acquisition system mapping algorithm based on the improved whale algorithm is mainly used to solve the mapping problem of high-speed data acquisition systems in three-dimensional on-chip networks. The algorithm analyzes the problems of the whale optimization algorithm, such as easy to fall into local optimality and low convergence accuracy, and uses an adaptive random sine function to optimize the local search ability of the whale algorithm (WOA). The WOA is combined with the genetic algorithm (GA) through the levy flight function to enhance the global search ability of the WOA. The specific flow chart of the algorithm of the present invention is as follows Figure 6 shown.

[0057] Step 1: Establish a thermal balance model to calculate the power consumption of each node in the 3DNoC network. The formula is as follows:

[0058] In the formula, represents the total consumption at the node xyz(i, j, l). Ehorizontal is the energy consumed by the physical link transmission of adjacent resource nodes per unit bit in the horizontal direction, E vertical is the energy consumed by the physical link transmission of a unit bit in the vertical direction between adjacent resource nodes; b ith With b itv They represent the data transmission of the routing node in the horizontal and vertical directions, E switch It is the energy consumed by each bit of data in the switching structure.

[0059] Step 2: Establish the optimization objective function. Multiply the total power consumption of the node by the heat value coefficient of the node, and the formula is as follows:

[0060] The coefficient matrix in the above formula is It is obtained by adjusting the scale and normalizing the heat value matrix, which can reflect the thermal effect of 3D NoC and is obtained by Access Noxim simulation. The fitness of the mapping solution is obtained by adding the heat mean value and the heat mean square error in a certain proportion of the proportional coefficient. The calculation formula is as follows:

[0061]

[0062] In the above formula, the NoC network size is n×m×z. The first term is It is used to evaluate the balance of NoC heat. The second item is the average node heat value x avg To evaluate the total heat generation of NoC, the two items are multiplied by the weight parameter α and added to finally obtain the evaluation function.

[0063] Step 3: Use the improved whale algorithm to implement on-chip network mapping. The specific process of the algorithm is as follows:

[0064] Each whale individual represents a feasible solution. Each dimension of the solution corresponds to the acquisition node, storage node, transmission node and control node in the high-speed data acquisition system. The specific solution is in the form of x i =[α1,α2,α3,…,α n ], where α n It means that the task node numbered n in the high-speed data acquisition system is mapped to the node numbered α on the 3Dmesh architecture NoC. The algorithm design process is as follows:

[0065] S1, initialization. Determine the number of iterations as T, set the population size as M, and generate M random sequences x i=[α1,α2,α3,…,α n ], i∈(0,n] and calculate the fitness of each individual;

[0066] S2, update parameters and mark the prey position. Mark the individual with the best fitness as prey, and record the fitness and position X of the individual. *t =[β1,β2,β3,…,β n ]. Update the decision parameters P and A, where P∈[0.1] is a random number in the interval, and the update formula of A is as follows:

[0067] A=2×a×ra

[0068] Where the parameter r is a random value in the range [0,1], and the coefficient a decreases linearly from 2 to 0 according to the number of iterations.

[0069] S3, when the decision parameter P<0.5, |A|<1, the whale population implements the strategy of surrounding prey. i =[α1,α2,α3,…,α n ],i∈(0,n] to the prey position X *t =[β1,β2,β3,…,β n ] close, that is, each dimension α of the current individual i According to the strategy of surrounding the prey, the prey corresponds to the dimension β i The formula is as follows:

[0070]

[0071] Where C is the bracketing step length, which takes a random value in the interval [0,2]. After executing this strategy, jump to S7.

[0072] S4, when the decision parameter P < 0.5, |A| ≥ 1, the whale population implements the prey search strategy. Randomly select any individual in the population as the search target, and the randomly selected individual is marked as X rand =[θ1,θ2,θ3,…,θ n ], each whale moves towards X rand Update the position according to the adaptive random difference formula to explore the solution space to find unvisited areas. The formula is as follows:

[0073] X(t+1)=X i (t).|sin(R1)|+R2.|sinR1|.|α.X rand -β.X i (t)|

[0074] Where R1 is a random number in [0,2π], R2 is a random number in [0,π], α and β are adaptive parameters, respectively related to Xrand and X i Multiply by (t). After executing this strategy, jump to S7.

[0075] In S5, when the decision parameter P ≥ 0.5, the whale population executes the bubble net attack strategy. Each whale individual x i = [α1, α2, α3, …, α n , i ∈ (0, n] updates its position by randomly selecting one of the encircling or spiral shrinking strategies with a 50% probability. The formulas for shrinking or encircling the prey are as follows:

[0076]

[0077] where e is the base of the natural logarithm, the spiral equation constant is taken as 1, and l is a random number in the [1, 1] - interval. After executing this strategy, jump to S7.

[0078] In S6, when the random number generated by the levy formula is greater than the evolution factor G, the whale population executes the evolution strategy. According to the process of the genetic algorithm, selection, crossover, mutation, and repair are carried out in sequence according to the fitness of the whale individuals, and screening is performed under the condition of ensuring that the population size remains unchanged before and after iteration for global search. After executing this strategy, jump to S7.

[0079] In S7, the individuals are repaired and the individual fitness is calculated. According to the principle that each dimension cannot be the same and is a real number not exceeding the number of NoC network nodes. The elements in the individual x i = [α1, α2, α3, …, α n , i ∈ (0, n] are all taken as integers, and then the repeated numbers are replaced with the unoccupied NoC node numbers. After the repair is completed, the fitness of each individual is calculated, and the new prey position X *t+1 is marked according to the fitness. The iteration count times is incremented by one. If times < T, then jump to S2 to continue execution; otherwise, output the currently recorded optimal individual X *t = [β1, β2, β3, …, β n and the optimal score Fitness(X *t ).

[0080] From Figure 7From the convergence trend comparison diagram, it can be observed that the local search ability in the early stage of iteration is significantly improved. Even if the improved whale algorithm falls into the local optimal solution in the early stage of iteration, it can still jump out of the local optimal solution through genetic evolution operation. This feature significantly improves the stagnation phenomenon of the algorithm. And because the exploration ability of the whale algorithm is improved by using the random adaptive sine formula, the evaluation value of the improved whale algorithm is better than that of the whale algorithm under the same number of iterations, which significantly improves the quality of data acquisition system mapping. Therefore, the improved whale algorithm can solve the mapping problem of high-speed data acquisition system to a certain extent.

Claims

1. A high-speed data acquisition system mapping method based on an improved whale algorithm, characterized in that the steps include: Step 1: Establish a thermal balance model to calculate the power consumption of each node in the 3DNoC network. The formula is as follows: In the formula, represents the total consumption at the node xyz(i, j, l), E horizontal is the energy consumed by the physical link transmission of a unit bit in the horizontal direction between adjacent resource nodes, E vertical is the energy consumed by the physical link transmission of a unit bit in the vertical direction between adjacent resource nodes; b ith With b itv They represent the data transmission of the routing node in the horizontal and vertical directions, E switch It is the energy consumed by a unit bit of data in the switching structure; Step 2: Multiply the total power consumption of the node by the heat value coefficient of the node to establish the optimization objective function. The formula is as follows: The coefficient matrix in the above formula is It is obtained by adjusting the scale and normalizing the heat value matrix. The heat value matrix can reflect the thermal effect of 3DNoC and is obtained by Access Noxim simulation. The fitness of the mapping solution is obtained by adding the heat mean value and the heat mean square error in a certain proportion of the proportional coefficient. The calculation formula is as follows: In the above formula, the NoC network size is n×m×z, and the first term It is used to evaluate the balance of NoC heat. The second item, the average node heat value xavg, is used to evaluate the total heat generation of NoC. The two items are multiplied by the weight parameter α and added to finally obtain the evaluation function. Step 3: Use the improved whale algorithm to implement on-chip network mapping. The specific process of the algorithm is as follows: In the improved whale algorithm, each whale individual represents a feasible solution. Each dimension of the solution corresponds to the acquisition node, storage node, transmission node and control node in the high-speed data acquisition system. The specific solution is in the form of x i =[α1,α2,α3,…,α n ], where α n It indicates that the task node numbered n in the high-speed data acquisition system is mapped to the node numbered α on the 3Dmesh architecture NoC; The improved whale algorithm design process is as follows: S1, determine the number of iterations to be T, set the population size to be M, and generate M random sequences x i =[α1,α2,α3,…,α n ] and calculate the fitness of each individual; S2, mark the individual with the best fitness as prey, and record the fitness and position X of the individual *t =[β1,β2,β3,…,β n ], update the decision parameters P, A, where P∈[0.1] is a random number in the interval, and the update formula of A is as follows: A=2×a×ra Where the parameter r is a random value in the range [0,1], and the coefficient a decreases linearly from 2 to 0 according to the number of iterations; S3, when the decision parameter P<0.5 and |A|<1, the whale population implements the strategy of surrounding prey. Each whale individual x i =[α1,α2,α3,…,α n ],i∈(0,n] to the prey position X *t =[β1,β2,β3,…,β n ] close, that is, each dimension α of the current individual i According to the strategy of surrounding the prey, the prey corresponds to the dimension β i Closer, the formula is as follows: Where C is the bracketing step length, which randomly takes values ​​in the interval [0,2]. After executing this strategy, jump to S7; S4, when the decision parameter P < 0.5 and |A| ≥ 1, the whale population executes the prey search strategy and randomly selects any individual in the population as the search target. The randomly selected individual is marked as X. rand =[θ1,θ2,θ3,…,θ n ], each whale moves towards X rand Update the position according to the adaptive random difference formula to explore the solution space to find unvisited areas. The formula is as follows: X(t+1)=Xi(t).|sin(R1)|+R2.|sinR1|.|α.X rand -β.X i (t)| Where R1 is a random number in [0,2π], R2 is a random number in [0,π], α and β are adaptive parameters, respectively related to X rand and X i (t) multiply, and jump to S7 after executing this strategy; S5, when the decision parameter P ≥ 0.5, the whale population implements the bubble net attack strategy, and each whale individual x i =[α1,α2,α3,…,α n ], i∈(0,n] updates its position with a probability of 50% by choosing one of the strategies of surrounding or spiral contraction. The formula for contraction or surrounding prey is as follows: Where e is the base of the natural number logarithm, is the spiral equation constant and takes 1, l is a random number in the [1,1]-interval, and after executing this strategy, jump to S7; S6, when the random number generated by the levy formula is greater than the evolution factor G, the whale population executes the evolution strategy. According to the process of the genetic algorithm, selection, crossover, mutation, and repair are carried out in sequence according to the fitness of the individual whales. Under the condition that the population size before and after the iteration remains unchanged, screening is carried out to perform a global search. After executing this strategy, jump to S7; S7, according to the principle that each dimension cannot be the same and all are real numbers not exceeding the number of NoC network nodes, the individual x i =[α1, α2, α3, …, α n , where all elements in i∈(0, n] are taken as integers, and then the repeated numbers are replaced with the numbers of unoccupied NoC nodes. After the repair is completed, the fitness of each individual is calculated, and the new prey position X *t+1 is marked according to the fitness. The iteration times is incremented by one. If times < T, then jump to S2 to continue execution, otherwise output the currently recorded optimal individual X *t =[β1, β2, β3, …, β n and the optimal score Fitness(X *t ).

2. According to claim 1, a high-speed data acquisition system mapping method based on an improved whale algorithm is characterized in that: 2 is a detailed description of the S4 adaptive random difference formula mentioned in step 3: The adaptive random sine formula is obtained by substituting the adaptive parameter w' and the sine random numbers R1 and R2 into the differential expression, and the improved random adaptive sine differential expression is as follows: X(t+1)=X i (t).|sin(R1)|+R2.|sinR1|.|α.X rand -βX i (t)| Where X rand To randomly select an individual, R1 is a random number in [0,2π], R2 is a random number in [0,π], and the expressions of α and β are: w' is an adaptive parameter that changes with the number of iterations. The adaptive function formula is as follows: Where w min and w max are the minimum and maximum values ​​of the weight coefficient respectively, l is the curvature coefficient which is generally a constant, w' refers to the weight size at the tth iteration, the adaptive weight control algorithm controls the search step size at different times, and uses the random adaptive sine formula for perturbation to enhance the randomness of the prey search strategy.

3. According to claim 1, a high-speed data acquisition system mapping method based on an improved whale algorithm is characterized in that: 3 is a detailed description of the S6 evolution strategy mentioned in step 3: The evolutionary strategy mechanism includes: levy decision, selection operation, crossover operation and mutation operation; The Levy decision is to coordinate the cooperation between the whale algorithm and the genetic algorithm by using the mathematical characteristics of the Levy function. The Levy flight formula is as follows: Where β is a constant with a range of [0,2], μ is a shift parameter, and v>0 controls the size of the distribution. Both obey the normal distribution. The evolution factor G is used to determine whether levy flight is a "long step" and thus make decisions on population behavior; The selection operation refers to the process of selecting superior individuals from a group and eliminating inferior individuals; The crossover operation is an operation to generate new individuals by replacing and recombining part of the structures of two parent individuals; The mutation operation mainly involves changing certain gene values ​​on the coding string of a specific individual in the population, spontaneously generating random changes and replacing one or more genes on each chromosome.

Citation Information

Cited By

  • In-field transfer robot communication method and device based on mesh communication

    CN120935538A

  • Communication method and device for in-field carrying robot based on mesh communication

    CN120935538B