Generalised and instance-specific modelling for biological systems

https://doi.org/10.1016/S1364-8152(98)00104-2Get rights and content

Abstract

Biological, ecological and environmental systems are difficult to model. Due to the complexity of the relevant entities their behaviour is hard to capture. Moreover, due to the evolutionary behaviour of the biological entities and assumption-dependent non-linearities, models are valid in only a limited condition range and time frame. This paper introduces the concepts of generalised and instance-specific models, and their relevance for biological, ecological and environmental systems. For each concept, a modelling approach from the field of AI is introduced. The concepts are illustrated in two modelling applications in the waste water domain, one for opportunistic modelling and one for resource allocation.

Introduction

Biological, ecological and environmental (BEE) processes are becoming more and more important in policy- and decision-making. The recent Kyoto summit has defined strategic targets to be realised by countries world-wide. On a smaller scale, regional officials have to predict, monitor and control environmental phenomena such as smog, manure, eutrophication and acid rain on local ecosystems, while on a micro-scale, such as waste- and surface water systems, operators have to maintain operational set-points by controlling complex processes. Moreover, biological processes are more and more applied on an industrial scale.

Decision and policy quality strongly depend on model validity. The complexity of BEE systems complicates the development of reliable models for use in the real world. This often leads to the deployment of models that are simplified and stripped versions of the models that would be required. Artificial intelligence techniques can be used to improve the modelling of BEE processes.

In the next section we explore problems in traditional modelling approaches for the complex systems that are found in natural and artificial biological systems. Then, in Section 3, we discuss differences between generalised and instance-specific models, and present two artificial intelligence approaches for these kinds of problems. Section 4presents for both types of solutions an example, as implemented in an activated sludge waste water system. Section 5concludes by discussing these approaches, and generalising the concepts to biological, ecological and environmental systems in general.

Section snippets

Context

Modelling BEE systems introduces a number of specific problems. A model is a representation of a (real-world) entity that mimics some aspects of the behaviour of the referent entity. Realistic systems consist of many individual entities of different kinds, each with their own behaviour. Growth, decline and extinction behaviour of organisms, as well as the behaviour of large numbers of biological processes, depend on many variables that show autonomous dynamics. Even for describing simple

AI modelling techniques

In policy and decision making models help to optimise the decision quality. The models are typically used for simulation and optimisation. Models may play different roles, depending of the way of modelling. For our paper the difference between a generalised model and an instance-specific model is important. The former employs techniques known as compositional modelling (Section 3.1), the latter are realised through case based reasoning (Section 3.2).

Case studies

This section presents two case studies in the domain of waste water treatment. In Section 4.1we give a short introduction of this domain. Section 4.2presents an overview of the WaterCIME project. In this project, the modelling concepts presented in this paper have been developed and field tested in two applications. These applications are presented in Section 4.3Section 4.4.

Conclusions

Ecological systems require flexible approaches for modelling and control to cope with typical aspects of biological systems:

  • complexity

  • process dynamics

  • process evolution

  • the possibility to have different views on one process.

Existing modelling and control technology is not optimally suited to cope with the typical problems in this domain. Two typical problems where AI technology can improve the functionality of existing software are discussed in this paper: modelling of biological processes by

Acknowledgements

The authors thank the WaterCIME partners for their close co-operation during the project, and the EU for supporting this research. Special thanks to Thorsten Abels and Peter Dannenmann (both from DASA-RI) and Christoph Bernatzky (of BEB).

References (13)

  • B. Falkenhainer et al.

    Compositional modelling: finding the right model for the job

    Artificial Intelligence

    (1991)
  • A. Aamodt et al.

    Case based reasoning: foundational issues, methodological variations and system approaches

    AI Communications

    (1994)
  • A.P.J. Breunese et al.

    Libraries of reusable models: theory and application

    Simulation

    (1998)
  • Broeze, J., Verdenius, F., Sloof, M., Rekswinkel, E., van der Sluis, H., 1997. Advanced modelling of waste water...
  • Henze, M., Gujer, W., Mino, T., Matsuo, T., Wentzel, M.C., Marais, G.v.R., 1995. Activated sludge model No. 2. IAWPRC...
  • Kohonen, T., 1995. Self-organizing Maps. Springer-Verlag,...
There are more references available in the full text version of this article.

Cited by (16)

  • Modeling of 'Gala' apple fruits diameter for improving the accuracy of early yield prediction

    2013, Scientia Horticulturae
    Citation Excerpt :

    On the other side, the prediction based on heat accumulation, which is calculated as the sum of the daily minimum and the maximum temperature from the date of full flowering to harvest shows to be relative successful tool (Streif, 1996; Stanley et al., 2000). Moreover, due to the evolutionary behavior of the biological individuals and assumption-dependent non-linear characteristics, models are valid in only a limited condition range and time frame (Verdenius and Broeze, 1999). On the other hand, more advanced piecewise power function and spline exponential function promised sufficient accuracy for predicting fruit mass from the fruit diameter (De Silva et al., 1997).

  • Integrating case-based and fuzzy reasoning to qualitatively predict risk in an environmental impact assessment review

    2009, Environmental Modelling and Software
    Citation Excerpt :

    King et al. (1999) used CBR to aid the design of process sequences aimed at separating ternary mixtures containing azeotropes. Verdenius and Broeze (1999) devised a CBR system to control the activated sludge process at a real-world wastewater treatment plant. Kalapanidas and Avouris, 2001 proposed the NEMO CBR system to support short-term prediction of NO2 maximum concentration levels in Athens.

View all citing articles on Scopus
View full text