In order to improve the automation and intelligence level of underground fluid sampling, this paper proposes a kind of underground fluid automatic sampling device, and carries out the structural design of the sampling device, the control system design and the field experiment test. According to the action process and movement characteristics of the underground fluid stratified sampling device, the control system needs to use multiple electromagnetic control valves to control the switching of the oil circuit of different actuators respectively. In order to improve the control state and response speed of the stratified sampling device system, a fuzzy identification algorithm is chosen to identify the control model, and the MIMO robust generalized predictive controller is used as the robust adaptive controller of the system to realize the low-flow and low-disturbance acquisition of underground fluids at the same monitoring point and at different depths. In the field sampling, the average values of DO at sampling depths of 1m, 2m, 3m, 4m, and 5m for manual sampling, vertical sampling, and fuzzy adaptive device sampling under the 1-2 sampling plumb line were 7.98mg/L, 7.86mg/L, 8.25mg/L, 7.83mg/L, and 7.77mg/L, respectively. The deviation of dissolved oxygen content at the same sampling point in the three ways is small and the trend of change is consistent at different depths. It shows that the fuzzy adaptive stratified sampling device system designed in this paper can be applied to the sampling of subsurface fluids with dissolved oxygen as the detection target.

In this paper, finite element simulation of heat transfer process is carried out using Cu composites reinforced with TiB2 of different particle sizes. Based on the FEA data, the BP neural network algorithm is integrated and optimized by the MEA algorithm to establish the FEA-MEA-BP performance prediction model. The results of thermal conductivity analysis show that the correction factor of the simulated thermal conductivity value of TiB2/Cu composites can be calculated using the finite element method as 2.3. Compared with the actual value measured by the LINSEISLFA1600 laser thermal conductivity meter, the fluctuation of the simulated thermal conductivity results from the experimental results is no more than 10% between 50~200°C, and the simulation performance has a high degree of accuracy. Taking 304L stainless steel as a sample, the RMSE, MAE and R² are improved to different degrees compared with other models, so the performance of the FEA-MEA-BP model is excellent in terms of the accuracy of prediction.

The article is based on the need for music education innovation in colleges and universities to optimize the traditional piano skill training through Monte Carlo algorithm. Taking the finger as the research entry point, based on the physiological structure of the hand, the reduced-density Monte Carlo method is used to carry out the mechanical design of the finger trainer and plan the finger training movement mode. Through kinematic simulation experiments to understand the feasibility of the piano finger training device in this paper. Analyze the error sensitivity of position and posture on the finger training device. Finally, the teaching experiment method is utilized to explore the training effectiveness of the Monte Carlo-based piano finger training device in this paper. This paper has good usability. When the position error of the mechanism varies in the range of -40mm~40mm, the position error gradually decreases in the X-axis and Z-axis, and the position error in the direction of Y-axis remains stable. The attitude error of the mechanism gradually increases with the increase of the X-axis rotation angle. The output accuracy gradually increases during the rotation from -5° to 5° around the Y-axis. The angular attitude around the Z-axis has no significant effect on the output accuracy. The two groups did not have significant differences in the four dimensions of piano playing skills before the experiment. After the teaching experiment, the experimental group was much better than the control group, and the posttest results of the two groups produced significant differences, and the pre-test and post-test results of the experimental group possessed very significant differences. The Monte Carlo optimization-based piano finger training device has a significant effect on the improvement of students’ piano skills.

Driven by artificial intelligence and deep learning technology, this study proposes an intelligent course recommendation system for art and design education. By constructing XMMC, a joint extraction model of knowledge entities and relations based on deep learning, the accurate analysis of course knowledge structure is realized. Key features such as user preference, content semantics and social influence are extracted by combining multi-feature ranking models such as collaborative filtering, topic modeling and course hotness. Finally, based on the deep reinforcement learning algorithm DDPG, a dynamic recommendation strategy is designed to optimize the recommendation effect. The experiments are based on Coursera Course, Caltech-UCSD Birds 200 and Education Recommendation datasets, and the results show that the improved DDPG model achieves 49.11%, 70.05% and 59.23% course coverage on the three datasets, respectively, which is better than the traditional algorithms Time SVD and CDAE with significant improvement. We constructed the art education course category with the number of topics as 5. In the practical application, the recommended list generated by the system is highly consistent with the course heat analysis, in which the course “Introduction to 3D Modeling and Blender” ranks the first with 6729 average playbacks, which verifies that the recommendation strategy can effectively improve the fitness of the pushed content and the current course progress of the students. It verifies that the recommendation strategy can effectively improve the compatibility between the pushed content and the students’ current course progress.

Mining the dynamic association between psychological state changes and sports performance is one of the core tasks of physical education towards scientific teaching. In this paper, the data of psychological change indexes of student athletes were collected by scales and the indexes variability was tested. Combined with the principal component analysis to extract the principal component factors of the psychological change index data, construct the correlation coefficient matrix, and calculate the multiple linear regression equations of psychological change and sports performance. The gray correlation model based on the whitening weight function was used to analyze the gray correlation between psychological change and athletic performance, and calculate the influence of the two.Among the 9 psychological indicators, 4 dimensions, such as social evaluation anxiety, had a significant difference with P<0.01. P<0.05 for 2 dimensions such as competition preparation anxiety, there was a difference. In the principal component analysis, the negative and positive psychological dimensions were extracted as principal components, including the 7 psychological indicator components excluding the 2 dimensions. Judging from the regression coefficients and gray correlation calculation results, the 3 psychological indicators of cognitive state anxiety, state self-efficacy, and injury anxiety had the greatest influence on sports performance. Targeted alleviation of cognitive and injury anxiety and improvement of self-confidence can optimize students' sports performance.

In this study, a multimodal analysis framework based on GCN is constructed to address the needs of interaction behavior analysis and creativity assessment of programming games for 0-6 years old children. A stack noise reduction self-coding neural network is used to recognize human gestures in images, and the feature representation of interaction behaviors is realized based on GCN, and the effectiveness of the method is proved by the experimental results on the two-player interaction behavior library. Construct a creativity evaluation system applicable to programming game scenarios for young children, and recruit 80 students aged 0-6 years old to carry out experiments. The students were classified using the GCN-based interaction behavior analysis model, and the weights of creativity evaluation indexes were determined by AHP. The fuzzy comprehensive evaluation method was used to evaluate and score the factors of creativity of the three categories of students, and the test results were verified with the help of the gray correlation method. The comprehensive evaluation scores of the three types of students are 2.006, 3.507 and 5.026, respectively, in which the creativity level of excellent learners is the highest and reaches the excellent grade. The normalized gray comprehensive correlation vector (0.3224, 0.3727, 0.3049) is close to the AHP weight vector (0.328, 0.357, 0.315) with a good assessment effect, and the research results provide a new technical path for behavior analysis and creativity development assessment in early childhood programming education.

The study of student behavior data is a necessary way to enhance the digitalization process of the curriculum system of business administration. This paper collects students’ online platform learning behavior and other data, and carries out data cleaning and other pre-processing on them. Using the density function and entropy discretization algorithm to divide the continuous student data into intervals, and study the course learning characteristics of students in different attribute intervals. On this basis, optimize the curriculum system of business administration majors in colleges and universities, and judge its application value through comparative experiments. Obtain students’ satisfaction data on the curriculum system of business administration majors in colleges and universities, and analyze the direction of continuous improvement. Through preprocessing and descriptive analysis, it can be judged that the student behavior data conforms to the characteristics of continuous data, and can be classified using the discretization algorithm. The student behavior data are discretized into 3 major categories and 11 subcategories, and the attribute characteristics of each category can represent the behavioral characteristics and curriculum needs of different students. The mean value of the overall satisfaction of the optimized business administration curriculum system is 3.567 points, and the scores of all dimensions are higher than those before optimization, and there is no gender difference in satisfaction (P>0.05). The entropy-based discretization algorithm can effectively support the optimization of business administration professional course system.

This paper discusses the application of AI color analysis technology in oil painting teaching, combined with experiments to verify its effect on improving teaching quality. Firstly, the core algorithm of AI color analysis technology is analyzed, and the implementation scheme of digital image sharpening preprocessing is proposed based on the RGB color model, and the edge and color information of the image is extracted based on the improved Canny operator. Improved GAN completes the reconstruction of the oil painting image, and the characteristic colors of the oil painting are extracted using the optimized K-means clustering algorithm. The oil painting images are selected for color feature analysis, and the color matching scheme is improved based on the color feature results to construct the color analysis process based on AI technology in oil painting teaching. Finally, students from art colleges were selected as the research subjects, and a control experiment was designed to investigate the effect of AI color analysis in teaching. The p-value of the five factors of the experimental group and the control group’s post-test scores of creativity of modeling, application of color, color richness, emotional tendency of color and expression of the theme are all less than 0.05, and the average scores of the experimental group in these five aspects, 3.66, 3.74, 3.85, 3.77, 3.34, are all significantly larger than those of the control group, which indicates that the experimental group using AI color analysis to assist teaching has significantly widened the gap between the control group and the experimental group in terms of the use of color. It shows that the experimental group using AI color analysis to assist teaching has a significant gap with the control group in the use of color.

This study focuses on the computational analysis of business data, constructs a market prediction framework that integrates K-means clustering, feature standardization and improved N-BEATS model, and verifies its effect on the accuracy of enterprise strategic decision-making based on multi-source data. The study selects real-time transaction data and weather data from 800 merchants under Alibaba, extracts key features through standardization and correlation analysis, and improves the model by introducing topological features and multi-attention mechanism, which significantly optimizes the time series prediction accuracy and reduces the RMSE by 18.6%. The empirical analysis for tissue paper category shows that the forecast error rate of the time series decomposition method is only 0.58%, which is better than the traditional trend method and seasonal index method. Through the regression analysis of 328 business managers’ questionnaires, data-driven analysis β=0.617, p<0.001 and innovative forecasting β=0.594, p<0.001 have a significant positive effect on strategic decisionmaking accuracy and consensus.

In the context of digital economy, the cultivation of rural revitalization talents urgently needs interdisciplinary collaboration and intelligent support. Taking Ningbo Future Rural College as a practice carrier, this study proposes an interdisciplinary platform construction model that integrates knowledge graph and deep reinforcement learning. The dynamic semantic association network is constructed through knowledge graph representation learning (TTransE), which is combined with a hierarchical reasoning intelligent framework to realize personalized learning and career recommendation. The experiment is based on 850 student sample data, and uses principal component analysis (PCA) dimensionality reduction with Pearson’s correlation coefficient to validate feature relevance and construct a breadth-first association knowledge graph. The model achieved the highest 97.79% accuracy with 90.89% F1 value in entity recognition, and the skill assessment score was improved to 80.69, which was significantly higher than the control group’s 67.10. The model significantly improved students’ thinking ability (8.58 vs. 6.20), skill level (7.96 vs. 5.64), and innovation and adaptability (7.84 vs. 5.71) in this paper. The application of the proposed research methodology in Ningbo Future Rural College is effective and has certain promotion value for rural revitalization talent cultivation across specialties, which can help professional decision makers to formulate and implement cultivation policies.