Jeffery D. Weir

An Adaptive Particle Swarm Optimization With Multiple Adaptive Methods

Particle swarm optimization (PSO) has attracted much attention and has been applied to many scientific and engineering applications in the last decade. Most recently, an intelligent augmented particle swarm optimization with multiple adaptive methods (PSO-MAM) was proposed and was demonstrated to be effective for diverse functions. However, inherited from PSO, the performance of PSO-MAM heavily depends on the settings of three parameters: the two learning factors and the inertia weight. In this paper, we propose a parameter control mechanism to adaptively change the parameters and thus improve the robustness of PSO-MAM. A new method, adaptive PSO-MAM (APSO-MAM) is developed that is expected to be more robust than PSO-MAM. We comprehensively evaluate the performance of APSO-MAM by comparing it with PSO-MAM and several state-of-the-art PSO algorithms and evolutionary algorithms. The proposed parameter control method is also compared with several existing parameter control methods. The experimental results demonstrate that APSO-MAM outperforms the compared PSO algorithms and evolutionary algorithms, and is more robust than PSO-MAM.

An intelligent augmentation of particle swarm optimization with multiple adaptive methods

Over the last two decades, the newly developed optimization technique - Particle Swarm Optimization (PSO) has attracted great attention. Two common criticisms exist. First, most existing PSOs are designed for a specific search space thus an algorithm performing well on a diverse set of problems is lacking. Secondly, PSO suffers premature convergence. To address the first issue, we propose to augment PSO via the fusion of multiple search methods. An intelligent selection mechanism is developed based on an effectiveness index to trigger appropriate search methods. In this research, two search techniques are studied: a non-uniform mutation-based method and an adaptive sub-gradient method. We further improve the proposed PSO using adaptive Cauchy mutation to prevent premature convergence. As a result, an augmented PSO with multiple adaptive methods (PSO-MAM) is proposed. The performance of PSO-MAM is tested on 43 functions (uni-modal, multi-modal, non-separable, shifted, rotated, noisy and mis-scaled functions). The results are compared in terms of solution quality and convergence speed with 10 published PSO methods. The experimental results demonstrate PSO-MAM outperforms the comparison algorithms on 36 out of 43 functions. We conclude, while promising, there is still room for improving PSO-MAM on complex multi-modal functions (e.g., rotated multi-modal functions).

Decentralized operation strategies for an integrated building energy system using a memetic algorithm

The emerging technology in net-zero building and smart grids drives research moving from centralized operation decisions on a single building to decentralized decisions on a group of buildings, termed a building cluster which shares energy resources locally and globally. However, current research has focused on developing an accurate simulation of single building energy usage which limits its application to building clusters as scenarios such as energy sharing and competition cannot be modeled and studied. We hypothesize that the study of energy usage for a group of buildings instead of one single building will result in a cost effective building system which in turn will be resilient to power disruption. To this end, this paper develops a decision model based on a building cluster simulator with each building modeled by energy consumption, storage and generation sub modules. Assuming each building is interested in minimizing its energy cost, a bi-level operation decision framework based on a memetic algorithm is proposed to study the tradeoff in energy usage among the group of buildings. Two additional metrics, measuring the comfort level and the degree of dependencies on the power grid are introduced for the analysis. The experimental result demonstrates that the proposed framework is capable of deriving the Pareto solutions for the building cluster in a decentralized manner. The Pareto solutions not only enable multiple dimensional tradeoff analysis, but also provide valuable insight for determining pricing mechanisms and power grid capacity.

AHPS2: An optimizer using adaptive heterogeneous particle swarms

Abstract Particle swarm optimization (PSO) has suffered from premature convergence and lacked diversity for complex problems since its inception. An emerging advancement in PSO is multi-swarm PSO (MS-PSO) which is designed to increase the diversity of swarms. However, most MS-PSOs were developed for particular problems so their search capability on diverse landscapes is still less than satisfactory. Moreover, research on MS-PSO has so far treated the sub-swarms as cooperative groups with minimum competition (if not none). In addition, the size of each sub-swarm is set to be fixed which may encounter excessive computational cost. To address these issues, a novel optimizer using Adaptive Heterogeneous Particle SwarmS (AHPS2) is developed in this research. In AHPS2, multiple heterogeneous swarms, each consisting of a group of homogenous particles having similar learning strategy, are introduced. Two complementary search techniques, comprehensive learning and a subgradient method, are studied. To best take advantage of the heterogeneous learning strategies, an adaptive competition strategy is proposed so the size of each swarm can be dynamically adjusted based on its group performance. The analyses of the swarm heterogeneity and the competition models are presented to validate the effectiveness. Furthermore, comparisons between AHPS2 and state-of-the-art algorithms are grouped into three categories: 36 regular benchmark functions (30-dimensional), 20 large-scale benchmark functions (1000-dimensional) and 3 real-world problems. Experimental results show that AHPS2 displays a better or comparable performance compared to the other swarm-based or evolutionary algorithms in terms of solution accuracy and statistical tests.

A Practical Procedure for Customizable One-Way Sensitivity Analysis in Additive Value Models

Although a wide variety of sensitivity analysis methods for additive value models appear in the literature, the most commonly used and easily understood is one-way sensitivity analysis. We address concerns from practice with this traditional method and build the mathematical foundation for a new and robust one-way sensitivity analysis approach that enables the decision maker(s) to gain a greater ownership of the model. A traditional one-way sensitivity analysis varies one attribute weight while keeping all others proportionally constant; although useful, it also has limitations, particularly in group decision making. Our proposed methodology, customizable one-way sensitivity analysis (COSA), retains the simplicity of the traditional method but extends it to a more powerful form. COSA allows the decision maker(s) to tailor the sensitivity analysis to the desired decision context by assigning weight coefficients of elasticity to the attribute weights. These parameters let the decision maker choose how the adjusted weight change is distributed throughout the model. Furthermore, we provide a detailed discussion about performing sensitivity analysis on complex hierarchies with more than one tier and include an applied example to demonstrate COSAs usefulness. Overall, a consistent standard for the one-way sensitivity analysis of additive value models is provided for the decision analysis community.

A recommendation system for meta-modeling

A meta-learning based recommendation system for meta-modeling is proposed.Novel meta-features for geometrical characterization on black-box problems are proposed.Model-based meta-learners generally outperforms instance-based meta-leaners.Singular value decomposition boosts the performance of the recommendation system.Experimental results indicate the proposed system significantly improves the modeling efficiency and facilitates model selection. Various meta-modeling techniques have been developed to replace computationally expensive simulation models. The performance of these meta-modeling techniques on different models is varied which makes existing model selection/recommendation approaches (e.g., trial-and-error, ensemble) problematic. To address these research gaps, we propose a general meta-modeling recommendation system using meta-learning which can automate the meta-modeling recommendation process by intelligently adapting the learning bias to problem characterizations. The proposed intelligent recommendation system includes four modules: (1) problem module, (2) meta-feature module which includes a comprehensive set of meta-features to characterize the geometrical properties of problems, (3) meta-learner module which compares the performance of instance-based and model-based learning approaches for optimal framework design, and (4) performance evaluation module which introduces two criteria, Spearmans ranking correlation coefficient and hit ratio, to evaluate the system on the accuracy of model ranking prediction and the precision of the best model recommendation, respectively. To further improve the performance of meta-learning for meta-modeling recommendation, different types of feature reduction techniques, including singular value decomposition, stepwise regression and ReliefF, are studied. Experiments show that our proposed framework is able to achieve 94% correlation on model rankings, and a 91% hit ratio on best model recommendation. Moreover, the computational cost of meta-modeling recommendation is significantly reduced from an order of minutes to seconds compared to traditional trial-and-error and ensemble process. The proposed framework can significantly advance the research in meta-modeling recommendation, and can be applied for data-driven system modeling.

Testing local search move operators on the vehicle routing problem with split deliveries and time windows

The vehicle routing problem (VRP) is an important transportation problem. The literature addresses several extensions of this problem, including variants having delivery time windows associated with customers and variants allowing split deliveries to customers. The problem extension including both of these variations has received less attention in the literature. This research effort sheds further light on this problem. Specifically, this paper analyzes the effects of combinations of local search (LS) move operators commonly used on the VRP and its variants. We find when paired with a MAX-MIN Ant System constructive heuristic, Or-opt or 2-opt* appear to be the ideal LS operators to employ on the VRP with split deliveries and time windows with Or-opt finding higher quality solutions and 2-opt* requiring less run time.

A Three-Phase Approach to Solving the Bidline Problem

This paper describes a three-phase approach to solving the bidline generation problem within airline flightcrew scheduling. Phase 1 builds “patterns” from existing pairings. Phase 2 builds bidlines from the “patterns” found in Phase 1 and solves a set partitioning problem to generate a final schedule. If Phase 2 fails to cover enough of the scheduled work, Phase 3 is used to fill in the uncovered pairings. Along with this methodology is a new rule set that tries to improve the work–rest schedule for flightcrews by trying to take into account circadian rhythms. This new rule set is an attempt at addressing some of the flightcrewss quality of life issues.

Robust goal programming using different robustness echelons via norm-based and ellipsoidal uncertainty sets

The notion of robust goal programming (RGP) using cardinality-constrained robustness via interval-based uncertainty was first examined over a decade ago. Since then, the RGP methodology has not been widely researched, specifically when considering different uncertainty sets to implement. Within this context, this paper compares interval-based and norm-based uncertainty sets using cardinality-constrained robustness. Strict robustness using ellipsoidal uncertainty sets is also examined in the RGP realm. The aforementioned methods are demonstrated for a simple instance from the literature, and the results are summarized. Conclusions are made regarding the proposed RGP models when likened to a similar RGP model seen in the literature. Further, the suitability of each RGP model is offered when a decision makers risk preference or computing availability are taken into consideration. Inferences are made regarding the effectiveness of each uncertainty set in the context of solutions that are relatively unaffected by data uncertainty – that is, robust solutions.

Tooth-to-Tail Impact Analysis: Combining Econometric Modeling and Bayesian Networks to Assess Support Cost Consequences Due to Changes in Force Structure

Current constraints in the fiscal environment are forcing the Air Force, and its sister services, to assess force reduction considerations. With significant force reduction comes the need to model and assess the potential impact that these changes may have on support resources. Previous research has remained heavily focused on a ratio approach for linking the tooth and tail ends of the Air Force cost spectrum and, although recent research has augmented this literature stream by providing more statistical rigor behind tooth-to-tail relationships, an adequate decision support tool has yet to be explored to aid decision-makers. The authors of this research directly address this concern by introducing a systematic approach to perform tooth-to-tail policy impact analysis. First, multivariate linear regression is applied to identify relationships between the tooth and tail. Then, a novel decision support system with Bayesian networks is introduced to model the tooth-to-tail cost consequences while capturing the uncertainty that often comes with such policy considerations. Through scenario analysis, the authors illustrate how a Bayesian network can provide decision-makers with (i) the ability to model uncertainty in the decision environment, (ii) a visual illustration of cause-and-effect impacts, and (iii) the ability to perform multi-directional reasoning in light of new information available to decision-makers.