A neural computing approach to the Markov decision problem is presented. The method makes use of historical records of rewards as inputs and average long-run incomes per epoch as targets for training a backpropagation network to associate the two quantities, without a priori knowledge of the state transition probabilities. Estimation of the long-run income, given a new reward matrix as input, is interpreted as a statistical-association learning problem. Following the training, the relevant functional relationship between the income (output) and the reward (input), that has been learned by the network, can be used to compute an unknown expected income. The present study examines network generalization to new inputs, considering the effects of different topological designs and the characteristics of training samples on neural computing accuracy. Successful performance of the best design, which is determined directly from the underlying mathematical model of the Markov decision problem, is demonstrated by a computer simulation experiment. 相似文献
The effects of fluctuating demand on production and inventory levels are important in manufacturing resource planning. Thus, the focus of this presentation is on aggregate production planning of manufacturing resources in order to satisfy stochastic demand for a family of products to minimize total costs that include production and inventory holding costs over a rolling horizon.
If it is assumed that, in a commercial setting, the demands are fixed, then the production plans generated by a mathematical programming procedure are not responsive to the actual fluctuations of stochastic demand in each time period.
The situation discussed here considers the case where demands are normally distributed with means and variances that are sequentially revised as new observations of demand are received over time. This assumption allows the probabilistic constraint to be converted to an equivalent linearly-constrained deterministic model. Extensions to the normality assumption are discussed. Also other ideas such as optimal control theory, learning and adaptive signal processing extensions are discussed as well. 相似文献
The “Central Atoms” model presented by the authors in an earlier paper is extended to ternary silicate and alumino-silicate
melts. The model is applied to the CaO-FeO-SiO2 and the CaO-Al2O3-SiO2 system. Use is made of the parameters from the relevant binaries only. The agreement between experimental and calculated
isoactivity curves is good in all cases.
Formerly Research Student, Department of Metallurgy, Indian Institute of Science, Bangalore-560 012, India 相似文献
We propose a sequential test procedure for transient detections in a stochastic process which can be expressed as an autoregressive moving average (ARMA) model. Preliminary analysis shows that if an ARMA(p,q) time series exhibits a transient behavior, then its residuals behave as an ARMA(Q,Q) process, where Q ≤ p + q. Based on this fact, we derive a new sequential test to determine when a transient behavior occurs in a given ARMA time series. Simulation experiments conducted in this study show that the proposed test can detect the occurrence of a transient in the ARMA model. We also apply the proposed method to detect transient changes in the pH of an erythromycin salt. 相似文献
Results of the photoelectrochemical studies carried out with sintered discs of pure Fe2O3 and of Fe2O3 doped with TiO2, SnO2, ZrO2 and Ta2O5 are compared and discussed. Flat band potential, saturation current and minority charge carrier diffusion length are higher for TiO2-doped Fe2O3 samples than for other doped specimens. Ta2O5-doped Fe2O3 samples and SrFe2O4 have been found to be photo-inactive. The results are analysed in terms of depletion layer theory. 相似文献
The concentration of deuterium at the surface of cathodically charged high strength steels AISI 1062, 4037, and 4140 has been
determined by secondary ion mass spectrometry (SIMS). The beneficial effects of pickling in NAP (a mixture of nitric, acetic,
and phosphoric acids) to remove surfacebound deuterium have been observed.
formerly with AMCA International Limited, Kanata, ON, Canada 相似文献
Mild steel samples charged with hydrogen were stored in EDTA, imidazole and a commercial inhibitor (with an active amino group) and analyzed for hydrogen. These reagents were found to decrease hydrogen loss by diffusion. Imidazole and the commercial inhibitor were also found to minimize the entry of hydrogen into the steel. The surface film consisting of Fe? N and Fe? N? HS bonded structures is thought to minimize hydrogen loss from the samples. Amines such as imidazole are suitable and more convenient than liquid nitrogen as sample storage media. 相似文献
A comprehensive anodic potentiodynamic polarization analysis of Ag-Pd binary alloys in chloride solutions has revealed composition
dependent active-passive polarization profiles. The results are interpreted in terms of the transition metal properties of
Pd and the electron donor acceptor theory of passivation. 相似文献