Quantum computing (QC) has recently gained significant traction across various fields, spurring the integration of swarm intelligence algorithms into the quantum realm to boost their search capabilities. The bat algorithm (BA), modeled on bats’ echolocation-based prey-hunting behavior, is a swarm intelligence algorithm, but it often gets stuck in local minima under certain conditions. To solve the problems, this research puts forward a hybrid predictive mutation operation and adaptive frequency adjustment strategy from a QC perspective to optimize the BA for better performance in complex problems. The method evolves the BA system into a highly uncertain complex nonlinear system via the state superposition principle, adds an optimization factor to balance global exploration and local exploitation, builds a rigorous mathematical model based on QC, and conducts a convergence analysis proof. Three parameter-control-mode experiments are designed to explore key parameter impacts. Numerical experiments on classic benchmarks and the three-dimensional trajectory planning for UAVs show that the proposed APMQF-QBA method outperforms comparison algorithms. Overall, the new strategy effectively improves the BA, avoiding local extrema and accelerating convergence, offering a more efficient solution for relevant fields.

Multi-view subspace clustering has attracted much attention because of its effectiveness in unsupervised learning. The high time consumption and hyper-parameters are the main obstacles to its development. In this paper, we present a novel method to effectively solve these two defects. First, we employ the bisecting k-means method to generate anchors and construct the hierarchical bipartite graph, which greatly reduce the time consumption. Moreover, we adopt an auto-weighted allocation strategy to learn appropriate weight factors for each view, which can avoid the influence of hyper-parameters. Furthermore, by imposing low rank constraints on the fusion graph, our proposed method can directly obtained the cluster indicators without any post-processing operations. Finally, numerous experiments verify the superiority of proposed method.

It is generally known that the error function is one of the key factors that determine the convergence, stability and generalization ability of neural networks. For most feedforward neural networks, the squared error function is usually chosen as the error function to train the network. However, networks based on the squared error function can lead to slow convergence and easily fall into local optimum in the actual training process. Recent studies have found that, compared to the squared error function, the gradient method based on the entropy error function measures the difference between the probability distribution of the model output and the probability distribution of the true labels during the iterative process, which can be more able to handle the uncertainty in the classification problem, less likely to fall into a local optimum and can learn to converge more rapidly. In this paper, we propose a batch gradient method for Sigma-Pi-Sigma neural networks based on the entropy error function and rigorously demonstrate the weak and strong convergence of the new algorithm in the batch input mode. Finally, the theoretical results and effectiveness of the algorithm are verified by simulation.

Joint clustering and dimensionality reduction methods have great potential in solving high-dimensional problems. Feature extraction and feature selection are both feature reduction methods. Compared to feature extraction, feature selection retains the original features and has advantages in certain aspects. To achieve optimized clusters while selecting distinct features, this paper imposes the L2,0-norm on the discriminative projected matrix based on the trace difference model. To solve the joint optimization model, we use the coordinate descent method to solve for the discriminative projected matrix and the cluster indicator matrix alternately. Extensive experiments show that the proposed model improves accuracy, NMI, and purity compared to other related clustering models. The proposed algorithm also demonstrates good convergence and stability across different datasets.

Density peaks clustering (DPC) algorithm determines the cluster centers by calculating the local density and relative distance, and then assigns the non-center points to the cluster where the nearest high-density point is located. In this paper, an improved density peaks clustering algorithm based on mutual nearest neighbor distance (MD-DPC) is proposed to solve the problem of uncertainty and influence of the cut-off distance and poor fault tolerance of one-step allocation strategy. Firstly, the local density is calculated by the mutual nearest neighbor distance without cut-off distance to overcome the influence of cut-off distance parameter on the clustering results. Secondly, a measurement criterion is defined to consider the density of the region where the data points are located. According to this criterion, the data points are divided into core region and non-core region, and a two-step allocation strategies is proposed to overcome the problem of poor fault tolerance of the one-step allocation strategy of DPC. The allocation strategy of the data points in the core region is the same as the original DPC algorithm, and the data points in the non-core region are assigned to the cluster with the highest likelihood. Finally, in order to validate MD-DPC, we test it on synthetic and real-world datasets, and compare it with DPC, DBSCAN, k-means, KNN-DPC and DPCSA methods. Experimental results suggest that MD-DPC can effectively find clusters.

For sparse subspace clustering methods, it is crucial to develop a good representation matrix to capture the data structure. In this paper, we incorporated the label information into sparse representation and proposed a new semi-supervised sparse subspace clustering method, named semi-supervised sparse subspace clustering with manifold regularization (S ^4 CMR). When developing the sparse self-expressive matrix, the S ^4 CMR method utilized the label information to constrain the development of expressiveness coefficients. The local manifold regularization was also integrated to enhance clustering stability and local consistency. By utilizing the Alternating Direction Method of Multipliers (ADMM), the convex optimization problem associated with linear constraints can be easily resolved. The developed similarity matrix can provide strong discriminant information, making it more effective for semi-supervised tasks. The effectiveness of the proposed algorithm is demonstrated through experiments on benchmark data sets, such as motion segmentation and image clustering.

Multi-label feature selection, an essential means of data dimension reduction in multi-label learning, has become one of the research hotspots in the field of machine learning. Because the linear assumption of sample space and label space is not suitable in most cases, many scholars use pseudo-label space. However, the use of pseudo-label space will increase the number of model variables and may lead to the loss of sample or label information. A multi-label feature selection scheme based on constraint mapping space regularization is proposed to solve this problem. The model first maps the sample space to the label space through the use of linear mapping. Second, given that the sample cannot be perfectly mapped to the label space, the mapping space should be closest to the label space and still retain the space of the basic manifold structure of the sample space, so combining the Hilbert-Schmidt independence criterion with the sample manifold, basic properties of constraint mapping space. Finally, the proposed algorithm is compared with MRDM, SSFS, and other algorithms on multiple classical multi-label data sets; the results show that the proposed algorithm is effective on multiple indicators.

High order neural networks have strong nonlinear mapping ability, but the network structure is more complex, which restricts the efficiency of the network, and the relevant theoretical analysis is still not perfect up to now. To solve these problems, an online gradient learning algorithm model of Pi-Sigma neural network with a smooth set lasso regular term is proposed. Since the original lasso regular term contains absolute values and is not differentiable at the origin, it causes experiment oscillations and poses a great challenge to the convergence analysis of the algorithm. We use grinding technology to overcome this deficiency. The main contribution of this paper lies in the adoption of online learning algorithm, which effectively improves the efficiency of the algorithm. At the same time, strict theoretical proofs are presented, including strong convergence and weak convergence. Finally, the effectiveness of the algorithm and the correctness of the theoretical results are verified by numerical experiments.

Density peaks clustering (DPC) algorithm has attracted the attention of scholars because of its simplicity and efficiency. However, it certainly has some disadvantages. On the one hand, the cut-off distance of DPC is artificially set, which greatly affects the clustering results of the DPC. On the other hand, the one-step allocation strategy is not robust and has poor fault tolerance. In this paper, we propose an improved density peaks clustering based on sparrow search algorithm(SSA-DPC) to solve the problems. First, the cut-off distance is optimized by the sparrow search algorithm with the ACC index as the object function to reduce the impact of cut-off distance on clustering results. Second, the idea of mutual nearest neighbor is introduced to divide the dataset into high-density region and low-density region, and different allocation strategies are adopted for different regions to overcome the problem of poor fault tolerance of one-step allocation strategy in DPC. Finally, in order to validate SSA-DPC, we test it on synthetic and real-world datasets, and compare it with DPC, SNN-DPC, DBSCAN, k-means, KNN-DPC and DPCSA methods. Experimental results suggest that SSA-DPC can effectively find clusters.

The expensive time consumption and hyper-parameters are two main drawbacks of most existing multi-view graph clustering methods. Especially in large-scale data clustering, these two defects are more serious. Aiming at these two problems, we propose a novel auto-weighted multi-view clustering method based on the hierarchical bipartite graph to effectively address these two limitations. Similar to the idea of an anchor graph, firstly, the bisecting k-means method is used instead of traditional method to generate a hierarchical anchor points set. And then, the hierarchical bipartite graph can be constructed between the original points and the anchor points of last layer. Since we only consider the anchor points on the last layer rather than using a larger number of anchors to obtain the bipartite graph, our proposed method will greatly reduce the time consumption. Moreover, the automatic learning strategy was adopted to select the appropriate weights for each view in our proposed approach, which remove the weight-related hyper-parameters and effectively avoid the vexing problem of hyper-parameters in multi-view graph clustering. Finally, experiments were carried out on synthetic data and different sizes benchmark data respectively to verify the validity of our method.

In this paper, a new variant of convolutional neural network (CNN) based on L1 regularization is proposed. The main consideration is to generate a sparse weight matrix, i.e., to generate a sparse neural network model. To some extent, L1 regularization prevents the occurrence of overfitting and effectively improves the learning efficiency of the network. Unfortunately, the L1 regularization term is non-smooth, which leads to the occurrence of oscillations in experiments and poses a great challenge to the theoretical analysis. So, we overcome this drawback by using polishing techniques and rigorously prove the monotonicity of the error function and the strong and weak convergence theorems of the algorithm. Finally, numerical simulations on several data sets support our theoretical results and the superiority of the proposed algorithm.

In order to develop a fast, efficient and intelligent tool to recognize anti-hypertensive peptides (AHTPs), an identification model based on multi-source characteristics and deep learning was constructed for the recognition of AHTPs. Novel enhanced grouped amino acid composition (NEGAAC), reduced dipeptide composition (RDPC), dipeptide deviation from expected mean (DDE), amino acid physicochemical properties-based distance transformation (AAP-DT) and BLOSUM62 encoding were used for feature extraction of peptide sequences. In addition, bidirectional gated recurrent units (BiGRU) were used for deep learning of protein characteristics, so as to effectively identify AHTPs. Under 10-fold cross-validation, the recognition accuracy of the recognition model based on multi-source features and deep learning reaches 96.78% and 98.72% on the benchmark data set and independent data set.

In view of the problems of dimensional sensitivity, poor stability and low convergence accuracy of bat optimization algorithm, a strategy of grouping bats is proposed, and a hybrid search strategy is adopted to avoid skipping the optimal solution. An inertia weighting strategy is added to improve the population diversity. Six benchmark functions are used for simulation. The results show that the proposed DGMBA algorithm outperforms ELBA, BA, FPA and PSO algorithms. The rank test of the algorithm shows that the algorithm is stable and effective.

Differential evolution (DE) algorithm is one of the most effective and efficient heuristic approaches for solving complex black box problems. But it still easily suffers from premature convergence and stagnation. To alleviate these defects, this paper presents a novel DE variant, named enhanced adaptive differential evolution algorithm with multi-mutation schemes and weighted control parameter setting (MWADE), to further strengthen its search capability. In MWADE, a multi-schemes mutation strategy is first proposed to properly exploit or explore the promising information of each individual. Herein, the whole population is dynamically grouped into three subpopulations according to their fitness values and search performance, and three different mutant operators with various search characteristics are respectively adopted for each subpopulation. Meanwhile, in order to ensure the exploration of algorithm at the later evolutionary stage, a weight-controlled parameter setting is proposed to suitably assign scale factors for different differential vectors. Moreover, a random opposition mechanism with greedy selection is introduced to avoid trapping in local optima or stagnation, and an adaptive population size reduction scheme is devised to further promote the search effectiveness of algorithm. Finally, to illustrate the performance of MWADE, thirteen typical algorithms are adopted and compared with MWADE on 30 functions from IEEE CEC 2017 test suite with different dimensions, and the effectiveness of its proposed components are also investigated. Numerical results indicate that the proposed algorithm has a better search performance.

Differential evolution (DE) is one of the most popular and widely used optimizers among the community of evolutionary computation. Despite numerous works having been conducted on the improvement of DE performance, there are still some defects, such as premature convergence and stagnation. In order to alleviate them, this paper presents a novel DE variant by designing a new mutation operator (named "DE/current-to-pbest_id/1") and a new control parameter setting. In the new operator, the fitness value of the individual is adopted to determine the chosen scope of its guider among the population. Meanwhile, a group-based competitive control parameter setting is presented to ensure the various search potentials of the population and the adaptivity of the algorithm. In this setting, the whole population is randomly divided into multiple equivalent groups, the control parameters for each group are independently generated based on its location information, and the worst location information among all groups is competitively updated with the current successful parameters. Moreover, a piecewise population size reduction mechanism is further devised to enhance the exploration and exploitation of the algorithm at the early and later evolution stages, respectively. Differing from the previous DE versions, the proposed method adaptively adjusts the search capability of each individual, simultaneously utilizes multiple pieces of successful parameter information to generate the control parameters, and has different speeds to reduce the population size at different search stages. Then it could achieve the well trade-off of exploration and exploitation. Finally, the performance of the proposed algorithm is measured by comparing with five well-known DE variants and five typical non-DE algorithms on the IEEE CEC 2017 test suite. Numerical results show that the proposed method is a more promising optimizer.

为解决粒子群算法(Particle Swarm Optimization,PSO)易受初值影响、迭代后期早熟收敛、局部寻优结果不稳定等问题,提出混沌映射粒子种群初始化的方案.即在算法前期根据混沌映射初值敏感性、随机性等特征,基于6种混沌映射对粒子群算法分别进行初始化,进而增加种群多样性和解的覆盖.引入动态惯性权重系数,提高算法收敛速度.通过6个测试函数进行仿真实验,对比不同混沌映射数据结果,实验证明该算法能在不改变原有时间复杂度的基础上较好地提高算法的收敛速度和寻优精度.

A three-dimensional dynamic heat transfer mathematical model of the process when cored wire feed in molten iron is established based on finite volume method (FVM). The calculation area is meshed with triangles and quadrilaterals to determine nodes and control volumes, and implicit time integration method is used to ensure the stability of calculating process. For exposing the dynamic heat transfer behavior, the variation of temperature field and explosion characteristics of cored wires are studied. In addition, the melt loss rate of the top end of cored wire and the correlation among melt explosion depth, molten iron temperature and feeding speed of cored wires are theoretically calculated. More importantly, the influence of different structures of cored wires on the absorption rate of magnesium are considered. The calculation results are in good agreement with the experimental data, which indicate that the existing theoretical model has good validity and can provide theoretical guidance for spheroidization process in molten iron.

针对支持向量回归(SVR)模型参数选择困难以及在碳期货价格预测中模型误差高的问题,提出一种基于改进粒子群算法-支持向量回归(TSA-PSO-SVR)的期货价格预测模型.通过改进粒子群算法惯性权重实现局部搜索和全局搜索能力的平衡,引入被囊群算法(TSA)对粒子群位置更新公式进行优化,利用改进的粒子群算法(TSA-PSO)找出最优参数有效解决支持向量回归参数选择盲目性的问题;将得到的最优参数应用于期货价格预测模型.选取福建碳交易市场的碳交易价格进行预测,与支持向量回归(SVR)、差分自回归移动平均模型(ARIMA)、长短期记忆模型(LSTM)模型作对比,实验结果表明TSA-PSO-SVR模型有效克服了高预测误差和参数选择随机性的问题,并具有较高的泛化能力.

为了克服樽海鞘群算法(Salp Swarm Algorithm,SSA)求解准确性不足和易过早收敛的缺点,提出了一种多策略改进的樽海鞘群算法(MISSA).引入 Baker混沌映射生成樽海鞘群的初始种群,以提高初始个体的均匀性;将T分布策略应用到食物源位置公式中,对原始位置进行随机干扰,引导樽海鞘个体向最优解空间运动;在跟随者位置更新公式中引入不完全Γ函数的自适应权重,以改善算法的局部和全局搜索能力.将改进算法在 8 个测试函数上进行仿真实验,并与不同的群智能算法进行了比较.结果表明,改进算法具有更好的全局和局部搜索性能以及更高的搜索精度.

In order to solve the problem of trajectory shift, a PSO-DRBFNNILC strategy is designed. The first RBFNN is introduced to estimate the output of the ILC system; second RBFNN is built to adaptively adjust the learning gain matrix in the input update law. PSO algorithm is used to find the optimal search step for the update of the weight, center and radius of the activation function. Convergence analysis shows that the estimation error of the weight of the network and the tracking error of the ILC system are bounded. The effectiveness of the control strategy is verified by numerical simulation.