In light of increasing natural disasters, especially earthquakes, there is a pressing need for innovative technologies that can greatly enhance the efficiency and effectiveness of rescue operations. Breakthrough methodologies for data collection and analysis can support the design of disruptive architectures capable of gathering and analyzing information from unstable and damaged data networks. This can aid in planning fast and targeted recovery operations, maximizing the number of lives saved. This paper proposes computational architecture that utilizes fuzzy rule-based systems with real-time sensor data to enable rapid and interpretable decision-making processes during critical search and rescue missions. Preliminary simulations indicate that fuzzy systems are suitable for efficient use in critical scenarios, such as those related to post-earthquakes.
Fuzzy logic has enhanced decision-making by enabling more effective modeling of uncertainty in human judgments through membership functions. However, current approaches to constructing membership functions often over-look active collaboration between decision-maker and decision analyst, leading to a lack of interpretability and personalization. This paper introduces a socio-technical methodology, the Deck of Cards Method, to co-construct trapezoidal fuzzy numbers tailored to decision-maker's semantics. We illustrate the method's working through the performance evaluation example in a corporate context.
The paper proposes a multi-attribute decisionmaking method that integrates a basic uncertain linguistic information (BULI) language model grounded in prospect theory (PT) with a two-layer weight mechanism based on recursive least squares (RLS). On the one hand, linguistic term sets facilitate the transformation of linguistic information, but their internal conversion processes often lack interpretability. As a first attempt, this study introduces prospect theory into a uniform set of linguistic terms for mapping. The objective is to incorporate human psychological expectations into the decisionmaking model via nonlinear mapping, thereby more accurately simulating the influence of psychological factors such as loss aversion in decision-making and enhancing the rationality and interpretability of the model. On the other hand, traditional methods for determining attribute weights typically assign fixed values to weights, limiting their ability to flexibly adapt to actual requirements. The RLS method, by incorporating a “two-layer” structure (time and confidence), enables dynamic adjustment and optimization of attribute weights. This ensures that the final decision-making outcomes comprehensively reflect the integrated influence of various dimensions. Finally, we demonstrated the feasibility of the proposed method through a case study on subway passenger satisfaction.
Normalizing anomaly scores as probability estimates has become a standard practice. Scores turn into interpretable values and become suitable to build ensembles. While this is undoubtedly useful, it introduces a conceptual imprecision and imposes unnecessary constraints. Algorithms usually output raw scores that are measures of a data space property, and do not necessarily reflect a probability. By modeling the anomaly (a) quality of a point as a probability, it naturally follows to portray its normal (n) quality as complementary (for all a,n a + n = 1). However, this causes an unnecessary tension between both variables, a detriment to the information that scores store and even potential misinterpretations. We propose an alternative normalization where anomaly and normal are antagonistic fuzzy sets independently modeled, i.e., a, n is an element of [0; 1]. Thus, they are more consistently interpreted as measures of vagueness rather than uncertainty. This allows, for example, to acknowledge the existence of samples that are neither normal nor anomalous (e.g., noise). Beyond overcoming a conceptual problem, our proposal has the practical benefit of unifying scores while minimizing information loss. We show, evaluate and compare its effects with 47 popular field dataset benchmarks. Results confirm that our approach retains more information and leads to better performances. Anomaly membership degrees yield ensembles with enhanced overall ranking, better precision-recall balance, and higher precision in top scores.
Energy fraud in power distribution networks poses significant economic and technical challenges, impacting both distribution companies and legitimate consumers. While existing approaches for fraud detection have demonstrated promise, they often rely on complex, data-intensive models that lack interpretability. This paper addresses these challenges by proposing a novel fuzzy characterization of electricity consumption time series to facilitate fraud detection and interpretation. The proposed method incorporates the semantics used by domain experts, particularly the identification of fraud patterns as “a drop in energy consumption followed by a period of low or constant consumption.” By employing linguistic labels and fuzzy set modeling, this approach captures expert knowledge in a formalized and interpretable manner. The efficacy of the proposed method is demonstrated through experimental evaluations, highlighting its potential to enhance fraud detection in smart grids while improving interpretability and trust in automated decisions.
Mobile robot navigation is a significant research challenge, especially in scenarios where efficient learning is crucial. Deep reinforcement learning (DRL) offers a promising alternative to traditional control strategies, yet its effectiveness is often limited by sparse reward structures and simple episode termination criteria. This work proposes two modifications to enhance DRL-based path planning. First, we revise the episode termination process to ensure that each robot accumulates rewards until it reaches its individual terminal condition, thereby preventing premature episode endings and promoting improved learning. Second-and most importantly for this study-we introduce a reward shaping mechanism that leverages fuzzy logic and the Adaptive Neuro-Fuzzy Inference System (ANFIS) to produce continuous, context-aware sub-rewards that provide detailed performance feedback to each agent throughout its navigation task. Simulation results show that the revised episode termination significantly improves learning efficiency, while the fuzzy logic-based reward shaping notably enhances reward accumulation and overall performance. The fuzzy logic method effectively encodes the knowledge and manages uncertainty, offering a flexible and interpretable framework for decisionmaking in complex environments. These findings demonstrate the potential of incorporating fuzzy logic into DRL frameworks to achieve faster convergence and superior navigation performance in mobile robot systems.
In this contribution, the Borda rule has been extended to the case of reciprocal preferences through the deck of cards method. Initially, each voter arranges the alternatives in a weak order and, subsequently, introduces a number of cards between consecutive equivalence classes representing the corresponding preference intensities. A consistency condition generates the number of cards between non-consecutive equivalence classes. Taking this information into account, a reciprocal preference on the set of alternatives satisfying a strong transitivity condition is associated with each voter. Then, a score is given to each alternative for each voter: the sum of intensities of preference between that alternative and the others. Finally, the alternatives are ranked by means of the sum of the individual scores.
The study of aggregation functions, either from a theoretical point of view or for their interesting applications, is a hot topic. In this paper we propose a new method for constructing aggregation functions on the set of Zadeh's discrete Z-numbers based on total orders. This method is characterised by using aggregation functions defined on the set of discrete fuzzy numbers whose support is a closed interval of the finite chain L-n = {0, 1, center dot center dot center dot, n}. Furthermore, it is shown that this construction method preserves important properties of the chosen initial aggregation functions.
Photovoltaic power forecasting (PPF) plays a crucial role in advancing renewable energy. With the growth grid-connected PPF systems, ensuring their safe and stable operation is becoming a key challenge. Therefore, it is imperative to accurately forecast photovoltaic power. This study presents an ensemble model based on long short-term memory (LSTM) and extended belief rule base (EBRB), so called LSTM-EBRB ensemble model, for PPF, where the LSTM and EBRB serve as base learners of the stacking ensemble framework, respectively, so that the proposed model can combine the deep learning capability of the LSTM and the fuzzy information processing capability of the EBRB to effectively capture long-term dependencies and uncertainties in PPF. To validate the effectiveness of the proposed model, experiments are conducted using the real dataset from Yumara solar system photovoltaic plant at the Desert Knowledge Australia Solar Center. The results demonstrate that the proposed model can achieves high accuracy better than six other forecasting models in the terms of three evaluation metrics: root mean square error, mean absolute error, and coefficient of determination.
The Hybrid Fuzzy Load Balancing for Cloud Computing (Hybrid-FLBCC) approach introduces a fuzzy model integrated with machine learning algorithms, enabling dynamic feature selection of input variables and supporting the generation of dynamic base rules to predict optimal Virtual Machine migrations. As an evolutionary extension of the Int-FLBCC system, it leverages the concept of Flexible Computing to enhance server consolidation in Cloud Computing environments by refining the inference process based on the utilization levels of Physical Machines. The feature selection methodology identifies the most relevant variables for consolidation, facilitating the definition of improved decision-making criteria. This refinement results in more interpretable fuzzy rule sets, simplifies system implementation, and may contribute to reducing computational resource usage. The promising results highlight the potential of hybrid approaches for intelligent decision-making within cloud infrastructures.
Understanding and modeling the policies of other agents in multi-agent systems (MAS) is crucial for enabling effective interaction in complex, dynamic environments. This paper introduces the Fuzzy-CLAM method, a novel approach in the agent policy modeling field that integrates fuzzy logic with contrastive learning-based agent modeling (CLAM). Fuzzy-CLAM provides a transparent and explainable decision-making process by applying fuzzy logic to map agents' decision logic into if-then rules. A structure learning mechanism is employed to dynamically adapt fuzzy rules to changes in the policy representation distribution, ensuring robustness and reducing redundancy. Experimental results demonstrate that Fuzzy-CLAM achieves a competitive level of learning performance while offering high explainability for agents’ decision-making processes. These insights further suggest potential avenues for refining and advancing agent policy modeling techniques in MAS.
In this paper we study the relationship between the inclusion and the similarity measures in type-2 fuzzy sets. In particular, we show how to obtain measures of inclusion from measures of similarity.
This paper presents a real-world application of approximate reasoning to sludge dewatering. One of the constraint is that the same rule base must be applicable in a network of heterogeneous wastewater treatment plants. In particular, they differ on how they are equipped of sensors or not. We propose a solution based on approximate reasoning to process both precise, i.e. measured instantly with an accurate sensor or with a delay of 24 hours in an analysis laboratory, and imprecise inputs, that come instantly from a low-cost sensor or even from human perception. We tested on a small model of a decanter centrifuge to study the behavior of the fuzzy inference system. The results are presented at the end of the paper and show that the fuzzy inference system tends to an optimal state even if the decanter centrifuge is disturbed.
Automation of industrial processes can have a real impact on the improvement of production or greater control over the maintenance of tools necessary for its smooth operation. In particular, solutions based on computer vision have a number of applications here, both in the context of production optimization and fault prevention. In this work, we present a method for identifying defects in metal surfaces, which are components of many industrial systems. The proposed solution is based on a semantic segmentation approach using a nested encoderdecoder architecture. What distinguishes the method from others is the implementation of a dynamic fuzzy-based cost assignment module, which manipulates the cost function during training to enhance the obtained results. With the help of a properly prepared fuzzy controller, which verifies the course of model adaptation to a specific case, the strategy can outperform the basic approach.
Accurate river water level prediction is crucial for disaster management and water resource planning. This study introduces Flex Bag, a hybrid framework integrating fuzzy c-means clustering with machine learning models. By combining global relationships from base models with localized insights from FCM features, Flex Bag enhances predictive accuracy. Experiments on the Niyodo River dataset show a reduced root mean square error (34.09) compared to Cascaded-ANFIS (35.51), highlighting its robustness in rainfall-runoff modeling. Visual analyses validate the framework's effectiveness in capturing spatial-temporal patterns. Flex Bag offers a scalable solution for hydrological challenges and sets a foundation for handling real-world issues like missing and noisy data.
This paper presents a model-reference control approach for nonlinear systems in the Takagi-Sugeno form and applies it to photovoltaic power plants to provide instantaneous reserve. The proposed scheme is applied to a photovoltaic power plant with two stages separated into an aggregated photovoltaic array and DC/DC converter. The coefficients of the modelreference controller are calculated using a set of proposed linear matrix inequality (LMI) based conditions. In addition to the LMI constraint for the model reference design, the paper also introduces region constraints. It is shown how a suitable combination of the proposed constraints can specifically improve the controller performance. Simulation studies verify the applicability of the design approach.
To the best of our knowledge, almost all general type-2 (GT2) fuzzy inference systems (FISs) currently used in practice rely on discretizing the domains of the primary and secondary membership functions (MFs) and have type-2 (T2) inputs, antecedents, and - if applicable - consequents whose secondary MFs are convex and normal. Only a few recent studies are concerned with non-convex, normal secondary MFs and their applications in T2 FISs. Other authors have measured the similarity between GT2 fuzzy sets (FSs) with non-normal secondary MFs but have not used them in T2 FISs. In this paper, we clarify that the class of convex upper semicontinous FSs on [0], [1] yields a bounded lattice in the sense of Garrett Birkhoff. This result implies that the class of GT2 FSs whose secondary MFs are convex and upper semicontinuous, but not necessarily normal, constitutes a bounded lattice as well. We employ the latter in order to propose GT2 FISs having inputs, antecedents, and consequents that are GT2 FSs with not necessarily normal secondary MFs. We finish by presenting a simple, but novel application in computing with words (CWW).
Out-of-distribution (OOD) detection refers to the task of identifying whether a given input belongs to classes that were unknown during training (i.e., OOD data). Due to the unavailability and diversity of OOD data, this task involves significant uncertainty and poses substantial challenges. Most existing solutions for OOD detection focus on designing a score function to determine whether a given data point is OOD. However, these methods often fail to effectively address the uncertainty inherent in OOD detection. Notably, fuzzy sets and fuzzy logic exhibit significant advantages in handling uncertainty. Despite this, the application of fuzzy sets to OOD detection remains largely unexplored. In this paper, we propose a novel score function, termed the positive fuzzy similarity score (PFS), which is based on fuzzy sets and cosine similarity, to enhance OOD detection performance. Through extensive experiments, we evaluate our proposed score function against existing OOD detection methods using three evaluation metrics. The results demonstrate the superiority of our method in addressing the challenges of OOD detection.
We present an enhanced fuzzified variant of the classical Frankot-Chellappa algorithm for reconstructing surfaces from normal vector fields by using a multiplicative fuzzification process and an adaptive defuzzification strategy. By incorporating local spread metrics and noise-level estimation, the adaptive defuzzification process dynamically adjusts reconstruction weights, significantly improving robustness to noise. Experimental results demonstrate that the proposed method achieves superior accuracy compared to a previous fuzzy approach, particularly under noisy conditions.
This paper introduces ADP-T2LIMG, a new datadriven interval type-2 fuzzy time series (FTS) forecasting method applied in load forecasting applications. The model integrates interval type-2 fuzzy logic with an autonomous data partitioning (ADP) algorithm to address the challenges of uncertainty and non-linearity in energy demand data. This new model leverages the parameters from ADP's output, utilizing this enriched information to more accurately design and define the fuzzy sets in the forecasting model, aligning them more closely with the data distribution. The proposed model is evaluated on five datasets, including energy consumption data from the PJM and GEFCom 2012 datasets, using a sliding window methodology for crossvalidation. Experimental results demonstrate that ADP-T2LIMG outperforms state-of-the-art forecasting techniques across all datasets in terms of root mean square error (RMSE). Statistical analysis further confirms the model's superiority, with ADPT2LIMG consistently achieving the highest rankings.
