The rapid development of natural language processing technology makes machine translation play an increasingly important role in cross-lingual information exchange. In this paper, we propose an English long text translation paradigm based on the self-attention mechanism and introduce various improvement strategies to enhance the model performance. The model’s ability to process English long text is improved by introducing multi-head attention and hierarchical self-attention modules. The long text translation paradigm is optimized by using techniques such as residual linkage, layer normalization and dynamic memory network. A series of experiments are conducted to verify the effectiveness of the improved model on the English long text translation task. The English long text translation paradigm constructed in this paper outperforms the Transformer model and other related variants on both CPU and GPU. And Transformer outperforms this paper’s model in terms of n-gram accuracy in real translation experiments. The BLEU scores of the improved model on News and other datasets are significantly improved compared with the original baseline model, which verifies the effectiveness of the improvement strategy of this paper and provides a reference for the solution of the problem of English long text translation.

The lateral quantitative control of paper machine is the key to the quality control of paper machine. In this paper, the paper machine transverse quantitative control system is introduced, for the transverse quantitative control and the model exists of strong coupling, large time lag, multi-dimensional number of characteristics, combined with the predictive control theory, put forward based on the dynamic matrix control and Gram polynomials of the intelligent prediction method. Meanwhile, the speed chain control device of the paper machine control system is designed based on the self-immunity controller and simulated and analyzed. The simulation results show that the slope of the model size and computation in this paper’s method is 1.97, which is smaller than that of the traditional MPC’s 2.85, and has more computational efficiency without affecting the predictive control effect, which is more suitable for online operation. At the same time, the speed chain control system applying the self-resistant control algorithm is better than the traditional PID control in terms of steady state performance, dynamic performance and anti-disturbance performance. The method proposed in this paper facilitates the predictive control and speed chain anti-disturbance of the lateral dosing control system of paper machine and promotes the improvement of paper quality.

In recent years, hyperspectral imaging technology has a large application prospect in quality inspection in the tobacco industry. The study is based on near infrared spectroscopy technology and partial least squares regression method to establish mathematical analysis model of tobacco adulteration ratio of four components, such as expanded tobacco, stalked tobacco, large threaded tobacco and small threaded tobacco, and carry out internal and external inspection. At the same time, TLBO algorithm is used in the optimization of ELM tobacco purity grade determination model to realize the design of tobacco purity monitoring method, and then build the real-time monitoring system of tobacco blending ratio and purity. Tobacco with different purity grades were selected for experimental testing and model comparison analysis. The results show that the constructed PLS model can accurately predict the adulteration content of the four components in tobacco, and the correlation coefficients between the predicted and actual values are above 0.95 (p < 0.01), and the relative deviation of the prediction is below 3%.The accuracy of the TLBO-ELM model for identifying tobacco with different grades of purity is 88%, and the classification accuracy in the validation set is improved by 9.32% compared with the ELM model, which is within the acceptable range. It shows a better classification effect than PLS-DA in an acceptable range, which proves that the proposed method can be used for discriminating and monitoring the purity of tobacco. The monitoring system in this paper can be used in the analysis of tobacco blending ratio and purity detection.

The combination of technology and art in film and television special effects can greatly enhance the visual impact of film and television animation, and improve its commercial and artistic value. The study elaborates on the generation of special effects in 3D modeling technology in film and television production, and based on the application of rigid body special effects, it proposes a highly efficient rigid body crushing mode for optimization in response to the problems such as low real-time performance in rigid body crushing simulation. The model is a particle-based real-time simulation method of object crushing under the impact of external forces, using the discrete unit method to represent the inter-particle force, and proposes an inverse crushing mechanism, which realizes the particle-based DEM simulation on the GPU. Experimental results show that the simulation method of rigid body crushing constructed in this paper can meet the simulation requirements in different scene scales, and the rendering rate in small-scale and large-scale scenes is 90~155FPS and 40~50FPS, respectively, which is not only realistic but also real-time, and can meet the requirements of film and television production.

As the world’s No. 1 sport with wide popularity and high degree of attention, there exists a great application demand and development potential for applying artificial intelligence to soccer sports training. In this paper, Yolov5s-CBAM target detection network is utilized to identify the human body posture of target athletes in soccer sports training, and HRNet network is used to detect the location information of key points of target human skeleton and identify the skill movements of soccer players. Subsequently, the TDS-Fast DTW algorithm is applied to evaluate the skill movements to establish a skill recognition and evaluation system for soccer sports athletes. It is verified that the soccer player skill movement recognition model proposed in this paper outperforms other comparative models, with the checking rate reaching 99.12%, and the evaluation scores of the model on the skill movements of the athletes are not different from those of the manual evaluation scores (P>0.05). It is also found that the application of the system in actual soccer training matches can fully meet the needs of soccer training. The system in this paper can accurately assess the technical movements of soccer sports athletes to meet the needs of scientific training, and at the same time, it can meet the needs of coaches to timely grasp the understanding of the level of technical movements of soccer athletes and improve the quality of training.

Taijiquan is a kind of sport that can be used as a national ϐitness program, and its effect on the training effect of adolescent physical coordination has important research value. In this paper, particle swarm optimization algorithm is applied to the optimization of taijiquan training program, and independent samples test and analysis of variance (ANOVA) are used to investigate the quantitative impact of taijiquan training on adolescents’ physical coordination. The results show that the particle swarm optimization algorithm can effectively improve the effect of taijiquan training, and the algorithm convergence and other properties have obvious superiority compared with other algorithms. At the same time, after the experiment, all the physical coordination test indexes of the experimental group were signiϐicantly improved compared with the pre-test and the control group, which explains the important role of taijiquan training in the physical coordination training of adolescents.

Based on the wide application of collaborative filtering algorithm in the current field of graduate employment, this paper introduces it into the employment recommendation mechanism of senior college students and takes it as one of the auxiliary means to formulate the employment policy for senior college students. By studying the implementation effect of employment policy, so as to explore the adaptability of employment policy. Through the time series prediction method based on neural network, the prediction model of employment policy adaptability of higher vocational tertiary students is constructed. Compare the prediction performance of this paper’s prediction model with other models, predict the employment policy implementation effect through this paper’s model, and finally, construct an evaluation system of employment policy prediction results to evaluate the model prediction results. The prediction fit of the model of this paper is 0.8644, and the average relative prediction error is 0.35%, which is the best performance among all prediction models. In the prediction of the employment of higher vocational college students in province A, the number of employment of higher vocational college graduates is positively correlated with the average annual income level and the market share of graduates, and negatively correlated with the total number of gaps between faculty and students in the institutions and the amount of education expenditure. The overall score of the employment policy implementation effect predicted by the prediction model in this paper is 88.8, which is a good evaluation result.

The study constructs a prediction model to predict the mental health status of innovative entrepreneurs. The real data of mental health assessment of innovative entrepreneurs in S province in 2023 is chosen as the data source. The recursive random forest feature elimination method is used to select the features of the mental health status prediction model. The pre-selection-elimination mechanism was used to construct the mental health state prediction model. The prediction models constructed by support vector machine algorithm, decision tree algorithm and random forest algorithm were trained and evaluated respectively. The AUC value and accuracy corresponding to the random forest algorithm are 0.9126 and 86.39%, respectively, which are better than the other two comparison models. Among the 17 mental health characteristic variables selected in this paper, emotional stress and self-acceptance degree have the greatest influence on the prediction model based on the random forest algorithm.

Social media as a new type of media has become an important channel for people to obtain information and communicate, which brings new opportunities and challenges for English education. In this study, the Markov chain model is improved by using the weighted ward system clustering method and the fuzzy set theory to improve the prediction performance of the model. Then the ARIMA model and the improved Markov chain model are combined to construct an improved time series prediction model to realize the prediction of the interaction efficiency of social media in English education. The performance of the improved prediction model is superior compared to other comparative models, providing reliability for the subsequent prediction results. The prediction results show that the interactive efficiency of social media interaction data in English education shows an upward trend over time, and the number of readings and playbacks of English courseware resources as well as video resources increases from 18477 and 18147 to 88629 and 84571 in six months. The predicted results of this study indicate that social media has good interactive efficiency in English education, which can be utilized in the future to expand the dimension of education, build an English education platform, expand the teaching space and extend educational thinking, and play a percolating role in English education.

Since the 21st century, the rapid development of artificial intelligence technology, artificial intelligence in many fields have achieved remarkable research results and applications, the integration of AI technology and music has also gradually become an emerging research field. In this paper, first of all, the generation principle of vocal interpretation AI model is studied, in order to realize the digital conversion of vocal interpretation this paper constructs a converter model so as to facilitate the application of artificial intelligence algorithm model. In this paper, in order to match the generated opera vocal music with the given opera performance background, the rhythmic relationship between opera and vocal interpretation is established, and the relationship between motion salience and note intensity is constructed. On this basis, the generator model is changed to a model with a loop structure, and the music theory is mathematically modeled to propose an adversarial network model based on improved multi-track sequence generation. Finally, for the prediction problem in the vocal interpretation AI model, this paper is optimized based on support vector regression. Through empirical analysis, the improved model in this paper has a smaller gap with the real dataset on the metrics of pitch use, pitch shift, note interval and polyphony rate within the track. Meanwhile, the TD distances of this paper’s improved model on the three datasets are 0.655, 0.784, and 0.685, respectively, which is the smallest in the experimental data, and the quality of the improved model’s vocal music generation is excellent. The pitch distribution of this paper’s improved model and the original vocal data basically match, indicating that this paper’s model has better effect on pitch adjustment. In addition, the improved model of this paper generates vocal music with better musicality effect, which has higher musicality while avoiding the generation of more invalid notes. The research work of the paper proves the feasibility of the AI model for opera vocal interpretation and provides a new solution for the current field of vocal music generation.