Vous êtes sur la page 1sur 12

Model Based Manufacturing

Consolidated Review of Research and Applications

8. Planning and Scheduling in the Process


Industry
Mario Stobbe, Thomas Löhl, Christian Schulz and Sebastian Engell
Process Control Laboratory, University of Dortmund

8.1. Introduction
8.1.1 Distinction between planning and scheduling
Usually, planning means the generation of production tasks for longer periods of time, given
forecasts for prices and product demands. In contrast, scheduling means the assignment of
resources to activities, sequencing of activities and determination of starting respectively ending
times for the execution over a short period of time.
Planning and scheduling are closely related as the decisions made at the planning level have a
strong influence on scheduling. Ideally, the availability of resources would be taken into account
fully on the planning level and thus scheduling would just be a part of planning. The complexity
of the overall problem, however, in most cases excludes the possibility of detailed planning over
long horizons and thus the two tasks are treated sequentially. From a methodological point of
view, there is not much difference between the two problems.

8.1.2 Characteristics of the problem


This section gives some characteristics of the scheduling and planning problem rather than
proposing a classification, as any set of listed constraints and properties defines a unique
problem which has to be treated in a different manner.
Scheduling and planning problems can roughly be divided into flow-shop, job-shop or open-shop
problems. For chemical processing plants, this characterisation is not sufficient since additional
constraints and specific characteristics have to be taken into account which result from features
such as

• shared resources

• continuous units

• limited connectivity by piping networks.

Moreover, the way the plant is operated as well as features of the recipes may impose
characteristics, which have to be taken into account:

• Change-over procedures

• Maintenance procedures
• Recycling streams

• Scalable batch sizes

The properties of the substances of material cause additional constraints such as limited storage
time or limited storage capacity due to special requirements concerning how the material must be
stored.
The variety of problems makes it difficult to define a general model and to solve the problem by
a general algorithm.

8.2 Scheduling Methods


Numerous contributions to the scheduling and planning problem were made by different research
communities as e. g. Operations Research and Artificial Intelligence. This led to different
models, different terminologies and impedes practitioners from exploiting this knowledge.
Most contributions to scheduling and planning problems, no matter wether originating from the
Operations Research (OR) or the Artificial Intelligence (AI) community, aim either at new
methods which were recently developed or architectures for scheduling respectively planning
systems. Only a few contributions provide information about implementation issues or mention
if the method is used in practice at all. Therefore, this section only outlines recent solution
methods and assesses the approaches with respect to modelling effort and known advantages and
disadvantages.
Several reviews have to be mentioned:

• [Rek96] gives a comprehensive overview of solution techniques and a description of the


elements of the scheduling problem.

• In [Wie97] the applicability of some solution methods is critically reviewed.

• A review of scheduling systems is given in [ME90]

8.2.1 Heuristics
Most scheduling problems belong to the class of NP-complete problems even when
simplifications in comparison to practical problems are introduced. Therefore it was often
argued, that only the use of problem specific heuristics can lead to efficient solution procedures.

Examples
A critical path method based on a heuristic algorithm is proposed in [Jän84] for the short term
scheduling of multipurpose plants.
Heuristics drawn from serial flowshop and parallel network problems are used to solve the
sequencing problem [KR89]. Namely dispatching rules, bottleneck-, local search- and best-fit
sequencing.
In [KEP+94] a heuristic approach for multiproduct batch and semicontinuous plants is reported,
where finite intermediate storage is available. In the initial step, orders are scheduled
sequentially due to their priorities. By employing heuristics, for example by aggregating orders
of the same product to reduce switchover costs, the schedule is improved.
Advantages and Disadvantages
Heuristics have the disadvantage that no measure of quality can be given to judge the reached
solutions. The main advantage is that the algorithms incorporate problem specific knowledge
which often leads to good solutions, which are obtained in an acceptable amount of time.

8.2.2 Mathematical Programming


Types of models
For scheduling problems, mixed integer formulations are widely used either with linear
constraints and a linear objective function only (MILP) (e. g. [KK88], [PG94]) or with non-linear
constraints and/or a non-linear objective function (MINLP) (e. g. [SG91a]).
Only in very rare cases, pure linear problems with real valued variables only (LP) or pure integer
problems (IP) arise. For a comprehensive overview of MILP and MINLP formulations and
algorithms used in chemical engineering see [Flo95].

Solution methods
For MILP problems, general purpose solver software like CPLEX and OSL is available which
uses variants of the branch-and-bound algorithm. The solution process can be further enhanced
using Langrangian relaxation and problem specific branching heuristics.
Benders Decomposition (BD) (e. g. [WR91a], [WR91b]) is often used to solve convex MINLP
problems. This technique decomposes the MINLP problem into a NLP sub-problem and a MILP
master-problem. The NLP problem is solved to provide an upper bound to the overall problem
whereas the MILP problem is solved to provide a lower bound. This procedure is iteratively
repeated until the gap between the lower and the upper bound is sufficiently small.
For continuous plants, a solution based on Outer Approximation (OA) and BD is proposed in
[PG94]. OA is quite similar to BD but uses the primal representation of the MILP instead of the
dual representation used in BD.
DICOPT++ [VG90] is an example of a general purpose software to solve MINLP. It uses the
OA/ER/AP approach which is an Outer Approximation Method, extended by an augmented
penalty function and handling of nonlinear constraints as inequalities.

Advantages
The main advantage of mathematical approaches is the global view on all assignment and
sequencing decisions. Thus, all choices are made simultaneously and the optimal solution can be
obtained for convex problems. Even if the solution process terminates before reaching the
optimum, bounds on the optimal solution are provided for convex problems. Thus, the quality of
the solution can be evaluated. For non-convex problems, it is still a research interest to ensure the
global optimality of a solution.

Disadvantages and problems


Although general algorithms are very powerful, a feasible solution often cannot be obtained in a
reasonable time (e. g. [LSE98]). Thus specific solution algorithms have to be used. Another
important problem is related to the task of modelling. The user has to state the problem in
abstract terms which do not allow for a intuitive modelling.
Furthermore, the same problem can be modelled in different ways which strongly influence the
solution process. A well known example is the treatment of time: [KPS88] proposed a
formulation based on the idea of uniform discretization of time. Time is treated as slots with
equal duration and events occur at the start of the slots. The discrete time representation has been
proven to be an efficient way for short term scheduling for academic problems. However, when
the time horizon or the problem size increases, the resulting problem contains a large number of
discrete variables. Instead, continuous representations of time can be used to enhance the
performance by decreasing the number of discrete variables. As the number of discrete variables
is not the only criterion, which determines the computational complexity of the scheduling
problem but the problem structure itself is important [SPS93], it is an expert task to decide which
treatment of time performs best for a specific problem.
The modelling task becomes even more difficult when decomposition strategies are used. In
[DSP97], a rolling horizon approach is reported. A small part of the planning horizon is
modelled in detail while the rest is represented in an aggregate manner. After the detailed part is
solved to optimality, the procedure is repeated for the planning horizon which was not taken into
account in detail. The results for the detailed part are only considered as far as they provide
values for decision variables. This allows for reassessing all continuous variables to compensate
inaccuracies introduced by the aggregate model. Several other methods for problem reduction
are discussed in [BSP+95].
As is evident from the above discussion, mathematical models are hardly reusable and after
slight changes of the problem the chosen algorithms may perform poor. To allow for an easier
modelling, some approaches aim at a uniform description of scheduling problem such as the
State Task Network (STN) [KPS88] and the Resource Task Network (RTN) [Pan94]. The STN
is a network which consists of states representing material, and tasks which represent operations
which transform at least one input state into one output state. [Zha95] points out that the STN
does not provide an uniform description of all important cases. For example, the STN does not
allow for treating cleaning operations as tasks as no material is transformed. In contrast, the RTN
models the problem in terms of tasks and resources viewing material as well as equipment and
utility as resources, thus providing a uniform model of the scheduling problem. Even though the
planning and especially the scheduling problem exist in many different forms with very plant
specific constraints which a single modelling approach can hardly cover, these approaches are
promising.

Recent progress and application


Recent progress is made in developing new algorithms, reducing the size of models as well as in
the application of methods in practice.
The Parallel Algorithms and Software for Mixed Integer Programming in Industrial Scheduling
project (PAMIPS), which is supported by the European Union, scheduling problems contributed
by BASF were reported to be still difficult to solve, but not impossible any more [Dan95]. The
PAMIPS approach tackles the problem combining parallel computing with interior point
methods which were reported to be very efficient in solving the LP-part of the problem.
In [BG96], industrial scheduling problems with a total number of variables up to 26000
including 11000 binary variables and 32000 constraints were solved to near optimality within 40
minutes with the standard OSL Solver using heuristics to reduce the number of values assigned
to the variables and decomposition strategies.

Recently, a new model was reported in [IF98] which dramatically decreases the total number of
variables. Compared to modelling approaches described in [Zha95], examples were solved with a
standard solver orders of magnitude faster.
8.2.3 Genetic Algorithms
Genetic algorithms mimic the natural evolution on an abstract level. This stochastic search
techniques explore different regions of the search space simultaneously. Since the publications of
Goldberg [Gol89] and Davis [Dav91], genetic algorithms are used to solve scheduling problems.
Various examples of application of genetic algorithms in manufacturing industry have been
published [Hus95],[Bie95],[KOY95], whereas for the process industry, up to now very few
publications consider the application to scheduling in the process industry
[MJI96],[CT94],[LSE98].
The modelling can be divided into the choice of an abstract representation and the representation
of constraints on the one hand and the actual design of search operators and their
parameterisation on the other.

Types of models
Two fundamental approaches are used to map the problem into an abstract formulation which is
suitable for the genetic search, indirect and direct coding.
While the indirect coding represents a template for the generation of a schedule, the direct coding
represents a complete schedule.

In the indirect coding, templates describe dispatching rules or a sequence of operations [Dav91].
This representation does not provide to a point in the search space directly, thus these templates
must be interpreted in a second step. During the interpretation of these templates further problem
specific knowledge can be introduced to accelerate the search and enhance the solution quality.
The most common approach is the coding of the sequence of operations on the genetic level. The
actual scheduling is done mostly with local search techniques. If the problem is dominated by the
choice and the sequence of operations, this representation is straightforward. The major
drawback of this approach is that decisions on the scheduling level may affect the feasibility of
the generated schedule. Information on the type of the violated constraints must be reported to
the sequencing level and reasonable actions have to be taken [LSE98].
In direct coding, the representation on the genetic level directly refers to a point within the search
space. All Parameters and degrees of freedom are available on the genetic level of information
and are available for the stochastic search [Bru93].

Solution methods
The genetic operators, which control the stochastic search have to be designed very carefully
with respect to the chosen representation. Exploration and exploitation of the structure of the
search space should be balanced [KOY95]. Problem specific knowledge should be considered at
this level of abstraction to restrict the search space to regions, which are likely to yield good
solutions.

Advantages and Disadvantages


The strength of genetic algorithms is the exploration of different regions of the search space in
relatively short times. On the other hand, convergence to the local optimum usually is rather
slow. Compared to mathematical programming, the evaluation of the quality of a schedule can be
done by simulation, without mapping all aspects of the problem into equations and inequalities,
which makes genetic algorithms attractive as the optimisation mechanism in a simulation-based
scheduling system. Furthermore, multiple and complex objectives can easily be included. As
genetic algorithms work on a population of solutions, they can easily and naturally be distributed
on multi-processor machines. On the other hand, the quality of the solution is hard to assess
because no bounds on the cost function are provided, and the speed of convergence cannot be
predicted.

Recent progress
A strategy for tackling various constraints of real scheduling problems with a uniform
formulation of constraints is still a research topic [FF98]. The ongoing research tries to
generalize several approaches and evaluate the relative merits regarding the problem
representation.

8.2.4 Simulation
Compared to the global and often simplified view adopted by mathematical programming,
simulation provides a local insight into the consequences of all assignment, sequencing and
timing decisions and enables a detailed and fast analysis of a given schedule in detail at low
computational costs.
Simulation can be used to test candidate schedules that are proposed by the user. More promising
is the direction to combine a detailed simulation with a system that evaluates schedules on a less
detailed level, thus at low computational costs. The schedules are tested in detail for feasibility
by simulation. If any unexpected event arises, for example a prolongation of processing times,
this information is forwarded to the system and the solution process is restarted upon the new
information received. This leads to an iterative repetition of both the simulation and solution
step.

Besides the combination of mathematical programming and simulation which is proposed in


[CHK+94] and [Rek95], other techniques can be used to determine an initial solution, e. g. by
combining genetic algorithms and simulation with user inference by a leitstand, as proposed in
[SFL+97]. The simulator BASIP which is used in this approach provides different simulation
techniques (discrete, hybrid, continuous) based on a uniform model [FE98]. Thus, different
levels of detail can be easily be represented and simulation can be used on different levels of the
planning and scheduling process.

8.2.5 Constraint Programming


Constraint Programming techniques start with the premise that effective pruning of the search
space is as important as a good strategy to determine a feasible or optimal solution. Moreover, it
facilitates a clear distinction between problem solving and pruning and the application of
different solution strategies together with a separate pruning module. For these reasons,
Constraint Programming Techniques are mostly used in hybrid solution approaches.

Constraint Propagation
Constraint propagation is built on the following rather simple idea [Tsa93]: Remove inconsistent
values from the domain of each variable where "inconsistent values" means values for which it
can be proven that they cannot be a part of the overall solution. E.g., given the inequality X<Y
and the associated integer domains X =[10..50] and Y=[1..30], constraint propagation can narrow
the domains to X =[10..29] and Y=[11..30]. Moreover, different constraints that share the same
variable can communicate with each other through a so-called constraint store or blackboard:
Stating an additional constraint X+Y=23 would lead to X=[10,11] and Y=[12,13].
The reduction of domains is done by constraint propagators which are specific for each type of
constraint. This enables the encapsulation of efficient algorithms into constraints to deduce
information during the propagation. Examples for the application of efficient propagation
algorithms can be found in the scheduling area where disjunctive constraints of the form "either
operation A precedes operation B or B precedes A" have to be resolved [BP95]. The most
commonly known methods to deduce additional information for the disjunctive constraint
mentioned are the edge-finding algorithm based on an idea presented in [CP90] and the concept
of edge-finding within task intervals [CL96]. As these techniques have quite different
requirements concerning time and computation-space, the cost of additional constraint
propagation has to be traded against the reduction of search effort [Pap95].

Applications
A successful representative of constraint propagation, which is already used for solving planning
and scheduling problems in the manufacturing industries, is the ILOG Optimization Suite. It uses
search, problem reduction by constraint propagation and LP-Solvers to provide a cooperative
solving strategy [ILOG]. A predecessor of the ILOG Optimisation Suite, the ILOG Scheduler, is
reported to cope with the constraints typically arising in scheduling multipurpose plants
[DSC98]. The constrained based optimisation techniques provided by ILOG are based on former
work on incorporating OR-techniques into constraint programming [BPN95] and applying
resource constraints [Pap95].
Applications of constraint programming problems from the process industry are rare. An
approach specific to the needs of the process industry is proposed in [Ter96] based on the idea to
solve the combinatorial part of the problem with constraint directed search and to determine
afterwards if the remaining (N)LP problem can be solved. In the solution of the combinatorial
part, user intervention by making assignments to variables and stating heuristics for value and
variable ordering is included.

Recently, an interactive decision tool named SKYE using constraint techniques was developed
[DDK97]. It serves as an intelligent decision support tool comparable to the former mentioned
tool by Terpstra [Ter96]. An iterative repair method is introduced based on decomposing the
overall problem into problems with disjunctive constraints but the same set of variables. SKYE
distinguishes between soft constraints, which can be violated at any time without imposing
infeasibilities, and two types of hard constraints, which cannot be relaxed. The hard constraints
impose a distinction between two types of problems which are easy to solve alone, but lead to
hard problems when considered together. The first problem is solved in a qualitative manner to
give a first idea of the solution. Afterwards, the second problem is added and solved using the
solution of the first problem as a heuristic for assigning exact values to the variables. Thereafter,
the maximum set of soft constraints which can be fulfilled is determined by a stochastic
optimisation using simulated annealing. After evaluation of the result, the user can make some
modification e. g. change heuristics to gain a better solution.

Advantages and disadvantages


Constraint programming can be used for the implementation of flexible and efficient scheduling
systems [Pap94]. Flexibility is possible since the encapsulation of different algorithms into
propagators allows for the programming of reusable solvers which are nevertheless problem
specific. Efficiency results as collaborating solvers can be programmed [Wür97] which exchange
information through the constraint store.
In mathematical programs, a restricted set of constraints is formulated to describe the problem,
which is sometimes difficult and needs much expertise. In contrast, constraint programming is
not restricted to a set of constraints as a purely declarative model is used and each propagator
defines a separate view of the problem. Ideally, this leads to the possibility to consider any type
of constraint in a natural way. The effective treatment of different types of constraints, however,
requires additional programming work.
Constraint Programming is very useful when a large number of constraints has to be considered
as in scheduling or re-scheduling. Unfortunately, this approach has not yet been widely tried for
scheduling problems in the process industry.

For a comprehensive overview of scheduling systems based on constraint programming


including several case studies mostly concerning manufacturing see [ZF95].

8.3 Scheduling and planning in practice


In practice, there is still a gap between planning systems on the enterprise level and the short-
term scheduling on the process level. For this reason, full integration of the available information
on every level must be a target [ALS96] besides the development of new or the enhancement of
existing scheduling strategies.

Commercial products are mostly interactive scheduling tools which sometimes provide
simulation strategies or heuristics to support the user. Most products lack the incorporation of
advanced scheduling algorithms. At present, most scheduling work is still done by hand [ALS94,
DDK97].
Examples for commercial scheduling tools are:

• IDS Prof. Scheer: Within the context of the ARIS standard software, a leitstand (PI-2)
was developed for the specific needs of the process industries and is now used together
with SAP R/3. This used at AKZO Nobel.

• A similar tool is distributed by Numetrix. Planx is used to generate master schedules


which are used for short-term scheduling by Schedulex. Simulation of schedules is
possible. Planx and Schedulex are used at Bayer AG.

• Together with ILOG, SAP has recently announced the development of an SAP Advanced
Planner and Optimizer (APO) which is scheduled for general availability at the end of
1998.

8.4 Conclusion
In practice, implementations of scheduling techniques which were developed at academic
institutions in practice are rare. Some years ago, this could be attributed to the simplified
problems which were considered. Nowadays, due to improved algorithms and modelling
approaches as well as the availability of powerful computer systems, problems of practical size
and relevance can be solved. Nevertheless, successful implementations are limited to isolated
cases.
The applicability of the methods mentioned decreases from planning to scheduling as the
connection to the real process gets tighter imposing frequent changes in the environment as well
as uncertainty. Furthermore, the detailed view on the scheduling level emphasizes the differences
which raises the need for treating each problem separately. As modelling is still an art, the costs
of detailed modelling have to be traded against with the efforts that will be obtained.
In contrast to elder publications, several papers state that the human scheduler should not be
replaced but supported by a scheduling system. This is because there is no guarantee that a
system is able to deal with any unforeseen event as well as there is no guarantee that sufficient
and accurate data is available. Moreover, all mentioned scheduling techniques model the
problem in a deterministic manner, assuming a predictable and stable environment. This is
utopian, as non-predictable events occur frequently while operating the plant, e. g. slight
prolongation of operating times. To treat these events only by assuring robustness means to
operate the plant suboptimally. Stochastic models to handle such uncertainties are still in very
early stages of their development. For these reasons, experienced operators, being able to
compensate the lack of information, are necessary for the safe and economic operation of the
plant. Surprisingly, hardly any contribution deals with the behaviour of the user.

All in all, many scheduling problems of practical relevance can be solved by recent methods
under the assumption of a deterministic behaviour of the plant. The stochastic element at present
cannot be considered adequately. Moreover, the great variety of the scheduling problem does
excludes a general solution method. In contrast, the production planning problem lends itself
more to a general treatment as the details are not considered. However, the generation of
production task for longer periods of time also leads to large problems, thus planning still is a
challenge.

8.5 References
[AKS94] Aytug H., Koehler G.J., Snowdon J.L.: Genetic learning of dynamic scheduling
within a simulation environment. Computer and Operations Research, 21(8), pp
909-926, 1994
[ALS94] Allweyer T., Loos P., Scheer A.-W.: An Empirical Study on Scheduling in the
Process Industries. http://www.uni-sb.de/public/iwi-hefte/iwihef_1.html
[ALS96] Allweyer T., Loos P., Scheer A.-W.:Requirements and New Concepts for
Production Planning in the Process Industries. I-CIMPRO'96 Conference in
Eindhoven, Netherlands 1996
[Bie95] Bierwirth,C., „A generalized permutation approach to job-shop scheduling with
genetic algorithms", OR Spektrum 17(2-3), 1995:, pp. 87-92.
[BG96] Blömer F., Günther H.-O.: Scheduling of a Multi-Product Batch Process in the
Chemical Industry. Proceedings of the 2nd I-CIMPRO. Fransoo J.C., Rutten
W.G.M.M., eds. pp. 102-116, 1996
[BP95] P. Baptiste, C. Le Pape: A Theoretical and Experimental Comparison of
Constraint Propagation Techniques for Disjunctive Scheduling. In: Proceedings
of the 14th International Joint Conference on Artificial Intelligence, Montreal,
Quebec, 1995
[BPN95] Baptiste P., Le Pape C., Nuijten W.: Incorporating Efficient Operations Research
Algorithms in Constraint Based Scheduling. Proceedings of the 1st International
Joint Workshop on Artificial Intelligence and Operations Research, Timberline
Lodge, Oregon, 1995
[Bru93] Bruns, R. "Direct chromosome representation and advanced genetic operators for
production scheduling", Proc. of 5th Int. Conf. on GA, 1993, pp. 352-359
[BSP+95] Bassett M.H., Subrahmanyam S., Pekny J.F., Reklaitis G.V.: Techniques for
Problem Size Reduction in the Design of Large Scale Batch Chemical Plants
under Market Uncertainty. ESCAP5 Bled, Slovenia
[CHK+94] Clark S.M., Harper P., Kavuri S.,Joglekar G.S.: An Integrated System for the
Operation of Beverage Manufacturing Processes. Proceedings of FOCAPO-94,
Crested Butte, Colorado (1993)
[CT94] Cartwright H.M., Tuson A.L., "Genetic algorithms and flowshop scheduling:
Towards the development of a real-time process control system", AISB Workshop
on Evolutionary Computing, Leeds, 1994
[Dan95] Daniel R.:PAMIPS and Optimisation - When excellence is not good enough.
http://www.iwr.uni-heidelberg.de/iwr/agbock/usr/dieter/pamipspub1.html
[Dav91] Davis L.: Handbook of Genetic Algorithms, Van Nostrand Reinhold, New York
1991
[DDK97] Dockx C., De Boeck Y., Kürt M.: Interactive scheduling in the chemical process
industry. Computers chem Engng. Vol 21, No. 9, pp 925-945, 1997
[DSC98] Das B.P., Sha N., Chung P.W.H.: Off-line scheduling a simple batch process
production plan using the ILOG scheduler. Computers chem. Engng. Vol 22.
Suppl. S947-S950, 1998

[DSP97] Dimitriadis A.D., Shah N., Pantelides C.C.: RTN-based Rolling Horizon
Algorithms for Medium Term Scheduling of Multipurpose Plants. Computers
chem. Engng. Vol 21, Suppl., pp S1061-1066, 1997
[FE98] Fritz M., Engell S.: A Framework For Flexible Simulation of Batch Plants.
Automation Of Mixed Processes (ADPM), Reims, France, 1998
[FF98] Fonseca C.M. and Fleming P.J.: "Multiobjective Optimization and Multiple
Constraint Handling with Evolutionary Algorithms-Part I and II", IEEE Trans. on
Systems, Man, and Cybernetics, Part A, Vol 28, No 1, Jan 1998, pp. 26-47

[Flo95] Floudas C.A.: Nonlinear and Mixed-Integer Optimization -:Fundamentals and


Applications. (Topics in chemical engineering). Oxford University Press, New
York, Oxford. 1995
[Gol89] Goldberg D.E.: Genetic Algorithms in Search Optimization and Machine
Learning. Addison Wesley, 1989
[Hus95] Husbands, P. „Genetic Algorithms for Scheduling.", AISB Quaterly, No.89, 1994

[IF98] Ierapetritou M.G., Floudas C.A.: Short-Term Scheduling: New Mathematical


Models vs. Algorithmic Improvements. Computers chem. Engng. Vol. 22, pp 419-
426, 1998
[ILOG] ILOG Optimization Suite White Paper. Available at
http://www.ilog.com/html/products/optimization/tech_papers.htm
[Jän84] Jänicke W.: On the solution of scheduling problems for multipurpose batch
chemical plants. Computers chem. Engng., Vol. 8 No 6, pp 339-343 1984
[KEP+94] Kudva G., Elkamel A., Pekny J.F., Reklaitis G.V.: Heuristic Algorithm for
Scheduling Batch and Semicontinuous Plants with Production Deadlines,
Intermediate Storage Limitations and Equipment Changeover Costs.Computers
chem. Engng. Vol 18, No. 9, pp 859-875. 1994
[KK88] Ku H.M., Karimi I.A.: Scheduling in Serial Multiproduct Processes with Finite
Intermediate Storage: A Mixed Integer Formulation. Ind. Eng. Chem. Res., 27, pp
1840-1848, 1988
[KOY95] Kobayashi, S, Ono, I. and Yamamura, M.„An efficient genetic algorithm for job
shop scheduling problems", Proc. of the 6th Conf. On Genetic Algorithms
(ICGA'95), 1995, pp.506-511
[KPS88] Kondili E., Pantelides C.C., Sargent R.W.H.: A General Algorithm for Scheduling
Batch Operations. Proceedings of the 3rd International Symposium on Process
Systems Engineering, pp. 62-75, Sydney, 1988

[KR89] Kuriyna K., Reklaitis G.V.: Scheduling Network Flowshops so as to Minimize


Makespan. Computers chem Engng. Vol 13, No1/2 pp 187-200 1989
[LGP97] Lin S.-C., Goodman E.D., Punch,III W.F., "A Genetic Algorithm Approach to
Dynamic Job Shop Scheduling Problems", Proc. of the 7th Conf. on Genetic
Algorithms, Morgan Kaufann, San Francisco, 1997, pp.481-488
[LSE98] Löhl T., Schulz C., Engell S.: Sequencing of Batch Operations fo a Highly
Coupled Production Process: Genetic Algorithms Versus Mathematical
Programming. ESCAPE8

[ME90] Musier R. F .H., Evans L. B., Batch Process Management; Chem. Eng. Prog. June
1990, pp. 66-77
[Min86] Minoux M.: Mathematical programming. Translation of : Programmation
mathematique. John Wiley & Sons Ltd. 1986
[MJI96] Minseok Kim, Jae Hak Jung and In-Beum Lee, "Intelligent Scheduling and
monitoring for multi-product networked batch process", Comp. chem. eng., Vol.
20, Suppl., pp. S1149-S1154
[Pan94] Pantelides C. C.: Unified Frameworks for optimal process planning and
scheduling. Proc. Second Conference on Foundations of Computer-Aided
Operations (FOCAPO II), pp 235-274 , 1994
[Pap94] C. Le Pape: Constraint-Based Programming for Scheduling: a Historical
Perspective. In: Working Notes of the Operations Research Society Seminar on
Constraint Handling Techniques, London, UK, 1994
[Pap95] C. Le Pape: Three Mechanisms for managing Resource Constraints in a Library
for Constraint-Based Scheduling. In: Proceedings of the INRIA/IEEE Conference
on Engeneering, Technologies and Factory Autmation, Parsi, France, 1995

[PG94] Pinto J., Grossmann I.E.: An MILP model for short term scheduling of batch
plants. Presented at the AIChE Annual Meeting 1994
[Rek95] Reklaitis G.V.: Scheduling Approaches for the batch process industries. in ISA
Transactions 34 (1995) pp.349-358
[Rek96] Reklaitis G.V.: Overview of Scheduling and Planning of Batch Process
Operations. In Reklaitis G.V., Sunol A. K., Rippin D.W.T., Hortascu Ö.: Batch
Processing Systems Engineering. Springer, 1996
[SFL+97] Stobbe M., Fritz M., Löhl T. Engell S.: BaSiS: Simulationsbasierte
Produktionsplanung für rezeptgesteuerte Mehrproduktanlagen. 11. Symposium
Simulationstechnik ASIM'97, Dortmund, pp 290-295
[SG91] Sahinidis, N.V., Grossmann I. E.: MINLP Model for Cyclic Multiproduct
Scheduling on Continuous Parallel Lines. Comput. Chem. Eng., 15, pp 85-103,
1991
[Sha98] Shaw K.J.,

[SPS93] Shah N., Pantelides C.C., Sargent R.W.H.: A General Algorithm For Short-Term
Scheduling of Batch Operations-II: Computational Issues. Computers chem.
Engng. Vol. 17, No.2 pp229-244, 1993
[Ter96] Terpstra V., Batch Scheduling within the Context of Intelligent Supervisory
Control; PhD-Thesis, Delft University of Technology, 1996
[Tsa93] E. Tsang : Foundations of Constraint Propagation. Academic Press INC, 1993

[VG90] Viswanathan J., Grossmann I.E.: A Combined Penalty Function and Outer
Approximation Method for MINLP Optimization. Computers chem. Engng, Vol.
14, 1990
[Wie97] Wiers V.C.S.: A Review of the Applicability of OR and AI Scheduling
Techniques in Practice. Omega, Int. J. Mgmt. Sci. Vol. 25. No. 2, pp 145-153,
1997
[WR91a] Wellons M.C., Reklaitis G.V.: Scheduling of Multipurpose Batch Chemical
Plants. 1. Formation of Single-Product Campaigns. Ind. Eng. Chem. Res. 1991,
30, pp 671-688
[WR91b] Wellons M.C., Reklaitis G.V.: Scheduling of Multipurpose Batch Chemical
Plants. 2. Multiple-product campaign formation and production planning .Ind.
Eng. Chem. Res. 1991, 30, pp 688-705
[Wür97] J. Würtz: Constraint-Based Scheduling in Oz. Operations Research Proceedings
1996, Springer Veralg 1997
[ZF95] Zweben M., Fox S.M.: Intelligent Scheduling. Morgan Kaufmann (1995)
[Zha95] Zhang X..: Algorithms for optimal process scheduling using nonlinear models.
PhD-Thesis, University of London 1995

Vous aimerez peut-être aussi