Abstract
This paper focuses on the problem of monitoring/estimating process parameters in the insufficient case when only imprecise and uncertain information can be obtained, possibly due to limited precision and reliability of sensors in industries. To solve this problem, a constrained fuzzy evidential multivariate model is proposed as a soft sensor to monitor imprecise and uncertain process parameters. The most challenging task involved in the modeling is how to identify structure parameters of the monitor model, especially under sets of constraints. To tackle this challenge, we represent the imprecise and uncertain information as fuzzy belief functions in the evidence theory framework, and then propose a restricted fuzzy evidential Expectation-Conditional Maximization algorithm (RFE2CM) for maximum likelihood estimation from fuzzy belief functions under linear inequality constraints. Also, the convergence property of the restricted fuzzy evidential EM algorithm is discussed. In order to validate the performance of the proposed model and algorithm, some numerical simulations are conducted as well as an experimental simulation on a real ball mill in a power plant. The numerical and experimental simulation results show that the proposed model and algorithm can not only be feasibly applicable to monitor the process parameters in insufficient informatics cases, but also have high prediction accuracy with small mean square errors.
Similar content being viewed by others
Notes
The symbol “\(\rightarrow \)” means “approaching”. Notice that x cannot be one in practice, because “\(x = 1\)” means there is no gap inside the ball mill. This will never occur in practice.
Around the 4269th second, there are some fine coal powder overflowing from the inlet of ball mill and the ball mill thus achieving the over load condition. After the 4269th second, the operators reduce the load of feed coal slowly in order to make the level of coal powder maintain at a high-level case. Therefore, the level of coal powder associated to the running data before the 4269th second covers its whole domain in some sense.
References
Davidov O, Rosen S (2011) Constrained inference in mixed-effects models for longitudinal data with application to hearing loss. Biostatistics 12(2):327–340
Dempster AP (1967) Upper and lower probabilities induced by a multivalued mapping. Ann Math Stat 38(2):325–339
Dempster AP (1968) Upper and lower probabilities generated by a random closed interval. Ann Math Stat 39(3):957–966
Dempster AP, Laird NM, Rubin DB (1977) Maximum likelihood from incomplete data via the EM algorithm. J R Stat Soc B 39:1–38
Denoeux T (2011) Maximum likelihood estimation from fuzzy data using the EM algorithm. Fuzzy Sets Syst 183:72–91
Denoeux T (2013) Maximum likelihood estimation from uncertain data in the belief function framework. IEEE Trans Knowl Data Eng 25(1):119–130
Dykstra RL, Robertson T, Silvapulle MJ (eds) (2002) Statistical inference under inequality constraints (special issue). J Stat Plan Inference 107:1–2
Fessler JA, Hero AO (1994) Space-alternating generalized expectation-maximization algorithm. IEEE Trans Signal Process 42:2664–2677
Hoferkamp C, Peddada SD (2002) Parameter estimation in linear models with heteroscedastic variances subject to order restrictions. J Multivar Anal 82:65–87
Kim DK, Taylor JMG (1995) The restricted EM algorithm for maximum likelihood estimation under linear restrictions on the parameters. J Am Stat Assoc 430:708–716
Liew CK (1976) Inequality constrained least-squares estimation. J Am Stat Assoc 71:746–751
Little RJA, Rubin DB (1994) The ECME algorithm: a simple extension of EM and ECM with faster monotone convergence. Biometrika 81:633–648
Meng XL, van Dyk DV (1997) The EM algorithm–an old folk song sung to a fast new tune. J R Stat Soc Ser B 59:511–567
Meng XL, Rubin DB (1991) Using EM to obtain asymptotic variance-covariance matrices: the SEM algorithm. J Am Stat Assoc 86:899–909
Shafer G (1976) A mathematical theory of evidence. Princeton University Press, Princeton
Shi NZ, Zheng SR, Guo JH (2005) The restricted EM algorithm under inequality restrictions on the parameters. J Multivar Anal 92:53–76
Shi NZ, Jiang H (1998) Maximum likelihood estimation of isotonic normal means with unknown variances. J Multivar Anal 64:183–195
Smets Ph (1981) The degree of belief in a fuzzy event. Inf Sci 25(1):1–19
Smets Ph (1990) The combination of evidence in the transferable belief model. IEEE Trans Pattern Anal Mach Intell 12(5):447–458
Smets Ph (1991) Varieties of ignorance. Inf Sci 57–58:135–144
Su ZG, Wang PH, Yu XJ, Lv ZZ, Lu L (2011) Multi-model strategy based evidential soft sensor model for predicting evaluation of variables with uncertainty. Appl Soft Comput 11(2):2595–2610
Su ZG, Wang PH, Wang YF (2012) Interval-valued EM algorithm with application to estimating parameters. In: Proceedings of the world congress on engineering and computer science, vol. I, San Francisco, USA, pp 551–556
Su ZG, Wang PH, Song ZL (2013a) Kernel based nonlinear fuzzy regression model. Eng Appl Artif Intell 28(2):724–738
Su ZG, Wang YF, Wang PH (2013b) Parametric regression analysis of imprecise and uncertain data in the fuzzy belief function framework. Int J Approx Reason 54(8):1217–1242
Su ZG, Zheng SR, Wang PH (2014) Likelihood-based multivariate fuzzy model with linear inequality constraints. J Intell Fuzzy Syst 27(5):2194–2209
Su ZG, Wang PH (2014) Regression analysis of belief functions on interval-valued variables: comparative studies. Soft Comput 18(1):51–70
Wu CFJ (1983) On the convergence properties of the EM algorithm. Ann Stat 11:95–103
Yager RR (1982) Generalized probabilities of fuzzy events from fuzzy belief structures. Inf Sci 28(1):45–62
Yager RR (1996) On the normalization of fuzzy belief structure. Int J Approx Reason 14(2–3):127–153
Yen J (1990) Generalizing the Dempster-Shafer theory to fuzzy sets. IEEE Trans Syst Man Cybern 20(3):559–569
Zangwill WI (1969) Nonlinear programming: a unified approach. Prentice-Hall, Englewood Cliffs
Zheng SR, Guo JH, Shi NZ, Tian GL (2012) Likelihood-based approaches for multivariate linear models under inequality constraints for incomplete data. J Stat Plan Inference 142:2926–2942
Acknowledgments
The authors are grateful to the contributions of Professor Thierry Denoeux to our work. The authors also thank the significant contributions of the editors and the two anonymous referees.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors declare that they have no conflict of interests.
Additional information
Communicated by V. Loia.
This work is supported by the National Natural Science Foundation of China (No. 51106025, 51476028) and in part by the Fundamental Research Funds for the Central Universities (No. 2242014R30002).
Appendices
Appendix 1: Calculations of \(\mathbf {z}^{( q)}={\mathtt {E}}_{\varvec{\psi } ^{(q)}} ( {\left. \mathbf {z} \right| {pl}})\),\(\varvec{\alpha } _i^{( q)} ={\mathtt {E}}_{\varvec{\psi } ^{(q)}} ( {\left. {\mathbf {x}_i \mathbf {{x}'}_i } \right| {pl}_i })\) and \(\beta _i^{( q)} ={\mathtt {E}}_{\varvec{\psi } ^{(q)}} ( {\left. {\mathbf {x}_i } \right| {pl}_i })\) in E-step of RFE2CM algorithm.
Let us assume that P is the distribution of a univariate normal random variable \(X_{ij}\) with mean v, standard deviation \(\sigma \) and p.d.f. g(x). A piece of (marginal) FBA \(m_{ij}\) on \( X_{ij}\), holding \(r_{i}\) focal elements \(F(m_{ij})\) = {\(\tilde{A}_1 \), \(\tilde{A}_2 \), ..., \(\tilde{A}_{r_i } \)}, is used to represent our partial and uncertain information about the realization \(x_{ij}\). Let \(\forall \tilde{A}=( {a,b,c})\in F( {m_{ij} })\)be a triangular fuzzy number, with membership function:
Denoting by g(x) the p.d.f. of \(X_{ij}\) with parameter \(\varvec{\psi }\) = (v, \(\sigma \))’, we have
which can be completed by
where \(\Phi \)(.) denotes the c.d.f. of the standard normal distribution, and \(x_{ij}^{*}\) denotes (\(x_{ij}\) – v)/ \(\sigma \) for all \(x_{ij}\). It is easy to show that
Eq. (62) makes it possible to complete the Eq. (70).
Let us now compute the expectation of \(X_{ij}\) given \({pl}_{ij} \)for the conditional density of \(X_{ij}\). We have
where the denominator is given by (61), and the numerator is
which can be computed using Eq. (62)and
Therefore, we have
Because \({\varvec{\bar{\mathrm{x}}}}=( {{\varvec{\bar{\mathrm{x}'}}}_1 ,\varvec{{\bar{\mathrm{x}}'}}_2 ,\ldots ,{\varvec{\bar{\mathrm{x}'}}}_t })^\prime \) and \(\mathbf {z}={{\varvec{\bar{\mathrm{U}}}}'D\bar{x}}\), we have
We finally compute
The numerator in (68) is
which can be computed using (65) and
Then, we have
Appendix 2
Proof of Theorem 1
The Lagrange function corresponding to optimization problem (31) can be given by
where \(\varvec{\lambda } \) is a non-negative l-dimensional vector.
Let \(\varvec{\theta } \) be the solution to (31), we have
If B \(_{1}\) \(\varvec{\theta } \) \(>\) a \(_{1}\) and B \(_{2}\) \(\varvec{\theta } \) = a \(_{2}\), then (74) implies
where \(\varvec{\lambda } \) \(_{1}\) and \(\varvec{\lambda } \) \(_{2}\) are, respectively, \(l_{1}\)- and \(l_{2}\)-dimensional vectors such that \(l_{1}\)+ \(l_{2}=l\).
Therefore, we get \(\varvec{\lambda } \) \(_{1}\) = 0 and \(\varvec{\lambda } \) \(_{2}\) \(\ge \) 0. From (72), we have
Multiplying B \(_{2}\) on both sides of (76), we have
From (77), we can obtain the solution of \(\varvec{\lambda } \) \(_{2}\) as following:
By inserting (78) into (77), we can obtain the final solution to (31):
In addition, by Kuhn–Tucker Theorem, we have B \(_{1}\) \(\varvec{\theta } \) - a \(_{1} \quad >\)0 and \(\varvec{\lambda } \) \(_{2} \ge \) 0, i.e.,
\(\square \)
Proof of Proposition 1
According to Theorem 1, the MLEs of parameters with linear inequalities is
We have
Then
where \(\mathbf {K}_1 =( {\varvec{\Lambda }^{( q)}})^{-1/2}( {\mathbf {I}_k -\mathbf {{B}'}_2 ( {\mathbf {B}_2 \mathbf {{B}'}_2 })^{-1}\mathbf {B}_2 })( {\Lambda ^{( q)}})^{-1/2}{{\varvec{\bar{\mathrm{U}}}}'{\mathbf {D}}}^{( q)}{\varvec{\bar{\mathrm{U}}}}\), and\(\varvec{K}_2 =( {\varvec{\Lambda }^{( q)}})^{-1/2}\mathbf {{B}'}_2 ( {\mathbf {B}_2 \mathbf {{B}'}_2 })^{-1}\mathbf {a}_2 \). \(\square \)
Rights and permissions
About this article
Cite this article
Hao, Ys., Su, Zg., Wang, Ph. et al. Constrained fuzzy evidential multivariate model identified by EM algorithm: a soft sensor to monitoring imprecise and uncertain process parameters. Soft Comput 21, 1619–1642 (2017). https://doi.org/10.1007/s00500-015-1948-2
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00500-015-1948-2