Four-layer Feedforward Network Research Articles

Predictive control of seismic responses of a novel seismic isolation – non-seismic isolation composite structure system based on Levenberg-Marquardt algorithm

In order to reduce the seismic responses of adjacent structures, this paper proposes a novel seismic isolation – non-seismic isolation composite structure system, which is made up of a seismic isolation structure and a non-seismic isolation structure adjoined via controllers, and whose seismic responses are controlled with good effect by the principle of interacting control between the seismic isolation and non-seismic isolation structures. With a novel five-layer seismic isolation – non-seismic isolation composite structure system as an example, this paper builds a calculation model for the structure, derives the kinematic equations for the interacting control within this structural system, and adopts linear quadratic regulator (LQR) optimal control to analyze the control effect of this system and study the effect of time lag on the interacting control within this system. To address the issue of time lag in the interacting control between structures, an improved BP algorithm (Levenberg-Marquardt) is further adopted to set up a four-layer feedforward network to forecast the seismic responses of this novel composite structure system; the network structure is trained using the data acquired from LQR control under the action of the first 500 sampling points of the seismic motion taft, and the structure undergoes predictive control using the untrained seismic motion data of El Centro. The research results show that the interacting control of such a novel seismic isolation – non-seismic isolation composite structure system can achieve a good result, with the displacement response of the topmost layers of both structures decreasing by above 50 %; time lag may have an adverse effect on the control of the composite structure system, so that the structure’s seismic response increases by 20 % ∼ 40 %. The issue of time lag in interacting control can be addressed by adopting neutral network for predictive control, so that the structure’s seismic responses are 11 % ∼ 25 % lower than when it is not predicted. Levenberg-Marquardt algorithm demonstrates a strong learning ability and a curve approximation ability. The time history curves of the structure’s seismic responses under predictive control of this algorithm are highly overlapped with the time history curves under LQR optimal control with zero time lag.

Artificial neural networks for modeling the reverse osmosis unit in a wastewater pilot treatment plant

This paper presents experimental and modeling data from a membrane-based wastewater treatment (WWT) pilot plant. The effluents from various upstream steps of a can-manufacturing plant were combined and subjected to a pretreatment process, which consisted of coalescing filters, coagulation and gravity settling, and sand activated carbon and polishing filtration, and a pressure-driven membrane process, such as reverse osmosis (RO). The performance of the RO membrane was evaluated and experiments were conducted using continuous wastewater flow. The complete membrane separation scheme was validated with a closed loop cell through several experiments, in which the concentration of the antiscaling agent and the pH were varied to determine the optimal operational conditions. Detailed parametric studies for these continuous flow experiments were conducted, and the permeate flow rates in the RO membrane system were experimentally measured. The experimental flow data were correlated and analyzed using an artificial neural network (ANN). A four-layer feed-forward network with a back-propagation algorithm was used to train the ANN models. After the training process was completed, the experimental flow data was used to assess the prediction capabilities of the networks based on the RO permeate water flow rate. This research showed that the RO unit results in the acceptable removal of 96.1% of the total dissolved solids and a maximum effluent recovery close to 72%. The predicted and experimental flow data were well correlated, and a determination coefficient between 0.97 and 0.99 was achieved.

A pulse neural network reinforcement learning algorithm for partially observable Markov decision processes

AbstractThis paper considers learning by a pulse neural network and proposes a new reinforcement learning algorithm focusing on the ability of pulse neuron elements to process time series. The conventional integrator neuron element is modeled in terms of the average firing rate of the biological neuron. But the pulse neuron is a modeling of the input–output relation of the time‐series pulse (spike) and the decay of the internal state (internal potential). The application of such neural networks has been considered in recent engineering studies. It is known in particular that a pulse neuron with a high decay rate acts as a coincidence detector. The proposed model combines pulse neuron elements with different decay rates, which facilitates the processing of the time‐series input information and the discrimination of fuzzy states in a partially observable Markov decision process. The proposed network is a four‐layered feedforward network in which the pulse neuron elements forming the two hidden layers provide a pseudo‐representation of the state in the environment. The elements generate a secondary reinforcement signal which results in learning similar to the conventional reinforcement scheme based on the state evaluation function. A computer experiment verifies that the proposed model works effectively in an environment which is strongly partially observable. © 2005 Wiley Periodicals, Inc. Syst Comp Jpn, 36(3): 42–52, 2005; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.10645

A neuro-fuzzy scheme for simultaneous feature selection and fuzzy rule-based classification.

Most methods of classification either ignore feature analysis or do it in a separate phase, offline prior to the main classification task. This paper proposes a neuro-fuzzy scheme for designing a classifier along with feature selection. It is a four-layered feed-forward network for realizing a fuzzy rule-based classifier. The network is trained by error backpropagation in three phases. In the first phase, the network learns the important features and the classification rules. In the subsequent phases, the network is pruned to an "optimal" architecture that represents an "optimal" set of rules. Pruning is found to drastically reduce the size of the network without degrading the performance. The pruned network is further tuned to improve performance. The rules learned by the network can be easily read from the network. The system is tested on both synthetic and real data sets and found to perform quite well.

Self-organization model on receptive field of neuron with asymmetric time window of synaptic modification

The spatial-temporal response properties of some simple neurons in visual pathway arise basically prior to birth. In the absence of visual experience, how do these neurons develop in visual system? Based on Wimbauer network with delay, a four-layer feed-forward network model is proposed, which is characterized by modifying the Hebb learning rule through introducing the asymmetric time window of synaptic modification found recently in neurobiology. The model can not only generate by self-organization more diversified spatial-temporal response characteristics of neuronal receptive field than earlier models but also provide some explanations for the possible mechanism underlying the development of receptive fields of contrast polarity sensitive neurons found in visual system of vertebrate. Thus the proposed model may be more widely applicable than Linsker model and Wimbauer model.

A model of invariant object recognition in the visual system: learning rules, activation functions, lateral inhibition, and information-based performance measures.

VisNet2 is a model to investigate some aspects of invariant visual object recognition in the primate visual system. It is a four-layer feedforward network with convergence to each part of a layer from a small region of the preceding layer, with competition between the neurons within a layer and with a trace learning rule to help it learn transform invariance. The trace rule is a modified Hebbian rule, which modifies synaptic weights according to both the current firing rates and the firing rates to recently seen stimuli. This enables neurons to learn to respond similarly to the gradually transforming inputs it receives, which over the short term are likely to be about the same object, given the statistics of normal visual inputs. First, we introduce for VisNet2 both single-neuron and multiple-neuron information-theoretic measures of its ability to respond to transformed stimuli. Second, using these measures, we show that quantitatively resetting the trace between stimuli is not necessary for good performance. Third, it is shown that the sigmoid activation functions used in VisNet2, which allow the sparseness of the representation to be controlled, allow good performance when using sparse distributed representations. Fourth, it is shown that VisNet2 operates well with medium-range lateral inhibition with a radius in the same order of size as the region of the preceding layer from which neurons receive inputs. Fifth, in an investigation of different learning rules for learning transform invariance, it is shown that VisNet2 operates better with a trace rule that incorporates in the trace only activity from the preceding presentations of a given stimulus, with no contribution to the trace from the current presentation, and that this is related to temporal difference learning.

Capabilities of a four-layered feedforward neural network: four layers versus three

Neural-network theorems state that only when there are infinitely many hidden units is a four-layered feedforward neural network equivalent to a three-layered feedforward neural network. In actual applications, however, the use of infinitely many hidden units is impractical. Therefore, studies should focus on the capabilities of a neural network with a finite number of hidden units, In this paper, a proof is given showing that a three-layered feedforward network with N-1 hidden units can give any N input-target relations exactly. Based on results of the proof, a four-layered network is constructed and is found to give any N input-target relations with a negligibly small error using only (N/2)+3 hidden units. This shows that a four-layered feedforward network is superior to a three-layered feedforward network in terms of the number of parameters needed for the training data.

MCFC: A Fuzzy Neural Network Model for Speech Recognition

A connectionist fuzzy classifier, called CFC, was proposed and shown to perform well in speech recognition. In this article a new learning algorithm with better on-line learning ability and speech recognition performance is developed to replace the original learning algorithm of the CFC model. To distinguish the two learning algorithms, the CFC executing the new learning algorithm is called modified CFC MCFC. Both of the learning algorithms execute a one-pass learning scheme, which is far faster than the backpropagation-based learning algorithms. In addition, both the CFC and MCFC employ the same four-layered feed forward network structure to implement a “weighted Euclidean distance” fuzzy classification procedure that can be realized by digital hardware with a massively parallel architecture. Some experiments and comparisons for MCFC, CFC, and some other neural network models are also made. The experimental results show that the MCFC has better accuracy and stability for speech recognition than the original CFC, probabilistic neural network, fuzzy ART, and fuzzy ARTMAP, especially in a noisy environment.