A Reinforcement Learning Scheme of Fuzzy Rules with Reduced Conditions

2000 ◽  
Vol 4 (2) ◽  
pp. 146-151 ◽  
Author(s):  
Hiroshi Kawakami ◽  
◽  
Osamu Katai ◽  
Tadataka Konishi ◽  
Keyword(s):  
Fuzzy Inference ◽  
Learning Algorithm ◽  
Fuzzy Rule ◽  
Fuzzy Rules ◽  
Q Value ◽  
Huge Number ◽  
Q Learning ◽  
Control Rules ◽  
A Value ◽  
Rule System

This paper proposes a new method of Q-learning for the case where the states (conditions) and actions of systems are assumed to be continuous. The components of Q-tables are interpolated by fuzzy inference. The initial set of fuzzy rules is made of all combinations of conditions and actions relevant to the problem. Each rule is then associated with a value by which the Q-values of condition/action pairs are estimated. The values are revised by the Q-learning algorithm so as to make the fuzzy rule system effective. Although this framework may require a huge number of the initial fuzzy rules, we will show that considerable reduction of the number can be done by adopting what we call Condition Reduced Fuzzy Rules (CRFR). The antecedent part of CRFR consists of all actions and the selected conditions, and its consequent is set to be its Q-value. Finally, experimental results show that controllers with CRFRs perform equally well to the system with the most detailed fuzzy control rules, while the total number of parameters that have to be revised through the whole learning process is considerably reduced, and the number of the revised parameters at each step of learning increased.

Self-Learning Fuzzy Logic Controller using Q-Learning

2000 ◽  
Vol 4 (5) ◽  
pp. 349-354 ◽  
Author(s):  
Min-Soeng Kim ◽  
◽  
Sun-Gi Hong ◽  
Ju-Jang Lee
Keyword(s):  
Fuzzy Logic ◽  
Fuzzy Logic Controller ◽  
Expert Knowledge ◽  
Learning Algorithm ◽  
A Priori ◽  
Fuzzy Rule ◽  
Fuzzy Rules ◽  
Q Learning ◽  
Continuous Output ◽  
Continuous Actions

Fuzzy logic controllers consist of if-then fuzzy rules generally adopted from a priori expert knowledge. However, it is not always easy or cheap to obtain expert knowledge. Q-learning can be used to acquire knowledge from experiences even without the model of the environment. The conventional Q-learning algorithm cannot deal with continuous states and continuous actions. However, the fuzzy logic controller can inherently receive continuous input values and generate continuous output values. Thus, in this paper, the Q-learning algorithm is incorporated into the fuzzy logic controller to compensate for each method’s disadvantages. Modified fuzzy rules are proposed in order to incorporate the Q-learning algorithm into the fuzzy logic controller. This combination results in the fuzzy logic controller that can learn through experience. Since Q-values in Q-learning are functional values of the state and the action, we cannot directly apply the conventional Q-learning algorithm to the proposed fuzzy logic controller. Interpolation is used in each modified fuzzy rule so that the Q-value is updatable.

scholarly journals An ELM-based single input rule module and its application in power generation

2020 ◽  
Vol 11 (1) ◽  
pp. 359
Author(s):  
Chong Tak Yaw ◽  
Shen Young Wong ◽  
Keem Sian Yap
Keyword(s):  
Power Generation ◽  
Extreme Learning Machine ◽  
Fuzzy Inference ◽  
Learning Algorithm ◽  
Fuzzy Rule ◽  
Fuzzy Rules ◽  
Benchmark Datasets ◽  
Single Input ◽  
Learning Machine ◽  
Hidden Neurons

Extreme Learning Machine (ELM) is widely known as an effective learning algorithm than the conventional learning methods from the point of learning speed as well as generalization. In traditional fuzzy inference method which was the "if-then" rules, all the input and output objects were assigned to antecedent and consequent component respectively. However, a major dilemma was that the fuzzy rules' number kept increasing until the system and arrangement of the rules became complicated. Therefore, the single input rule modules connected type fuzzy inference (SIRM) method where consociated the output of the fuzzy rules modules significantly. In this paper, we put forward a novel single input rule modules based on extreme learning machine (denoted as SIRM-ELM) for solving data regression problems. In this hybrid model, the concept of SIRM is applied as hidden neurons of ELM and each of them represents a single input fuzzy rules. Hence, the number of fuzzy rule and the number of hidden neuron of ELM are the same. The effectiveness of proposed SIRM-ELM model is verified using sigmoid activation functions based on several benchmark datasets and a NOx emission of power generation plant.  Experimental results illustrate that our proposed SIRM-ELM model is capable of achieving small root mean square error, i.e., 0.027448 for prediction of NO<sub>x</sub> emission.

scholarly journals Aplikasi Pengambilan Keputusan dengan Metode Tsukamoto pada Penentuan Tingkat Kepuasan pelanggan

CAUCHY ◽  
2015 ◽  
Vol 4 (1) ◽  
pp. 10 ◽  
Author(s):  
Venny Riana Riana Agustin ◽  
Wahyu Henky Irawan
Keyword(s):  
Decision Making ◽  
Fuzzy Logic ◽  
Fuzzy Inference ◽  
Weighted Average ◽  
Fuzzy Rules ◽  
Quality Of Services ◽  
Inference System ◽  
A Value ◽  
Fuzzy Logic Analysis

Tsukamoto method is one method of fuzzy inference system on fuzzy logic for decision making. Steps of the decision making in this method, namely fuzzyfication (process changing the input into kabur), the establishment of fuzzy rules, fuzzy logic analysis, defuzzyfication (affirmation), as well as the conclusion and interpretation of the results. The results from this research are steps of the decision making in Tsukamoto method, namely fuzzyfication (process changing the input into kabur), the establishment of fuzzy rules by the general form IF a is A THEN B is B, fuzzy logic analysis to get alpha in every rule, defuzzyfication (affirmation) by weighted average method, as well as the conclusion and interpretation of the results. On customers at the case, in value of 16 the quality of services, the value of 17 the quality of goods, and value of 16 a price, a value of the results is 45,29063 and the level is low satisfaction

Self-Tuning for Fuzzy Rule Generation Based upon Fuzzy Singleton-type Reasoning Method

1999 ◽  
Vol 3 (3) ◽  
pp. 200-206 ◽  
Author(s):  
Yan Shi ◽  
◽  
Masaharu Mizumoto ◽  
Keyword(s):  
Clustering Algorithm ◽  
Learning Algorithm ◽  
Fuzzy Model ◽  
Fuzzy Rule ◽  
Fuzzy Rules ◽  
Rule Generation ◽  
Nonlinear Functions ◽  
Tuning Parameters ◽  
Tuning Method ◽  
Self Tuning

Using fuzzy singleton-type reasoning method, we propose a self-tuning method for fuzzy rule generation. We give a neurofuzzy learning algorithm for tuning fuzzy rules under fuzzy singleton-type reasoning method, then roughly design initial tuning parameters of fuzzy rules based on a fuzzy clustering algorithm before learning a fuzzy model. This should reduce learning time and fuzzy rules generated by our approach are reasonable and suitable for the identified model. We demonstrate our proposal’s efficiency by identifying nonlinear functions.

scholarly journals EAQR: A Multiagent Q-Learning Algorithm for Coordination of Multiple Agents

Complexity ◽  
2018 ◽  
Vol 2018 ◽  
pp. 1-14 ◽  
Author(s):  
Zhen Zhang ◽  
Dongqing Wang
Keyword(s):  
Sensor Network ◽  
Repeated Games ◽  
Learning Algorithm ◽  
The Other ◽  
Q Value ◽  
Multiple Agents ◽  
Distributed Sensor ◽  
Q Learning ◽  
Network Problem ◽  
Distributed Sensor Network

We propose a cooperative multiagent Q-learning algorithm called exploring actions according to Q-value ratios (EAQR). Our aim is to design a multiagent reinforcement learning algorithm for cooperative tasks where multiple agents need to coordinate their behavior to achieve the best system performance. In EAQR, Q-value represents the probability of getting the maximal reward, while each action is selected according to the ratio of its Q-value to the sum of all actions’ Q-value and the exploration rate ε. Seven cooperative repeated games are used as cases to study the dynamics of EAQR. Theoretical analyses show that in some cases the optimal joint strategies correspond to the stable critical points of EAQR. Moreover, comparison experiments on stochastic games with finite steps are conducted. One is the box-pushing, and the other is the distributed sensor network problem. Experimental results show that EAQR outperforms the other algorithms in the box-pushing problem and achieves the theoretical optimal performance in the distributed sensor network problem.

Artificial Bee Colony-Fuzzy Q Learning for Reinforcement Fuzzy Control (Truck Backer-Upper Control Problem)

2016 ◽  
Vol 24 (01) ◽  
pp. 123-136 ◽  
Author(s):  
Sima Saeed ◽  
Aliakbar Niknafs
Keyword(s):  
Fuzzy Control ◽  
Fuzzy System ◽  
Food Source ◽  
Artificial Bee Colony ◽  
Fuzzy Inference ◽  
Food Resource ◽  
Q Value ◽  
Inference System ◽  
Q Learning ◽  
Bee Colony

A new method for reinforcement fuzzy controllers is presented by this article. The method uses Artificial Bee Colony algorithm based on Q-Value to control reinforcement fuzzy system; the algorithm is called Artificial Bee Colony-Fuzzy Q learning (ABC-FQ). In fuzzy inference system, precondition part of rules is generated by prior knowledge, but ABC-FQ algorithm is responsible to achieve the best combination of actions for the consequence part of the rules. In ABC-FQ algorithm, each combination of actions is considered a food source for consequence part of the rules and the fitness level of this food source is determined by Q-Value. ABC-FQ Algorithm selects the best food resource, which is the best combination of actions for fuzzy system, using Q criterion. This algorithm tries to generate the best reinforcement fuzzy system to control the agent. ABC-FQ algorithm is used to solve the problem of Truck Backer-Upper Control, a reinforcement fuzzy control. The results have indicated that this method arrives to a result with higher speed and fewer trials in comparison to previous methods.

scholarly journals Fuzzy Inference Based Connection Admission Control in ATM Networks

1997 ◽  
Vol 1 (1) ◽  
pp. 14-22
Author(s):  
Kiyohiko Uehara ◽  
◽  
Kaoru Hirota ◽  
Keyword(s):  
Admission Control ◽  
Upper Bound ◽  
Fuzzy Inference ◽  
Learning Algorithm ◽  
Fuzzy Rules ◽  
Connection Admission Control ◽  
Atm Networks ◽  
Back Propagation Algorithm ◽  
Inference Method ◽  
Connection Admission

A connection admission control (CAC) method is proposed for asynchronous transfer mode (ATM) networks by applying the fuzzy inference and learning algorithm of neural networks. In order to guarantee the allowed cell loss ratio (CLR) in CAC, the upper bound of CLR must be used as the criterion for judging whether an incoming call can be accepted or not. For estimating the upper bound of CLR from observed CLR data, fuzzy inference, based on a weighted mean of fuzzy sets, is adopted. This inference method can effectively estimate the possibility distribution of CLR by applying the error back-propagation algorithm with the proposed energy functions in learning and provide the upper bound of CLR efficiently from the distribution. A self-compensation mechanism for estimation errors is also provided, which is simple enough to work in real time by taking advantage of the fuzzy inference method adopted. Fuzzy rules in the area with no observed data are generated by extrapolation from adjacent fuzzy rules in the area with observed data. This increases the multiplex gain, thereby guaranteeing the allowed CLR as much as possible. The simulation results show the feasibility of the proposed CAC method.

Optimization Design of Fuzzy Control Rules Based on Ant Colony Algorithm

2014 ◽  
Vol 716-717 ◽  
pp. 1662-1665
Author(s):  
Ya Lang Xing ◽  
He Xin ◽  
Jin Cheng Zhao
Keyword(s):  
Fuzzy Control ◽  
Optimization Design ◽  
Ant Colony Algorithm ◽  
Fuzzy Rule ◽  
Coefficient Matrix ◽  
Ant Colony ◽  
Fuzzy Rules ◽  
Rule Base ◽  
Control Rules ◽  
And Control

To avoid the fuzzy rules getting into “rule exploding” in fuzzy control system, a fuzzy control rules optimization algorithm based on compatibility coefficient is proposed. The method defines the compatibility coefficient of fuzzy rules, and the compatibility coefficient matrix is used to be the heuristic information in ant colony algorithm. Ant colony algorithm is used to optimize designed complete fuzzy rule base. Simulation results show that the fuzzy rules have good compatibility and control performance.

FUZZY CONTROLS BY PRODUCT-SUM-GRAVITY METHOD DEALING WITH FUZZY RULES OF EMPHATIC AND SUPPRESSIVE TYPES

1994 ◽  
Vol 02 (03) ◽  
pp. 305-319 ◽  
Author(s):  
M. MIZUMOTO
Keyword(s):  
Fuzzy Control ◽  
Membership Function ◽  
Fuzzy Inference ◽  
Good Control ◽  
Fuzzy Rules ◽  
Gravity Method ◽  
Control Rules

This paper shows that emphatic effects on fuzzy inference results are observed under product-sum-gravity method by using fuzzy control rules whose consequent part is characterized by a membership function whose grades are greater than 1. Suppressive effects are also realized by employing fuzzy control rules whose consequent part is characterized by a negative-valued membership function. It is shown that good control results are obtained by using the fuzzy control rules of emphatic and suppressive types.

Co-Evolution of Fuzzy Controller for the Mobile Robot Control

2004 ◽  
Vol 8 (4) ◽  
pp. 356-361 ◽  
Author(s):  
Kwang-Sub Byun ◽  
◽  
Chang-Hyun Park ◽  
Kwee-Bo Sim
Keyword(s):  
Mobile Robot ◽  
Fuzzy Systems ◽  
Fuzzy Inference ◽  
Fuzzy Controller ◽  
Fuzzy Rule ◽  
Fuzzy Rules ◽  
Rule Base ◽  
Membership Functions ◽  
Mobile Robot Control ◽  
Evolution Algorithms

In this paper, we design the fuzzy rules using a modified Nash Genetic Algorithm. Fuzzy rules consist of antecedents and consequents. Because this paper uses the simplified method of Sugeno for the fuzzy inference engine, consequents have not membership functions but constants. Therefore, each fuzzy rule in this paper consists of a membership function in the antecedent and a constant value in the consequent. The main problem in fuzzy systems is how to design the fuzzy rule base. Modified Nash GA coevolves membership functions and parameters in consequents of fuzzy rules. We demonstrate this co-evolutionary algorithm and apply to the design of the fuzzy controller for a mobile robot. From the result of simulation, we compare modified Nash GA with the other co-evolution algorithms and verify the efficacy of this algorithm.

Sign in / Sign up

Export Citation Format

Share Document