GPU加速的并行协同进化算法：PMSM参数识别与温度监测
194 浏览量
更新于2024-08-27
收藏 1.58MB PDF 举报
本文献标题《GPU-Accelerated Parallel Coevolutionary Algorithm for Parameters Identification and Temperature Monitoring in Permanent Magnet Synchronous Machines》聚焦于永磁同步电机(PMSM)的参数识别与温度监控技术。作者们提出了一种层次化的、利用图形处理器(GPU)技术加速的并行协同进化免疫粒子群优化算法（G-PCIPSO），该算法在工业信息技术领域（IEEE Transactions on Industrial Informatics）发表，卷11，第5期，2015年10月。
G-PCIPSO算法的设计灵感来源于生物免疫系统，它包含两个级别的进化过程：上层抗体（Abs）由下层种群的粒子通过免疫克隆选择操作进行优化，增强了搜索效率。这种算法采用内存共享机制，允许不同层次的粒子群体之间信息交流，促进全局搜索。此外，文中提出了免疫疫苗增强操作，引导最优解（Pbest）粒子探索未被充分利用的区域，从而提高参数估计的精度和收敛速度。
算法的核心优势在于利用GPU的并行计算能力，通过Compute Unified Device Architecture (CUDA)平台实现了算法的优化实现。CUDA是一种专为GPU设计的并行计算模型，使得算法能够高效地处理大量数据和复杂的数学运算，显著提升了算法的执行速度和性能。对于永磁同步电机来说，这不仅有助于减少计算时间，还能实时准确地监测电机内部温度，对电机健康管理和故障预防具有重要意义。
这篇研究论文探讨了如何利用GPU技术提升参数识别算法的性能，并将其应用于实际的电机系统中，以实现更精确的参数估计和有效的温度监控，这对于工业自动化和电机维护具有重要的理论价值和实践应用潜力。
1222 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 11, NO. 5, OCTOBER 2015
3) High-performance computing ability of GPU is fully
utilized to speedup the search process of particles and
Abs. Fast and exact parameter estimation and tempera-
ture monitoring for PMSM under changing conditions is
realized.
This paper is organized as follows. In Section II, a brief intro-
duction of PMSM model is provided and challenges of design-
ing parameter estimators for PMSM is analyzed. G-PCIPSO
algorithm for parameters estimation and temperature moni-
toring of PMSM is proposed in Section III, where algorithm
principle, mathematical model, and implementation procedure
are addressed in detail. Section IV presents parameter estima-
tion mechanism modeling based on the proposed G-PCIPSO.
Experiment results are given in Section V. Finally, conclusion
and future work are summarized in Section VI.
II. PMSM M
ODEL
The mathematical model of the PMSM in dq-axis frame is
given as
U
d
U
q
=
⎡
⎢
⎣
R + L
d
d
dt
−ωL
q
ωL
d
R + L
q
d
dt
⎤
⎥
⎦
i
d
i
q
+
0
ωψ
(1)
where
d
dt
is a differential operator, ω is the electrical angu-
lar velocity and R, L
d
,L
q
, and ψ are the motor winding
resistance, d-axis and q-axis inductances, and magnet flux,
respectively. U
d
,U
q
,i
d
, and i
q
are dq-axis stator voltage and
current. The motion equation can be expressed in discrete form
as follows when the machine is on the steady-state model [4]
U
d(k)
U
q(k)
=
R −ω(k)L
q
ω(K).L
d
(K) R
i
d(k)
i
q(k)
+
0
ω(k).ψ(k)
(2)
where R, L
d
,L
q
, and ψ are unknown parameters. From (2),
it is clear that there are four parameters to be identified, while
both rank numbers are two and coupled with each other [7],
[8]. Generally, i
d
is set to be zero for decoupling the flux and
torque control and a very short time of i
d
< 0 should be injected
to obtain a full-rank reference model inspired by [7] and [22],
which are shown as follows:
⎡
⎢
⎢
⎣
U
d0(k
0
)
U
q0(k
0
)
U
d(k
1
)
U
q(k1)
⎤
⎥
⎥
⎦
=
⎡
⎢
⎢
⎣
R −ω(k0)L
q
00
ω(k0).L
d
(k0) R 00
00R −ω(k1)L
q
00ω(k1).L
d
(k1) R
⎤
⎥
⎥
⎦
×
⎡
⎢
⎢
⎣
i
d0(k0)
i
q0(k0)
i
d1(k1)
iq
1(k1)
⎤
⎥
⎥
⎦
+
⎡
⎢
⎢
⎣
0
ψω(k
0
)
0
ψω(k
1
)
⎤
⎥
⎥
⎦
. (3)
Equation (3) is the full-rank model of PMSM and will be
used for the design of objective function of multiparameter
estimator. id
0
and id
1
means the id-axis is injected with the
current equal and unequal to zero, respectively. It is a chal-
lengeable work to estimate these parameters simultaneously
since the PMSM operates under various operating conditions.
Noise, disturbance, and parameter variations are unavoidable
and immeasurable in many applications. Therefore, it is very
important to develop an efficient model parameter tracking
approach for PMSM multiparameter identification.
III. GPU-A
CCELERATED PARALLEL IMMUNE
COEVOLUTIONARY PSO
A. Principle of Basic PSO Algorithm
According to the social intelligent behavior of bird flock-
ing, the particles fly toward better searching areas during the
evolutionary process. Assuming in a d-dimensional solution
space, each particle i is composed of two vectors, which are the
velocity vector V
i
= {V
i1
, V
i2
,...,V
id
} and the position vec-
tor X
i
= {X
i1
, X
i2
,...,X
id
}, t he searching procedure can
be given as
V
id
(t +1)=ωV
id
+ c
1
∗ rand
1
() (P best
id
(t) − X
id
(t))
+ c
2
∗ rand
2
() (gBest
d
(t) − X
id
(t)) (4)
X
id
(t +1)=X
id
(t)+V
id
(t +1). (5)
In (4) and (5), P best
id
represents the best position found by
ith particle up to now and gBest
d
is the best particle among the
entire population. c
1
and c
2
are the acceleration coefficients,
ω is the inertia weight factor decreasing linearly, and rand
1
and rand
2
are two uniformly distributed numbers generated
randomly within [0, 1], respectively.
B. Principle of G-PCIPSO Algorithm
The general steps of G-PCIPSO are stated as follows.
Algorithm 1. G-PCIPSO
Step1: Initialize t he normal sub-swarms P
i
and antibodies
memory; set up parameters for G-PCIPSO.
Step2: Transfer data from CPU to GPU.
// sub-processes in “for” are done in parallel
Step3:fori =1 to I//1 ≤ i ≤ I, I is the number of normal
sub-swarms;
Perform the process of PSO for sub-swarms Pi
for j =1to N//1 ≤ j ≤ N, N is the number of parti-
cles //sub-swarms Pi
update particle
i
velocity using the equation (4)
update particle
i
position using the equation (5)
Evaluate the fitness value of particle
i
;
end for
end for
Step4:fori =1to k//1 ≤ i ≤ k, k is the number of Pbests
Perform the vaccine-enhanced for Pbest
i
based on the
equations (9)–(11) and the Fig. 2.
end for
剩余10页未读，继续阅读
711 浏览量
447 浏览量
233 浏览量
138 浏览量
106 浏览量
188 浏览量
270 浏览量
230 浏览量
149 浏览量
weixin_38644780
- 粉丝: 2
- 资源: 886
我的内容管理
展开
最新资源
- VectorMetaballs(iPhone源代码)
- get-a-life-elm
- leetcode-daily：针对LeetCode每日一题进行记录
- myprofile:型材乔治
- 基于JAVA的数字化题库系统
- Frontend-I-Digital-House
- atom-watcher:观看文件更改并实时重新加载 Atom-Shell 应用程序
- 研究生管理信息系统.rar
- MiPortfolio
- pinlog:PinLog是一个功能强大的驱动程序和库，集成了多个日志记录系统。 它分为最小的独立服务单元，易于使用
- bitrise-cli-webui:Bitrise CLI的简单本地Web UI-进行中
- 音乐大赛活动网页模板
- 翻译解读-crx插件
- CakePhp3-PhotoCrop:使用 jquery JCrop 并将画布元素的文件上传到本地 webroot 文件夹的插件
- algorithm:算法挑战
- RTCM3.3 yy.zip
