This paper proposes a Hierarchical Self-Evolutionary PSO (HSEPSO) Approach for UAV Path Planning to address the challenges faced by traditional Particle Swarm Optimization (PSO), such as high sensitivity to parameters, the tendency to become trapped in local optima, and slow convergence in later stages. Additionally, existing improvements to PSO lack the ability to dynamically adjust evolution strategies based on the current state of particles. HSEPSO employs a hybrid clustering strategy combining K-Means and DB-SCAN for population initialization, followed by population division based on clustering results. This ensures diversity within the population while enabling particles to focus their search on regions more likely to contain the optimal solution. The algorithm also dynamically adjusts the learning factors and inertia weights through a nonlinear adaptive update strategy, effectively balancing global search and local exploitation. Moreover, based on the real-time state of the particles, HSEPSO incorporates different evolutionary strategies to accelerate convergence, optimize the search for solutions, and enhance algorithm robustness. Experimental results demonstrate that, compared to traditional PSO and other improved algorithms (such as MFIPSO, SDPSO, and SA2PSO), HSEPSO shows notable improvements in optimization performance, convergence speed, and robustness.

One of the major challenges in designing an effective intrusion detection system for identifying abnormal behavior in data is the large number of original features, which can significantly impact the classification performance. A high-dimensional feature space often leads to increased computational complexity, decreased model efficiency, and the risk of overfitting. To overcome this challenge, this study presents an intrusion detection system that integrates an optimized feature selection process. The proposed system employs an Enhanced Hybrid Genetic Algorithm (EHGA) during the classification phase to perform feature selection and dimensionality reduction. The EHGA combines the strengths of genetic algorithms with other optimization techniques to improve recognition efficiency while maintaining classification accuracy. The system’s performance is assessed through extensive simulation experiments conducted on widely used intrusion detection datasets. The experimental results demonstrate that the EHGA-based feature selection process effectively reduces data dimensionality, enhances prediction accuracy, and improves the overall detection performance of the intrusion detection system. This method significantly enhances the feature selection performance of the intrusion detection system, offering a promising solution for improving intrusion detection in cybersecurity applications.

The key to portfolio optimization (PO) lies in balancing risk and return, which presents a complex, multi-dimensional challenge that requires a robust and stable heuristic algorithm. The Artificial Bee Colony (ABC) algorithm is highly regarded for its exceptional search capabilities and insensitivity to parameters, especially in optimizing the Sharpe Ratio, thus providing effective PO strategies. However, despite its excellent performance in searching, the search strategy of the ABC algorithm remains relatively simple, failing to fully exploit discovered solutions for further optimization. This indicates significant room for improvement. To address these issues, an ABC with random neighborhood search and the Metropolis algorithm (ABCNM) is proposed in this study. The neighborhood search method replaces the original search method, and a greedy random mechanism is introduced based on the neighborhood search to dynamically adjust the search range. Additionally, to accelerate the search process, we redesign the scout bee mechanism of the ABC algorithm by incorporating the Metropolis algorithm. Through ablation experiments, we found that introducing a randomized neighborhood search significantly improves the optimization capability of the ABC algorithm, and the improved scout bee mechanism not only enhances the algorithm’s search performance but also increases its robustness. In simulated experiments using stock market data, the algorithm achieves a high Sharpe ratio in both low-volatility and high-volatility market environments.

Path planning is a critical component in UAV mission execution. While traditional optimization algorithms are mature and effective, they typically rely on expert knowledge and manual tuning, resulting in high development barriers and limited adaptability. With the growing capabilities of large language models (LLMs) in natural language understanding and code generation, we explore the autonomous generation of UAV path planning algorithms using DeepSeek, a general-purpose LLM developed in China. We propose a prompt-driven framework that guides DeepSeek to generate both structural descriptions and implementation code for optimization algorithms. A simulation environment is built to evaluate the generated strategies in terms of feasibility, logical consistency, and baseline performance, with comparisons to classical approaches. Experimental results show that DeepSeek is capable of producing executable and modular optimization strategies, demonstrating strong potential for intelligent algorithm design. Instead of aiming to surpass existing state-of-the-art methods, this work focuses on reducing development costs and enhancing accessibility in control-oriented algorithm design. Our study presents a novel perspective on automated algorithm generation and highlights the practical applicability of LLMs in robotics and intelligent control.

To address the limitations of existing particle swarm optimization algorithms in solving discrete problems, this paper proposes a discrete particle swarm optimization algorithm for vehicle routing problem with time windows. This algorithm enhances the velocity and position updating mechanism using an set-based strategy, making it more efficient in the discrete search process. Additionally, to further balance the algorithm’s global search and local development capabilities, this paper introduces a chaotic mapping algorithm and a nonlinear dynamic adjustment parameter strategy to enhance the algorithm’s convergence speed.

Session-based recommendation aims to provide item recommendations to users based solely on anonymous session data, without relying on the users’ historical preferences. Traditional session-based recommendation models often treat sessions as sequences, recommending items based on the relevance of user preferences in the current session to the last clicked item. While this approach has shown some effectiveness, it often overlooks complex transitions between items and higher-order information, thus underestimating the global sequence information of the session. To address these issues, we propose a Graph Self-Attention Residual Connection Neural Network (GSA-RCN) model for session-based recommendation. GSA-RCN uses directed graphs to construct session sequences and then graph neural networks to get complex item transformations. Through a combination of graph neural networks, self-attention mechanisms, and residual connections, it associates user global preferences with current interests to represent the session. Through extensive experiments conducted on two real-world datasets, GSA-RCN exhibits a noteworthy enhancement in recommendation performance compared to other state-of-the-art methods.

Path planning is an indispensable part of the application of intelligent mobile robots. Q-learning is a type of reinforcement learning without prior knowledge and is currently a popular path planning method. Although there are numerous research articles on Q-learning for mobile robot path planning, the difficulties of stability and convergence speed of its path planning can hinder the effectiveness of mobile robot applications in reality. This paper presents an improved Q-learning algorithm incorporating the A* algorithm (IQLA). First, a goal heuristic function is introduced to perform an initialisation operation on the Q-table, which guides the intelligent body to move toward the target for accelerating the convergence of the algorithm. Second, an adaptive exploration factor is proposed to improve the action selection strategy, which balances the exploration and exploitation of the environment via intelligence. Furthermore, an adaptive distance search range is proposed to accelerate the convergence of the algorithm. Finally, the A* heuristic evaluation function with weights is incorporated into the Q-learning reward and punishment functions to guide the algorithm to quickly converge to the optimum. This paper validates the performance and effectiveness of the IQLA algorithm over comparison algorithms for mobile robot path planning through simulation and comparison experiments.

Competitive swarm optimizer (CSO) is a powerful variant of particle swarm optimization (PSO) that has been demonstrated to be effective in solving large-scale optimization problems. In this paper, we propose Momentum-incorporated Competitive Swarm Optimizer (MOCSO), which extends CSO by introducing a momentum mechanism inspired by gradient-based optimization algorithms. The key idea behind MOCSO is to improve convergence speed and solution quality by incorporating the exponential moving average of historical differences between competing particles into the velocity update formula. The momentum coefficient in MOCSO is adaptively adjusted based on feedback from the search process. Experimental results on both small-scale and large-scale benchmark problems confirm that MOCSO significantly improves the performance of CSO, making it faster and more robust in terms of solution quality.

The present data collection techniques can generate thousands or even more features in a dataset. However, an excessive number of redundant features can negatively impact the learning speed and classification performance of machine learning models. Selecting important features from high-dimensional data is a challenge. To address this challenge, We propose a cooperative evolutionary algorithm based on feature clustering to partition the feature subspace for feature selection. Firstly, a clustering method based on feature similarity is employed to partition the subspaces at a lower computational cost. Then, an initialization strategy based on feature-label correlation is proposed to accelerate the convergence of the population. Finally, to reduce the dimensionality of the feature subset while ensuring the classification efficiency of the algorithm, a mutation operator based on the optimal individual is introduced to obtain higher quality solutions. The algorithm is applied to 14 classic datasets and compared with 7 advanced algorithms. Experimental results demonstrate the algorithm's ability to obtain good feature subsets.

Session-based recommendation aims to make recommendations for anonymous users based on limited session data. However, traditional session-based recommendation methods fail to capture complex item transitions and simply represent the user's last clicked item as a short-term preference, neglecting the global sequential information of the session. This approach struggles to consider transitions between contexts and cannot accurately capture the user's true intentions. To address these issues, this paper proposes a session recommendation method based on self-attention residual connections and graph neural Hawkes (SRGNH). This method introduces a duallayer network structure consisting of graph neural self-attention residual connection layers and graph neural Hawkes layers, designed to learn users' long-term and short-term preferences, respectively. SRGNH employs a Gated Graph Neural Network (GGNN) to capture complex interactions between nodes, obtaining latent vectors for each item. It incorporates self-attention networks and residual connections to effectively utilize low-level inspired information for capturing users' long-term preferences. The graph neural Hawkes layer combines the Hawkes process with GGNN to capture the relationship between user item clicks over continuous time, accurately representing users' short-term preferences. To better represent user intent, we linearly combine users' long-term and short-term preferences in the end. Experimental results demonstrate that the proposed SRGNH outperforms other recommendation models on the Diginetica, Yoochoose1/64, and Yoochoose1/4 datasets.

Accurately categorizing communities within a social network is a crucial aspect of community detection, carrying significant practical relevance. To achieve a higher quality of community division, we combine reinforcement learning methods to learn the distribution characteristics of community nodes during the iteration process of the algorithm, which guides the nodes to migrate to other communities and enhances the algorithm's global search capability. Through a new$\epsilon$-gradient-greedy strategy, which can obtain a node's gradient information relative to its neighbors and achieve higher performance in local search. To speed up the adaptability of the algorithm at the beginning of the iteration and to alleviate the resolution limitation imposed by modularity optimization, this paper employs a triangular subnetwork-based weight assignment method for balancing the weights of each edge in the network. Experimental results on real-world and synthetic network datasets demonstrate that our method's community identification precision outperforms recent community detection algorithms, exhibiting higher accuracy, higher resolution, and adaptability to various network characteristics and structural changes.

In order to mitigate premature convergence commonly faced by conventional Particle Swarm Optimization (PSO) and enhance the algorithm's global search cparticles can swiftly navigatingapability in UAV path planning, this paper proposes a simulated annealing and dimensional learning augmented particle swarm optimization algorithm (SDPSO). Firstly, learning factors and inertia weights are dynamically adjusted in the search process to achieve a balance between global exploration and local exploitation. Subsequently, the simulated annealing (SA) algorithm is utilized in the early search phase to help the algorithm escape from local optima and enhance its ability to discover the global optimal solution, while retaining the fast convergence characteristic of PSO. Moreover, to rectify the challenge of particle oscillation appeared in the search process, SDPSO embeds a dimensional learning strategy (DLS), which enables all dimensions of each particle to learn useful information from the global optimal. Experimental results demonstrate that incorporating SA in the first 30 iterations of the algorithm not only enhance the capability of jumping out from local optima, but also maintains the rapid convergence characteristic of PSO. Comparative experiments conducted in two distinct environments reveal that SDPSO exhibits advantages in terms of optimization capability, convergence rate, and robustness when compared to other algorithms.

Multimodal optimization aims at efficiently finding multiple optimal solutions of a problem. Owing to the population‐based search mechanism, evolutionary algorithms (EAs) are becoming increasingly popular in solving multimodal optimization problems (MOPs). Most existing work focuses on designing and incorporating niching techniques into EAs so that multiple subpopulations can be formed and assigned to locate different optima. To further enhance the exploration and exploitation abilities of existing EAs, this paper developed a multimodal level‐based learning strategy. The basic idea is that individuals should be treated differently according to their positions in the subpopulation. In the evolutionary process, a subpopulation is formed for each candidate solution by grouping its neighboring solutions. Then, individuals in the subpopulation are sorted according to their fitness. Subsequently, the multimodal level‐based learning strategy applies different mutation operators to different individuals according to their rankings. Experiments are conducted on a set of benchmark problems to verify the efficacy of the multimodal level‐based learning strategy. The results show that the proposed learning strategy can significantly enhance the performance of the existing algorithm. In addition, the algorithm integrated with the proposed strategy is applied to the task of finding multiple roots of nonlinear equation systems (NESs). The results indicate that with the support of the proposed learning strategy, the integrated algorithm compares favorably with state‐of‐the‐art root finding algorithms.

Permutation polynomials (PPs) and their inverses have applications in cryptography, coding theory and combinatorial design theory. In this paper, we make a brief summary of the inverses of PPs of finite fields, and give the inverses of all PPs of degree ≤ 6 over finite fields Fq for all q and the inverses of all PPs of degree 7 over F2(n). The explicit inverse of a class of fifth degree PPs is the main result, which is obtained by using Lucas' theorem, some congruences of binomial coefficients, and a known formula for the inverses of PPs of finite fields.

Recently, Melhem presented a "new" class of low-diameter interconnection (LDI) networks, (IEEE Trans. Computers, Vol. 56, No. 4, pp. 502-510). We note that LDI networks are the same as the previously known generalized de Bruijn graphs, point out an error in the decomposition of LDI networks into permutations, and find that the correct decomposition scheme is an instance of arc coloring for coset graphs. Hence, we pursue a number of general results on arc coloring of coset graphs that can be applied to this particular decomposition problem as well as within many other contexts, including complete arc coloring and normality of coset graphs.