162 results about "Multilayer perceptron" patented technology

This invention relates to the field of depression monitoring and intervention technology, and discloses a depression monitoring and intervention system based on an internet hospital and heart rate variability (HRV). The system includes: a multimodal physiological signal acquisition module for simultaneously acquiring raw HRV signals, skin conductance fluctuations, and respiratory rate data; a signal processing module for preprocessing and feature extraction of the raw signals to obtain multidimensional feature vectors; and a depression level inference module that constructs a multilayer perceptron neural network, taking as input the multidimensional feature vector containing multidimensional HRV features, skin conductance fluctuations, and respiratory rate statistics, and outputting probability values ​​for no depression, mild depression, and moderate to severe depression levels. This invention establishes an intelligent management mechanism for internet hospitals for the first time, collecting multidimensional physiological data of patients with depression in real time through wearable monitoring units. Based on the dynamic trends of physiological data characteristics, it can serve as an auxiliary diagnostic tool for psychiatry, especially suitable for early screening and remote follow-up of depression, reducing the consumption of medical resources.

This invention discloses a method for optimizing the topology of a 3D on-chip network (SoC) for high-speed data acquisition systems. Addressing the problem that general-purpose 3D SoC architectures are difficult to adapt to the communication characteristics of high-speed data acquisition systems, leading to high transmission latency and redundant link resources, this method models the network topology adjustment process as a Markov decision process and employs deep reinforcement learning to achieve adaptive optimization of the topology. The method uses a general-purpose 3D SoC as the initial architecture, constructing a state representation that includes node connection features, average network latency, longest path latency, and link area; defining a pruning and regrowth action space; designing a reward function that integrates changes in average latency, longest path latency, and link area; and using a deep Q-learning network with a multilayer perceptron structure for policy training. To improve learning efficiency, an action candidate set is constructed and combined with a two-layer greedy policy to achieve fast and effective search. Through iterative learning and topology updates, a 3D SoC topology that better suits the communication load of high-speed data acquisition systems can be obtained, reducing average transmission latency and link area overhead while ensuring connectivity.

This invention discloses a machine learning-based method for parameter prediction and fabrication of compound silicon heterojunction solar cells, belonging to the field of silicon heterojunction solar cells. The J-V parameters of the solar cell are used as input features, and fabrication and performance parameters are used as target labels. A variational autoencoder is used to learn the latent representation of the data, and adversarial examples are generated by calculating the gradient of the input features relative to the model loss. A multilayer perceptron network incorporating a multi-head self-attention mechanism and L2 regularization is constructed, and adversarial training is performed using an enhanced training set to accurately predict the optimal UV / ozone treatment time and cell performance parameters. The intrinsic hydrogenated amorphous silicon / silicon oxide composite passivation layer is fabricated using the optimal UV / O3 treatment time. This invention, driven by machine learning, effectively captures the complex nonlinear relationships between parameters, improves the prediction accuracy of process parameters, and shortens the R&D cycle. It also effectively suppresses atomic diffusion between intrinsic hydrogenated amorphous silicon and the molybdenum oxide transport layer, improving cell performance.

The application discloses a kind of based on agent model's forming product quality prediction method and system, belong to product simulation field.This method: data model, design based on process parameter+material attribute double-flow input data model for quality prediction;Process twin agent model, by designing multilayer perceptron algorithm, cross attention mechanism is introduced in middle hidden layer to simulate process-material parameter between cross-domain coupling and high-order interaction influence relationship, realize product quality prediction;System linkage optimization, build process parameter perception acquisition device, construct material parameter database, build heterogeneous system cooperation and interaction interface, design system linkage optimization method;Based on real-time acquisition process data and database material attribute data continuously update agent model input parameter.The application effectively overcomes the drawbacks that traditional finite element simulation response time is long and model cannot be combined with actual production, realizes the real-time dynamic accurate prediction of forming equipment product quality.

This invention discloses a method for evaluating the freshness of scallops based on a multilayer perceptron (MLP), belonging to the field of rapid detection technology for aquatic product freshness. Using refrigerated scallops with different preservation treatments as the object, volatile odor response signals from an electronic nose are collected. An MLP machine learning prediction model is constructed and optimized to achieve rapid and non-destructive prediction of the volatile basic nitrogen (TVB-N) content of scallops under different treatment groups. Addressing the problems of large differences in volatile components of scallops under various preservation treatment systems and the poor universality of a single model, this invention significantly improves the accuracy and universality of the model's TVB-N prediction for scallops under different treatments through model input feature optimization, hidden layer structure adjustment, and joint training with multiple treatment samples. It is suitable for rapid detection of the freshness of refrigerated scallops under various treatments, solving the shortcomings of traditional detection methods such as time consumption, sample destruction, and difficulty in meeting the needs of rapid on-site detection. The method is simple to operate, highly efficient, and widely applicable, making it suitable for quality control applications in the cold chain distribution and large-scale production of aquatic products.

This invention provides a system and method for synthesizing novel viewpoints in complex scenes (e.g., outdoor scenes). [Solution] In some implementations, the system and method may include or use a machine learning model that can learn from unstructured and / or unconstrained collections of images, such as “wild” photographs. In particular, exemplary implementations of this disclosure can learn volume scene density and luminance represented by a machine learning model such as one or more multilayer perceptrons (MLPs).

This invention provides a method for high-speed train bearing fault diagnosis and selection based on multi-model comprehensive evaluation. The method first obtains bearing fault feature vectors, then divides the training and test sets using 8:2 stratified sampling, mitigating sample imbalance bias through a sample weighting strategy. Next, it constructs a diagnostic system comprising five models, including random forest and support vector machine, and employs optimal hyperparameters for parallel training. Subsequently, a multi-dimensional evaluation system for accuracy and efficiency is established, and model reliability is verified using a confusion matrix. Finally, model selection is tailored to specific scenarios: multilayer perceptrons are used for real-time monitoring, gradient boosting trees for high-precision scenarios, and random forests or K-nearest neighbors for lightweight deployment. This achieves scientific multi-model selection, balancing diagnostic accuracy, real-time performance, and stability, adapting to the maintenance needs of high-speed trains, and demonstrating strong engineering practicality.

This application discloses a particle trajectory reconstruction method, apparatus, device, and medium based on flow-sensing implicit neural representation, relating to the field of computer technology. The method includes: constructing a spatiotemporal position embedding network based on Eulerian flow field information of the flow field surrounding an aircraft, utilizing a multilayer perceptron and implicit neural representation model to capture the spatiotemporal distribution and motion patterns of particles in the flow field around the aircraft; constructing a multi-scale residual learning module using Taylor expansion to balance trajectory learning at different time scales; constructing an initial particle trajectory reconstruction model using the spatiotemporal position embedding network and the multi-scale residual learning module; training the initial particle trajectory reconstruction model using target constraints; generating a target motion trajectory using the obtained target particle trajectory reconstruction model and based on the position information and integration time of the target particle in the flow field; where the integration time is the length of the target particle's motion time in the flow field. This improves the efficiency of particle trajectory reconstruction in complex flow fields.

This invention discloses an intelligent dynamic integration method and system for mobile office applications, addressing issues such as fragmented mobile office applications, inefficient information integration, and low recommendation matching accuracy. The method includes collecting user attributes and application usage data; establishing a user attribute mapping layer with a hash table structure and assigning initial weights to applications; constructing a neural network recommendation model with a multilayer perceptron structure; obtaining real-time application weights after supervised learning training and inputting updated data; sorting and grouping applications according to their weights; and then accurately pushing applications to users through the mapping layer. The system includes corresponding acquisition, mapping, modeling, classification, and push units, and also provides electronic devices and storage media for implementing the method. This invention achieves personalized and accurate matching between applications and user needs, improves the real-time performance and stability of recommendations, adapts to dynamic changes in enterprise organizational structure and user needs, and significantly improves mobile office efficiency.

This invention relates to the field of physical field evolution prediction technology, specifically to an intelligent prediction method for the spatiotemporal evolution of high-dimensional physical fields in forming and manufacturing. The method includes encoding preprocessed historical data of the high-dimensional physical field into a low-dimensional latent representation; extracting historical evolution feature sequences from the latent representation sequence; and using a multilayer perceptron to learn the external environmental interference influence features of the corresponding process environment parameters. The historical evolution feature sequences and the external environmental interference influence features are adaptively fused using a multi-head attention mechanism to obtain the predicted value of the latent representation for the next moment. The predicted value of the latent representation for the next moment is then decoded and reconstructed to obtain the high-dimensional physical field prediction result for the next moment. This invention achieves accurate prediction of the evolution of high-dimensional physical fields affected by environmental disturbances.

This invention discloses a hardware acceleration method and accelerator for Transformer based on mixed-precision quantization and Huffman coding. The acceleration method includes: using a genetic algorithm to obtain several configuration schemes for mixed-precision quantization of Transformer network layers; performing mixed-precision quantization on each Transformer network layer based on each quantization configuration scheme to obtain a corresponding KL divergence; training a multilayer perceptron to obtain a quantization configuration prediction network using the quantization configuration scheme and the corresponding KL divergence as the output label and input feature, respectively; receiving a user-set target KL divergence value, using the quantization configuration prediction network to obtain the corresponding quantization configuration scheme, and performing mixed-precision quantization on each network layer based on the quantization scheme; and using Huffman coding to encode and compress all quantization weights before on-chip storage. This invention can reduce storage and computational overhead while maintaining model accuracy.

This invention provides a drug interaction prediction method based on a bidirectional cross-view attention network, belonging to the field of drug interaction prediction technology. The method first constructs a Morgan fingerprint similarity view of the drug, an original DDI view, and a multi-scale diffusion view based on personalized PageRank. Then, a graph convolutional network with shared weights is used to co-encode the multiple views, generating a unified drug embedding representation. Finally, a bidirectional cross-view attention mechanism is used to achieve fine-grained interaction and alignment between the structural and attribute views, and interaction prediction is completed via a multilayer perceptron. This invention effectively alleviates the sparsity problem of the DDI network through multi-scale topology enhancement and achieves complementary enhancement between views using bidirectional cross-view attention, significantly improving the accuracy and generalization ability of drug interaction prediction. It can provide reliable technical support for drug development screening and clinical combined drug safety assessment.

The application provides a humanoid robot facial expression mapping and calibration method, the facial expression mapping method constructs a self-supervised discrete expression data set, trains a multilayer perceptron model based on the data set, realizes preliminary mapping from expression parameters to rudder control signals, extracts expression parameters in a real person expression data set, generates robot rudder control signals through the model, constructs a self-supervised time sequence expression data set based on a redirection method, trains a long short-term memory network model, and realizes time sequence mapping of continuous expressions. The facial expression calibration method uses a visual capture device to obtain robot facial expression parameters in real time, gradually optimizes the rudder control signal, and makes the expression parameter response approach the target value. Through data-driven modeling and visual feedback optimization, the application realizes high-fidelity mapping from semantic expression parameters to multi-rudder collaborative control and systematic calibration, effectively improving the naturalness, accuracy and long-term stability of the robot facial expression.

This invention discloses a design method and system for high-frequency transformers, belonging to the field of high-frequency transformer optimization design technology. The method includes: acquiring design parameter data of the high-frequency transformer to be optimized, a dataset of historical transformer design parameters, and a dataset of historical transformer actual performance indicators; acquiring a convolutional neural multilayer perceptron deep learning model based on the above data to obtain an initial particle population and predicted particle fitness; if there is a particle fitness greater than a preset historical particle fitness, updating the initial particle population, obtaining a first particle population, and then acquiring the optimization design parameter data of the high-frequency transformer to be optimized; acquiring the simulation experimental deviation value of the simulation data and experimental data based on it, and if it is less than or equal to a preset threshold, completing the design of the high-frequency transformer. The high-frequency transformer design method provided by this invention can achieve high-precision and high-efficiency optimization of high-frequency transformer design parameters.

This invention discloses an artificial intelligence-based method for predicting drug interactions based on external entity association mapping, comprising the following components: (1) Extraction of intrinsic properties of drug molecules: The spatial geometry and atomic properties of molecules are processed using a Uni-Mol pre-trained model to obtain drug molecule characterization. (2) External entity association mapping: Projection blocks are constructed using a knowledge graph, and complex association patterns of external entities corresponding to drugs are captured through graph neural networks and global perception logic. (3) Deep fusion of multi-source heterogeneous information: Information is efficiently integrated using attention mechanisms and multilayer perceptrons to form the final fused characterization. (4) Prediction and decoding of interactions: The specific types of interactions are accurately decoded and predicted using a decoder architecture. (5) Interactive visualization and decision support: Visualization tools and AI decision support are provided to help researchers reduce trial-and-error costs and make scientific decisions.

This invention discloses a gene regulation inference method guided by topological data analysis and gene network embedding. It combines TDA and GNN to enhance the inference capability of gene regulation networks. By capturing the topological structure of the gene regulation network graph through TDA features, the model's ability to model gene expression is enhanced. The TDA features and GAT embedding representations are effectively integrated through gating fusion. This fusion mechanism enables the model to adaptively adjust node embeddings based on global topological characteristics, which not only improves the accuracy of gene interaction representation but may also enhance the accuracy of regulatory relationship prediction. The traditional GAT architecture is extended through a four-layer graph attention mechanism. Each layer uses residual connections to alleviate the gradient vanishing problem and improve training stability. In addition, independent multilayer perceptron branches are designed for transcription factors and target gene embeddings. This deep architecture can achieve more expressive feature transformations and capture subtle patterns in gene regulation networks.

This invention belongs to the field of text retrieval technology, specifically referring to an automatic retrieval system for digital economy texts. The system includes a basic task instance identification module, a key instance association mining module, a tag local score estimation module, a global tag configuration module, a noise tag configuration generation module, and a precise digital economy text retrieval application module. This solution uses two dedicated multilayer perceptrons to calculate semantic and syntactic dependency weights and fuse them to obtain the final association strength, achieving precise capture of digital economy instances and their associations. It combines global tag configuration to quantify the tightness of instance tag associations, and uses the noise tag configuration generation module to transform the global optimal tag configuration problem into a binary classification problem. By constructing a contrastive loss function and optimizing it through backpropagation, it efficiently solves the problem of computational explosion, achieving concept-based precise retrieval and intelligent decision support, providing users with an efficient digital economy information acquisition experience.

The application relates to a bearing fault recognition algorithm suitable for small sample conditions in the field of intelligent manufacturing and industrial fault diagnosis. The whole method process involves vibration signal data acquisition, data cropping, conversion into a frequency domain graph, large-core multi-scale feature extraction, local similarity comparison, global similarity comparison, and fusion of local and global classification stages, and can cope with the real-time performance, high efficiency and deployability of fault recognition under complex working conditions. Among them, the large-core multi-scale feature extraction realizes unified feature extraction modeling; the local module realizes fine-grained feature matching by combining cosine similarity and a multilayer perceptron; the global module models the semantic association between the support set and the query set by using two-dimensional position coding and cross-attention mechanism. Finally, accurate fault classification is realized by fusing the similarity scores of the local and global modules. The method has high accuracy, strong generalization ability and low computational complexity, and is particularly suitable for online diagnosis and edge deployment in industrial scenes.

This invention belongs to the interdisciplinary fields of marine engineering, marine geotechnical engineering, and artificial intelligence. Specifically, it relates to a method for predicting the seabed response around pile foundations based on a deep learning time-series model. The method includes generating multi-point time-series physical field data of multi-directional wave and seabed dynamic response through CFD-FEM coupled simulation, followed by random spatial sampling processing; enhancing wave nonlinear characteristics using multi-order difference; extracting spatiotemporal dynamic features using TimeDistributed-LSTM; constructing a four-dimensional spatiotemporal tensor based on relative coordinates and Euclidean distance; aggregating local and global features using a local multilayer perceptron and max pooling; and finally, using a multi-output decoding network to predict the seabed pore water pressure response in parallel. This invention eliminates dependence on fixed grids and measurement points, exhibits strong spatial adaptability and generalization capabilities, and can achieve real-time, high-precision prediction of the dynamic response of the seabed around pile foundations.

This invention discloses a drug-target interaction prediction method based on graph neural networks, belonging to the field of bioinformatics. It addresses the problem of inaccurate interaction prediction by acquiring molecular structure data of the target drug and target data of the target target; constructing a molecular neural network graph of the target drug and a target neural network graph of the target target, obtaining atomic node vectors, atomic edge vectors, target node vectors, and target edge vectors; constructing molecular feature vectors of the target drug and target feature vectors of the target target, thereby obtaining a drug-target interaction prediction matrix; setting positive and negative samples and training using a multilayer perceptron network model to obtain a drug-target interaction prediction model; and using the drug-target interaction prediction model to identify the target drug and target target, determining the interaction between them. This invention achieves accurate prediction of the interaction between the target drug and target target.

The application discloses a contrastive semi-supervised learning seismic fault detection method. The method comprises the following steps: obtaining synthetic seismic data and field seismic data, wherein the synthetic seismic data is labeled, and the field seismic data does not contain labels; designing a convolutional neural network with an autoencoder structure; using a gradient descent-based optimization method to train the neural network with labeled synthetic data to obtain a pre-trained model and complete the pre-training process; copying the trained neural network as a teacher network and setting its learning rate to 0, while setting the original neural network as a student network; entering the training step, randomly cutting an n*n*n seismic block X0 from the field seismic data, and performing strong enhancement on the seismic block X0 to obtain an enhanced seismic block X1; performing weak enhancement on the same seismic block to obtain an enhanced seismic block X2; inputting the seismic block X1 into the student network and inputting the seismic block X2 into the teacher network, projecting the outputs of the two through a multilayer perceptron, then calculating the cosine similarity loss, and updating the student network and the teacher network with the loss; and repeating the training step until a specified number of executions is reached.

This invention relates to a target localization and counting method based on point domain feature learning. The method includes: inputting the image to be detected into a backbone network to extract a three-level feature pyramid comprising top-level features, intermediate-level features, and bottom-level features; inputting the three-level feature pyramid into a multilayer perceptron module to capture the global long-distance dependence of its top-level features and obtain global information features; inputting the three-level feature pyramid into a visual centering mechanism module to aggregate local region features of its top-level features, obtain local information features, and connect them with global information features along the channel dimension to obtain mapped top-level features; then, combining these with intermediate-level and bottom-level features to form a mapped three-level feature pyramid; inputting this pyramid into a global centralization adjustment module; outputting an initial feature map to a multi-perceptron module; extracting channel features and spatial features and concatenating them to obtain a multi-perceptron feature map; then, passing through a regression head to output the predicted target count of the image to be detected; and finally, passing through a classification head to output the predicted target localization of the image to be detected.

This application belongs to the field of fetal brain age prediction technology, and relates to a method and system for predicting fetal brain age of the cerebellar vermis based on multimodal feature fusion. It employs the MST-Mamba segmentation network, and achieves synergy between local detail capture and global semantic modeling by embedding local-global aggregators at each level of the encoder. Simultaneously, a dynamic channel fusion unit is deployed at the jump connection between the encoder and decoder to avoid problems such as boundary ambiguity, missed classification, and misclassification. Through three parallel branches of a multi-granularity morphology-texture collaborative perception architecture, it simultaneously extracts two types of explicit features (macro-geometric and topological morphology) and two types of implicit features (micro-texture), achieving a comprehensive representation of the developmental features of the cerebellar vermis. After standardizing and calibrating the explicit features, they are spliced ​​and fused with the implicit features along the channel dimension to solve the problems of insufficient multimodal feature fusion and lack of calibration. Finally, prediction is performed using a multilayer perceptron regression head, ensuring the accuracy and stability of the brain age prediction results from the source.

The present application relates to a method and system for predicting optical module failure based on deep learning and index fusion, the method comprising: acquiring real-time multi-dimensional monitoring index time series data of the optical module and preprocessing, fusing the preprocessed data and sending it into a prediction model to obtain the probability of optical module failure; the prediction model comprises a convolutional neural network and a multilayer perceptron; the training process comprises: obtaining the multi-dimensional monitoring time series data of the optical module that has failed within a preset time in a first preset time window; randomly obtaining the multi-dimensional monitoring time series data of the optical module that has never failed within a second preset time window; performing multiple time series feature extraction, activation processing and normalization processing to obtain deep time series features; then performing multiple feature integration and nonlinear transformation to obtain deep abstract time series features; further predicting to obtain the probability of optical module failure; compared with the prior art, the present application has the advantages of high failure prediction accuracy.

A method and apparatus for Chinese medical named entity recognition based on multi-level adaptive semantic enhancement, the method comprising: (1) representing Chinese text as T={C1、C2、···、C N}, construct character C i Features, including character features, boundary features, radical features, and pinyin features; (2) Character-level C is generated through the ERNIE-Health pre-trained model. i The features are transformed into vector representations, including character feature vectors e. c Boundary eigenvector e b , radical feature vector e r Pinyin feature vector e p (3) Input the four character-level features into the character-level adaptive semantic enhancement module. Use convolutional layers to compress the character features, perform nonlinear transformation through gating mechanism and ReLU activation function, dynamically adjust semantic weights, and use multilayer perceptron for decompression to obtain enhanced character-level features; (4) Input the enhanced character-level features into the sentence-level adaptive semantic enhancement module, and adaptively learn the contribution of different characters in the sentence through compression and decompression mechanism; (5) Input the enhanced features after multi-level adaptive semantic enhancement module into BiLSTM-CRF module for label prediction. This invention can better capture semantic differences in context, solve the limitations of existing methods in feature weight allocation, and improve the overall performance of CNER task.

The specification discloses an image classification method and system based on enhanced small sample feature decoupling, aiming to solve the problem that fine-grained features and coarse-grained features cannot be effectively extracted to obtain sufficient local information and global information. The application comprises: based on the to-be-tested data, obtaining enhanced to-be-tested data through a small sample-based data enhancement model; obtaining the enhanced to-be-tested data through a multilayer perceptron and a first classifier; decoupling the enhanced to-be-tested data to obtain fine-grained features and coarse-grained features; connecting the fine-grained features and the coarse-grained features through a feature connector, and classifying based on the connected features through a trained second classifier to obtain a classification prediction result. The application can quickly extract fine-grained features and coarse-grained features, effectively solve the conflict between fine-grained feature extraction and coarse-grained feature extraction, improve the accuracy of image recognition, and solve the problem of insufficient training samples.

A small-sample intelligent prediction method for photovoltaic (PV) power generation based on the fusion of time and frequency domain channels is proposed. This method utilizes raw high-dimensional, multivariate time-series data collected by sensors to construct and preprocess a sample dataset. The dataset is then divided, and an improved generative adversarial network (GAN) is used, combined with gated recurrent units and an attention mechanism, to generate potential representations of sequences from noise. Temporal continuity is ensured through a supervised network, and quality is improved through Wasserstein adversarial training. After fixing the generator parameters, multiple samples are synthesized with the training set to form an extended training set. Next, an intelligent prediction model for PV power generation is constructed, including time and frequency channel feature extraction modules, a cross-channel feature fusion module, and a prediction module based on an improved multilayer perceptron. The model is then trained and optimized. Finally, the model is deployed to a PV production system to perform online inference and prediction on real-time acquired time-series data and output the results. This method is simple to implement, has low implementation costs, high prediction accuracy, and good robustness.

This invention discloses an optimization method and system for a predictive model of diabetic complications. It involves collecting routine laboratory test data from diabetic patients, performing preprocessing such as missing value removal, label merging, and SMOTE resampling to construct a balanced dataset. The average importance of indicators is evaluated using multiple machine learning models to select a subset of key features. Random forest, XGBoost, support vector machine, and multilayer perceptron are selected as base classifiers, and after hyperparameter optimization, a stacked ensemble learning model is constructed to further improve predictive performance. Finally, the model is packaged as an API service and embedded into a hospital information system to achieve high-precision, low-cost, and real-time risk prediction of diabetic complications. Experimental results show that this invention achieves an accuracy of 98.5% and an AUC of 99.76% in predicting diabetic nephropathy complications, outperforming single models.

This invention belongs to the field of rotating machinery fault diagnosis and discloses a fault diagnosis working condition domain generalization method and system based on working condition-independent feature representation. The method includes collecting vibration signals and corresponding working condition information of the target equipment under multiple working conditions in both normal and fault states; cutting the vibration signals into segments of preset length; mapping the working condition information into working condition vectors of the same length as the vibration signal segments; generating modulation parameters using a learnable multilayer perceptron through the working condition vectors; adjusting the vibration signals and their intermediate features step-by-step according to time steps; mapping the vibration signals to a working condition-independent feature representation space; classifying faults based on the mapped feature representations; and outputting fault diagnosis results. This invention, by embedding working condition information into a learnable model, guides the vibration signals to a normalized feature space independent of working condition changes, thereby achieving adaptation to unseen vibration signals under new working conditions and solving the cross-working-condition generalization problem.

This invention discloses a scheduling method for dependent tasks based on a combination of DRL and HSA, including: a step of constructing a directed unordered graph based on dependent tasks; a reinforcement learning-based dependent task scheduling algorithm designed for edge computing scenarios, modeling the dependent task scheduling problem as a Markov decision process, modeling and representing the edge computing scenario and dependent tasks, and combining deep reinforcement learning algorithms, using multilayer perceptrons and cross-attention mechanisms to provide execution strategies and node selection strategies based on task execution paths; for edge computing scenarios where there are large differences in computing power due to node heterogeneity, a heuristic scheduling algorithm based on list scheduling is designed. In scenarios where there are large differences in node computing power, especially where nodes with hardware accelerators cause large differences in the execution time of dependent tasks on different nodes, the algorithm provides task scheduling priorities based on task-node matching differences, and selects execution nodes based on these priorities.