Method of moments for estimating uncertainty distributions
- Xiaosheng Wang1Email author and
- Zixiong Peng2
https://doi.org/10.1186/2195-5468-2-5
© Wang and Peng; licensee Springer. 2014
Received: 9 February 2014
Accepted: 13 February 2014
Published: 6 March 2014
Abstract
Uncertainty theory is a branch of mathematics for modeling human uncertainty, and uncertain statistics is a methodology for collecting and interpreting expert’s experimental data by uncertainty theory. In order to estimate uncertainty distributions via experts’ experimental data, this paper will present a method of moments and design a numerical method to find moment estimate of unknown parameters.
Keywords
Introduction
Probability theory is a branch of mathematics concerned with analysis of random phenomena. The probability distribution describes the range of possible values and the probability of a random variable. In order to determine the probability distribution, random statistics (i.e., the classical mathematical statistics) was proposed as a methodology for collecting and interpreting the test data with correlative information in a system by probability theory. On the other hand, the fuzzy set theory via membership function initiated by Zadeh [1] is a mathematical model of vague qualitative or qualitative data, frequently generated by means of natural language. In order to determine the membership function, fuzzy statistics was presented including fuzzy point estimation by a fuzzy decision-making approach and so on.
However, a lot of surveys showed that some information and knowledge represented usually by human language like ‘about 1,000 km,’ ‘roughly 60 kg,’ ‘high speed,’ and ‘small size’ are neither randomness nor fuzziness. When the sample size is too small (even no-sample) to estimate a probability distribution, we have to invite some domain experts to evaluate their belief degree that each event will occur. Since human beings usually overweigh unlikely events, the belief degree may have much larger variance than the real frequency. Perhaps some people think that the belief is a subjective probability or fuzzy membership degree. However, the examples [2] tell us that it is inappropriate because probability theory and fuzzy set theory may lead to counterintuitive results in this case.
In order to model this type of imprecise quantities, uncertainty theory was founded by Liu [3] in 2007 and refined by Liu [4] in 2010 and became a branch of mathematics based on the normality, duality, subadditivity, and product axioms. Based on Liu’s uncertainty theory, some basic and important theoretical work of uncertainty theory such as uncertain process [5], uncertain calculus [6], uncertain differential equation [6], uncertain logic [7], uncertain inference [8], and uncertain risk analysis [9] have been established. Meanwhile, as an application of uncertainty theory, Liu [10] proposed a spectrum of uncertain programming and applied uncertain programming to system reliability design, facility location problem, vehicle routing problem, project scheduling problem, and so on. Other references related to uncertainty theory are Gao [11], Gao et al. [12], Peng and Iwamura [13], and Liu and Ha [14]. Nowadays, uncertainty theory was well developed on both theory section and practice section. For exploring the recent developments of uncertainty theory, the readers may consult the book [15].
One important issue in uncertainty theory is how to determine uncertainty distribution of an uncertain variable. In order to answer this question, uncertain statistics was firstly presented by Liu [4] in 2010, and is a methodology for collecting and interpreting expert’s experimental data by uncertainty theory. In uncertain statistics, Liu [4] first suggested an empirical uncertainty distribution and proposed a principle of least squares as the method for estimating the unknown parameters based on the expert’s experimental data. After that, Chen and Ralescu [16] employed uncertain statistics to estimate the travel distance between Beijing and Tianjin. In 2012, Wang et al. recast the Delphi method as a process to determine the uncertainty distributions [17] and presented a method of uncertain hypothesis testing for determining if the views of two domain experts are identical (i.e., wether or not they have the same uncertainty distribution) [18].
Based on experts’ experimental data and the empirical uncertainty distribution, the k th sample moment is defined in this paper. Then, a method of moments for estimating the unknown parameters of uncertainty distribution is proposed. The remainder of this paper is organized as follows. The next section is intended to introduce some concepts in uncertainty theory as they are needed. Some basic concepts of uncertain statistics are introduced in Section ‘Beginning of uncertain statistics.’ The uncertain moment method for estimating parameters is proposed in Section ‘Method of moments.’ A numerical method to find the moment estimate of unknown parameters is designed in Section ‘Numerical method.’ Finally, a conclusion is drawn in Section ‘Conclusions.’
Preliminaries
In this section, we will introduce some useful definitions about uncertain measure, uncertain variable, uncertain moment, and so on.
Let Γ be a nonempty set, and L be a σ-algebra over Γ. Each element Λ∈L is called an event. A number M(Λ) indicates the level that Λ will occur. Uncertain measure M was introduced as a set function satisfying the following three axioms (Liu [3]):
Axiom 1. M{Γ}=1.
Axiom 2. M{Λ}+M{Λ c }=1 for any event Λ.
Although probability measure satisfies the above three axioms, probability theory is not a special case of uncertainty theory because product probability measure does not satisfy the following product axiom (Liu [3]):
where Λ k are arbitrarily chosen events from L k for k=1,2,…, respectively.
provided that at least one of the two integrals is finite. The k th moment of an uncertain variable ξ is defined by E[ξ k ], where k is a positive integer.
In addition, Liu [4] proved a series of operational laws of independent uncertain variables for calculating the uncertainty distribution of monotone function of uncertain variables. By these operational laws, the uncertainty distributions of sum, difference, product, quotient, maximum, and minimum for n uncertain variables are easy obtained. The detail contents are shown in [4] or [15].
Beginning of uncertain statistics
Uncertain statistics is based on the expert’s experimental data rather than historical data. One question is how to obtain expert’s experimental data. Liu [4] designed a questionnaire survey for collecting expert’s experimental data, that is, we invite one or more domain experts who are asked to complete a questionnaire about the meaning of an uncertain variable ξ like ‘about 1,000 km’ individually.
We first ask the domain expert to choose a possible value x that the uncertain variable ξ may take. Then, we quiz him ‘how likely is ξ less than x?’ and denote his belief degree by α. Thus, we obtain an expert’s experimental data (x, α) from the domain expert. Repeating the above process, we obtain the expert’s experimental data.
where the unknown parameters θ i , i=1,2,…, p are the solution of the minimization problem.
where .
Method of moments
In this section, a method of moments based on expert’s experimental data is proposed to estimate the unknown parameters. Firstly, we present the k th moment of the empirical uncertainty distribution (2), which is the uncertainty distribution of some uncertain variable ξ.
The theorem is proved.
Definition 1. Let ξ be an uncertain variable. For any positive integer k , if the k th moment of ξ exists, then E[ξ k ] is called the k th theoretical moment of ξ.
is called the k th sample moment.
Secondly, we introduce a method of moments for estimating the unknown parameter, the description of the method is presented as follows.
A solution of the equations, θ1, θ2,…, θ p , is called a moment estimate. We denote the estimate values by , respectively.
From Equations 7 and 14, we have moment estimate values , and the moment estimate distribution is Φ(x)=0.2442x−0.0519.
Numerical method
In this section, we introduce Newton’s method to estimate the moment estimate value. A general algorithm is proposed to solve the parameter estimate problem as follows.
where are the sample moments, which are constants calculated by Equation 7.
The Equation 8 is equivalent to F(θ1, θ2,…, θ p )=0.
If the function F is differentiable and the Jacobian matrix of F is easily computed, the procedure can be summarized as follows:
Step 1. Estimate the k th sample moments by Equation (7), k=1,2,…, p.
Step 2. Give an initial value of parameters .
The integrals in the equation can be computed by some numerical methods.
Step 5. Repeat the third to fourth steps when |θk+1−θ k |>α and k<M, where α and M are given positive numbers.
Step 6. Report the last θ n as the solution, if |θ n −θn−1|≤α.
where i, j=1,2,…, p, and h1, h2,…, h p are given small positive numbers.
where . All of these integrals are computed by Simpson formula, and the initial value is (0.6,0.5).
Conclusions
Uncertain statistics is a methodology for collecting and interpreting expert’s experimental data by uncertainty theory. The method of moments in this paper is a new way to estimate the unknown parameters in uncertainty distribution. This method depends on the expert’s experimental data and is easy to be realistic in the real experiment.
Declarations
Acknowledgements
This work was supported by National Natural Science Foundation of China (No. 61073121) and Hebei Natural Science Foundation (No. G2013402063 and No. F2012402037).
Authors’ Affiliations
References
- Zadeh L: Fuzzy sets. Inform. Control 1965, 8: 338–353. 10.1016/S0019-9958(65)90241-XMathSciNetView ArticleMATHGoogle Scholar
- Liu B: Why is there a need for uncertainty theory? J. Uncertainty Syst 2012, 6(1):3–10.Google Scholar
- Liu B: Uncertainty theory. Berlin: Springer; 2007.View ArticleMATHGoogle Scholar
- Liu B: Uncertainty Theory: a Branch of Mathematics for Modeling Human Uncertainty. Berlin: Springer; 2010.View ArticleGoogle Scholar
- Liu B: Fuzzy process, hybrid process and uncertain process. J. Uncertain Syst 2008, 2(1):3–16.Google Scholar
- Liu B: Some research problems in uncertainty theory. J. Uncertain Syst 2009, 3(1):3–10.Google Scholar
- Li X, Liu B: Hybrid logic and uncertain logic. J. Uncertain Syst 2009, 3(2):83–94.Google Scholar
- Liu B: Uncertain set theory and uncertain inference rule with application to uncertain control. J. Uncertain Syst 2010, 4(2):83–98.Google Scholar
- Liu B: Uncertain risk analysis and uncertain reliability analysis. J. Uncertain Syst 2010, 4(3):163–170.Google Scholar
- Liu B: Theory and Practice of Uncertain Programming. Berlin: Springer; 2009.View ArticleMATHGoogle Scholar
- Gao X: Some properties of continuous uncertain measure. Int. J. Uncertainty, Fuzziness Knowl. Based Syst 2009, 17(3):419–426. 10.1142/S0218488509005954MathSciNetView ArticleMATHGoogle Scholar
- Gao X, Gao Y, Ralescu D: On Liu’s inference rule for uncertain systems. Int. J. Uncertainty, Fuzziness Knowl. Based Syst 2010, 18(1):1–11. 10.1142/S0218488510006349MathSciNetView ArticleMATHGoogle Scholar
- Peng Z, Iwamura K: A sufficient and necessary condition of uncertainty distribution. J. Interdiscip. Math 2010, 13(3):277–285. 10.1080/09720502.2010.10700701MathSciNetView ArticleMATHGoogle Scholar
- Liu Y, Ha M: Expected value of function of uncertain variables. J. Uncertainty Syst 2010, 4(3):181–186.Google Scholar
- Liu B: Uncertainty Theory. Berlin: Springer; . http://orsc.edu.cn/liu/ut.pdf
- Chen X, Ralescu D: B-spline method of uncertain statistics with application to estimating distance. J. Uncertain Syst 2012, 6(4):256–262.Google Scholar
- Wang X, Gao Z, Guo H: Delphi method for estimating uncertainty distributions. Inform.: Int. Interdiscip. J 2012, 15(2):449–460.MathSciNetMATHGoogle Scholar
- Wang X, Gao Z, Guo H: Uncertain hypothesis testing for two experts’ empirical data. Math. Comput. Modell 2012, 55: 1478–1482. 10.1016/j.mcm.2011.10.039MathSciNetView ArticleMATHGoogle Scholar
Copyright
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.