Information Technology Reference
In-Depth Information
quantum computing with an advantage in processing huge data sets. There-
fore, as an alternative, quantum computing has been introduced into the neural
network. Professor Kak [1] first proposed the concept of ”Quantum Neural Com-
putation”, and presented a new paradigm that combines neural computing and
quantum computing. Since then, a variety of quantum neural models emerged.
Perus discussed the mathematical analogies between the neural network theory
and quantum mechanics [2]. Similar works have also been done in [3][4] to state
new quantum neural networks. Learning is the fundamental feature of ANN. The
adjustable parameters of the feedforward neural network are tuned by specific
learning methods. In decades, gradient descent method is popular for learning
the feedforward neural network. However, gradient descent method falls into lo-
cal minima easily. Moreover, this method over-fits to the training data. In recent
years, many researchers turn their attention to the evolutionary algorithms (EA).
The global optimization capability enables them to be widely used for adjusting
the parameters of the feedforword neural network. Many scholars [5][6] studied
the applications of different EAs to optimizing the parameters. However, the slow
learning speed weakens the optimization eciency of EA. The iterative learning
steps may take a lot of time to train neural networks in many applications, es-
pecially complex problems. Additionally, more time would be spent on choosing
proper control parameters of EA. Different from the gradient descent method
and EA, a new learning algorithm called extreme learning machine (ELM) [7][8]
was proposed for single-hidden layer feedforward neural network (SLFN). ELM
randomly chooses the input weights and the hidden layer thresholds, and then
analytically identifies the output weights. It can be seen from many researches
[9][10] that all the parameters of SLFN need not be tuned iteratively and would
be obtained simply. Huang and Siew also extended ELM from SLFN to RBF
network [11].
This paper focuses on the assessment of syngas components in the Texaco
gasification process. A new quantum-inspired neural network is proposed, and
the extreme learning machine is further applied to identifying all the adjustable
parameters of the proposed QNN. The established ELM-QNN model is used
for the real-time measurement of the Texaco syngas components under different
operational situations.
2 Texaco Gasification Process
The Texaco gasification process [12] in a fertilizer plant of China is described
as below. The coal water slurry is pumped into the Texaco gasifier. The gasi-
fier is a two-compartment vessel, consisting of an upper refractorylined reaction
chamber and a lower quench chamber. Oxygen and slurry flow through an in-
jector nozzle into the reaction chamber. In the reaction chamber, they react to
produce the raw syngas and the molten slag. Subsequently, the raw syngas and
molten slag flow into the quench chamber where water cools and partially scrubs
the raw syngas. The water quench also converts the molten ash into glass-like slag
Search WWH ::
Custom Search