GPU-AcceleratedParallelCoevolutionaryAlgorithmforParametersIdentiﬁcationandTemperatureMonitoringinPermanentMagnetSynchronousMachines资源-CSDN文库

200 浏览量 2021-02-10 00:10:36 上传评论收藏 1.58MB PDF 举报

GPU-Accelerated Parallel Coevolutionary Algorithm for Parameters Identiﬁcation and Temperature Monitoring in Permanent Magnet Synchronous Machines 本文主要探讨了应用于永磁同步电机（Permanent Magnet Synchronous Machines, PMSM）参数识别和温度监控的GPU加速并行协同进化算法。为了深入理解文章所涉及的要点，我们将从以下几个方面展开详细讨论： 1. GPU加速技术与CUDA编程模型文章中提到的GPU加速技术，全称为图形处理器加速技术，主要用于并行计算以提高计算效率。文章中所使用的CUDA（Compute Unified Device Architecture）是NVIDIA推出的一个针对GPU计算的并行计算平台和编程模型，它允许开发者使用C、C++以及其他语言进行GPU并行编程。 2. 永磁同步电机（PMSM） PMSM电机因其高功率密度、响应速度快、效率高等特点被广泛应用在伺服控制、电动汽车、可再生能源发电等领域。为了支持系统的建模、优化控制、状态监控以及稳定性分析，需要精确估计电机参数。 3. 参数识别与温度监控对于PMSM电机来说，参数识别的准确性至关重要，因为这直接关系到电机模型的准确性和控制策略的有效性。同时，温度监控对于预防电机过热、保证电机可靠运行、延长电机寿命同样具有重要意义。传统上，这些任务可能需要长时间的计算和大量资源投入，而本文提出的算法可以有效地提高这一过程的实时性和准确性。 4. 免疫粒子群优化算法（PSO）粒子群优化（Particle Swarm Optimization, PSO）是一种启发式优化算法，通过模拟鸟群捕食行为而发展出来。在此算法中，每个粒子代表解空间中的一个潜在解。粒子通过跟踪个体经验极值和群体历史经验极值来更新自己的位置和速度。在此基础上，文章提出了一种基于免疫机制的PSO算法，通过模拟生物免疫系统中的抗体和抗原反应来改进传统PSO，以提高搜索效率和避免早熟收敛。 5. 协同进化算法协同进化算法是一种模拟生物进化中种群间共同进化过程的算法。在这种算法中，存在至少两个种群，这些种群在进化过程中相互影响，共同进化以求解问题。本文中的算法采用两层结构，将不同层级的抗体（代表解）进行协同进化，这有助于算法探索解空间的多个区域，提高优化质量和多样性。 6. 免疫克隆选择操作符与免疫疫苗增强操作符文中提到了免疫克隆选择操作符，这是一种模仿生物免疫系统中抗体克隆和选择过程的机制，它可以用来改进抗体的多样性并保持种群的遗传多样性。此外，还提出了免疫疫苗增强操作符，这个操作符的目的是引导粒子向未探索的区域移动，从而扩展搜索范围。 7. GPU加速并行协同进化免疫粒子群优化算法（G-PCIPSO）结合上述元素，文章提出了一种新的G-PCIPSO算法，这种算法通过利用GPU的强大计算能力来显著加快搜索过程，并通过并行化实现在多个参数识别和温度监控任务上的应用，最终实现在线实时监控，提升计算效率。文章通过对比传统基于CPU的串行执行方式，阐述了GPU加速并行计算技术在提高计算效率方面的巨大优势。文章提到的算法在多个参数识别和实时温度监控方面都能有效跟踪参数变化并进行监控，显著提升了系统的反应速度和监控的准确性。对于工业信息化和智能化领域的学者和工程师而言，这一研究成果极具参考价值，有助于推动智能电机系统的发展。

资源推荐

资源详情

资源评论

1220 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 11, NO. 5, OCTOBER 2015

GPU-Accelerated Parallel Coevolutionary Algorithm

for Parameters Identiﬁcation and Temperature

Monitoring in Permanent Magnet

Synchronous Machines

Zhao-Hua Liu, Xiao-Hua Li, Liang-Hong Wu, Shao-Wu Zhou, and Kan Liu

Abstract—A hierarchical fast parallel co-evolutionary immune

particle swarm optimization (PSO) algorithm, accelerated by

graphics processing unit (GPU) technique (G-PCIPSO), is pro-

posed for multiparameter identiﬁcation and temperature moni-

toring of permanent magnet synchronous machines (PMSM). It is

composed of two levels and is developed based on compute uniﬁed

device architecture (CUDA). In G-PCIPSO, the antibodies (Abs) of

higher level memory are selected from the lower level swarms and

improved by immune clonal-selection operator. The search infor-

mation exchanges between swarms using the memory-based shar-

ing mechanism. Moreover, an immune vaccine-enhanced operator

is proposed to lead the Pbests particles to unexplored areas.

Optimized parallel implementations of G-PCIPSO algorithm is

developed on GPU using CUDA, which signiﬁcantly speeds up

the search process. Finally, the proposed algorithm is applied to

multiple parameters identiﬁcation and temperature monitoring of

PMSM. It can track parameter variation and achieve tempera-

ture monitoring online effectively. Compared with a CPU-based

serial execution, the computational efﬁciency is greatly enhanced

by GPU-accelerated parallel computing technique.

Index Terms—Artiﬁcial immune system (AIS), condition mon-

itoring, graphics processing unit (GPU), parallel computing,

parameter identiﬁcation, particle swarm optimization (PSO), per-

manent magnet synchronous machines (PMSMs), temperature

monitoring.

I. INTRODUCTION

N RECENT years, permanent magnet synchronous motor

(PMSM) has been widely applied in servo control, electric

vehicles, and renewable energy power generation [1]–[5] for

its high power density, torque response, and efﬁciency. Precise

estimation of motor parameters is required to support system

modeling, optimization control, state monitoring, and stability

Manuscript received July 03, 2014; revised December 09, 2014 and February

04, 2015; accepted March 24, 2015. Date of publication April 16, 2015; date of

current version October 02, 2015. This work was supported in part by the China

Postdoctoral Science Foundation funded project under Grant 2013M540628

and Grant 2014T70767, in part by the National Natural Science Foundation of

China under Grant 61174140 and Grant 51374107, and in part by the Hunan

Provincial Natural Science Foundation of China under Grant 13JJ8014 and

Grant 14JJ3107. Paper no. TII-14-0700.

Z.-H. Liu, X.-H. Li, L.-H. Wu, and S.-W. Zhou are with the School of

Information and Electrical Engineering, Hunan University of Science and

Technology, Xiangtan 411201, China (e-mail: zhaohualiu2009@hotmail.com).

K. Liu is with the Department of Electronic and Electrical Engineering,

University of Shefﬁeld, Shefﬁeld S10 2TN, U.K.

Color versions of one or more of the ﬁgures in this paper are available online

at http://ieeexplore.ieee.org.

Digital Object Identiﬁer 10.1109/TII.2015.2424073

analysis in industrial application [6]. For instance, the temper-

ature can be monitored according to the variation of windings

resistance in the motor. However, PMSM is typically a non-

linear system, whose physical parameters are easily subjected

to changes with variation in operating conditions. Therefore,

technologies for parameter estimation of PMSM, especially

simultaneous estimation of multiparameters, such as winding

resistance, dq-axis inductances, and permanent magnet ﬂux,

become a challenging work in machine control [7].

Existing research mainly focuses on online estimation algo-

rithms, such as extended Kalman ﬁlter (EKF) [8]–[11], model

reference adaptive system (MRAS) [12], [13], recursive least-

square (RLS) methods [14]–[16], adaptive observer [17], [18],

and artiﬁcial neural network (ANN) [19], [20]. Each algorithm

has some inherent drawbacks of slow convergence, unsteadi-

ness, complex computing, system self-disturbances, and moni-

toring data growth. In [8], an EKF is proposed to estimate the

rotor s peed and position of PMSM. It suffers from unidenti-

ﬁed PMSM parameters because of noise sensitivity. Similarly,

an EKF for the estimation of the winding resistance and rotor

ﬂux linkage is proposed in [11]. The results indicate that this

method also suffers from noise and instability in engineering.

The MRAS estimators proposed in [12] and [13] cannot simul-

taneously estimate winding resistance, inductance, and rotor

ﬂux linkage accurately. The nominal values of other parameters

speciﬁed in the motor manual are required to estimate indi-

vidual parameters. Since the nominal value is not equal to the

actual operating value, and the working condition changes fre-

quently, those methods cannot ensure the convergence to actual

parameter value. The RLS estimator was discussed in [14]. It

also suffers from the noise characteristics and could not guar-

antee the convergence to actual values during experiments. In

order to obtain the observation values, an adaptive intercon-

nected observer for PMSM parameter estimation is discussed

in [18]. The algorithm may also be unfeasible when the param-

eters of PMSM vary in large ranges (such as empty running and

full-loaded running). Reference [19] presents an ANN to esti-

mate parameters of PMSM. Computation is complicated due to

the training of network weight. It is likely to get stuck in local

minima because of too much over-ﬁtness. As is pointed in [21],

the proposed method should be capable of identifying parame-

ters dynamically and adaptively in time-varying and nonlinear

systems.

See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

LIU et al.: GPU-ACCELERATED PARALLEL COEVOLUTIONARY ALGORITHM 1221

Inspired by biological computing, evolutionary computation

techniques are proposed to solve this problem. Genetic algo-

rithm (GA) and particle swarm optimization (PSO) algorithm

have attracted much attention in the investigation of PMSM

parameter identiﬁcation. It is because the evolutionary algo-

rithms have the ability to obtain a suitable set of parameter

values via optimizing objective function between the system

ideal model and the actual models. GA was proposed to identify

multiparameter of PMSM including stator winding resistance,

dq-axis inductances, and rotor ﬂux linkage. However, the accu-

racy and stability of these parameters estimation need to be

improved, since GA tends to lose diversity and easily get

trapped in local optima [22], [23] during the later evolution

process when solving complex problems. Some researchers

propose an intelligent system parameter identiﬁcation approach

by utilizing the PSO algorithm [24], [25]. Since PSO is easy to

implement on the computer and superior in convergence speed,

the intelligent estimator is effective in estimating the stator

resistance and disturbed load torque. However, it cannot exactly

estimate multiple parameters simultaneously (winding resis-

tance, dq-axis inductances, and permanent magnet ﬂux identiﬁ-

cation). The reason is that the basic PSO can be easily trapped

in local minima when solving complex nonlinear problem.

Hence, several algorithms are developed to improve the solu-

tion performances of PSO. A hybrid PSO with mutation was

proposed by Ahmed et al. in [26]. Juang et al. proposed another

hybrid PSO algorithm associated with the genetic operator in

[27]. The diversity of PSO is signiﬁcantly enhanced using the

GA operators. Compared with the basic PSO, a hybrid PSO

combined with wavelet dynamic mutating space was proposed

in [28], which can obtain dynamic optimization effect. Liang

et al. proposed a comprehensive learning PSO (CLPSO) in [29].

All the ﬂying directions of individuals were updated by ran-

domly selected particles during the iteration process. Although

the method in [29] was superior in keeping diversity, it did

not design a scheme to help the particles jump out of the

local optima when the whole population was homogeneous dur-

ing the later evolution process. The same defect also exists

in another improved CLPSO using different pbests [30]. A

parameter-adaptive regulation scheme and an elitist learning

strategy were introduced into the PSO by Zhan et al. [31], who

presented an adaptive PSO (APSO) that can accelerate the con-

vergence speed and jump out of the local optima. Nevertheless,

it still needs expert judgment on the evolutionary state, which

may be difﬁcult to determine. As the spatial distribution of

multiple populations was broad and outperformed a single

population in terms of diversity, other researchers proposed a

multipopulation scheme to improve the diversity of basic PSO

[32]–[35]. For instance, Bergh et al. [32] proposed a multiple

PSO algorithm, in which the whole population was split into

many small swarms. It showed a better performance than single

PSO. However, this method is redundant in iteration comput-

ing because there is no information interaction between these

isolated swarms. A coevolutionary PSO was discussed in [33],

where a truncated Gaussian distribution function is utilized

to accelerate the convergence speed. Reference [34] proposed

a dynamic multiswarm PSO embedded with an adaptively

adjusted scheme. In order to solve multiple-objective prob-

lems, Zhan et al. [35] proposed a multiple populations PSO,

which used an external shared archive and another two novel

operator, including velocity modiﬁcation and elitist learning

strategy to enhance the performance. However, coevolutionary

PSO always has mass data to analyze during machine operation.

The proposed methods, executed in conventional CPU, may

take considerably l ong time. In fact, it does not take full advan-

tage of the inherent parallelism of population-based intelligent

computing techniques. This disadvantage severely restricts the

practical applications of PSO in parameters estimation and con-

dition monitoring of PMSM. Therefore, the GPU-accelerated

massively parallel computing technique becomes a major com-

plementary way to speedup those population-based intelligent

algorithms.

Motivated by coevolution theory [36], artiﬁcial immune sys-

tem (AIS) [37], [38], and GPU parallel computing technique

[39], a novel parallel coevolutionary immune PSO algorithm,

accelerated by graphics processing unit (GPU) technique (G-

PCIPSO), is proposed for multiparameter identiﬁcation and

temperature monitoring of PMSM in this paper. The framework

of G-PCIPSO consists of one high-level memory population

and several bottom normal swarms based on parallel comput-

ing model of GPUs. In G-PCIPSO, the global best individuals

of normal swarms are regarded as antibody (Ab) in immune sys-

tem and memorized in the high-level memory, which serves as a

leader set. The memory is updated using the improved immune

clonal-selection operator. Then an immune vaccine-enhanced

operator is employed to accelerate the convergence speed

of Pbests. The information sharing mechanism is employed

for the information exchange between memory and different

swarms. Moreover, the proposed method is implemented on the

GPU using the compute uniﬁed device architecture (CUDA)

(NVIDIA Corporation). Finally, the proposed method is applied

to PMSM multiple parameter identiﬁcation and temperature

monitoring. It shows that the method’s performance is much

better than the existing improved hybrid PSOs in simultane-

ously estimating multiple electric parameters of the machine

system. In addition, the proposed method can not only track

the varied physical parameter but also realize temperature

monitoring online effectively. The experiments show that the

proposed approach demonstrates high efﬁciency compared with

a CPU-based serial execution.

Our main contributions can be summarized as follows.

1) The proposed hierarchical-based multiple population

cooperative scheme can improve the population search

performance since the immunity-based high-level mem-

ory can reserve valuable historical knowledge, and the

excellent solution information can spread between differ-

ent subswarms through the designed information sharing

updated scheme of particles in bottom level.

2) Immune concepts are introduced into PSO to overcome

the blindness in action of gBest particles (antibodies in

memory) and Pbest particles stochastic evolution in solv-

ing parameters identiﬁcation and temperature monitoring

of PMSM. Immune clonal-selection operating with an

adaptive wavelet mutation for gBest and an immune vac-

cine operator for Pbest are introduced. The proposed

parameter estimator is capable of identifying parame-

ters dynamically and adaptively for time-varying and

nonlinear PMSM system.

剩余10页未读，继续阅读

评论收藏

内容反馈

weixin_38644780

粉丝: 2
资源: 886

GPU-Accelerated Parallel Coevolutionary Algorithm for Parameters...

最新资源

GPU-Accelerated Parallel Coevolutionary Algorithm for Parameters...

Hands-On-GPU-Accelerated-Computer-Vision-with-OpenCV-and-CUDA-master.zip

GPU-Accelerated Vision for Robots with OpenCV and CUDA.pdf

藏经阁-LEVERAGING GPU-ACCELERATED ANA.pdf

Hands-On-GPU-Accelerated-Computer-Vision-with-OpenCV-and-CUDA:Packt发布的具有OpenCV和CUDA的动手GPU加速计算机视觉

藏经阁-leveraging gpu-accelerated analytics on top of apache spark.

Extending the graphic pipeline with new GPU-accelerated primitives

Spectre Circuit Simulator and Accelerated Parallel Simulator

GPU-accelerated 3D Stationary Wavelet-based Frame:3D Stationary wavelet-based transform using GPU-accelerated separable convolution-matlab开发

利用GPU加速的SIFT算法，实现影像匹配，并利用RANSAC进行粗差的剔除

Flow-through cell-based In Vitro Release Method for Triamcinolone acetonide

CoffeeScript - Accelerated JavaScript Development

GPU.Programming.in.MATLAB.0128051329

人工智能和机器学习库SwiftAI.zip

GMW 14458-2020 Copper-Accelerated Acetic Acid Salt Spray (CASS)

PyPI 官网下载 | pytorch-accelerated-0.1.2.tar.gz

ZynqNet: An FPGA-Accelerated Embedded Convolutional Neural Network

Fast compression of computer-generated holographic images based on a GPU-accelerated skip-dimension vector quantization method

最新版本的NVIDIA cuDNN FOR WIN 10

基于CORDIC的反正弦和反余弦计算的FPGA实现

BA无标度网络中的SIR模型

使用3DCNN和卷积LSTM进行手势识别学习时空特征

基于三次贝塞尔曲线的类汽车曲率连续路径平滑

基于机器学习的设备剩余寿命预测方法综述

基于维纳过程的退化模型，具有递归过滤算法，可用于估计剩余使用寿命

基于FPGA的奇异值和特征值分解的快速实现。

磁悬浮系统自适应模糊PID控制器的设计

最新资源