Lateral Control of Permanent Magnet Electrodynamic Suspension Vehicle Based on Improved Nonlinear Model Predictive Controller
-
摘要:
针对横向力不足、模型不确定和时变扰动环境下永磁电动悬浮汽车横向运动控制问题,提出一种改进非线性模型预测横向跟踪控制方法(NMPC-ESO-EKF)以实现车辆横向精准控制. 首先,提出通过偏转磁轮来补偿系统横向力的横向运行模式,以此建立横向非线性动力学模型;然后,建立含有约束条件的NMPC控制器,并构造扩张状态观测器(ESO)来观测系统内外扰动以补偿控制输入,同时引入扩展卡尔曼滤波器(EKF)消除传感器测量噪声对ESO观扰的影响;最后,搭建联合仿真平台和实验平台进行仿真与实验验证. 研究结果表明:永磁电动悬浮汽车在横向运行模式下,能有效实现左右横移运动;相较于PID-EKF控制,在定常数参考信号下,NMPC-ESO-EKF超调量降低98.90%,系统调节时间缩短47.78%;在方波参考信号下,系统平均超调量和平均跟踪误差分别降低了93.77%和36.13%;施加扰动后,系统横向位移波动幅值减小34.51%,恢复时间缩短42.08%,横向控制精度与抗扰能力大幅提升,为永磁电动悬浮汽车横向控制研究提供一定参考.
Abstract:Objective : The permanent magnet electrodynamic suspension (PMEDS) vehicle, with its integrated levitation and propulsion system, shows significant potential for applications in areas such as ultra-high-speed and heavy-load highways, as well as mountainous road transportation. However, the PMEDS vehicle is a lateral underdamped system, and its lateral open-loop system is extremely unstable. It is prone to instability when exposed to disturbances, such as lateral gusts, which seriously affect driving safety. Therefore, it is crucial to study lateral motion control under the conditions of insufficient lateral force, model uncertainty, and time-varying disturbances.Method : In order to achieve precise control of the lateral motion and enhance the lateral anti-disturbance ability of the PMEDS vehicle, a lateral operation mode was proposed, and a lateral nonlinear dynamic model was derived. Moreover, an improved nonlinear model predictive lateral tracking controller (NMPC-ESO-EKF) with constraints was established, and its effectiveness was verified by simulation and experiment. Firstly, the working principle of the radial annular Halbach permanent magnet wheel was introduced; the feasibility of compensating for the lateral force of the system by deflecting the magnetic wheels was analyzed, and the vehicle’s lateral operation mode via symmetrically deflecting the same-side magnetic wheels was proposed. Secondly, according to the principle of lateral operation, the lateral damping term was introduced to derive the lateral nonlinear dynamic model for the PMEDS vehicle. Then, based on the dynamic model, a nonlinear discrete state-space model was derived as the prediction model; a cost function and the amplitude constraint of the deflection angle were designed. According to the prediction model, cost function, and system constraints, the nonlinear model predictive controller (NMPC) was established; the control problem was transformed into a nonlinear programming problem in the finite time domain. Meanwhile, in order to improve the anti-disturbance ability of the NMPC, the extended state observer (ESO) was constructed to observe the internal and external disturbances of the system, compensating for the control input. In order to reduce the sensitivity of ESO to sensor measurement noise, an extended Kalman filter (EKF) was introduced to filter the raw signals. Moreover, the convergence of the overall control framework was demonstrated. Finally, the simulation and analysis of lateral constant trajectory tracking and lateral anti-disturbance under internal and external disturbances conditions were conducted using the Matlab/Simulink-Simscape joint simulation platform. Additionally, experiments were performed by using the lateral running experimental device and the principle prototype, covering short-distance constant trajectory tracking, square wave signal trajectory tracking, and the analysis of anti-disturbance ability under long-distance constant trajectory tracking.Result : The simulation results show that under the disturbance-free condition, the system can effectively realize the lateral trajectory tracking control function under PID, MPC, NMPC, and NMPC-ESO control strategies. The response speeds of the various controllers are relatively similar. The PID controller exhibits the largest overshoot, while MPC experiences some overshoot, though significantly smaller than that of PID. Both NMPC and NMPC-ESO have minimal overshoot, with their response curves closely resembling each other, indicating that ESO has no noticeable effect. Under external disturbance conditions, the addition of ESO significantly improves the anti-disturbance ability of NMPC. The lateral displacement fluctuations are minimal for NMPC-ESO, and the system recovery time is the shortest. When sensor measurement noise is introduced, the system can stably control within a range close to the target position under NMPC control, but there is still an error compared to the target position. Due to the excessive sensitivity of ESO to the noise, it overcompensates the deflection angle, leading to instability in the NMPC-ESO control after ESO’s application. To address this issue, an EKF is introduced to suppress the impact of internal noise on ESO’s disturbance estimation. Under both internal and external disturbance conditions, NMPC-ESO-EKF significantly improves the lateral trajectory tracking accuracy by 89.77% and greatly enhances the anti-disturbance ability, demonstrating the necessity of incorporating EKF.The experimental results indicate that the error between the values filtered by EKF and the true values is very small. PID-EKF exhibits significant overshoot and oscillation under various operating conditions, which poses safety concerns in practical engineering applications. Under short-distance constant reference signals, compared to PID-EKF control, NMPC-ESO-EKF reduces the overshoot by 98.90% to 2.12 mm, with the shortest system settling time of 4.7 s, which is reduced by 47.78%. Under square wave reference signals, the trajectory tracking curves of MPC-EKF, NMPC-EKF, and NMPC-ESO-EKF are all relatively stable, with NMPC-ESO-EKF showing the smallest average overshoot and tracking error, improving performance by 93.77% and 36.13%, respectively, compared to PID-EKF. Under long-distance constant reference signals, larger overshoots are observed across all controllers. After applying continuous disturbances, each controller can make the system return to the stable target value. After stabilization, the input deflection angle fluctuates around −10° to continuously resist external disturbance; the displacement fluctuation amplitude of MPC-EKF is the largest, with the longest recovery time; the anti-disturbance ability of NMPC-ESO-EKF is optimal, with the fluctuation amplitude reduced by 34.51% and the recovery time shortened by 42.08% compared to PID-EKF.Conclusion : The PMEDS vehicle, under the improved nonlinear model predictive control method, has enhanced its lateral stability and possesses the basic functionality of flexible lateral movement. It is suitable for scenarios such as lateral anti-disturbance during normal driving, active lane change, lateral hill climbing, and parallel parking. Up to now, significant breakthroughs have been achieved in areas such as the optimization of electromagnetic forces of magnetic wheels, structural design of vehicles, and single-degree-of-freedom control research. However, it is still in the theoretical research stage. In order to realize engineering applications as soon as possible, future efforts will focus on intelligent system perception, path planning, and multi-degree-of-freedom coupling trajectory tracking control research. -
近年来高速铁路发展迅猛[1],列车伴山而行、穿山而过,缩短城市间时空距离的同时也加大了铁路沿线环境安全监测的难度. 隧道口上方山体和轨旁边坡面临崩塌和溜滑等风险,威胁列车的安全运行[2].无线传感器网络(wireless sensor network,WSN)因其自组织能力,可被应用于山体和边坡的安全监测报警系统[3-4]. 部署的传感器节点可以实时感知并收集山体崩塌和滑坡的触发参量,例如降雨量、土壤湿度、孔隙水压力等[5]. 节点部署策略直接影响WSN的监测效果,通过优化节点部署方案,可以提高安全监测报警系统的性能,从而保障列车的安全运营.
WSN节点部署的研究目标主要有提高网络覆盖率、网络连通性以及网络生命周期等[6]. 文献[7]中提出一种自适应多策略人工蜂群算法,使用模拟退火与动态搜索改进人工蜂群算法,增强算法跳出局部最优的能力,从而提高网络覆盖率. 文献[8]中提出一种基于快速非支配排序的改进蚁狮算法,通过提高算法求解精度、种群多样性以及全局搜索能力,优化节点部署方案. 以上研究适用于部署二维平面的WSN,起伏地形下节点部署问题更加复杂. 文献[9]分析起伏地形覆盖问题,证明该问题为NP-hard完全问题.
针对起伏地形下的WSN节点部署,学者们基于虚拟力方法、Voronoi图、智能优化算法等进行了研究[10]. 文献[11]提出一种适用于起伏地形的WSN确定性部署算法,使用Voronoi图将监测区域分区,结合Delaunay方法建立部署节点间连通性,在保证网络连通性的同时提高网络覆盖率,但算法性能受地形影响大. 文献[12]通过经典分水岭算法保持地形拓扑特征,将起伏地形映射到二维平面,再利用代价因子构造感知概率,节点均匀分布,该算法在不规则地形下采用均匀随机分布方式部署节点,易产生冗余覆盖,增多节点部署数目. 文献[13]采用数字高程模型(digital elevation model,DEM)对起伏地形表面进行建模,提出一种基于网格扫描的贪婪节点部署算法,根据节点覆盖面积,降序选择节点部署位置;该算法仅以网络覆盖率为优化目标未考虑节点部署方案对WSN应用性能的影响. 文献[14]提出一种适用于起伏地形下WSN节点部署的改进海洋捕食者算法(improved marine predator algorithm,IMPA),通过引入随机对立学习与差分进化算子提高IMPA的全局搜索能力,并结合混合蛙跳算法进行局部搜索,提高求解精度,优化节点部署方案;然而,当网络目标覆盖率提升或搜索维度增加时,该算法的求解精度可能会下降,导致需要部署的节点数量增多. 文献[15]中提出一种增强灰狼优化器(enhanced grey wolf optimizer,EGWO),将灰狼种群分为内、外层进行搜索,并引入Tent映射以增强算法的局部与全局搜索能力,从而提高问题求解精度,优化起伏地形下的WSN覆盖率;但同样,在搜索维度增加时,该算法也面临求解精度下降和节点部署数量增加的问题.
上述研究成果虽然可以实现起伏地形下的WSN节点部署,但仍存在一些不足,如节点部署数目多导致网络成本高、部署难度大,并且缺乏与典型路由算法结合的网络生命周期性能测试分析. 此外,由于铁路沿线起伏地形复杂多样,不同地形粗糙度下的节点部署算法性能也存在一定差异. 为解决这些问题,提出一种适用于铁路沿线起伏地形的WSN节点部署算法. 该算法基于DEM与Delaunay方法对起伏地形进行建模,确定节点部署的解空间,以网络连通性为约束、网络覆盖率为目标,通过迭代方式,结合IMPA及遗憾最小化为准则,利用筛选函数完成WSN节点的部署.
1. 模型分析
1.1 起伏地形表面建模
地理信息系统中,多采用DEM表示地形特征,使用规则网格模型建立地形特征点集合DEM[16],如式(1)所示.
DEM={Qv,h}V,H, (1) 式中:Qv,h为特征点,表示节点可部署位置,是规则网格第v行、第h列交点;V、H分别为行、列最大数目.
Qv,h的编号为H(v−1)+h,特征点编号集合为D, V×H为网格精度,如图1(a)所示.
使用Delaunay方法将地形表面划分为三角平面,如图1(b)所示. 三角平面集合表示为Φ,三角平面边集合表示为e,网格精度需满足任意边的长度小于等于节点感知半径rc.
1.2 节点连通性模型
定义1:当两个传感器节点通视,且欧式距离小于等于通信半径rd,则称节点连通.
采用ModSAF算法判断两点通视,当两点连线地表切面的高程值不高于两点间直线时,则两点通视. 以图2为例,A、B为地表上不相连两点,其水平面垂直映射点为A'、B'. e'为集合e水平面垂直映射集合; b为线段A'B'与集合e'的交点集合,bδ为集合b中第δ个交点;zbδ为bδ对应线段AB上的高程值,z′bδ为bδ对应地表高程值. 若∀z′bδ⩽zbδ,则A、B通视,记为lA,B=1;否则A、B非通视,记为lA,B=0.当A、B符合定义1时,A、B连通,记为LA,B=1;否则,A、B非连通,记为LA,B=0.
定义2:网络连通性是指网络中任意两个传感器节点间至少存在一条通信路径.
定义3:网络连通度λ(⋅)是指网络中存在的通信路径总数与最大通信路径总数之比,如式(2)所示.
λ(N)={∑I∈NεI|N|2−|N|,|N|>1,0,|N|=1, (2) 式中:N为WSN部署节点集合,节点编号即部署位置编号,εI(0⩽εI⩽|N|−1)为节点I的通信路径数目,|∙|表示集合中元素数目.
1.3 节点覆盖模型
节点采用球形布尔感知模型,当节点I与点J间欧式距离dI,J⩽rc,且lI,J=1时,节点I覆盖点J,记为CI,J=1;否则,未覆盖,记为CI,J=0.
SI={φ|CI,aφ,1=1,CI,aφ,2=1,CI,aφ,3=1,φ∈Φ,φ∉K},为节点I覆盖三角平面中的集合. 其中:aφ,1、aφ,2、aφ,3为三角平面φ的顶点;K=⋃I∈NSI,为WSN覆盖集合.
网络覆盖率η是指WSN覆盖面积与起伏地形表面总面积之比[19],如式(3)所示.
η(N)=∑φ∈Kqφ∑φ∈Φqφ×100\% , (3) 式中:qφ为三角平面φ的面积.
2. 改进海洋捕食者算法
海洋捕食者算法(MPA)是一种模拟海洋生物捕食的群智能算法,捕食策略分为莱维飞行与布朗运动[20],具有寻优能力强、调整参数少的特点. 但MPA在3个搜索阶段(高速比阶段、等速比阶段、低速比阶段)的搜索次数均固定为tmax/tmax33,易导致过早进入局部搜索或局部搜索不充分. 本节提出根据相邻搜索最优解适应度差值与最大搜索次数tmax联合控制各搜索阶段搜索次数的IMPA.
2.1 初始种群
种群初始解X0随机均匀分布在搜索空间上,如式(4)所示.
X0=xb+rand(xa−xb), (4) 式中:xa、xb分别为解空间上、下界,rand为(0,1)区间中均匀随机数向量.
猎物矩阵Y由n个维度为dm的猎物构成,如式(5)所示. 初始捕食者矩阵E=[XtopXtop⋯Xtop]n×dm,由顶级捕食者Xtop构成.
Y=[X1,1X1,2⋯X1,dmX2,1X2,2⋯X2,dm⋮⋮⋮Xn,1Xn,2⋯Xn,dm]n×dm. (5) 2.2 搜索策略
步骤1 高速比阶段. 通过布朗运动更新猎物,如式(6)所示.
Y(t)i=Y(t−1)i + pRA⊗si, (6) si=RB⊗(Ei−RB⊗Y(t−1)i), (7) 式中:Y(t)i为第t次搜索中第i个猎物,si为第i个猎物移动步长,Ei为第i个捕食者,p=0.5,RA为[0,1]区间内随机数向量,⊗表示向量中元素依次相乘,RB为基于布朗运动的随机数向量.
第t次搜索的最优解适应度差值为
FC(t)={Ftop(t),t = 1,Ftop(t)−Ftop(t−1),t > 1, (8) 式中:Ftop(t)为第t次搜索的最优解适应度.
当FC满足式(9),即搜索过程中连续M次FC=0时,进入步骤2,并更新FC=[11⋯1].
t∑j=t−M+1FC(j)=0,t>M. (9) 步骤2 等速比阶段. 通过布朗运动与莱维飞行更新猎物,如式(10)所示.
Y(t)i={Y(t−1)i + pRA⊗si,1⩽i⩽n2,Ei + pCF⊗si,n2 < i⩽n, (10) si={RL⊗(Ei−RL⊗Y(t−1)i),1⩽i⩽n2,RB⊗(RB⊗Ei−Y(t−1)i),n2 < i⩽n, (11) 式中:CF=(1−t/tmax)(2ttmax),为控制捕食者步长的自适应参数;RL为基于莱维飞行的随机数向量. 通过式(8)求解FC,当FC满足式(9)时,进入步骤3,并更新FC=[11⋯1].
步骤3 低速比阶段. 通过莱维飞行更新猎物,如式(12)所示. 通过式(8)求解FC,当FC满足式(9),终止搜索. 搜索过程中,t=tmax时终止搜索.
Y(t)i=Ei + pCF⊗si, (12) si=RL⊗(RL⊗Ei−Y(t−1)i). (13) 搜索过程中进行海洋记忆存储,保存优秀猎物;使用涡流或鱼类聚集装置效扰动猎物矩阵Y,使算法跳出局部最优[20].
3. WSN节点部署算法
综合考虑网络覆盖率、网络连通性以及网络生命周期部署传感器节点. 基于所提IMPA建立候选个体集,在IMPA搜索过程中对猎物进行取整. 以收益函数遗憾最小化为准则[21]衍生新个体,将最佳新个体并入集合N,通过迭代方式确定达到目标覆盖率ηt的节点部署方案.
3.1 基于IMPA构建候选个体集
节点部署问题解空间为集合D,基于所提IMPA建立候选个体集,搜索维度dm随η增加而减小,如式(14)所示.
dm=⌈(1−η(N)ηt)⌊CL2rc⌋⌊CW2rc⌋⌉, (14) 式中:CL、CW分别为起伏地形长、宽.
适应度函数F(⋅)如式(15)所示,适应度值最大的个体为最优解.
{F(Yi)=η(N∪Yi),s.t. λ(N∪Yi) = 1. (15) IMPA通过式(4) ~(13)求解最优解,并逐次提取最优解g(t)top,建立矩阵G=[g(1)topg(2)top⋯g(t)top],通过式(16)建立候选个体集Z.
Z={Gj1|ρ(N∪Gj1)⩾ρm}∪{Gj1|F(Gj1)⩾Fm}, (16) ρ(N)=∑I∈NξI|N|2−|N|, (17) 式中:ρ(⋅)为网络密度即网络中实际存在的连通边数目与最大可能边数目之比;Gj1为G中第j1个个体;ρm=∑|G|j1=1ρ(N∪Gj1)/|G|,为个体平均网络密度;Fm=∑|G|j1=1F(Gj1)/|G|,为个体平均适应度;ξI为节点I的连通边数目.
3.2 基于遗憾最小化衍生新个体
IMPA求解维度固定,易导致冗余节点增加,为优化节点部署数目、提高IMPA搜索结果利用率,基于遗憾最小化衍生新个体. 任取集合Z中两个个体Zj2与Zj3,生成节点候选集合W={Zj2∪Zj3}. W中节点为博弈参与者,策略空间Ω={ω1,ω2}. ω1=0,表示节点不加入新个体;ω2=1,表示节点加入新个体. 第k个参与者的策略σk∈Ω,策略组σ=(σ1,σ2,⋯,σ|W|),收益和函数如式(18)所示.
μ(σ)={|σ|∑k=1|βk|−||σ|∪k,α=1k≠α(SWk∩SWα)||σ|,βm=1,|σ|∑k=1(|βk|−2||σ|∪α=1α≠k(βk∩βα)|),βm>1, (18) 式中:βm=∑|σ|k=1σk,为新个体中节点数目;βk、βα分别为第k、α个参与者的覆盖集合;SWk、SWα分别为节点Wk、Wα的三角平面覆盖集合.
σk=1时,βk=SWk;否则βk=∅.μk(σ)=μ(σ)/|σ|,表示第k个参与者收益.
生成2|σ|个不重复策略组,第k个参与者在第T个策略组σ(T)中采取策略σk的遗憾如式(19)所示.
r(T)k(σk)=[μk(σk,σ(T)−k)−μk(σ(T))]+, (19) 式中:σ(T)−k为σ(T)中除σ(T)k以外的策略,[x]+=max{x,0}.
第k个参与者采用策略ωγ(γ∈[1,2])的概率为
Pk(ωγ)=Rk(ωγ)2∑γ0=1Rk(ωγ0), (20) 式中:Rk(ωγ)=∑2|σ|T=1r(T)k(ωγ),为第k个参与者在策略σk=ωγ时的累积遗憾.
参数者根据Pk选择策略,Pk(ω1)=Pk(ω2)时,生成多个策略组;Pk(ω1)>Pk(ω2)时,策略组σ∗中第k个参与者的策略σ∗k=ω1;Pk(ω1)<Pk(ω2)时,σ∗k=ω2. 当策略组σ∗满足式(21)时,σ∗达到纳什均衡,为最佳策略组;否则,更改σ∗中遗憾最大的参与者策略,直至σ∗达到纳什均衡. 最佳策略组σ∗衍生出新个体g∗={W⊗σ∗},构建新个体集合G∗.
μk(σ∗)⩾maxσk∈Ω(μk(σk,σ∗−k)). (21) 3.3 筛选函数
以网络覆盖率与网络密度的线性函数归一化值ηno(⋅)、ρno(⋅)为指标,构建筛选函数为
F1(G∗j4)=θtAηno(N∪G∗j4)+(1−θtA)ρno(N∪G∗j4), (22) 式中:G∗j4为新个体集合G∗j4中第j4个个体;θtA为第tA次迭代中网络覆盖率的权重,θtA随tA增大而增大,如式(23)所示.
θtA={θtA−1+(1−θtA−1)(η(N)ηt)dm−1,tA>1,θ0,tA=1. (23) 选择F1值最大的新个体并入集合N,即N=N∪argmaxG∗j4(F1(G∗j4)). 当η(N)<ηt时,更新覆盖集合SI,进入下一次迭代,tA=tA+1;否则,终止迭代,得到节点部署方案. 初始时,集合N中仅包含Sink节点所在特征点编号. 算法流程如下:
算法名称:WSN节点部署算法
输入:部署节点集合N,目标覆盖率ηt,tA=1
输出:部署节点集合N
1: while η(N)<ηt do
2: 基于IMPA建立候选个体集合Z;
3: for j2=1:|Z|−1 do
4: for j3= j2 + 1:|Z| do
5: W={Zj2∪Zj3},并生成2|σ|个策略组;
6: while μk(σ∗)<maxσk∈Ω(μk(σk,σ∗−k)) do
7: 通过式(19) ~ (21)更新σ∗;
8: end while
9: 根据最佳策略组衍生出新个体
10: end for
11: end for
12: 根据式(22)选择最佳新个体,更新集合N、S;
13: 更新tA =tA+1,计算权重θtA;
14: end while
4. 仿真分析
在Window 10系统上,使用MATLAB 2022a对算法进行仿真,CPU为Intel(R) Core(TM) i7-13700F,内存32 GB. 仿真参数如表1所示[22].
表 1 仿真参数及取值Table 1. Simulation parameters and values参数 取值 起伏地形:长/宽/高 m 100/100/50 DEM网格精度 21 × 21 IMPA种群大小 50 权重θ0 0.9 节点初始能量/J 1 数据包大小/bit 3200 电路能耗Eelec/(nJ•bit−1) 50 功放参数εfs/( pJ•bit−1•m−2) 10 功放参数εamp/(pJ•bit−1•m−4) 0.0013 距离阈值d0/m 87 将本文算法(tmax=500次,M=35次)与IMPA[14](tmax=100次)、EGWO[15](tmax=100次)、IMPA-FD(improved marine predator algorithm-fixed dimensions)(tmax=500次)、IMPA-AD(improved marine predator algorithm-adjustable dimensions)(tmax=500次,M=35次)进行对比分析. IMPA-FD是基于文献[11]的IMPA、维度dm=⌊CL/CL(2rc)(2rc)⌋⌊CW/CW(2rc)(2rc)⌋的迭代部署算法. IMPA-AD是基于本文IMPA、维度函数的迭代部署算法. rc=25 m,rd=2rc. 分析目标覆盖率ηt与地形粗糙度τ对算法性能的影响. 地形粗糙度τ即地表面积与其垂直投影面积之比[23],如式(24)所示. 图3为不同τ下的地形示意. 为降低偶然性,以测试场景独立执行20次的均值表示结果.
τ=∑φ∈ΦqφCLCW. (24) 1)目标覆盖率ηt对算法性能的影响
测试地形图3(d),ηt以2.5%为增量从80%递增至100%,分析ηt对节点部署数目与算法运行时长的影响,结果如图4所示. 由图4(a)可见,随ηt递增,节点部署问题求解复杂度增大,节点部署数目及其增量逐渐增加;IMPA、EGWO部署节点数目最高、增长幅度最大,这是因为随ηt提高搜索维度增大,算法求解精度降低;IMPA-FD部署节点数目次高,这是因为维度固定,但增加了冗余节点数目;本文算法、IMPA-AD部署节点数目较少,是因为维度函数提高了算法求解精度;相比于IMPA-AD,本文算法中衍生新个体的过程优化了节点部署与网络覆盖率之间的关系,有助于剔除冗余节点,进一步降低节点部署数目. 与对比算法相比,本文算法的节点部署数目降低2.9%~69.1%. 由图4(b)可见,随ηt递增,算法运行时长呈上升趋势,IMPA-AD、本文算法的运行时长上升幅度低于其他算法,受ηt影响小;IMPA-AD运行时长最短,这是因为本文提出的IMPA减少了搜索次数,且维度函数降低了问题求解复杂度. 本文算法因新个体衍生过程使其运行时长略高于IMPA-AD.
2)地形粗糙度τ对算法性能的影响
设置地形高度一定,ηt=100%,从节点部署数目、网络密度、网络生命周期以及算法运行时长方面分析算法在不同τ下的性能,结果如图5所示. WSN周期性收集数据,节点均需将感知数据发送至Sink节点[24]. 网络生命周期为首个节点能量耗尽时的网络运行轮次(单位:round). Sink节点坐标(50,0,25),采用DGABT路由算法进行数据传输[25].
图5(a)中,节点部署数目随τ递增,呈下降趋势. 这是因为地形高度一定时,τ增大后节点间空间关联性增强;与其他算法相比,本文算法节点部署数目降低3.1%~74.0%. 由图5(b)可见,随τ递增,本文算法、IMPA-AD的ρ呈上升趋势;IMPA-FD的ρ呈波动趋势;IMPA、EGWO的ρ呈下降趋势. 这是因为随τ递增,本文算法、IMPA-AD节点部署数目略有下降对最大可能边数目影响较大;IMPA-FD节点部署数目减少对网络边数目、最大可能边数目影响相近;IMPA、EGWO节点部署数目锐减对网络边数目影响较大. 由图5(c)可见,τ对网络生命周期具有一定的影响;本文算法对应的网络生命周期高出其他算法13.3%~286.5%,这是因为本文算法综合考虑η与ρ部署节点,使靠近Sink节点的网络密度高,提高节点能量利用率,且节点部署数目少,网络中需转发数据量小,降低数据传输能耗. 由图5(d)可见,算法运行时长随τ递增,呈下降趋势,且本文算法、IMPA-AD因τ增大后节点间空间关联性增强、搜索次数少,故而运行时长短于其他算法,EGWO、IMPA、IMPA-FD、IMPA-AD以及本文算法的运行时长方差分别为563.3、571.4、531.9、154.9、171.1 s2,可见本文算法、IMPA-AD受τ影响小.
5. 结 论
针对起伏地形下WSN节点部署数目多的问题,提出一种节点迭代部署算法,从仿真结果可见本文算法具有如下优点:
1) 有效降低节点部署数目. 相比于其他算法,本文算法在起伏地形一定时,不同ηt,节点部署数目降低2.9%~69.1%;在地形高度一定,ηt=100%时,不同τ,节点部署数目降低3.1%~74.0%.
2) 网络生命周期长. 当ηt=100%,不同τ时,本文算法的网络生命周期高出其他算法13.3%~286.5%.
3) 算法运行时长受ηt、τ影响较小.
综上所述,在不同测试地形下本文算法均表现出良好的性能,但仍存在算法运行时长高的缺点.
致谢:北京市高速铁路宽带移动通信工程技术研究中心(北京交通大学)开放课题基金资助(BHRC-2022-1).
-
表 1 永磁电动悬浮汽车模型参数
Table 1. Model parameters of PMEDS vehicle
编号 参数 数值 整车
参数整车质量m/kg
悬浮间隙h/mm
横向阻尼系数c/(N·s·m−1)18.1
10
17.2磁轮
参数极对数P
外径Ro/mm
内径Ri/mm
宽度d/mm
磁化角q/°
磁轮转速n0/rpm
磁轮磁阻力Fr0/N4
50
32.5
35
90
2000
17.33磁轮偏转角度范围α/° −20~20 表 2 系统响应结果对比
Table 2. Comparison of system response results
控制器 0~12 s 12 s后 平均误差/mm 性能提升/% 平均误差/mm 性能提升/% NMPC 71.31 34.41 NMPC-ESO 319.54 −348.10 183.07 −432.03 NMPC-EKF 57.99 18.70 6.12 82.20 NMPC-ESO-EKF 57.93 18.76 3.52 89.77 表 3 方波信号轨迹跟踪控制系统响应结果对比
Table 3. Comparison of system response results of square wave signal trajectory tracking control
mm 控制器 平均跟踪误差 平均超调量 PID-EKF 57.60 164.96 MPC-EKF 38.48 20.48 NMPC-EKF 36.82 15.69 NMPC-ESO-EKF 36.79 10.27 表 4 大距离定常数轨迹跟踪控制系统响应结果对比
Table 4. Comparison of system response results of long-distance constant trajectory tracking control
控制器 位移波动幅值/mm 恢复时间/s PID-EKF 40.89 6.82 MPC-EKF 45.39 10.46 NMPC-EKF 35.73 6.80 NMPC-ESO-EKF 26.78 3.95 -
[1] LIU B, SUN C, WANG B, et al. Adaptive speed planning of connected and automated vehicles using multi-light trained deep reinforcement learning[J]. IEEE Transactions on Vehicular Technology, 2022, 71(4): 3533-3546. doi: 10.1109/TVT.2021.3134372 [2] 邓自刚,刘宗鑫,李海涛,等. 磁悬浮列车发展现状与展望[J]. 西南交通大学学报,2022,57(3): 455-474,530.DENG Zigang, LIU Zongxin, LI Haitao, et al. Development status and prospect of maglev train[J]. Journal of Southwest Jiaotong University, 2022, 57(3): 455-474,530. [3] 林国斌,刘万明,徐俊起,等. 中国高速磁浮交通的发展机遇与挑战[J]. 前瞻科技,2023,2(4): 7-18.LIN Guobin, LIU Wanming, XU Junqi, et al. Opportunities and challenges for the development of high-speed maglev transportation in China[J]. Science and Technology Foresight, 2023, 2(4): 7-18. [4] 赵春发,刘浩东,冯洋,等. 五位姿参数下车载永磁体与永磁轨道之间的磁力特性研究[J]. 西南交通大学学报,2024,59(4): 804-811.ZHAO Chunfa, LIU Haodong, FENG Yang, et al. Magnetic force characteristics between on-board permanent magnet and permanent magnetic rail considering five pose parameters[J]. Journal of Southwest Jiaotong University, 2024, 59(4): 804-811. [5] 胡永攀,曾杰伟,王志强,等. 超高速永磁电动悬浮系统性能优化[J]. 西南交通大学学报,2023,58(4): 773-782.HU Yongpan, ZENG Jiewei, WANG Zhiqiang, et al. Performance optimization of ultra-high speed permanent magnet electrodynamic suspension system[J]. Journal of Southwest Jiaotong University, 2023, 58(4): 773-782. [6] QIN W, BIRD J Z. Electrodynamic wheel magnetic rolling resistance[J]. IEEE Transactions on Magnetics, 2017, 53(8): 8107407.1-8107407.7. [7] SHI H F, KE Z H, ZHENG J, et al. An effective optimization method and implementation of permanent magnet electrodynamic wheel for maglev car[J]. IEEE Transactions on Vehicular Technology, 2023, 72(7): 8369-8381. doi: 10.1109/TVT.2023.3245620 [8] WRIGHT J D. Modeling, analysis, and control of a radial electrodynamic wheel vehicle and analysis of an axial electrodynamic wheel[D]. Charlotte: The University of North Carolina at Charlotte, 2019. [9] FUJII N, NONAKA S, HAYASHI G. Design of magnet wheel integrated own drive[J]. IEEE Transactions on Magnetics, 1999, 35(5): 4013-4015. doi: 10.1109/20.800739 [10] FUJII N, HAYASHI G, SAKAMOTO Y. Characteristics of magnetic lift, propulsion and guidance by using magnet wheels with rotating permanent magnets[C]//Conference Record of the 2000 IEEE Industry Applications Conference. Thirty-Fifth IAS Annual Meeting and World Conference on Industrial Applications of Electrical Energy. Rome: IEEE, 2002: 257-262. [11] JUNG K S. Parametric design of contact-free transportation system using the repulsive electrodynamic wheels[J]. Journal of the Korea Academia-Industrial Cooperation Society, 2016, 17(3): 310-316. doi: 10.5762/KAIS.2016.17.3.310 [12] JUNG K S. Transfer system using radial electrodynamic wheel over conductive track[J]. The Korea Academia-Industrial Cooperation Society, 2017, 18(11): 794-801. [13] 刘新,邓自刚,梁乐,等. 基于斜置环形Halbach永磁轮的磁浮列车“悬浮-导向-推进” 一体化方案设计[J]. 机车电传动,2023(2): 90-96.LIU Xin, DENG Zigang, LIANG Le, et al. Levitation-guidance-propulsion integrated design for maglev trains based on oblique ring Halbach permanent magnet wheels[J]. Electric Drive for Locomotives, 2023(2): 90-96. [14] ZHANG Z, DENG Z G, ZHANG S, et al. Design and operating mode study of a new concept maglev car employing permanent magnet electrodynamic suspension technology[J]. Sustainability, 2021, 13(11): 5827.1-5827.20. [15] 徐俊起,林国斌,荣立军,等. 中低速磁浮列车悬浮控制技术成果及应用[J]. 铁道技术标准(中英文),2022(10): 34-39.XU Junqi, LIN Guobin, RONG Lijun, et al. Research achievements and application of levitation control technology for medium-low speed maglev train[J]. Railway Technical Standard (Chinese & English), 2022(10): 34-39. [16] DE BOEIJ J, STEINBUCH M, GUTIERREZ H. Real-time control of the 3-DOF sled dynamics of a null-flux maglev system with a passive sled[C]//2006 IEEE International Symposium on Industrial Electronics. Montreal: IEEE, 2006: 2549-2555. [17] ZHANG B J, KE Z H, LI Z Y, et al. Yawing stability and manipulative approach design for maglev car based on active disturbance rejection control[J]. Asian Journal of Control, 2024, 26(2): 1003-1016. doi: 10.1002/asjc.3246 [18] LI Z Y, KE Z H, SHI J H, et al. Investigation of RBF-SMC control strategy for vertical dynamics of maglev car considering temperature rise effects[J/OL]. IEEE Transactions on Intelligent Vehicles, 2024, 3456784.1-3456784.14(2024-09-10)[2024-09-28]. https://ieeexplore.ieee.org/document/10670497. [19] BIRD J, LIPO T A. Characteristics of an electrodynamic wheel using a 2-D steady-state model[J]. IEEE Transactions on Magnetics, 2007, 43(8): 3395-3405. doi: 10.1109/TMAG.2007.900572 [20] BIRD J, LIPO T A. Calculating the forces created by an electrodynamic wheel using a 2-D steady-state finite-element method[J]. IEEE Transactions on Magnetics, 2008, 44(3): 365-372. doi: 10.1109/TMAG.2007.913038 [21] PAUL S, BIRD J Z. A 3-D analytic eddy current model for a finite width conductive plate[J]. COMPEL: the International Journal for Computation and Mathematics in Electrical and Electronic Engineering, 2013, 33(1/2): 688-706. doi: 10.1108/COMPEL-03-2013-0083 [22] 闫玉盼,饶兵,刘砚菊,等. 基于非线性模型预测的四旋翼无人机轨迹跟踪控制[J]. 沈阳理工大学学报,2024,43(1): 36-43,49.YAN Yupan, RAO Bing, LIU Yanju, et al. Trajectory tracking control of quadrotor based on nonlinear model predictive control[J]. Journal of Shenyang Ligong University, 2024, 43(1): 36-43,49. [23] KAMEL M, ALEXIS K, ACHTELIK M, et al. Fast nonlinear model predictive control for multicopter attitude tracking on SO(3)[C]//2015 IEEE Conference on Control Applications (CCA). Sydney: IEEE, 2015: 1160-1166. [24] 许鹏,邢伯阳,刘宇飞,等. 基于扩张状态观测器和模型预测方法的四足机器人抗干扰复合控制[J]. 兵工学报,2023,44(增2): 12-21.XU Peng, XING (Bai| Bo)(Yang), LIU Yufei, et al. Anti-disturbance composite controller design of quadruped robot based on extended state observer and model predictive control technique[J]. Acta Armamentarii, 2023, 44(S2): 12-21. [25] HAN J Q. From PID to active disturbance rejection control[J]. IEEE Transactions on Industrial Electronics, 2009, 56(3): 900-906. doi: 10.1109/TIE.2008.2011621 [26] 王天威,黄军魁. 控制之美(卷2): 最优化控制MPC与卡尔曼滤波[M]. 北京:清华大学出版社,2023. [27] REIF K, UNBEHAUEN R. The extended Kalman filter as an exponential observer for nonlinear systems[J]. IEEE Transactions on Signal Processing, 1999, 47(8): 2324-2328. doi: 10.1109/78.774779 [28] LI J, XIA Y Q, QI X H, et al. On convergence of the discrete-time nonlinear extended state observer[J]. Journal of the Franklin Institute, 2018, 355(1): 501-519. doi: 10.1016/j.jfranklin.2017.11.019 [29] 陈松林,王鑫,何宗儒. 一种兼顾带宽拓展和噪声抑制的ESO参数整定方法[J]. 控制与决策,2018,33(10): 1908-1914.CHEN S L, WANG X, HE Z R. A parameter tuning method for extended state observer with balance of bandwidth expansion and noise suppression[J]. Control and Decision, 2018, 33(10): 1908-1914. [30] 郭永飞,张荣彬,姚植元,等. 基于ESO-MPC的核电厂协调系统优化控制研究[J]. 核动力工程,2024,45(6): 178-184.GUO Yongfei, ZHANG Rongbin, YAO Zhiyuan, et al. Research on optimization control of nuclear power plant coordination system based on ESO-MPC[J]. Nuclear Power Engineering, 2024, 45(6): 178-184. [31] HUANG R, PATWARDHAN S C, BIEGLER L T. Robust stability of nonlinear model predictive control based on extended Kalman filter[J]. Journal of Process Control, 2012, 22(1): 82-89. doi: 10.1016/j.jprocont.2011.10.006 -