DEVELOPING INTEGRATED ADVANCED SYSTEMS
12.10 DEVELOPING INTEGRATED ADVANCED SYSTEMS
Neural computing, expert systems, genetic algorithms, and fuzzy logic are effective ways to deal with complex problems efficiently. Each method handles uncertainty and
ambiguity differently, and these technologies can often be blended to utilize the best features of each, achieving impressive results. For example, a combination of neural computing and fuzzy logic can result in synergy that improves speed, fault tolerance,
and adaptiveness. There are many real-world applications of intelligent systems integration. These include the United Technologies Carrier product-reliability system, which integrates a rule-based system and a neural network (Deng and Tsacle, 2000; Moon et al., 1998), plastic molding control—integrating neural networks and fuzzy logic (Mapleston,
1999), construction price estimation—integrating expert system and neural network (Li and Love, 1999), forecasting—integrating genetic algorithms and fuzzy logic (Cox, 1999; Li and Kwan, 2003), motor control—integrating neural networks, expert systems,
and fuzzy logic (Bartos, 1999), and the prediction and optimization of a ceramic casting process—integrating neural networks and fuzzy logic (Kuo and Chen, 2004; Lam et al., 2000). The sections that follow show a few alternatives for integration of multiple methods.
FUZZY NEURAL NETWORKS Fuzzy neural networks combine fuzzy logic with artificial neural networks. The inte- gration can be either way. The input and output variables can be processed by the fuzzy logic before entering the neural networks for learning. This step is called fuzzification.
The neural network takes the fuzzified input and output to derive a model. The model is then converted back to the original input and output scales. This step is called defuzzification. The output of the "defuzzified" fuzzy system can further become the input to another intelligent system. This kind of integration can also be applied to fuzzy
. C H A P T E R 1 2 ADVANCED INTELLIGENT SYSTEMS 6 9 1
duces fuzzy rules (Dubois et al., 2003; Jeng et al., 1996). We describe an example of combining fuzzy logic with ANNs next.
EXAMPLE 6: INTERNATIONAL STOCK SELECTION
An international investment company uses a combination of fuzzy logic and artificial neural networks (called FuzzyNet) to forecast the expected returns from stocks, cash, bonds, and other assets to determine the optimal allocation of assets. Because the com- pany invests in global markets, it is first necessary to determine the creditworthiness of various countries, based on past and estimated performances of key socioeconomic ratios, and then select specific stocks based on company, industry, and economic data. The final stock portfolio must be adjusted according to the forecast of foreign exchange rates, interest rates, and so forth, which are handled by a currency exposure analysis. The integrated network architecture of the system is shown in Figure 12.19. The integrated system includes the following technologies:
• Expert system. The system provides the necessary knowledge for both country and stock selection (rule-based system). • Neural network. The neural network conducts forecasting based on the data included in the database.
F I G U R E 1 2 . 1 9 FUZZYNET ARCHITECTURE
Neural forecaster
• u> 0)
Country selection
Stock selection
Rule base
Source: F. Wong et al., "Neural Networks, Genetic Algorithms, and Fuzzy Logic for Forecasting," Proceedings, International Conference on Advanced Trading Technologies, New York, July 1992, p. 48.
P A R T I V INTELLIGENT DECISION SUPPORT SYSTEMS
Historical _ MFG: membership data for
function generator
generation of membership
^ Membership functions
functions
FIP: Fuzzy information processor
BPN: backpropagation neural net
Targets Fuzzy rules
(e.g., and initial
country weights of
rating or rules
stock Historical
rating) data for
teaming BPN net Current data for desired
targets
Output of
Combined output
individual rules
of rules
F I G U R E 1 2 . 2 0 INFORMATION FLOW I N F U Z Z Y N E T
Source: Adapted, with permission, from Financial Analysts Journal, Jan./Feb. 1992. Copyright 1992, Association for Investment M a n a g e m e n t and Research, Charlottesville, VA. All rights reserved.
• Fuzzy logic. The fuzzy logic component supports the assessment of factors for which there are no reliable data. For example, the credibility of rules in the rule base is given only as a probability. Therefore, the conclusion of the rule can be expressed either as a probability or as a fuzzy membership degree.
The rule base feeds into FuzzyNet (Figure 12.20) along with data from the data- base. FuzzyNet is composed of three modules: a membership function generator (MFG), a fuzzy information processor (FIP), and a backpropagation neural network (BPN).The modules are interconnected, and each performs a different task in the deci-
sion process. GENETIC ALGORITHMS AND NEURAL NETWORKS
The genetic learning method can perform rule discovery in large databases, with the rules fed into the conventional expert system or some other intelligent system. A typi- cal way to integrate genetic algorithms with neural network models is to use a genetic algorithm to search for potential weights associated with network connections. A good genetic learning method can significantly reduce the time and effort necessary to find the optimal neural network model. Kim and Han (2002) developed a hybrid system to conduct activity-based costing. Wang (2003) presents a hybrid intelligent method for
modeling EDM process. Integration of rules and case-based reasoning is also a good way of improving rule-based systems (Marling et al., 2002).
By using several advanced technologies it is possible to handle a broader range of information and solve more complex problems (see AIS in Action 12.10). This concept is valid not only in cutting-edge technologies but also in any integration of decision
. C H A P T E R 1 2 A D V A N C E D INTELLIGENT SYSTEMS
AIS IN ACTION 12.10
HYBRID EXPERT AND FUZZY LOGIC SYSTEM DISPATCHES TRAINS
The Carajas line is one of the busiest railway routes and
An innovative, rule-based expert system that uses
leading carriers of iron ore in the world. The 892-kilo-
fuzzy logic has transformed the culture of train opera-
meter-long single-track line connects Sao Luis harbor
tions. Operational rules are directly used in t h e ES.
with the Carajas iron ore mine in the state of Para in the Fuzzy logic techniques analyze train m o v e m e n t s and A m a z o n (Brazil). The line has b e c o m e even busier help the operators make the best possible decisions because a unique real-time knowledge-based system is
(priorities of trains, etc.). The module that generates t h e
increasing its productivity and reducing its operating
initial train-movement plans has helped increase t h e
costs, without compromising safety.
v o l u m e of iron ore transported by about 15 p e r c e n t
Train dispatchers try to k e e p the trains running
while saving about 1.6 liters of fuel per 1,000 metric tons
safely all day and all night while attempting to maxi- of ore transported. With system improvements, further mize the amount of iron ore transported per day, econo-
gains are expected.
mize on fuel consumption, and minimize train delays. For over 10 years, paper and pencil were used to solve
this difficult task.
Source: Modified from Vieira and Gomide (1996), pp. 51-53.
• CHAPTER HIGHLIGHTS • Machine learning is a family of methods that allow
• Backpropagation is the most popular paradigm in neural
machines to acquire knowledge for problem-solving by networks. Most business applications are handled by this showing them historical cases.
algorithm. A backpropagation-based neural network • Machine-learning methods can be classified into
consists of an input layer, an output layer, and a certain
supervised and unsupervised learning. Supervised
number of hidden layers (usually one). The nodes in one
learning methods derive knowledge from cases whose
layer are fully connected to the nodes in the next layer.
outcomes are known, while unsupervised learning
Learning is done through a trial-and-error process of
methods derive knowledge from cases whose outcomes adjusting the connection weights. are unknown.
• Genetic algorithms are a set of learning methods that
• Popular machine-learning methods include inductive emulate the natural evolution process. They include learning, case-based reasoning, neural networks, genetic
three basic operations: reproduction, crossover, and
algorithms, cluster analysis, and fuzzy logic.
mutation.
• Case-based reasoning is based on experience with • Reproduction is a process that creates the next- similar situations.
generation population based on the performance of
• In case-based reasoning, the attributes of an existing different cases in the current population. case are compared with critical attributes derived from
• Crossover is a process that allows elements in different
cases stored in the case library.
cases to be exchanged to search for a b e t t e r solution.
• Cases include ossified cases, paradigmatic cases, and
• Mutation is a process that changes an element in a case
stories. Different types of cases must be handled to search for a better solution. differently to maximize the effect of learning.
• Fuzzy logic deals with the kind of uncertainty that is • Case-based reasoning has advantages over rule-based
inherently human in nature. It allows numerical data to reasoning in that it can capture expert knowledge,
be converted into linguistic terms, such as young or better explain decisions, and build up incremental
good, for symbolic processing. learning capabilities.
• Fuzzy logic can be combined with other techniques, • Neural computing is a set of methods that emulates
such as rule induction and neural networks, to achieve the way the human brain works. The basic processing
better performance.
unit is a neuron. Multiple neurons are grouped into
• Fuzzy logic-based systems include two steps. T h e first,
layers and linked together. The knowledge is stored in
called fuzzification, converts numerical data into fuzzy
P A R T I V INTELLIGENT DECISION SUPPORT SYSTEMS
KEY WORDS • adaptive resonance theory (ART)
• parallel processing • analogical reasoning
• fuzzy logic
» pattern recognition • artificial neural networks (ANNs)
• fuzzy sets
• processing elements (PEs) • associative memory
• genetic algorithms
• recurrent network • axon
• hidden layer
• reproduction • backpropagation
• inductive learning
• self-organizing • black-box testing
• Kohonen self-organizing feature
• sigmoid (logical activation) • case base
maps
• learning algorithm
function
»case-based reasoning ( C B R ) .
• learning rate
• stories
• chromosome
• summation function • competitive filter associative
• machine learning
• supervised learning memory
• massive parallel processing
• momentum
• synapse
• threshhold value »defuzzification
»crossover
• mutation
• topologies »dendrites
• neural computing
• transfer function »explanation-based learning
• neural networks
• transformation (transfer) function »feedforward-backpropagation
• neurons ,
• unsupervised, learning • fuzzification
• ossified cases
•paradigmatic c a s e s .
• weights