Information Technology Reference
In-Depth Information
In this context, two situations of special interest are those corresponding to the
false alarms
and the
blooms not detected
. The former refers to predictions of bloom (concentration of
pseudo-nitzschia ≥ 100,000 cell/liter) which do not actually materialize (real concentration ≤
100,000 cell/liter). The latter, more problematic, occurs when a bloom exists but the model
fails to detect it. Another unwelcome situation takes place when the number of predictions
exceeds an absolute error of 100,000 cell/liter (labelled as incorrect predictions). In such a
situation, in which the rules governing a target process or system are unknown, the
prediction of the parameter values that determine the characteristic behaviour of the system
can be a problematic task. However, it has been found that a hybrid case-based reasoning
system can provide a more effective means of performing such predictions than other
connectionist or symbolic techniques (Fdez-Riverola & Corchado, 2003).
Related with this domain we describe the FSfRT (
Forecasting System for Red Tides
) system,
a hybrid model able to accurately forecast the concentrations of pseudo-nitzschia spp, the
diatom that produces the most harmful red tides causing amnesic shellfish poisoning (or
ASP). Our FSfRT system employs a case-based reasoning model to wrap a growing cell
structures network, a radial basis function network (Fritzke, 1994) and a set of Sugeno
fuzzy models (Jang et al. 1997) to provide an accurate prediction. Each of these techniques
is used at a different stage of the reasoning cycle of the decision support system to
retrieve historical data, to adapt it to the present problem and to automatically review the
proposed solution.
The forecasting system uses information from two main sources: (
i
) data coming from
several buoys and monitoring net used to create a succession of problem descriptors able to
characterize the current forecasting situation and (
ii
) data derived from satellite images
stored on a database. The satellite image data values are used to generate cloud and
superficial temperature indices which are then stored with the problem descriptor and
subsequently updated during the CBR operation. Figure 7 shows a schematic view of the
whole data managed by the FSfRT system.
In order to forecast the concentration of pseudo-nitzschia spp at a given point a week in
advance, a problem descriptor is generated on a weekly basis. A problem descriptor consists
of a sequence of sampled data values (filtered and pre-processed) recorded from the water
mass to which the forecast will be applied. The problem descriptor also contains various
other numerical values, including the current geographical location of the sensor buoys and
the collection time and date. Every week, the concentration of pseudo-nitzschia spp is added
to a problem descriptor forming a new input vector. The problem descriptor is composed of
a vector with the variables that characterise the problem recorded over two weeks. The
prediction or output of the system is the concentration of pseudo-nitzschia spp one week
later, as indicated in Table 3.
The cycle of forecasting operations (which is repeated every week) proceeds as depicted
in Figure 8. First a new problem instance is created from the pre-processed data cited
above. When a new problem is presented to the system, the GCS neuronal network is
used to obtain
k
more similar cases to the given problem (identifying the class to which
the problem belongs). In the reuse phase, the values of the weights and centers of the
neural network used in the previous forecast are retrieved from the knowledge base.
These network parameters together with the
k
retrieved cases are then used to retrain the
RBF network and to obtain an initial forecast of the concentration of pseudo-nitzschia spp
(see Figure 8). During this process the values of the parameters that characterise the
network are updated.
Search WWH ::
Custom Search