Local Search Optimization Research Articles

Active Learning for Well Control Optimization with Surrogate Models

SummaryOptimization of well controls in waterflooding operations has been established as an effective strategy to increase oil recovery efficiency and to reduce operating costs. Optimization problems, however, entail hundreds of full-scale reservoir simulation runs, which can be computationally expensive and impractical for field-scale applications. Fit-for-purpose surrogate models present more efficient alternatives that can be deployed for faster implementation and to enable real-time applications. In recent years, machine learning (ML)-based surrogate models have gained traction in building efficient proxy models to facilitate reservoir engineering and management workflows. For high-dimensional control variables that are frequently encountered in life cycle optimization, gradient-based local search methods have proven to be the optimization algorithms of choice, especially when adjoint models are available to provide the required gradient information in a computationally efficient manner. The use of local search optimization techniques implies that the surrogate models should be locally accurate and adapted to the optimization path that is traversed during the iterations. The adaptation can be accomplished through active learning, where the training of the surrogate model is performed online (throughout the iterations) and tied to the optimization iterations. This paper presents an active learning framework that is tuned to the local search nature of the optimization problem by updating an initially trained surrogate model based on new samples that are generated during the optimization iterations. Specifically, at each iteration, the full reservoir simulation model is performed to evaluate the true objective function and provide the new simulated data to update the surrogate model, thereby enhancing its local prediction accuracy. As the optimization progresses, older samples that become less relevant to the solution at the current iteration are dropped to preserve the local nature of the surrogate model. Various aspects of designing an active learning-based proxy model for well control optimization are discussed and the framework is applied to waterflooding problems to demonstrate its efficiency and superior performance compared to the traditional offline training approach.

A differential evolution with adaptive neighborhood mutation and local search for multi-modal optimization

In this paper, an adaptive neighborhood mutation based memetic differential evolution is proposed for multimodal optimization. In the proposed method, an adaptive neighborhood mutation (ANM) strategy is devised to allow the individuals to conduct a diverse search at the early stage of evolution and then gradually switch to an intensive search at the later stage of evolution. Further, the ANM is devised such that encouraging promising individuals for exploitation while unpromising individuals for exploration during evolution to appropriate search the multimodal space. In addition, an adaptive Gaussian based local search strategy, which considers the difference between the offspring and their paired individuals with successful replacements, is developed to appropriately improve promising individuals during evolution. The proposed method has been extensively accessed on a suite of twenty multimodal benchmark functions and the performance of which is compared with sixteen related multimodal algorithms. The results clearly demonstrate the superiority of proposed method as well as the merit of devised strategies.

Lithium Battery Remaining Useful Life Prediction Based on Multi-Kernel Support Vector Regression With Hybrid Optimization Algorithm

Abstract Accurately and reliably predicting the remaining useful life (RUL) of lithium battery is very important for the lithium battery health management system. However, most of the existing methods rely on complex multidimensional input features, which require a large number of sensors, increase the application cost and introduce redundant measurement errors. Therefore, this paper, only based on the battery capacity curve itself, proposes a method to construct a prediction model of support vector regression (SVR) by fusing multiple kernel functions. The linear equation coefficients of multiple kernel function combinations are optimized by the hybrid optimization algorithm. It is found that the hybrid kernel function can effectively overcome the problem that the single-kernel function is not capable of mapping the capacity fading trend of lithium battery. Hybrid optimization algorithm can avoid the problems of local optimization and global search ability deficiency. The proposed method is validated by experiments using the battery attenuation datasets from NASA, the University of Maryland, and a high-rate lithium battery in the laboratory stage. It can be seen from the experimental results that the prediction accuracy of this method is high. The mean prediction error, mean RMSE, and mean MAE are 2%, 0.0198, and 0.0157.

A niching backtracking search algorithm with adaptive local search for multimodal multiobjective optimization

• Propose a niching backtracking search algorithm with adaptive local search. • Backtracking search algorithm is applied in this field for the first time. • Develop a novel mutation based on the affinity propagation clustering. • Develop an adaptive local search strategy for the proposed algorithm. Multimodal multiobjective optimization problems,which widely exist in real-world applications, have multiple Pareto optimal sets in decision space corresponding to the same Pareto front in objective space. The key to handling such problems is locating and maintaining all Pareto optimal solutions in decision space simultaneously. The distribution of Pareto optimal solutions obtained by some existing multimodal multiobjective evolutionary algorithms is still not satisfactory. This paper proposes a niching backtracking search algorithm with adaptive local search to solve such problems. In the proposed algorithm, the affinity propagation clustering method as a parameter-free automatic niching technique is adopted to form multiple niches. A novel mutation based on affinity propagation clustering is then developed to search for more Pareto solutions within each niche. In addition, an adaptive local search strategy is designed in each niche to improve the search efficiency and the accuracy of Pareto optimal solutions. The proposed algorithm is compared with seven state-of-the-art multimodal multiobjective evolutionary algorithms on a multimodal multiobjective optimization test suite with 22 benchmark functions from CEC 2019 competition and a map-based practical application problem. The experimental results show that the proposed algorithm outperforms its competitors on 14 out of 22 benchmark functions in terms of the reciprocal of Pareto sets proximity and inverted generational distance in decision space metrics. Also, the proposed method is more effective and competitive than its competitors when solving the map-based practical application problem.

Practice Summary: Multilaser Load Balancing in Additive Manufacturing

In this paper, we address the challenges of load balancing and smoke interference in multi-laser systems in additive manufacturing using a local search optimization algorithm. Our algorithm is scheduled to be integrated into the GE Concept Laser M-line System, a state-of-the-art additive manufacturing printer sold by GE.

A Broyden-based algorithm for multi-objective local-search optimization

In multi-objective optimization, the direction vectors in the objective space that improve the current non-dominated set are named improvement directions. The Improvement Direction Mapping (IDM) methods apply a spatial transformation to map improvement directions in the objective space to search for directions in the variable space. Jacobian based transformations can perform the mapping, however, they require the analytic expressions of the objectives and their derivatives. Hence, they serve as a reference in academic problems but are impractical in real-world applications. Furthermore, they are commonly ill-conditioned and under-determined; consequently, they deviate the search as the current non-dominated set approaches to the Pareto set. This work proposes a solution via an iterative updating based on the Broyden method. The proposal promotes parallelism between the displacement of the non-dominated solutions in the objective space and improvement directions delivered by Chebyshev scalarizing functions. Compared to other transformations reported in the specialized literature, the proposed method demonstrates a better conditioning that provokes to err at a lesser extent in providing successful search directions. This impacts the algorithm performance, mainly in locations close to the Pareto set. These advantages are demonstrated using benchmark functions and metrics.

Deep CNN with Hybrid Binary Local Search and Particle Swarm Optimizer for Exudates Classification from Fundus Images.

Diabetic retinopathy is a chronic condition that causes vision loss if not detected early. In the early stage, it can be diagnosed with the aid of exudates which are called lesions. However, it is arduous to detect the exudate lesion due to the availability of blood vessels and other distractions. To tackle these issues, we proposed a novel exudates classification from the fundus image known as hybrid convolutional neural network (CNN)-based binary local search optimizer-based particle swarm optimization algorithm. The proposed method from this paper exploits image augmentation to enlarge the fundus image to the required size without losing any features. The features from the resized fundus images are extracted as a feature vector and fed into the feed-forward CNN as the input. Henceforth, it classifies the exudates from the fundus image. Further, the hyperparameters are optimized to reduce the computational complexities by utilization of binary local search optimizer (BLSO) and particle swarm optimization (PSO). The experimental analysis is conducted on the public ROC and real-time ARA400 datasets and compared with the state-of-art works such as support vector machine classifiers, multi-modal/multi-scale, random forest, and CNN for the performance metrics. The classification accuracy is high for the proposed work, and thus, our proposed outperforms all the other approaches.

Research on Gas Outburst Prediction Model Based on Multiple Strategy Fusion Improved Snake Optimization Algorithm With Temporal Convolutional Network

A gas outburst prediction model based on multiple strategy fusion and improved snake optimization algorithm (MFISO) and temporal convolutional network (TCN) is proposed to address the problems of low accuracy of deep learning prediction models for gas outburst in underground mines. By adopting the phase space reconstruction method, the time series of multiple complex influencing factors related to gas outburst were reconstructed and used as model inputs. Sine chaos mapping, spiral search strategy and snake dynamic adaptive weight are introduced to improve the snake optimization algorithm (SO), which enhances the local optimal escape capability and global search capability of the algorithm. The tangent-based rectified linear unit (ThLU) was used to improve the rectified linear unit (ReLU) of the standard TCN to strengthen the generalization capability of the model. The MFISO algorithm was used to optimize the relevant hyperparameters of the improved TCN model to optimize the accuracy of gas outburst prediction. The TCN, GRU, LSTM, SO-TCN, WOA-TCN, and PSO-TCN prediction models were selected to compare the prediction performance with the MFISO-TCN gas outburst prediction model, and the results showed that the mean absolute error (MAE), mean absolute percentage error (MAPE) and root mean square error (RMSE) of the MFISO-TCN model were 3.11%, 0.47% and 3.31% are lower than those of other models, which verifies that the method of this paper effectively intensifies the performance of gas outburst prediction model in underground mines.

Sum-Rate Optimization for IRS-Aided D2D Communication Underlaying Cellular Networks

The intelligent reflective surface (IRS)-assisted communication is recognized as a promising technology to enhance capacity and coverage of the network by controlling propagation. However, in IRS-aided device-to-device (D2D) communication underlaying cellular networks, co-channel interference is more complicated and challenging than traditional D2D communication networks due to the addition of extra reflection paths. To alleviate interference and improve sum-rate, we establish a multivariable resource optimization function based on IRS-assisted D2D communication in this paper. Then, a hybrid sub-optimal algorithm is proposed, which includes channel allocation based on game theory, power allocation and phase shift optimization. Specifically, a sub-channel allocation algorithm is introduced to maximize system sum-rate, where the problem of sub-channel allocation is formulated as a two-stage combinatorial auction model, and the approximate optimal solution is obtained by bidding and adjustment. Then, a sub-optimal solution for transmission power and phase shifts is obtained through local search and iterative optimization. Simulation results show that the proposed hybrid sub-optimal algorithm is feasible and effective. Compared to the other 5 schemes, it has best sum-rate performance and lowest power consumption.

Hybrid design of spectral splitters and concentrators of light for solar cells using iterative search and neural networks

The need for optically multi-functional micro- and nano-structures is growing in various fields. Designing such structures is impeded by the lack of computationally low-cost algorithms. In this study, we present a hybrid design scheme, which relies on a deep learning model and the local search optimization algorithm, to optimize a diffractive optical element that performs spectral splitting and spatial concentration of broadband light for solar cells. Using generated data set during optimization of a diffractive optical element, which is a one-time effort, we design topography of diffractive optical elements by using a deep learning-based inverse design scheme. We show that further iterative optimization of the reconstructed diffractive optical elements increases amount of spatially concentrated and spectrally split light. Our joint design approach both speeds up optimization of diffractive optical elements as well as providing better performance at least 57% excess light concentration with spectral splitting. The algorithm that we develop here will enable advanced and efficient design of multi-functional phase plates in various fields besides the application that we target in solar energy. The algorithm that we develop is openly available to contribute to other applications that rely on phase plates.

The multi-objective inspection path-planning in radioactive environment based on an improved ant colony optimization algorithm

More and more occupational workers work in radioactive environment. Although measures are taken to keep the radiation dose in a safe range, the workers will suffer more radiation during the overhauling of nuclear power plants. The dose they suffer during the overhauling of nuclear power plants account for 80% of the total annual dose so it is necessarily to plan a reasonable inspection path for them according to the safety principle of as low as reasonably achievable (ALARA). An improved ant colony optimization (IACO) algorithm is proposed to solve the multi-objective inspection path-planning problem in radioactive environment. To improve the performance of the algorithm, we not only combine ant colony optimization (ACO) algorithm and chaos optimization algorithm, but also introduce pheromone differentiated update strategy and local search optimization strategy. Additionally, 5 experimental simulation cases are conducted and the results are compared to those from particle swarm optimization (PSO) algorithm, chaos particle swarm optimization (CPSO) algorithm and tradition ACO algorithm. In the first three cases, the probability of IACO algorithm finding the optimal path is obviously greater than that of PSO algorithm and CPSO algorithm. IACO algorithm is more efficient and stable. By analyzing an inspection path-planning case with 35 target positions and a case with 44 target positions in a more complex radioactive environment, IACO algorithm could find the path with less effective dose that ACO algorithm cannot find. Therefore, the effectiveness and validity of IACO to solve multi-objective inspection path-planning problem in radioactive environment are verified by experimental simulations and it can help workers reduce radiation exposure.

Hybrid Tabu Search and Genetic Algorithm Imputation Approach for Incomplete Data

The common problem for data collection is happening missing value during the data collection and processing process that the quality of the data testing is decreased. A computational based technique for dealing with missing values, namely Genetic Algorithm Imputation (GAI). The usage was used to estimate the dataset's missing values. GAI generates the optimal set of missing values with the acquisition of information as a function of fitness to measure individual solutions' performance. GAI conducts continuous searching until the missing criteria value is found according to best fitness. So, it is trapped in optimal conditions temporarily. The improvement of GAI with tabu search is known as TS-GAI, that strength is two metaheuristic techniques modified at the mutase stage to distract the local optima's search. In applying missing values, this technique works better when many possible values are used instead of the mixed attribute having missing values. Because the new generation chromosome values generate many opportunities to make up for the missing values. The experimental results show that the TS-GAI shows better performance on 30% MV with a fitness value of 0.212. It converges at 159 iterations. Generally, TS-GAI is a faster iteration than simple GAI and it has a lower RMSE level than other imputation techniques.

A new hybrid model for MR elastomer device and parameter identification based on improved FOA

A new hysteresis model based on curve fitting method is presented in this work to portray the greatly nonlinear and hysteretic relationships between shear force and displacement responses of the magnetorheological (MR) elastomer base isolator. Compared with classical hysteresis models such as Bouc-Wen or LuGre friction model, the proposed model combines the hyperbolic sine function and Gaussian function to model the hysteretic loops of the device responses, contributing to a great decline of model parameters. Then, an improved fruit fly optimization algorithm (FOA) is proposed to optimize the model parameters, in which a self-adaptive step is employed rather than the fixed step to balance the global and local optimum search abilities of algorithm. Finally, the experimental results of the device under both harmonic and random excitations are used to verify the performance of the proposed hybrid model and parameter identification algorithm with the satisfactory results.

Coronavirus herd immunity optimizer with greedy crossover for feature selection in medical diagnosis

The importance of medical data and the crucial nature of the decisions that are based on such data, as well as the large increase in its volume, has encouraged researchers to develop feature selection (FS)-based approaches to identify the most relevant data for specific medical problems In this paper, two intelligent wrapper FS approaches based on a new metaheuristic algorithm named the coronavirus herd immunity optimizer (CHIO) were applied with and without the incorporation of a greedy crossover (GC) operator strategy to enhance exploration of the search space by CHIO. The two proposed approaches, CHIO and CHIO-GC, were evaluated using 23 medical benchmark datasets and a real-world COVID-19 dataset. The experimental results indicated that CHIO-GC outperformed CHIO in terms of search capability, as reflected in classification accuracy, selection size, F-measure, standard deviation and convergence speed. The GC operator was able to enhance the balance between exploration and exploitation of the CHIO in the search and correct suboptimal solutions for faster convergence. The proposed CHIO-GC was also compared with two previous wrapper FS approaches, namely, binary moth flame optimization with Lévy flight (LBMFO_V3) and the hyper learning binary dragonfly algorithm (HLBDA), as well as four filter methods namely, Chi-square, Relief, correlation-based feature selection and information gain. CHIO-GC surpassed LBMFO_V3 and the four filter methods with an accuracy rate of 0.79 on 23 medical benchmark datasets. CHIO-GC also surpassed HLBDA with an accuracy rate of 0.93 when applied to the COVID-19 dataset. These encouraging results were obtained by striking a sufficient balance between the two search phases of CHIO-GC during the hunt for correct solutions, which also increased the convergence rate. This was accomplished by integrating a greedy crossover technique into the CHIO algorithm to remedy the inferior solutions found during premature convergence and while locked into a local optimum search space.

An Improved Controlled Random Search Method

A modified version of a common global optimization method named controlled random search is presented here. This method is designed to estimate the global minimum of multidimensional symmetric and asymmetric functional problems. The new method modifies the original algorithm by incorporating a new sampling method, a new termination rule and the periodical application of a local search optimization algorithm to the points sampled. The new version is compared against the original using some benchmark functions from the relevant literature.

Combined Global and Local Search for Optimization with Gaussian Process Models

Gaussian process (GP) model based optimization is widely applied in simulation and machine learning. In general, it first estimates a GP model based on a few observations from the true response and then uses this model to guide the search, aiming to quickly locate the global optimum. Despite its successful applications, it has several limitations that may hinder its broader use. First, building an accurate GP model can be difficult and computationally expensive, especially when the response function is multimodal or varies significantly over the design space. Second, even with an appropriate model, the search process can be trapped in suboptimal regions before moving to the global optimum because of the excessive effort spent around the current best solution. In this work, we adopt the additive global and local GP (AGLGP) model in the optimization framework. The model is rooted in the inducing points based GP sparse approximations and is combined with independent local models in different regions. With these properties, the AGLGP model is suitable for multimodal responses with relatively large data sizes. Based on this AGLGP model, we propose a combined global and local search for optimization (CGLO) algorithm. It first divides the whole design space into disjoint local regions and identifies a promising region with the global model. Next, a local model in the selected region is fit to guide detailed search within this region. The algorithm then switches back to the global step when a good local solution is found. The global and local natures of CGLO enable it to enjoy the benefits of both global and local search to efficiently locate the global optimum. Summary of Contribution: This work proposes a new Gaussian process based algorithm for stochastic simulation optimization, which is an important area in operations research. This type of algorithm is also regarded as one of the state-of-the-art optimization algorithms for black-box functions in computer science. The aim of this work is to provide a computationally efficient optimization algorithm when the baseline functions are highly nonstationary (the function values change dramatically across the design space). Such nonstationary surfaces are very common in reality, such as the case in the maritime traffic safety problem considered here. In this problem, agent-based simulation is used to simulate the probability of collision of one vessel with the others on a given trajectory, and the decision maker needs to choose the trajectory with the minimum probability of collision quickly. Typically, in a high-congestion region, a small turn of the vessel can result in a very different conflict environment, and thus the response is highly nonstationary. Through our study, we find that the proposed algorithm can provide safer choices within a limited time compared with other methods. We believe the proposed algorithm is very computationally efficient and has large potential in such operational problems.

Particle Swarm Optimization Based on Fitness Considering Distance Information for Multiple Local Optimal Solutions Search

Particle Swarm Optimization Based on Fitness Considering Distance Information for Multiple Local Optimal Solutions Search

Towards accommodating deadline driven jobs on high performance computing platforms in grid computing environment

Grid computing is a connected computing infrastructure that furnishes reliable, stable, ubiquitous, and economic access to high-end computational power. The dynamic nature of the grid brings several challenges to scheduling algorithms that operate in queuing-based scheduling approach. This approach typically performs scheduling based on a certain fixed priority which leads to increase the delay for the running applications. Thus, the overall performance will be deteriorated sharply. The main aim of this study is to minimize the delay in the scheduler for the dynamic jobs. Therefore, this paper tackles dynamic scheduling issues by proposing Swift Gap (SG) mechanism. SG comprises of two stages by applying two mechanisms: A Backfilling Mechanism and Metaheuristic Local Search Optimization Mechanism. In the first stage, the job is placed in the earliest gap available in the local resources’ schedules, while the second stage optimizes the performance by checking all available gaps among resources’ schedules to find a better gap to place the job in. To further improve the performance, the Completion Time Scheme (CTS) is developed. CTS reduces the delay by placing the job in the gap that guarantees the best start time for the job, and the fastest resource available. The integration between SG and CTS (SG-CTS) is achieved by applying best start time rule in the first stage only, whereas the second stage includes both rules.SG-CTS is evaluated through simulation by using real workloads that reflect a real grid system environment. The findings demonstrate that SG-CTS improves the slowdown by 27 %, bounded slowdown by 25 %, tardiness by 21 %, waiting time by 16 % and response time by 7% compared to Conservative backfilling mechanism followed by Gap Search (CONS-GS). Finally, SG-CTS is evaluated against Deadline-Based Backfilling (DBF) algorithm. The evaluation revealed that SG-CTS performs better than DBF for slowdown and waiting time in HPC2N workload.

Electric vehicle charging and discharging scheduling strategy based on local search and competitive learning particle swarm optimization algorithm

It is great significance for environmental protection, energy conservation and emission reduction to replace fuel vehicles with EVs(electric vehicles).However, as a kind of random mobile load, large-scale integration into the power grid may lead to power quality problems such as line overload, line loss increase, voltage reduction and so on. In order to minimize the adverse effect of the disordered charging of EVs on the distribution grid, this paper takes the typical IEEE-33 node distribution system as the research object, a backward learning competitive particle swarm optimization (PSO) algorithm based on local search (SW-OBLCSO) is proposed. The SW-OBLCSO algorithm competitive learning and reverse learning mechanisms. In order to verify the performance of the algorithm, 4 common test functions are used, test functions compare the SW-OBLCSO algorithm with multiple optimization algorithms in different dimensions. The experimental results show that the proposed algorithm has outstanding performance in convergence speed and global search ability. This paper takes the minimum operation cost, the minimum environmental pollution, the minimum peak valley difference of load, the minimum node voltage offset rate, the minimum system grid loss and lowest charge cost as the optimization objectives; results shows that the proposed scheme can realize the transfer of charging load in time and space, so as to stabilize the load fluctuation of distribution grid, improve the operation quality of power grid, reduce the charging cost of users, and achieve the expected research objectives.

A simulation and optimisation package for emergency medical services

Emergency Medical Service (EMS) providers make decisions such as selecting ambulance station locations and ambulance dispatch policies; these decisions can be made effectively with the aid of optimisation and simulation models. We have developed JEMSS – a Julia (programming language) package for Emergency Medical Services Simulation – to allow for performance evaluation and optimisation of decisions or policies for problems arising in EMS operations and strategic planning. To the best of our knowledge this is the first free and open-source EMS simulation and optimisation package, which incorporates integer programming optimisation, heuristic local search, and simulation capabilities. The decision protocols used for ambulance dispatch and dynamic redeployment during a simulation run are flexible and easy to customize. Analytical optimisation models relevant to EMS decision-making are included, such as for station location and ambulance deployment problems. The simulations run quickly, requiring less than a second to simulate 100 days for a realistic city model. Data to simulate Auckland, New Zealand is included in the package and other cities will be added to enable researchers to conduct a more varied and fair comparison of newly developed EMS strategies. The JEMSS package includes a visualisation tool to view the simulation as it is running. The simulation model has been validated by comparing the simulation trace and output of existing EMS simulation and analysis software BartSim. This paper details the JEMSS design and presents a case study showing how to apply integer programming, local search, and simulation optimisation using JEMSS.