If machines are capable of understanding natural language, communications between humans and machines will be very efficient. However, natural language is often very vague or inaccurate, and it has been very difficult to construct such capable machines. When an instruction is given from a human to a system, the conventional way has been to utilize a priori knowledge to understand it. Where the instruction is expressed linguistically, it is very hard for the system to acquire a priori knowledge. This paper presents a system which is able to understand linguistic instructions given by a human operator and to extract intention from the linguistic instructions using a Fuzzy Classifier System(FCS). By extracting the intension, the system can generalize the instructions for different conditions. By applying the proposed system to an obstacle avoidance problem, simulations are done to show that the system can extract intentions, and the extracted rules have generality.

View full abstract

This paper presents a definition of estimative index for the possibility regression analysis. The estimative index quantitatively indicates the ability to explain of model and is indispensable for objectively estimating the model. An optimum model could be selected if there is an objective estimating index such as this estimative index when, for example, several different modeling for the same object have been attempted. Since such is an important matter of concern when it is tried to apply the possibility regression analysis to a practical problem, existence of the estimating index for a model such as estimative index is considerably significant. The estimative index defined here is aimed at indicating the ability to explain of possibility regression model upon noticing the fact that the possibility region for variable is estimated by possibility regression analysis.

View full abstract

Conditional probabilities are widely used, when we evaluate uncertainty of causalities. However, Peng & Reggia designate that conditional probability is different from the uncertainty of causation that we have in mind, since it is a probabilistic evaluation including events that the effect happens to be caused by another cause, not by the cause given as the condition. Thus they proposed causation event that the cause actually causes the effect, and conditional causal probability which is a conditional probability of causation event given the cause. They also applied them to an inverse problem of causalities, which is a problem to calculate the posterior probability that a given plural causes are arising, when a set of effects are observed. This paper proposes conditional causal possibility(CCP) which has similar semantics to the conditional causal probability, and describes the relation between CCP and conditional possibility(CP). It also studies methods to apply CP and CCP to an inverse problem of causalities, and indicates that improper cases may appear when CP is used and that they can be avoided if CCP is used instead.

View full abstract

Cooling load is the heat value of chilled water used for air-conditioning in a district heating and cooling system. Cooling load prediction is indispensable for operating a district heating and cooling system. For several reasons, the real data sets usually involve som outliers and missing data. In this paper, we focus on cooling load prediction problems in a district heating and cooling system. For dealing with such real data sets, a simplified robust filter which improves a numerical stability problem of a robust filter is proposed for filtering. Then RBF-NARMA model, which is a nonlinear autoregressive moving-average(NARMA)model through a radial basis function network(RBFN), is presented for time series prediction. The effectiveness of the prediction method proposed in this paper is demonstrated by applying both the proposed method and the recurrent NARMA model based method proposed by J.T.Connor et al.

View full abstract

In this paper, considering an efficiency condition in multiobjective linear programming problems, we propose effective efficiency tests, and apply these methods to extended problems with interval coefficients. In the conventional multiobjective programming literature, it is required to solve a linear programming problem in efficiency test procedure. Hence, in the case of our extended problems with interval coefficients, we need excessive computational demand for necessary efficiency tests. Therefore, development of effective efficiency test methods has been desired. In the former part of this paper, we consider the efficiency condition and its dual condition in ordinary multiobjective linear programming problems, and propose effective "efficiency tests" based on the extreme ray generation method. In the latter part of this paper, we apply the proposed efficiency tests to interval multiobjective linear programming problems and propose "necessary efficiency tests". These methods consist of the branch and bound algorithm and the efficiency test methods proposed in the former part of this paper. Numerical examples are shown to explain these proposed methods.

View full abstract

Many control systems have been changed from manual controls to automatic controls. Fuzzy control has played an important role in these changes. However, because knowledge is gained through everyday working experience and intiution, it is difficult for experts(control operators)to describe their own knowledge and skills in linguistic terms. Yet this linguistic description is still a necessary step in the construction of fuzzy control ruled. One solution to this problem is inductive learning and many algorithms have been developed to extraxt or generate rules from operator data. ID3, a very famous inductive learning algorithm, is an easy and powerful algorithm. It can make small and useful decision trees. But unfortunately, it cannot deal with real numbers such as control data. It can only deal with attributes and classes. There are some papers on fuzzy ID3, which can deal with fuzzy number attributes. However, there are no reports about a version of ID3 which can deal with classes in real numbers. In this paper, we propose a new fuzzy ID3 which can deal with fuzzy classes given in real numbers. This new fuzzy ID3 uses a fuzzy clustering algorithm FCM to deal with the real numbers. Next, We apply it to input/output data obtained from behavior of experts in the inverted pendulum control and extracted fuzzy control rules from the data. Finally, we show simulation results of the inverted pendulum control by the fuzzy rules and discuss the performance of the results.

View full abstract

In this paper, we propose a method to generate Japanse hiragana style characters that reflect user's personality or individuality. Input device is a pen. Our method uses input coordinates(sampling points), pen pressure and pen speed as input information. These information transform the hiragana style variously. In our method, B-spline curves interpolate the selected sampling points, and make smooth input curve. Fuzzy spline curves are applied for the selecting, and enable the shape of sampling-points-queue to be preserved approximately. Moreover, we propose the pen-moving rules. The smooth input curve is divided into several parts by applying the pen-moving rules.the each divided part is transformed into hiragana style part. Our proposed method was implemented on the calligraphic system. Hiragana style characters were generated by using the system and they showed that the proposed method is effective.

View full abstract

In this paper, we examine a linear programming problem with multiple decision makers and consider allocation of an optimal value of an objective function among the decision makers. Especially, we deal with a linear programming problem with fuzzy parameters from the viewpoint of experts'imprecise or fuzzy understanding of the nature of parameters in a problem-formulation process and consider a fuzzy linear programming game arising from the linear programming problem. Constructing fuzzy goals of coalitions with respect to payoff in order to suitably reflect fuzzy environments in which the linear programming game is formulated, we define a solution concept maximizing minimal fuzzy goal and a solution concept maximizing the sum of fuzzy goals, and develop computational methods for obtaining the solutions. Finally using a numerical example, we illustate how to arise a fuzzy linear programming game and to derive the solutions.

View full abstract

The paper studies inverse casual reasoning which derives possible causes from uncertain causation knowledge and evidence(results), and proposes two new approaches, one based on probability theory and the other on possibility theory. There are some conventional approaches that can deal with the inverse causal reasoning with uncertain evidence. Reasoning with Jeffrey's rule, approximate reasoning with subjective Bayesian method, and Bayesian network approach are those based on probability theory. As for possibility theory, inverse problem of fuzzy relational equations and a method employing the idea of Jeffrey's rule could be applicable. However, it has been designated that conditional probabilities and conditional possibilities, one of which are used in all approaches mentioned above, are inappropriate to express uncertainty of causation recognized by human, and that conditional causal probabilities and conditional causal possibilities should be used instead. The paper first discusses conventional approaches of inverse causal reasoning from uncertain evidence with conditional probabilities and conditional possibilities. Then, it proposes two new approaches, one employs conditional causal probabilities and the other conditional causal possibilities. These two approaches are developed based on the same idea, though they use different measures meaning probabilities and possibilities to express the uncertainty. It also discusses how to give probability or possibility data when the proposed approaches are applied to inverse causal problems.

View full abstract

Load forecasting has always been the essential part of efficient power system planning and operation. The objective of this paper is to propose load forecasting technique based on similarity, which is useful for power system operator in case traditional approach requiring modeling cannnot be utilized to forecast power load. This paper presents Fuzzy Neural Network(FNN)for next day peak load forecasting based on peak load for similar day. In order to obtain accurate next day peak load, the FNN is used to estimate correction factor to modify power load on similar day. The FNN can be trained to perfect fuzzy logic rules and to find optimal membership functions for input and output variables. This technique has the advantage of dealing with not only the nonlinear part of power load curve but also with load forecasting on weekend. The suitability of the proposed approach is illustrated through an application to actual load data of the Okinawa Electric Power Company in Japan.

View full abstract