文章快速检索     高级检索
  大地测量与地球动力学  2020, Vol. 40 Issue (4): 331-334, 345  DOI: 10.14075/j.jgg.2020.04.001

引用本文  

袁德宝, 张振超, 张军, 等. 最优化分数阶算子EGM(1, 1)模型在变形监测预报中的应用[J]. 大地测量与地球动力学, 2020, 40(4): 331-334, 345.
YUAN Debao, ZHANG Zhenchao, ZHANG Jun, et al. Application of Optimized Fractional Order EGM (1, 1) Model in Deformation Monitoring and Forecasting[J]. Journal of Geodesy and Geodynamics, 2020, 40(4): 331-334, 345.

项目来源

国家自然科学基金(51474217);内蒙古自治区自然科学基金(2018MS03047);内蒙古农业大学教育教学改革研究重点项目(JGZD201815)。

Foundation support

National Natural Science Foundation of China, No. 51474217; Natural Science Foundation of Inner Mongolia Autonomous Region, No. 2018MS03047;Key Projects of Education and Teaching Reform of Inner Mongolia Agricultural University, No.JGZD201815.

通讯作者

张振超,硕士生,主要研究方向为变形监测数据处理与InSAR形变监测,E-mail:zhangzckd@163.com

Corresponding author

ZHANG Zhenchao, postgraduate, majors in deformation data processing and InSAR deformation monitoring, E-mail:zhangzckd@163.com.

第一作者简介

袁德宝, 博士, 副教授, 主要从事GPS卫星定位与导航方面的教学与研究,E-mail:yuandb@cumtb.edu.cn

About the first author

YUAN Debao, PhD, associate professor, majors in teaching and research of GPS satellite positioning and navigation, E-mail:yuandb@cumtb.edu.cn.

文章历史

收稿日期:2019-04-07
最优化分数阶算子EGM(1, 1)模型在变形监测预报中的应用
袁德宝1     张振超1     张军2     张建1     
1. 中国矿业大学(北京) 地球科学与测绘工程学院,北京市学院路丁11号,100083;
2. 内蒙古农业大学理学院,呼和浩特市昭乌达路306号,010018
摘要:针对传统灰色模型在形变监测中数据序列拟合和预测精度不理想的情况,提出粒子群算法优化的分数阶算子EGM(1, 1)模型。通过粒子群算法选择拟合EGM(1, 1)平均相对误差最小的分数阶次,构建最优分数阶算子EGM(1, 1)模型。用典型的变形监测数据验证优化模型,结果表明,优化模型对变形监测数据的拟合和预测都达到较高的精度,说明优化模型在变形监测数据的处理中具有可行性和有效性。
关键词分数阶算子灰色模型粒子群变形监测

灰色模型自提出以来,在理论研究和实践应用方面都得到了发展。灰色系统理论中目前应用最广泛的为GM(1, 1)模型,其具有所需样本少、原理简单、运算简便等优点。但由于其固有的缺陷,模型的建模精度存在较大偏差,稳定性也不足[1]。很多学者对GM(1, 1)模型进行了改进,主要方式包括构建无偏模型[2]、改进背景值的构造方式[3-4]、调整或修改模型的边界条件[5]、优化灰作用量[6]、对原始数据进行函数变换[7]、优化模型参数估计方法[8]以及优化残差序列[9]等。以上优化模型中,GM(1, 1)模型建模的主要方式为对原始数据序列进行一阶累加,求解模型,再一阶累减还原得到模型的拟合和预测数据。由于在实际问题中原始数据间存在不等价随机性的问题,吴利丰[10]提出分数阶算子灰色预测模型,通过分数阶精确调节累加数之间的数量级,为不同的原始数据序列选择特定的分数阶算子构建灰色预测模型,能明显提高模型对数据的拟合和预测效果。

目前,灰色预测模型在变形监测领域中得到非常广泛的应用[11],但是基于分数阶建模思想构建分数阶算子的灰色预测模型在变形监测中的研究和应用还较少。鉴于此,本文构建了粒子群算法优化的分数阶算子EGM(1, 1)模型,通过典型的变形监测算例对比分析得出,优化模型对变形监测数据具有较高的拟合和预测效果,具有实际应用价值。

1 分数阶算子EGM(1, 1)模型

X(0)={x(0)(1), x(0)(2), …, x(0)(n)}为原始数据序列,rR+X(r)={x(r)(1), x(r)(2), …, x(r)(n)}是X(0)={x(0)(1), x(0)(2), …, x(0)(n)}的r阶累加生成算子,其中,${X^{(r)}}(k) = \sum\limits_{i = 1}^k {\frac{{\varGamma (r + k - i)}}{{\varGamma (k - i + 1)\Gamma (r)}}{x^{(0)}}(i), {\rm{ }}k = 1, 2, \cdots , n} $z(r)(k)=(z(r)(2), z(r)(3), …, z(r)(n)), 其中, ${z^{(r)}}(k) = \frac{{{x^{(r)}}(k) + {x^{(r)}}(k - 1)}}{2}, {\rm{ }}k = 2, 3, \cdots , n$X(-r)=(x(-r)(1), x(-r)(2), …, x(-r)(n))为X(0)r阶累减生成算子,其中,$\begin{array}{l} {X^{({\rm{ - }}r)}}(k) = \sum\limits_{i = 1}^{k{\rm{ - }}1} {{{( - 1)}^i}\frac{{\varGamma (r + 1)}}{{\varGamma (i + 1)\varGamma (r - i + 1)}}{x^{(0)}}(k - i)} , {\rm{ }}k = 1, 2, \cdots , n \end{array}$

分数阶算子EGM(1, 1)模型x(r-1)(k)+az(r)(k)=b中的参数向量$\boldsymbol{\hat a} = {[a, b]^{\rm T}}$可以运用最小二乘法估算:

$ \boldsymbol{\hat a} = {({\boldsymbol{B}^{\rm T}}\boldsymbol{B})^{ - 1}}{\boldsymbol{B}^{\rm T}}\boldsymbol{Y} $ (1)

式中,

$ \boldsymbol{Y} = \left[ {\begin{array}{*{20}{c}} {{x^{(r - 1)}}(2)}\\ {{x^{(r - 1)}}(3)}\\ \vdots \\ {{x^{(r - 1)}}(n)} \end{array}} \right]{\rm{ , }} \boldsymbol{B} = \left[ {\begin{array}{*{20}{c}} { - {z^{(r)}}(2)}&1\\ { - {z^{(r)}}(3)}&1\\ \vdots & \vdots \\ { - {z^{(r)}}(n)}&1 \end{array}} \right] $ (2)

$ \begin{array}{c} {x^{(r - 1)}}(k) = {x^{(r)}}(k) - {x^{(r)}}(k - 1) =\\ \sum\limits_{i = 1}^k {\frac{{\varGamma (r + k - 1)}}{{\varGamma (k - i + 1)\varGamma (r)}}{x^{(0)}}(i)} - \\ \sum\limits_{i = 1}^{k - i} {\frac{{\varGamma (r + k - i - 1)}}{{\varGamma (k - i)\varGamma (r)}}{x^{(0)}}(i)} , k = 2, 3, \cdots , n \end{array} $

则:

$ {z^{(r)}}(k) = \frac{{\sum\limits_{i = 1}^k {\frac{{\varGamma (r + k - i)}}{{\varGamma (k - i + 1)\varGamma (r)}}{x^{(0)}}(i) + \sum\limits_{i = 1}^{k - 1} {\frac{{\varGamma (r + k - i)}}{{\varGamma (k - i + 1)\varGamma (r)}}{x^{(0)}}(i)} } }}{2} $ (3)
$ \boldsymbol{Y} = \left[ {\begin{array}{*{20}{c}} {\sum\limits_{i = 1}^2 {\frac{{\varGamma (r + 2 - i)}}{{\varGamma (2 - i + 1)\varGamma (r)}}{x^{(0)}}(i) - \sum\limits_{i = 1}^1 {\frac{{\varGamma (r + 2 - 1 - i)}}{{\varGamma (2 - i)\varGamma (r)}}{x^{(0)}}(i)} } }\\ {\sum\limits_{i = 1}^3 {\frac{{\varGamma (r + 3 - i)}}{{\varGamma (3 - i + 1)\varGamma (r)}}{x^{(0)}}(i) - \sum\limits_{i = 1}^2 {\frac{{\varGamma (r + 3 - 1 - i)}}{{\varGamma (3 - i)\varGamma (r)}}{x^{(0)}}(i)} } }\\ \vdots \\ {\sum\limits_{i = 1}^n {\frac{{\varGamma (r + n - i)}}{{\varGamma (n - i + 1)\varGamma (r)}}{x^{(0)}}(i) - \sum\limits_{i = 1}^{n - 1} {\frac{{\varGamma (r + n - 1 - i)}}{{\varGamma (n - i)\varGamma (r)}}{x^{(0)}}(i)} } } \end{array}} \right] $
$ \boldsymbol{B} = \left[ {\begin{array}{*{20}{c}} { - \frac{1}{2}[\sum\limits_{i = 1}^2 {\frac{{\varGamma (r + 2 - i)}}{{\varGamma (2 - i + 1)\varGamma (r)}}{x^{(0)}}(i) + \sum\limits_{i = 1}^1 {\frac{{\varGamma (r + 1 - i)}}{{\varGamma (2 - i)\varGamma (r)}}{x^{(0)}}(i)} } ]}&1\\ { - \frac{1}{2}[\sum\limits_{i = 1}^3 {\frac{{\varGamma (r + 3 - i)}}{{\varGamma (3 - i + 1)\varGamma (r)}}{x^{(0)}}(i) + \sum\limits_{i = 1}^2 {\frac{{\varGamma (r + 2 - i)}}{{\varGamma (3 - i)\varGamma (r)}}{x^{(0)}}(i)} } ]}&1\\ \vdots & \vdots \\ { - \frac{1}{2}[\sum\limits_{i = 1}^n {\frac{{\varGamma (r + n - i)}}{{\varGamma (n - i + 1)\varGamma (r)}}{x^{(0)}}(i) + \sum\limits_{i = 1}^{n - 1} {\frac{{\varGamma (r + n - i - 1)}}{{\varGamma (n - i)\varGamma (r)}}{x^{(0)}}(i)} } ]}&1 \end{array}} \right] $

式(4)为分数阶算子EGM(1, 1)模型x(r-1)(k)+az(r)(k)=b的白化微分方程:

$ \frac{{{\rm d}{x^{(r)}}}}{{{\rm d}t}} + a{x^{(r)}} = b $ (4)

$\boldsymbol{Y}、\boldsymbol{B} 、\boldsymbol{\hat a}$如上文所述,$ \boldsymbol{\hat a} = {[a, b]^{\rm T}} = {({ \boldsymbol{B} ^{\rm T}} \boldsymbol{B})^{ - 1}}{ \boldsymbol{B}^{\rm T}} \boldsymbol{Y}$,则分数阶算子EGM(1, 1)模型的白化微分方程$\frac{{{\rm d}{x^{(r)}}}}{{{\rm d}t}} + a{x^{(r)}} = b$的时间响应函数为:

$ {\hat x^{(r)}}(t) = [{x^{(r)}}(1) - \frac{b}{a}]{{\rm e}^{ - at}} + \frac{b}{a} $ (5)

分数阶算子EGM(1, 1)模型x(r-1)(k)+az(r)(k)=b的时间响应序列为:

$ \begin{array}{c} {\hat x^{(r)}}(k) = [{x^{(0)}}(1) - \frac{b}{a}]{{\rm e}^{ - a(k - 1)}} + \frac{b}{a} \\ k = 2, 3, \cdots , n \end{array} $ (6)

模型还原值:

$ \begin{array}{l} {{\hat x}^{(0)}}(k) = {({{\hat x}^{(r)}}){( - r)}}(k)=\\ \sum\limits_{i = 0}^{k - 1} {{{( - 1)}^i}} \frac{{\varGamma (r + 1)}}{{\varGamma (i + 1)\varGamma (r - i + 1)}}\\{{\hat x}^{(r)}}(k - i) , k = 2, 3, \cdots , n\\ {\hat x^{(0)}}(1) = {x^{(0)}}(1) \end{array} $ (7)
2 粒子群算法选择EGM(1, 1)模型最优分数阶次

粒子群(particle swarm optimization, PSO)算法具有概念易理解、调整参数少、编程易实现等优点,在神经网络训练和函数优化等领域得到广泛的应用[12]

PSO算法的基本思路为:在搜索空间中初始化为一群随机粒子(随机解),然后通过迭代找到最优解。每一次迭代中,所有粒子都有一个由被优化的函数决定的适应度。基于适应度,粒子本身找到的最优解称为个体极值(Pbest), 整个种群目前找到的最优解为全局极值(Gbest)。在找到这2个最优值时,每个粒子基于个体和全局极值更新其速度和位置。每个粒子的速度和位置的更新表示为:

$ V_{id}^{K + 1} = wV_{id}^k + {c_1}{\alpha _1}(P_{\rm best} - X_{id}^k) + \\ {c_2}{\alpha _2}(G_{\rm best} - X_{id}^k),X_{id}^{k + 1} = X_{id}^k + V_{id}^{k + 1} $

式中,w为惯性权重因子,决定粒子先前速度对当前速度的影响程度,起到平衡算法全局搜索和局部搜索的作用,c1c2为控制粒子速度的加速因子,随机取2左右的值,α1α2为随机产生的在[0, 1]之间变化的加速度权重系数。每个粒子跟踪其先前的最佳位置和全局最佳位置,不断更新其速度和位置,直到迭代次数超过最大值。

最小平均相对误差下对分数阶算子EGM(1, 1)模型的最优分数阶次最优化问题为:

$ \begin{array}{c} \min f(r) = \\ \frac{1}{{n - 1}}\sum\limits_{k = 2}^n {\left| {\frac{{{{\hat x}^{(0)}}(k) - {x^{(0)}}(k)}}{{{x^{(0)}}(k)}}} \right|}, {\rm{ }}r \in {R^ + } \end{array} $ (8)

本文粒子群算法的运行参数设定如下:学习因子c1c2=2,动态惯性权重因子w=0.8,算法最大迭代次数为3 000,微粒范围r∈[0, 1],群体个体数目为50,精度设定为0.000 001。

粒子群算法优化的分数阶EGM(1, 1)模型的建模求解步骤如下:

1) 随机初始化粒子群中每个粒子的位置Xid1和速度Vid1, 可取Pbest=1,即EGM(1, 1)模型;

2) 将粒子中的Pbest设置为当前位置,Gbest设置为初始群体中最佳粒子的位置;

3) 计算模型分数阶r=Pbest时的分数阶算子EGM(1, 1)模型的拟合平均相对误差;

4) 更新粒子群中每个粒子的Vid1Xid1

5) 对每个粒子,如果其粒子适应度优于Pbest的适应度,将Pbest设置为新位置;如果粒子适应度优于Gbest的适应度,将Gbest设置为新位置;

6) 输出Gbest,即r的最优值,分数阶算子EGM(1, 1)模型的分数阶算子为r的最优值时,构建的模型拟合原始数据序列的平均相对误差值最小。

3 模型精度检验

由以上模型可以求出原始序列的拟合值为:

$ {\hat X^{(0)}}(k) = \left\{ {{{\hat x}^{(0)}}(1), {{\hat x}^{(0)}}(2), \cdots , {{\hat x}^{(0)}}(n)} \right\} $ (9)

$\varepsilon (k) = {x^{(0)}}(k) - {\hat x^{(0)}}(k), {\rm{ }}k = 2, \cdots , n$。残差可表示为:

$ E(k) = \left\{ {\varepsilon (2), \varepsilon (3), \cdots , \varepsilon (n)} \right\} $ (10)

相对误差序列为:

$ {\rm PE}({T_k}) = \frac{{\varepsilon (k)}}{{{x^{(0)}}(k)}} \times 100\% , {\rm{ }}k = 2, 3, \cdots , n $ (11)

平均相对误差为:

$ {\rm MAPE} = \frac{1}{{n - 1}}\sum\limits_{k = 2}^n {\left| {{\rm PE}(k)} \right|} $ (12)
4 优化模型实例分析应用 4.1 算例1

以新建的呼和浩特至准格尔铁路工程的沉降观测数据为例。以DK41+550号沉降监测点2015-07-02~13的12期累计沉降量数据作为建模数据,分别构建EGM(1, 1)模型和优化分数阶的EGM(1, 1)模型,并用构建的模型预测2015-07-14~17的4期沉降情况。2种模型的拟合和预测精度如表 1所示。

表 1 2种模型的拟合和预测结果与监测值的对比 Tab. 1 Fitting and predicting results of two models and comparison with monitoring values

EGM(1, 1)模型的时间响应序列为:

$ \begin{array}{c} {{\hat x}^{(1)}}(k) = - 106.207{{\rm e}^{0.063(k - 1)}} - 112.157\\ k = 2, 3, \cdots , n \end{array} $ (13)

优化分数阶EGM(1, 1)模型(r=0.480 492)的时间响应序列为:

$ \begin{array}{c} {{\hat x}^{(0.480 \;492)}}(k) = - 268.466{{\rm e}^{ - 0.013(k - 1)}} - 274.416\\ k = 2, 3, \cdots , n \end{array} $ (14)

从结果可以看出, EGM(1, 1)模型的最优阶次为0.480 492。优化模型的拟合和预测结果的精度相比EGM(1, 1)模型有了很大程度的提高,其构建的模型更符合原始数据数列的发展规律。

4.2 算例2

向家坡滑坡位于渝黔高速公路K13+500~960段。以文献[13]中滑坡监测点JB5的监测值为例,对其空间位移进行建模和预测。用前7期数据构建模型,预测后2期滑坡的位移量。精度评价结果如表 2所示。

表 2 2种模型的拟合和预测结果与监测值的对比 Tab. 2 Fitting and predicting results of two models and comparison with monitoring values

EGM(1, 1)模型的时间响应序列为:

$ \begin{array}{c} {{\hat x}^{(1)}}(k) = 168.694{{\rm e}^{0.102(k - 1)}} - 136.394\\ k = 2, 3, \cdots , n \end{array} $ (15)

优化分数阶EGM(1, 1)模型(r=0.244 999)的时间响应序列为:

$ \begin{array}{c} {{\hat x}^{(0.244\;999)}}(k) = 123.777{{\rm e}^{0.046(k - 1)}} - 156.077\\ k = 2,3, \cdots ,n \end{array} $ (16)

可以看出, 优化分数阶的EGM(1, 1)模型的拟合和预测精度明显优于EGM(1, 1)模型。

5 结语

最优阶次的选取是分数阶算子灰色预测模型的关键。本文通过粒子群算法选择拟合原始变形监测数据平均相对误差最小的分数阶次,构建粒子群算法优化的分数阶算子EGM(1, 1)模型。典型的变形监测算例分析表明,相比传统的灰色预测模型,优化模型对变形监测数据的拟合效果和预测精度更优,在变形监测领域中具有可行性和应用价值。

参考文献
[1]
尚军亮, 方敏. 一种优化的高精度灰色GM(1, 1)预测模型[J]. 电子与信息学报, 2010, 32(6): 1 301-1 305 (Shang Junliang, Fang Min. New Optimized Method of High-Precision Grey GM(1, 1) Forecasting Model[J]. Journal of Electronics and Information Technology, 2010, 32(6): 1 301-1 305) (0)
[2]
吉培荣, 黄巍松, 胡翔勇. 无偏灰色预测模型[J]. 系统工程与电子技术, 2000, 22(6): 6-7 (Ji Peirong, Huang Weisong, Hu Xiangyong. An Unbiased Grey Forecasting Model[J]. Systems Engineering and Electronics, 2000, 22(6): 6-7 DOI:10.3321/j.issn:1001-506X.2000.06.002) (0)
[3]
Li X C, Tao Y, Zheng Y. Model GM(1, 1, β) and Its Applicable Region[J]. Grey Systems: Theory and Application, 2013, 3(3): 266-275 DOI:10.1108/GS-04-2013-0005 (0)
[4]
李俊峰, 戴文战. 基于插值和Newton-Cores公式的GM(1, 1)模型的背景值构造新方法与应用[J]. 系统工程理论与实践, 2004, 24(10): 122-126 (Li Junfeng, Dai Wenzhan. A New Approach of Background Value-Building and Its Application Based on Data Interpolation and Newton-Cores Formula[J]. Systems Engineering-Theory and Practice, 2004, 24(10): 122-126 DOI:10.3321/j.issn:1000-6788.2004.10.021) (0)
[5]
党耀国, 刘思峰, 刘斌. 以x(1)(n)为初始条件的GM模型[J]. 中国管理科学, 2005, 13(1): 132-135 (Dang Yaoguo, Liu Sifeng, Liu Bin. The GM Models That x(1)(n) be Taken as Initial Value[J]. Chinese Journal of Management Science, 2005, 13(1): 132-135 DOI:10.3321/j.issn:1003-207X.2005.01.025) (0)
[6]
徐华锋, 刘思峰, 方志耕. GM(1, 1)模型灰色作用量的优化[J]. 数学的实践与认识, 2010, 40(2): 26-32 (Xu Huafeng, Liu Sifeng, Fang Zhigeng. Optimum Grey Action Quantity for GM(1, 1)[J]. Journal of Mathematics in Practice and Theory, 2010, 40(2): 26-32) (0)
[7]
Zhang J, Ran M F, Han G D, et al. Grey Modeling Based on the Transformation of Aarc cot x+B Function[J]. Grey Systems: Theory and Application, 2015, 5(2): 157-164 DOI:10.1108/GS-09-2014-0031 (0)
[8]
Cui J, Liu S F, Zeng B, et al. A Novel Grey Forecasting Model and Its Optimization[J]. Applied Mathematical Modeling, 2013, 37(6): 4 399-4 406 DOI:10.1016/j.apm.2012.09.052 (0)
[9]
孙辰军, 王翠茹, 张江维, 等. 残差灰色预测模型的改进与应用[J]. 统计与决策, 2005(5): 19-20 (Sun Chenjun, Wang Cuiru, Zhang Jiangwei, et al. Improvement and Application for Grey Prediction Model of Residual Error[J]. Control and Decision, 2005(5): 19-20 DOI:10.3969/j.issn.1002-6487.2005.05.010) (0)
[10]
吴利丰.分数阶灰色预测模型及其应用研究[D].南京: 南京航空航天大学, 2015 (Wu Lifeng. Fractional Order Grey Forecasting Models and Their Application[D]. Nanjing: Nanjing University of Aeronautics and Astronautics, 2015) http://cdmd.cnki.com.cn/Article/CDMD-10287-1016752024.htm (0)
[11]
王鸣翠, 于胜文, 张帅帅, 等. 基坑变形非等时距灰色预测模型程序设计及应用[J]. 测绘地理信息, 2015, 40(1): 41-44 (Wang Mingcui, Yu Shengwen, Zhang Shuaishuai, et al. Programming and Application for Non-Equal Interval Grey Model of Foundation Pit Deformation[J]. Journal of Geomatics, 2015, 40(1): 41-44) (0)
[12]
倪庆剑, 邢汉承, 张志政, 等. 粒子群优化算法研究进展[J]. 模式识别与人工智能, 2007, 20(3): 349-357 (Ni Qingjian, Xing Hancheng, Zhang Zhizheng, et al. Research Progress of Particle Swarm Optimization[J]. Pattern Recognition and Artificial Intelligence, 2007, 20(3): 349-357 DOI:10.3969/j.issn.1003-6059.2007.03.010) (0)
[13]
尹光志, 张卫中, 张东明, 等. 基于指数平滑法与回归分析相结合的滑坡预测[J]. 岩土力学, 2007, 28(8): 1 725-1 728 (Yin Guangzhi, Zhang Weizhong, Zhang Dongming, et al. Forecasting of Landslide Displacement Based on Exponential Smoothing and Nonlinear Regression Analysis[J]. Geotechnical Mechanics, 2007, 28(8): 1 725-1 728) (0)
Application of Optimized Fractional Order EGM (1, 1) Model in Deformation Monitoring and Forecasting
YUAN Debao1     ZHANG Zhenchao1     ZHANG Jun2     ZAHNG Jian1     
1. College of Geoscience and Surveying Engineering, China University of Mining and Technology, D11 Xueyuan Road, Beijing 100083, China;
2. College of Science, Inner Mongolia Agricultural University, 306 Zhaowuda Road, Hohhot 010018, China
Abstract: In view of the unsatisfactory fitting and prediction accuracy of deformation monitoring data series, we propose a fractional order EGM (1, 1) model, optimized by particle swarm optimization, to fit and predict deformation monitoring data. We use particle swarm optimization (PSO) to select the fractional order, which fits the minimum average relative error of EGM (1, 1), and the optimal fractional order EGM (1, 1) model is constructed. We use typical deformation monitoring data to validate the optimization model. The results show that the optimization model achieves high accuracy in fitting and predicting deformation monitoring data. It shows that the optimization model is feasible and effective in processing deformation monitoring data.
Key words: fractional order operator; grey model; particle swarm optimization; deformation monitoring