Low-level Optimizations Research Articles

Automatic Generation of Truffle-based Interpreters for Domain-Specific Languages.

Numerous language workbenches have been proposed over the past decade to ease the definition of Domain-Specific Languages (DSLs). Language workbenches enable language designers to specify DSLs using high-level metalanguages and to generate their implementation (e.g., parsers, interpreters) and tool support (e.g., editors, debuggers) automatically. However, little attention has been given to the performance of the resulting interpreters. In many domains where performance is key (e.g., scientific and high-performance computing), this forces language designer to hand-craft ad-hoc optimizations in the interpreter implementations, or to lose compatibility with tool support. In this paper, we propose to systematically exploit the domain-specific information of language specifications to derive optimized Truffle-based language interpreters executed over the GraalVM. We implement our approach on top of the Eclipse Modeling Framework (EMF) by complementing its existing compilation chain with Truffle-specific information, which drives the GraalVM to benefit from an optimized just-in-time compilation. A key benefit of our approach is that it leverages existing language specifications and does not require additional information from language designers who remain oblivious of Truffle's low-level intricacies and JIT optimizations in general while staying compatible with tool support. We evaluate our approach using a representative set of four DSLs and eight conforming programs. Compared to the standard interpreters generated by EMF running on the GraalVM, we observe an average speed-up of x1.14, ranging from x1.07 to x1.26. Although the benefits vary slightly from one DSL or program to another, we conclude that our approach yields substantial performance gains while remaining non-intrusive of EMF abstractions.

Sustainable recycle network of heterogeneous pharmaceuticals with governmental subsidies and service-levels of third-party logistics by bi-level programming approach

Reverse logistics can facilitate sustainable production, and plays an important role in protecting environment and saving resources. In this paper, the management problem of pharmaceutical reverse supply chain (PRSC) is studied, where the producer, the third party logistics company (3 PL) and the government are jointly involved in recycling the leftover pharmaceutical to make a trade-off among economic, environmental and social benefits. Different from the existing models, this paper formulates such a management problem as an integrated nonlinear bi-level programming (NBLP) model, where the producer is the leader, the 3 PL company is the follower, and the government encourages donation of pharmaceuticals and more pharmaceuticals to be recycled by subsidy and tax deduction. In our model, the recovered quantity depends on incentives and service-level of the 3 PL, and the producer as the leader pays the recovery fee to the 3 PL by taking into account heterogeneity of pharmaceuticals. To solve this complicated model, we first convert the lower-level optimization model into parametric complementarity constraints, the original model then is further transformed into a series of standard smooth problems by a partially smoothing method. An efficient algorithm is developed to find an equilibrium solution of the NBLP model. Case study and sensitivity analysis demonstrate that the proposed method in this paper can efficiently provide optimal policies for the complicated pharmaceutical recovery system, and a number of important managerial implications are revealed: (1) The unit sale price of pharmaceuticals, the 3 PL’s capacity and the tax deduction of the government can generate different impacts on the profits and the optimal collected quantities of the leftover pharmaceuticals. (2) Classification recycle of the pharmaceuticals is important to the scientific management of reverse supply chain. (3) The developed model and algorithm can efficiently improve sustainability of the recycle system of pharmaceuticals.

Two‐level area‐load modelling for OPF of power system using reinforcement learning

Load modelling is essential to the planning and operation of a power system. This study proposes a two-level hierarchical framework of real-time area-load modelling for optimal power flow (OPF). The upper-level problem is a parameter identification for an area-load model improved via using a weighting strategy decayed with time, whereas the lower-level optimisation is a dynamic OPF considering N − 1 static security constraints. In the framework, the error of the lower-level optimisation is added into the upper-level model, which guides a search direction for the load modelling toward minimising the error between the online measurement and equivalent model output as much as possible. An improved method is proposed based on function optimisation by reinforcement learning to identify the parameters of the area-load model online in real time. Simulation studies verify the effectiveness of the proposed framework, algorithm and improved strategies.

A population‐based approach to the bi‐level multifollower problem: an application to the electricity retail market

AbstractDynamic tariffs are expected to be implemented as commercial offers of electricity retailers in smart grids, conveying price signals aimed at shaping usage patterns, with potential benefits to enhance grid efficiency and reduce end‐users’ costs. Retailers and consumers have divergent goals. While the objective of the retailer is to maximize profits, the purpose of consumers is to minimize their electricity bill. The interaction between the retailer and consumers can be modeled by means of bi‐level (BL) programming: the retailer sets the prices to be charged to consumers and these react by scheduling flexible appliances according to those prices and their comfort requirements. In this work, two hybrid BL optimization approaches are proposed to solve this problem, considering one leader (retailer) and multiple followers (consumers). Two population‐based approaches were developed, a genetic algorithm (GA) and a particle swarm optimization (PSO) algorithm, to deal with the upper level problem, both encompassing an exact mixed‐integer linear programming solver to address the lower level optimization problem. Different scenarios were generated, comprising one leader (retailer imposing different price schemes) and three followers (with different consumer profiles). Typical residential appliances were considered, with different operation cycles. Also, diverse tariff structures set by the retailer were analyzed. The performance of the two algorithms was compared. Results revealed a consistent superiority of PSO over GA.

Bilevel Optimization for Cost Function Determination in Dynamic Simulation of Human Gait.

Predictive simulation based on dynamic optimization using musculoskeletal models is a powerful approach for studying human gait. Predictive musculoskeletal simulation may be used for a variety of applications from designing assistive devices to testing theories of motor control. However, the underlying cost function for the predictive optimization is unknown and is generally assumed a priori. Alternatively, the underlying cost function can be determined from among a family of possible cost functions, representing an inverse optimal control problem that may be solved using a bilevel optimization approach. In this study, a nested evolutionary approach is proposed to solve the bilevel optimization problem. The lower level optimization is solved by a direct collocation method, and the upper level is solved by a genetic algorithm. We demonstrate our approach to solve different bilevel optimization problems, including finding the weights among three common performance criteria in the cost function for normal human walking. The proposed approach was found to be effective at solving the bilevel optimization problems. This approach should provide practical utility in designing assistive devices to aid mobility, and could yield insights about the control of human walking.

Nested evolutionary algorithms for computationally expensive bilevel optimization problems: Variants and their systematic analysis

Abstract Bilevel optimization problems involve a hierarchical model where an upper level optimization problem is solved with a constraint on the optimality of a nested lower level problem. The use of evolutionary algorithms (EAs) and other metaheuristics has been gaining attention to solve bilevel problems, especially when they contain non-linear/black-box objective(s) and/or constraint(s). However, EAs typically operate in a nested mode wherein a lower level optimization is executed for each upper level solution. Evidently, this process requires excessive number of function evaluations, which might become untenable if the underlying functions are computationally expensive. In order to reduce this expense, the use of approximations (also referred to as surrogates or meta-models) has been suggested previously. However, the previous works have focused only on the use of surrogates for the lower level problem, whereas the computational expense of the upper level problem has not been considered. In this paper, we aim to make two contributions to address this research gap. The first is to introduce an improved nested EA which uses surrogate-assisted search at both levels in order to solve bilevel problems using limited number of function evaluations. The second is the revelation and a systematic investigation of a previously overlooked aspect of bilevel search – that the objective/constraints at the upper and lower levels may involve different computational expense. Consideration of this aspect can help in deciding a suitable strategy, i.e., in which level is the use of surrogates most appropriate for the given problem. Towards this end, four different nested strategies – with surrogate at either level, none or at both levels, are compared under various experimental settings. Numerical experiments are presented on a wide range of problems to demonstrate the efficacy and utility of the proposed contributions.

A Bilevel Optimization Approach for Joint Offloading Decision and Resource Allocation in Cooperative Mobile Edge Computing.

This paper studies a multiuser cooperative mobile edge computing offloading (called CoMECO) system in a multiuser interference environment, in which delay-sensitive tasks may be executed on local devices, cooperative devices, or the primary MEC server. In this system, we jointly optimize the offloading decision and computation resource allocation for minimizing the total energy consumption of all mobile users under the delay constraint. If this problem is solved directly, the offloading decision and computation resource allocation are generally generated separately at the same time. Note, however, that they are closely coupled. Therefore, under this condition, their dependency is not well considered, thus leading to poor performance. We transform this problem into a bilevel optimization problem, in which the offloading decision is generated in the upper level, and then the optimal allocation of computation resources is obtained in the lower level based on the given offloading decision. In this way, the dependency between the offloading decision and computation resource allocation can be fully taken into account. Subsequently, a bilevel optimization approach, called BiJOR, is proposed. In BiJOR, candidate modes are first pruned to reduce the number of infeasible offloading decisions. Afterward, the upper-level optimization problem is solved by ant colony system (ACS). Furthermore, a sorting strategy is incorporated into ACS to construct feasible offloading decisions with a higher probability and a local search operator is designed in ACS to accelerate the convergence. For the lower-level optimization problem, it is solved by the monotonic optimization method. In addition, BiJOR is extended to deal with a complex scenario with the channel selection. Extensive experiments are carried out to investigate the performance of BiJOR on two sets of instances with up to 400 mobile users. The experimental results demonstrate the effectiveness of BiJOR and the superiority of the CoMECO system.

Network Reconfiguration for Increasing Transportation System Resilience Under Extreme Events.

Evacuating residents out of affected areas is an important strategy for mitigating the impact of natural disasters. However, the resulting abrupt increase in the travel demand during evacuation causes severe congestions across the transportation system, which thereby interrupts other commuters' regular activities. In this article, a bilevel mathematical optimization model is formulated to address this issue, and our research objective is to maximize the transportation system resilience and restore its performance through two network reconfiguration schemes: contraflow (also referred to as lane reversal) and crossing elimination at intersections. Mathematical models are developed to represent the two reconfiguration schemes and characterize the interactions between traffic operators and passengers. Specifically, traffic operators act as leaders to determine the optimal system reconfiguration to minimize the total travel time for all the users (both evacuees and regular commuters), while passengers act as followers by freely choosing the path with the minimum travel time, which eventually converges to a user equilibrium state. For each given network reconfiguration, the lower-level problem is formulated as a traffic assignment problem (TAP) where each user tries to minimize his/her own travel time. To tackle the lower-level optimization problem, a gradient projection method is leveraged to shift the flow from other nonshortest paths to the shortest path between each origin-destination pair, eventually converging to the user equilibrium traffic assignment. The upper-level problem is formulated as a constrained discrete optimization problem, and a probabilistic solution discovery algorithm is used to obtain the near-optimal solution. Two numerical examples are used to demonstrate the effectiveness of the proposed method in restoring the traffic system performance.

Jointly optimizing microgrid configuration and energy consumption scheduling of smart homes

Abstract In this paper, we formulate joint optimization of microgrid configuration and energy consumption scheduling as a leader-follower Stackelberg game to model the coordination between two rational decision makers of microgrid configuration and energy consumption scheduling. The microgrid configuration decision, as the leader, is modeled as an upper-level optimization problem for optimal installed numbers of wind turbines, photovoltaic (PV) units, micro-turbines and batteries. The energy consumption scheduling, as the follower, is modeled as a lower-level optimization problem, which responds to decisions of the upper level in order to determine the optimal appliance scheduling. A bi-level nonlinear programming model is formulated for the Stackelberg game. To solve this optimization model, four bi-level hierarchical algorithms with the combination of different evolutionary algorithms are implemented and compared. A case study of microgrid configuration of smart building is employed to demonstrate the feasibility and advantage of the proposed game-theoretic model.

Tackling Large-Scale and Combinatorial Bi-Level Problems With a Genetic Programming Hyper-Heuristic

Combinatorial bi-level optimization remains a challenging topic, especially when the lower-level is an ${\mathcal {NP}}$ -hard problem. In this paper, we tackle large-scale and combinatorial bi-level problems using GP hyper-heuristics, i.e., an approach that permits to train heuristics like a machine learning model. Our contribution aims at targeting the intensive and complex lower-level optimizations that occur when solving a large-scale and combinatorial bi-level problem. For this purpose, we consider hyper-heuristics through heuristic generation. Using a GP hyper-heuristic approach, we train greedy heuristics in order to make them more reliable when encountering unseen lower-level instances that could be generated during bi-level optimization. To validate our approach referred to as GA+AGH, we tackle instances from the bi-level cloud pricing optimization problem (BCPOP) that model the trading interactions between a cloud service provider and cloud service customers. Numerical results demonstrate the abilities of the trained heuristics to cope with the inherent nested structure that makes bi-level optimization problems so hard. Furthermore, it has been shown that training heuristics for lower-level optimization permits to outperform human-based heuristics and metaheuristics which constitute an excellent outcome for bi-level optimization.

Tidal Current and Level Uncertainty Prediction via Adaptive Linear Programming

Short-term uncertainty prediction modeling of tidal power generation supports power systems in reserve and regulation markets. In tidal power generation via various tidal energy harvesting technologies, tidal current and level are the most influential factors. This paper addresses a nonparametric prediction interval (NPI)-based uncertainty model thereof. The proposed model adapts a bi-level optimization formulation, based on extreme learning machine (ELM) prediction engine and quantile regression (QR). The quantile probabilities are asymmetrically and adaptively chosen in the upper level optimization to make prediction intervals sharper for a specific reliability level (RL). Besides, the training process of ELM is improved by adaptively selecting ELM's hidden neurons via upper level optimization. The lower level optimization finds ELM's output weighting coefficients through linear programming of QR. The heuristic optimization, consisting of gray wolf optimizer and simplex method, is designed to facilitate the NPI with high exploration and exploitation capabilities in upper level optimization. The performance of the proposed NPI is examined using empirical data recorded in three different sites, located in North America. The results of case studies show that the proposed NPI can provide sharper PIs in comparison to the well-tailored rival models whilst a prespecified RL criterion is met.

Generation of high-performance code based on a domain-specific language for algorithmic skeletons

Parallel programming can be difficult and error prone, in particular if low-level optimizations are required in order to reach high performance in complex environments such as multi-core clusters using MPI and OpenMP. One approach to overcome these issues is based on algorithmic skeletons. These are predefined patterns which are implemented in parallel and can be composed by application programmers without taking care of low-level programming aspects. Support for algorithmic skeletons is typically provided as a library. However, optimizations are hard to implement in this setting and programming might still be tedious because of required boiler plate code. Thus, we propose a domain-specific language for algorithmic skeletons that performs optimizations and generates low-level C++ code. Our experimental results on four benchmarks show that the models are significantly shorter and that the execution time and speedup of the generated code often outperform equivalent library implementations using the Muenster Skeleton Library.

Bilevel Programming Weight Minimization of Composite Flying-Wing Aircraft with Curvilinear Spars and Ribs

This paper studies weight minimization of a composite flying-wing aircraft using bioinspired arbitrarily shaped spars and ribs, known as SpaRibs, for the internal structural layout design. A generalized mathematical model is developed to parameterize the shape of SpaRibs using a four-node nonuniform rational basis spline curve. Additionally, for reducing weight, composite structures and lift redistribution using multiple control surfaces are considered. A stacking sequence table is used to parameterize the wing laminate configuration with ply drops for flexible wing design. For the optimization problem including topology as well as discrete and size design variables, a bilevel programming optimization is employed by using a previously developed parallel particle swarm optimization and a gradient-based optimization, respectively, for the upper- and lower-level optimization problems. The upper-level optimization problem is employed to satisfy such design requirements as the flutter constraints for achieving a specific body freedom flutter mode. The lower-level optimization is used to optimize the control surface rotations to minimize the wing root bending moment, which is considered to be a surrogate for wing weight. Different internal structural layouts and flutter constraints are studied for comparison. Optimization studies show that using SpaRibs for aircraft wing box internal structural layout design can reduce the wing weight by 21.8% as compared to the base model and by 7.3% as compared to the wing box with traditional straight spars and ribs. The additional specific flutter result constraints lead to a 4.2% increase in the wing weight for aircraft design with SpaRibs as compared to the one obtained only with the lower bound on the flutter speed.

OLTC-Induced False Data Injection Attack on Volt/VAR Optimization in Distribution Systems

This paper presents a new class of false data injection attacks (FDIAs) on volt/VAR optimization (VVO), which may result in abnormal voltage conditions along the radial medium voltage (MV) distribution feeder with an on-load tap changer (OLTC), capacitor banks (CBs), solar photovoltaic (PV) systems, and smart meters. In comparison with existing FDIAs against voltage control that do not consider the VVO process, we propose a new attack strategy with which the adversary can maliciously change the distribution feeder voltage profile by misleading the VVO function through stealthily injecting false data into smart meter measurements that are used for the VVO. The proposed attack strategy is formulated as a bilevel optimization problem using mixed integer linear programming (MILP). Injected false load data that raise or lower the tap position of the OLTC are calculated at the upper level while the VVO process is guaranteed to correctly operate with false data at the lower level. The bilevel optimization problem is finally reformulated to a single-level optimization problem based on Karush–Kuhn–Tucker conditions of the lower level optimization problem. A simulation study is carried out in an IEEE 33-bus distribution system with an OLTC, CBs, PV systems, and smart meters, and our results demonstrate the feasibility and capability of the proposed attack approach in terms of voltage level, attack effort, and PV penetration rate.

4K Real Time Software Solution of Scalable HEVC for Broadcast Video Application

Scalable high-efficiency video coding (SHVC) is the scalable extension of the high-efficiency video coding (HEVC) standard. SHVC enables spatial, quality, bit-depth, color gamut, and codec scalability. The architecture of the SHVC encoder is based on multiple instances of the HEVC encoder where each instance encodes one video layer. This architecture offers several advantages of being modular and close to the native HEVC coding block scheme. However, the close-loop SHVC architecture requires the complete decoding of the reference lower layer frames to decode a higher quality layer, which considerably increases the complexity of both encoder and decoder processes. In this paper, we propose an end-to-end 4K real-time SHVC solution, including both software encoder and decoder, for video broadcast applications. The SHVC codec relies on low-level optimizations for specific Intel ×86 platform and parallel processing to speed up the encoding and decoding processes. The proposed encoder enables real-time processing of 4Kp30 video in 2× spatial scalabilities on the 4 × 10 cores Intel Xeon processor (E5-4627V3) running at 2.6 GHz. In addition, the SHVC decoder enables to decode, respectively, the lower quality layer in full HD (1920 × 1080p30) resolution, an advanced RISC machine (ARM) Neon mobile platform, and the enhancement layer in UHD (3840 × 2160p30), on a fitted laptop, with 4 cores Intel i7 processor running at 2.7 GHz. Finally, experimental results have shown that the proposed solution can reach a high rate-distortion performance close to the reference SHVC reference software model (SHM) with a speedup of 37 and 66 in intra and inter coding configurations.

Compiling with Continuations and LLVM

LLVM is an infrastructure for code generation and low-level optimizations, which has been gaining popularity as a backend for both research and industrial compilers, including many compilers for functional languages. While LLVM provides a relatively easy path to high-quality native code, its design is based on a traditional runtime model which is not well suited to alternative compilation strategies used in high-level language compilers, such as the use of heap-allocated continuation closures. This paper describes a new LLVM-based backend that supports heap-allocated continuation closures, which enables constant-time callcc and very-lightweight multithreading. The backend has been implemented in the Parallel ML compiler, which is part of the Manticore system, but the results should be useful for other compilers, such as Standard ML of New Jersey, that use heap-allocated continuation closures.

Two-stage design optimization based on artificial immune system and mixed-integer linear programming for energy supply networks

A two-stage optimization approach based on an artificial immune system (AIS) and mixed-integer linear programming (MILP) was developed to efficiently solve large-scale structural design problems of energy supply networks and obtain multiple and diverse design candidates. By focusing on a hierarchical relationship between design and operation variables, a structural design problem, formulated using MILP, is decomposed into an upper-level design problem and a lower-level operation problem. The upper-level design problem is solved using an AIS, in which multiple and diverse sets of suboptimal solutions are searched in a short computation time. In the lower-level optimization, design variables are fixed at the values searched in the upper-level optimization and operation variables are optimized using MILP. Moreover, the lower-level optimization for multiple sets of design variables is separately conducted using parallel computing. The developed approach was applied to the structural design of an energy supply network, consisting of candidates of cogeneration units and heat pump water heating units under power and heat interchange, for a housing complex with four dwellings. The diversity and energy-saving performance of multiple design candidates were analyzed. The computational efficiency was also demonstrated in comparison to the results obtained using only a commercial MILP solver.

A Branch and Bound-Based Algorithm for the Weak Linear Bilevel Programming Problems

Most real-world optimization problems are hierarchical involving non-cooperative objectives. Many of these problems can be formulated in terms of the first (upper level) objective function being minimized over the solution set mapping of the second (lower level) optimization problem. Often the upper level decision maker is risk-averse. The resulting class of problem is named weak bilevel programming problem. This paper presents a new algorithm which embeds a penalty function method into a branch and bound algorithm to deal with a weak linear bilevel programming problem. An example illustrates the feasibility of the proposed algorithm.

Further on Inner Regularizations in Bilevel Optimization

A crucial difficulty in pessimistic bilevel optimization is the possible lack of existence of exact solutions since marginal functions of the sup type may fail to be lower semicontinuous. So, to overcome this drawback, we have introduced, in Lignola and Morgan (J Optim Theory Appl 173(1):183–202, 2017), suitable inner regularizations of the lower level optimization problem together with relative viscosity solutions for the pessimistic bilevel problem. Here, we continue this research by considering new inner regularizations of the lower level optimization problem, which not necessarily satisfy the constraints but that are close to them, and by deriving an existence result of related viscosity solutions to the pessimistic bilevel problem.

Hierarchical Scheduling Scheme for AC/DC Hybrid Active Distribution Network Based on Multi-Stakeholders

This paper presents a hierarchical multi-stage scheduling scheme for the AC/DC hybrid active distribution network (ADN). The load regulation center (LRC) is considered in the developed scheduling strategy, as well as the AC and DC sub-network operators. They are taken to be different stakeholders. To coordinate the interests of all stakeholders, a two-level optimization model is established. The flexible loads are dispatched by LRC in the upper-level optimization model, the objective of which is minimizing the loss of the entire distribution network. The lower-level optimization is divided into two sub-optimal models, and they are carried out to minimize the operating costs of the AC/DC sub-network operators respectively. This two-level model avoids the difficulty of solving multi-objective optimization and can clarify the role of various stakeholders in the system scheduling. To solve the model effectively, a discrete wind-driven optimization (DWDO) algorithm is proposed. Then, considering the combination of the proposed DWDO algorithm and the YALMIP toolbox, a hierarchical optimization algorithm (HOA) is developed. The HOA can obtain the overall optimization result of the system through the iterative optimization of the upper and lower levels. Finally, the simulation results verify the effectiveness of the proposed scheduling scheme.