AirWare
air quality assessment & management
Reference and User Manual
AirWare   On-line Reference Manual
  Release Level 6.1
  Release Date 2012 05
  Revision Level 1.0
Last modified on:   Thursday, 29-Nov-12 19:57 CET

Optimization Strategy

The Emission control optimization is based on the application of emission control technologies to individual sources or classes of emission sources.

These combination of sources and control technologies define Optimization Scenarios Within the two-phase optimization approach, the first phase generates a set of alternatives for the subsequent discrete multi-criteria DSS. For the generation of the alternatives (candidate solutions), several alternative strategies can be selected/used, which represents different trade-offs between efficiency/performance (generating large sets of "random" trials fast without feedback as to their "quality") and increasing level of "sophistication", to increase the relative number of feasible and non-dominated solution in this set at the cost of performance, using more complex search strategies and embedded feedback.

Alternative Strategies

    Plain Monte Carlo

    is based on a two step random selection procedure:

    • step 1 select the instrument in a binary, yes/no decision biased by the optional weight specified by the user to modify the probability of an instrument to be applied.
    • step 2 then selects an applications level (constrained between MIN and MAX levels defined), again randomly.

    Adaptive

    the first trial on the first source/technology combination yields a starting point; this point has SOME distance from UTOPIA. We can NOW iterate with the application% to get closer to UTOPIA = improvement (we have chosen a point between MIN and MAX; now HALF the larger of the distances to MIN and MAX, evaluate; if it improved, continue in the same direction always halving the remaining distance to MIN or MAX UNTIL the results is WORSE than the previous one, THEN change direction UNTIL we reach some cut-off defined by (a) number of trials or (b) relative improvement [should be scenario configuration parameters]

    Heuristic (three types/options)

    as above for individual (currently only: boilers) sources; for area (and later LINE sources) INSTEAD of applying the reduction pro-rata to ALL members of the class, apply to INDIVIDUAL MEMBERS 100% until the OVERALL, class specific MAX (as a % of the total baseline emission for that class) has been reached. Apply to the members based on a pre-defined RANKING based on

    1. SIZE (possible contribution)
    2. COST efficiency
    3. PROXIMITY (to receptor/sensitive locations)

    Learning

    The changes the weights progressively, based on a continuous analysis of performance/results. As a preparatory step, this requires a preference structure defined: as set of criteria with their optimization direction (min, max, min the deviations from a target), setting of constraints.

    • step 1: plain Monte Carlo; for each trial, we also record its "performance" vis a vis the constraints: feasible or infeasible solution.
    • ste2: a coincidence matrix is built progressively; for three levels of performance, (feasible, near feasible, infeasible) it records the applicable level for each technology: none (0), small (<50%), large (>50%).
    • step 3: based on the coincidence matrix built progressively, the weights for the selection of technologies, and similar (internal) weights for the application rates are progressivily adapted according to the values in the coincidence matrix.
      Weights for technologies (and application rates) that have a higher-than-averge contribution to feasible solution are increased, those that have a higher-than-average coincidence with infeasible solutions are decreased.


© Copyright 1995-2016 by:   ESS   Environmental Software and Services GmbH AUSTRIA | print page