In order to improve the performance of the sparrow search algorithm (SSA), in this paper, a novel series of SSA variants is proposed by combining SSA with improved Tent chaos mutation (IT), Lévy flights mutation (LF), elite opposition-based learning mutation (EOBL), variable radius mutation (VR) and the combination of IT, LF, EOBL, and VR, namely, ITSSA, LFSSA, EOBLSSA, VRSSA, and CMSSA, respectively. Initially, the performance of these variants is evaluated on a comprehensive set of 31 benchmark test functions. Moreover, the performance of the best algorithm among these variants is compared with 19 state-of-the-art optimization algorithms to validate its performance on 31 benchmark test functions. The convergence and computational complexity of the best variant are also analyzed to test exploration, exploitation, and local optima avoidance. It is then employed on eight real-world constrained engineering problems to further verify its robustness. The experimental results reveal that the best algorithm of SSA variants outperforms other competitors and is highly effective in solving real-life cases.

The orthotropic steel bridge deck weld structure would easily cause fatigue cracking under the repeated action of vehicle load. This paper took the steel box girder in a bridge as a research object, researched the mechanical properties of the steel plate and the microstructure of the welded joint, then designed the fatigue specimens of the deck plate and did the fatigue test. The Δσ-N curves and stress amplitudes of the weld details of the deck plate with U-rib and diaphragm under different probabilities of survival were obtained. After extended the Δσ-N curves to the long life range, the fatigue damage calculation equation of the detail was proposed, and the cut-off limit under the 50% and 97.7% probability of survival were 81.50 MPa and 53.11 MPa, respectively. Based on the actual vehicle load spectrum and simplified finite element model of the steel box girder section, the stress amplitude of the details of the weld joint was calculated. The calculation result shows that the maximum stress amplitude of the concerned point was 38.29 MPa, less than the cut-off limit. It means that the fatigue strength of the details of the weld joint meet the requirement of the fatigue design.

In order to better predict ultimate bearing capacity of perfobond leiste shear connection (PBL), the six specimens were designed for push-out test, and the prediction models were built based on an Improved Adaptive Genetic Algorithm (IAGA) and Back Propagation neural network (BPNN) algorithm. With the finite element model established, it was found that the effects of different parameters on the ultimate bearing capacity of PBL vary greatly if using a single parameter method. The calculation results showed that transverse reinforcement diameter, hole diameter of steel plate, the thickness of steel plate and the strength grade of concrete were the four key factors affecting the ultimate bearing capacity of PBL. In order to overcome the disadvantages of BPNN, such as slow convergence speed and easy to fall into local optimization, an improved adaptive genetic algorithm is used to optimize the initial weights and thresholds of BPNN. The comparison shows that the algorithm is superior to the standard genetic algorithm and other heuristic algorithms, in terms of convergence speed, global search ability and robustness. The IAGA-BPNN prediction model was established. Using the experimental data obtained from both the fatigue test and the references as samples to train the prediction model, the results show that the IAGA-BPNN algorithm proposed in this paper can accurately predict the ultimate bearing capacity of PBL, with an average error of 1.69%. The comprehensive sensitivity analysis (CSA) method adopts to explore the relative contribution of each key factors and the interaction between the key factors, and the analysis results show that the ultimate bearing capacity of PBL increases significantly with the increase of the thickness of steel plate and hole diameter of steel plate. The accuracy and stability were better than formulas for calculating the ultimate bearing capacity and the BP neural network prediction algorithm.

Human language is multimodal, including textual, visual and acoustic information. The task of multimodal sentiment analysis is to use human multimodal information for sentiment recognition. Among the three modalities, text contains richer information than other modalities. With the development of pre-trained representation models on text, most of multimodal sentiment analysis methods use text as primary information and the other modalities as supplementary information. The existing methods suffer from the following limitations: 1) inherent heterogeneity of multimodal data, which makes multimodal fusion difficult as different modalities reside in different feature spaces; 2) asynchronism caused by the inconsistent sampling rates of the time series data of different modalities. To alleviate the heterogeneity and asynchronism of multimodal data, we propose HMAI-BERT, a hierarchical multimodal alignment and interaction network-enhanced BERT. In HMAI-BERT, to improve the efficiency of multimodal interaction, we introduce a memory network to align the different multimodal representations before fusion. After multimodal alignment, we propose a modal update method to address the problem of asynchronism, where each modality is reinforced by interacting with other modalities. In addition, we introduce a fusion module to integrate the three reinforced modalities, and a sentiment enhanced memory to enhance multimodal representation. Our experiments on two public datasets show that the proposed HMAI-BERT outperforms the state-of-the-art methods.

Specular reflection algorithm (SRA) was a single population meta-heuristic algorithm inspired by the physical function of mirror. However, similar to most of meta-heuristic algorithms, it had the disadvantages of weak population diversity, stagnation in local optimal and low convergence rate. In order to overcome these shortcomings, a chaotic multi-specular reflection optimization algorithm considering shared nodes (CMSRAS) was proposed by the combination of population strategy with shared node, improved Tent chaos strategy and Gaussian mutation strategy. Initially, a single population SRA was extended to the multi-population with shared node and the population was initialized by improved Tent chaos sequence to improve the quality of the initial solution and the population diversity, and to enhance the global search ability. Meanwhile, to strengthen the local search ability and the convergence accuracy, the Gaussian mutation and improved Tent chaotic disturbance strategy were introduced into SRA. And then the influence law and sensitivity analysis of the CMSRAS algorithm between the initial setting parameters were obtained based on the response surface method and the Sobol's method. Finally, compared with both 12 state-of-the-art algorithms and 8 well-known advanced algorithms, the performance of CMSRAS was evaluated on a comprehensive set of 32 benchmark problems. In addition, CMSRAS was applied to solve the complex optimization problems of engineering structure. The results demonstrated that proposed CMSRAS algorithm outperformed most competitive algorithms and efficiently solve the real-life global optimization problems.