The electric power industry is an important basic industry of the country, and among all the electric power equipment, the distribution lines are directly facing the end-users, which is an important infrastructure to serve the people’s livelihood. In this study, we first transformed the distribution line engineering quality defect acceptance problem into a sequential decision-making problem, and constructed an improved reinforcement learning network model DDQN based on it, and introduced a reward function into the model to improve the intelligent adjustment ability of the intelligent bodies in the model to the data related to the distribution line, so as to improve the detection performance of the DDQN model in the distribution line engineering quality defect acceptance. The results show that the improved DDQN model is highly feasible and effective in the detection of quality defects in distribution line engineering compared with other comparative models. The simulation test of distribution line engineering quality defects found that the accuracy of the DDQN model-based distribution line engineering quality defects acceptance technique in detecting line quality defects is 95%. It is verified that the accurate and reliable distribution network line engineering quality defect acceptance technology based on the improved DDQN model is conducive to guaranteeing the safe and stable operation of the power grid system.

In this paper, OpenCV technology is used to produce the distribution network defects dataset, which can be used as a training set, validation set, and test set in the ratio of 6:2:2. Combining the dataset and the Transformer framework, the S-Transformer based distribution network key quality defect identification model is constructed together. At this level, the degree of equipment deterioration is fitted, the distribution network intelligent operation and maintenance optimization strategy is formulated, and the experimental method is applied to evaluate the distribution defect identification and intelligent operation and maintenance. The identification rate of S-Transformer network for the six collected distribution network equipment defects is 0.9~0.95, which accurately controls the potential dangers, and is conducive to the subsequent intelligent equipment operation and maintenance of the distribution grid and its management and control, compared to the Compared with the traditional operation and maintenance program, the operation and maintenance program in this paper can reduce the operation and maintenance time by 52 hours per month, which greatly provides the efficiency of operation and maintenance labor.

Financial sharing has become an important trend in the process of enterprise development in the era of big data. This topic centers on the research of the application of cloud computing technology in financial shared services, and introduces machine learning algorithms into financial risk early warning. Financial and non-financial indicators are selected to construct the financial analysis index system, K-tuning and mean value algorithm is used to realize the risk level division, SVM algorithm is used to construct the financial risk early warning model, the parameters are continuously adjusted according to the model accuracy rate, and the model is applied to the benefit analysis. Dividing the samples into four financial risk levels of none, low, medium and high can more accurately reflect the specific situation of enterprise finance. It is proved through experiments that the financial risk prediction performance of SVM model in this paper far exceeds the logistic regression model and Gaussian plain Bayesian model, the accuracy rate is improved by 9.7% and 18.6% respectively, and the average accuracy rate in the test set reaches more than 93%. Therefore, it is feasible as well as of great research value to apply cloud computing technology in artificial intelligence to the research field of risk warning of financial shared services.

The construction of ecological civilization is a fundamental plan related to the sustainable development of economy and society, and the dispute settlement mechanism of environmental damages is its innovative and important content. Starting from the environmental legal dispute resolution mechanism, the article analyzes the legal basis of environmental dispute mediation and the process related to pre-litigation mediation. Considering environmental legal dispute resolution as a kind of multi-objective decision-making optimization problem, a multi-objective decision-making optimization model for environmental legal disputes is constructed with the objective functions of legal effectiveness, legal applicability and subject interest rate. Then adaptive inertia weights and dynamic image Pareto solution set updating strategy are introduced to improve the multi-objective particle swarm algorithm, and combined with information entropy-based TOPSIS decision-making to realize the optimal solution selection for environmental legal dispute resolution. In the multi-objective decision-making optimization model, the improved multi-objective particle swarm algorithm achieves the optimum for a total of 15 data, and the simulation time in solving the optimal solution of the 10*10*5 case problem is only 2.314s, and the optimal solution of environmental legal dispute resolution can be obtained based on different objective functions. Environmental legal dispute resolution needs to aim at effectiveness, applicability and subject’s interests, introduce appropriate punitive damages, realize the effective connection between administrative law and criminal law, and promote the high efficiency of environmental legal dispute resolution.

This paper improves the deep residual network, proposes 3DResNet network and carries out particle swarm optimization, constitutes the PSO-3DResNet model, and designs the coal mill fault diagnosis model based on PSO-3DResNet model. The technical parameters, common fault types and fault characteristics of the coal mill are analyzed, and the relationship between the input and output parameters of the coal mill is decomposed by the residual-based condition monitoring method. Combining the numerical simulation model of coal mill and historical operation data, the typical fault condition monitoring of coal mill is constructed. Compare the classification accuracy of each model on the working state of blast furnace wind mouth, and get the anomaly detection performance of each model. The PSO-3DResNet model is analyzed to monitor the normal operating state of the coal mill, and the model is tested using the historical current and outlet wind temperature anomaly data of the coal mill. When the coal mill is in an abnormal state, the estimated residuals of the current abnormal condition fluctuate within [-16,3] with a small range, and the weighted average residuals of the current abnormal condition index remain within [-4,1].

Artificial Intelligence Generated Content (AIGC), as a computer technology mainly characterized by intelligent content generation, has caused significant changes in film and television performances and creations, and has greatly broadened the creation and development space of film and television performances. In this paper, we use motion capture technology to obtain the character movement data in film and television performances, and combine it with the skeletal motion data generation algorithm to realize the mapping of skeletal motion data. Using ResNet-122 as the backbone network, a 3D action pose estimation model is constructed by combining multi-view and multi-feature fusion networks. Based on the 3D action pose estimation sequence, the character animation generation model is constructed by combining GAN and action detail attention mechanism, and the action detail feature loss function is designed to improve the generalization ability of the animation generation model. In order to verify the effectiveness of the above method, data analysis is carried out through simulation verification. The average value of PCP3D index of the 3D action pose estimation model is 98.37, which is 0.28 percentage points higher than the sub-optimal model, and the average joint position error is only 16.07 mm. The animation generation model combining GAN and the action detail attention mechanism has the values of animation generation diversity and richness index of 5.104 and 3.997, respectively, and the animation generation diversity and richness indexes of the animation generation model combining GAN and the action detail attention mechanism are 5.104 and 3.997, respectively. 3Ds MAX software can map the generated animation sequences into the virtual space, providing assistance for optimizing the motion design of film and television performances.

The emotional curve of a story is the core embodiment of the reading value of a novel, and good novels tend to have similar patterns of emotional changes, which are explored in novels by combining artificial intelligence technology. After collecting modern Chinese novel texts, Chinese word segmentation and de-duplication are performed to complete the novel text preprocessing. In view of the limitations of convolutional neural network (CNN) and recurrent neural network (RNN) in text feature extraction, this paper proposes a multi-channel convolutional and bi-directionally gated recurrent unit (BiGRU) deep learning model, Pt-MCBGA, to mine the emotional polarity in the text and analyze the emotional trend of modern Chinese novels. After a series of comparison experiments, it is demonstrated that the model performance achieves a relatively excellent performance, and the recall rate on the two datasets is improved to 83.53% and 83.69%, respectively. According to the Pt-MCBGA model, the sentiment analysis of the modern Chinese novel The Legend of the Eagle Shooting Heroes finds that the novel is dominated by positive sentiment, with both positive and negative sentiment values being relatively high, and that the characters are rich in emotions and have great emotional ups and downs.

In wireless sensor networks in industrial control systems, wireless communication security is challenged due to the broadcast nature of the wireless channel, where information is more easily eavesdropped by illegal nodes on the network. The article establishes a secure communication system based on ZigBee wireless communication technology applied to wireless sensor networks in industrial control systems. In order to improve the secure communication performance of wireless sensor networks, this paper combines the Merkle tree with the μTesla protocol to establish a key management scheme for wireless communication. Then from the node trust degree, the node two-way authentication mechanism for data transmission is constructed by combining the digital signature algorithm. For the effectiveness of the secure communication mechanism of wireless sensor networks, this paper carries out data analysis through performance testing. The key management scheme takes about 17.37 μs and 3.24 μs to add and revoke a key, respectively, and the local optimal value of user time consumption is 7.26 s when the connectivity frequency is 12 min and the revocation threshold is 60. The average value of the node bidirectional authentication mechanism can reach 96.17% for the accuracy of identifying the malicious nodes in the wireless sensor network, and the bit error rate is lower than 0.5 % for the communication transmission with the mesh topology. The bit error rate is less than 0.1%. The introduction of Merkle tree and digital signature algorithms into the construction of secure communication mechanisms in wireless sensor networks can significantly improve the data transmission security performance of industrial control systems.

Tang poetry, as a treasure of ancient Chinese literature, contains a wealth of natural imagery, which not only add to the picture sense of Tang poetry, but are also important carriers of the poet’s emotions and thoughts. The study outlines the nature imagery from the perspective of Tang poetry, as well as the key elements and intrinsic connections among them, and borrows k-means clustering to categorize the nature imagery groups. In addition, the study improves the principal component model by using index homogenization, homogenization, and entropy weighting, so that it achieves the best dimensionality reduction effect while guaranteeing the integrity of the data of Tang poetry text.The F1 value of SVM and KNN classifiers for classifying the natural imagery and emotional expression of Tang poetry text is more than 0.9 after dimensionality reduction of the method in this paper, which is a good classification performance. Cluster analysis divides the natural imagery of Tang poetry into astronomical imagery, landscape imagery, and animal imagery, which account for 38%, 53%, and 9%, respectively. “Old times – bright moon”, “Thinking – slanting sun”, “Looking back – west wind”, “the end of the world – west wind” natural discourse is more likely to form word clusters in the natural imagery of Tang poetry. The analysis of principal component model shows that poets are more willing to express their emotions through natural imagery, and the proportion of neutral emotional expression is 5.17% to 7.43%.

This paper proposes a vocal music teaching system architecture integrating multimedia technology, aiming to enhance the intuitiveness, interactivity and personalization of vocal music teaching through technical means. The system is equipped with virtual reality and voice interaction technologies to realize the digital presentation of the functional modules of the architecture. In addition, in order to evaluate the teaching effectiveness of the system, a number of evaluation indicators are designed. The fuzzy comprehensive evaluation algorithm is used as the main method, supplemented by hierarchical analysis method, to comprehensively evaluate the teaching effectiveness. Multimedia technology can improve students’ vocal ability and mastery of theoretical knowledge, in which the vocal ability is improved by 5.98% to 10.48% compared with the control class, and at the same time, there is a promotion effect on students’ positive interest in vocal learning. The students’ recognition of the system in terms of technology application, learning interaction experience, learning content and process, and teaching effect ranged from 4.077 to 4.608, with a high degree of recognition. The experts’ comprehensive evaluation of the classroom effectiveness of vocal music teaching under the system of this paper is 93.437, which is highly satisfactory. This study not only provides new technical support for vocal music teaching, but also provides a scientific assessment method for teaching evaluation, which is of great significance to improve the level of vocal music teaching.