699results about How to "Increase training speed" patented technology

The invention relates to the technical field of electric power short-term load prediction, in particular to an industrial park short-term load prediction method and system based on a Diffusion-Transform-AR model. Acquiring non-stationary historical load data and meteorological data of the industrial park, and performing missing value processing on the data; performing correlation analysis on the data by using the maximum information coefficient, selecting meteorological data with relatively high correlation, and reprocessing time data; performing time sequence decomposition on the load data by using a seasonal trend decomposition method to obtain trend term, seasonal term and residual term features; constructing a diffusion Transform prediction model, and taking the residual term, the meteorological data and the time data as input prediction residual; constructing an autoregression prediction model and taking the trend term as an input prediction trend; according to the method, the advantages of various models are combined, time sequence features are deeply mined, the prediction precision and the model generalization ability are effectively improved, and the method is suitable for complex industrial park load prediction scenes.

The invention discloses a quantitative training method and system for a deep neural network, and the method comprises the steps: obtaining an original neural network model and a verification data set, and generating a quantitative priority through hierarchical sensitivity evaluation and topology analysis; on the basis of the layer sensitivity matrix, differentiated quantization parameters are configured for each layer, and quantization constraints are established; constructing an error compensation mechanism, and generating an error compensation strategy; executing dual-scale adaptive quantization training, and respectively adopting an independently optimized quantization strategy for forward propagation and back propagation; dynamically adjusting quantization parameters according to the training state and the resource constraint; and evaluating model performance and resource efficiency. According to the method, through multi-dimensional sensitivity evaluation, a dynamic quantification strategy and system error compensation, the precision and efficiency of quantification training are improved, and meanwhile it is guaranteed that deployment resource constraints are met.

The invention provides an air-ground integrated mining area environment monitoring method and system based on hyperspectral imaging, and relates to the technical field of mining area hyperspectral imaging determination, and the method comprises the steps: obtaining hyperspectral image data and spectral data of atmosphere, water, soil and landform of a mining area, and carrying out the preprocessing; interpreting the preprocessed hyperspectral image data and spectral data of the atmosphere, the water body, the soil and the landform respectively to obtain hotspot feature vectors of the atmosphere, the water body, the soil and the landform respectively; and inputting the hotspot feature vectors of the atmosphere, the water body, the soil and the landform into an anomaly monitoring model to obtain an environment anomaly monitoring result, performing data processing and model prediction identification on the atmosphere, the water body, the soil and the landform of the mining area by adopting an artificial intelligence means, and accelerating the network training speed by utilizing a customized transfer learning strategy to obtain an environment anomaly monitoring result. The model generalization is improved; and finally, mining area prediction is carried out by using on-site imaging spectrum data, so that intelligent monitoring of the mining area environment is realized.

The invention discloses an efficient auditing method and system based on an edge computing technology, and relates to the technical field of auditing methods and systems.The efficient auditing method comprises the following steps that S1, data collection and preprocessing are conducted, specifically, business data are collected in real time through various sensors and interfaces at edge nodes, and preprocessing is conducted; s2, edge side data analysis: performing real-time analysis on the preprocessed data by utilizing edge computing resources; s3, feature extraction and model training; s4, constructing a risk assessment and early warning model based on edge analysis and a model training result; s5, edge and cloud collaborative auditing; the edge node compresses and encrypts the processing result and the key feature data and then uploads the data to the cloud; and S6, visualizing an audit result and generating a report. According to the method, data acquisition and preprocessing, analysis and model training are realized through edge calculation, the auditing efficiency and accuracy are improved, a risk assessment and early warning model is constructed, the edge and the cloud are coordinated, and comprehensive, timely and reliable auditing support is provided for enterprises.

The invention belongs to the technical field of model training, and particularly relates to a large model increment training method and system based on dynamic sparsification. The method comprises the following steps: S1, evaluating the parameter importance of a large model by adopting a multi-dimensional evaluation method, and reflecting the contribution of parameters to the performance of the large model; s2, performing dynamic sparsification control on the large model structure by adopting a self-adaptive structure adjustment method; s3, performing incremental training management on the large model by constructing a capacity-adjustable empirical playback buffer pool and proposing a multi-objective loss function; and S4, carrying out model compression optimization on the large model by adopting a mixed precision quantification strategy. The method has the characteristics that the network structure can be dynamically adjusted by evaluating the importance of model parameters in real time, so that the efficient training is realized while the model performance is ensured.

The invention discloses a large language model training method and system based on elastic federal low-rank adaptive fine tuning, which adopts federal aggregation based on singular value decomposition (SVD) to reconstruct a local LoRA module of a client into a local increment matrix, then aggregates the local increment matrix into a global increment matrix, and generates a LoRA module of the next round through SVD, thereby effectively eliminating aggregation errors of traditional FedLoRA. According to the method, the LoRA rank is dynamically distributed for each weight matrix, and the balance between statistics and hardware efficiency is realized. In order to further reduce communication overhead and maintain model precision, a self-adaptive matrix freezing mechanism is introduced, and whether freezing is carried out or not is judged according to training stability of a weight matrix. According to the method, challenges of traditional FedLoRA in the aspects of training efficiency, communication overhead, model precision and the like are solved.

The invention discloses a KAN network-based water quality parameter remote sensing inversion method and system, and the method comprises the following steps: 1, obtaining field detection water quality parameters of a set region, and carrying out the classification according to time; 2, acquiring and preprocessing a multispectral remote sensing image; 3, constructing a KAN model architecture, inputting the waveband reflectivity of the remote sensing image at a certain point, outputting the water quality parameter at the point, taking the label as the real detection data of the water quality parameter, and taking the loss function as the mean square error of the output and the label; 4, taking the water quality parameters and the wave band reflectivity data set as a training set, and training the KAN network by using a loss function; and 5, inputting the wave band reflectivity of the water quality parameter value to be detected into the trained KAN network to obtain the water quality parameter value of the whole water area. According to the method, the calculation cost and the model complexity can be effectively reduced, and the model result can be subjected to interpretable operation so as to better adapt to actual application requirements.

The invention discloses a lifelong target re-identification method based on descriptive text prompt prototype compensation, relates to the field of computer vision, and aims to display and solve the problem of feature drift caused by model increment update by introducing a prompt prototype compensation module. The module generates a multi-modal vision and text prototype for each target category through a vision language pre-training model, and maps old task features to a new task feature space through a cross attention mechanism so as to realize fine-grained semantic drift compensation. In addition, a boundary perception prototype correction module is designed to solve the problem of boundary confusion caused by updating of old prototypes. The module dynamically samples false features in compensation distribution to correct old prototypes with fuzzy boundaries, focuses on categories easy to confuse to maintain clear decision boundaries, optimizes cross-task and intra-task correction loss to reduce category confusion, effectively alleviates the problem of disastrous forgetting, and improves the adaptability and generalization ability of the model.

The invention discloses a power grid communication equipment fault identification method and system based on multi-source heterogeneous data characterization learning. The method comprises the following steps: firstly, collecting various complex multi-source heterogeneous data containing fault power grid communication equipment and preprocessing the complex multi-source heterogeneous data; extracting deep semantic features by using a pre-trained feature extraction network and obtaining a feature vector sequence; capturing time features of the feature vector sequence through a bidirectional long-short-term memory network, and obtaining complete sequence feature representation through a full connection layer; performing low-rank multi-modal fusion on the complete sequence feature representation in combination with cross-modal Transform to obtain three types of output splicing construction joint representation, and obtaining a fusion feature by using a self-attention mechanism; and finally, constructing a fault identification model by using the fusion features. Through innovative data preprocessing and data fusion technologies and in combination with an advanced deep learning model, fault-related characteristics can be automatically learned from multi-source heterogeneous data, and intelligent identification of the fault state of the power grid communication equipment is realized.

According to the radar moving target detection method based on small sample transfer learning and the attention mechanism, a small amount of actual radar echoes and high-fidelity scene simulation data are used as a data training set, and direct detection of a moving target under a strong clutter background can be realized. An attention mechanism is introduced by utilizing the point / line distribution characteristics of a moving target and a scene clutter in a space-time two-dimensional plane, the clutter sorting and recognition capability of the network can be ensured while the network training speed is improved, meanwhile, a small sample migration twin network is further combined, a small amount of actually measured data can be utilized to correct a target detection network, and the target detection accuracy is improved. The adaptability of the network to actual working conditions is improved, and the generalization ability is improved. According to airborne flight calibration test data analysis, under the condition of a low signal-clutter-noise ratio, compared with a traditional STAP + CFAR technology, the method has higher detection performance and lower false alarm rate. Meanwhile, based on network pre-training and a small sample transfer learning mechanism, the method has practicability in the aspect of early warning radar real-time processing and has high engineering application value.

The invention discloses a panoramic 3D occupancy prediction method based on 3D Gaussian sputtering. The method comprises the following steps: performing 2D feature extraction on a multi-view RBG image by using a 2D image feature extraction network; converting the 2D image features into 3D features through a deep prediction network; extracting the 3D features by using a 3D feature extraction network; performing 3D Gaussian sputtering on the 3D features, and then rendering the 3D features to a final 2D image; a loss function is set, and a 2D true value is used for supervision; and constructing a dense 2D semantic and deep truth value data set. According to the invention, effective supervision of 3D occupancy prediction is realized, and the training speed and prediction accuracy are improved.

The invention discloses a human face sensitive data protection method based on an elastic anti-interference privacy protection mechanism, belongs to the field of information processing, and provides a dynamic self-adaptive sampling strategy based on information entropy in a data preprocessing stage, and the sampling rate is adjusted according to the distribution characteristics of a human face data set and the weight of sensitive information. And the model performance is maximized on the premise of ensuring user privacy. In order to obtain a better privacy protection effect, an entropy adjustment differential privacy mechanism is designed, and Gaussian noise of different degrees is adaptively added according to an information entropy value of face data after a face data set is dynamically and adaptively sampled, so that refined privacy protection of face sensitive data under differential privacy constraint is realized. In a model training stage, an elastic anti-interference training method is provided, diversified adversarial samples are dynamically generated in different stages of model training, and the over-fitting risk of the model is reduced while the robustness of the model to various types of adversarial attacks is enhanced.

The invention belongs to the technical field of automatic driving, and discloses an autonomous driving decision near-end strategy optimization method based on distributed exploration enhancement. According to the invention, a distributed architecture of the reinforcement learning decision module of the autonomous vehicle is designed, through parallelization and distributed processing, multiple agents can execute strategy training in different environments at the same time, and strategy updating and empirical data are shared in real time. In addition, the distributed architecture allows the method to operate in a larger-scale data set and a complex environment, and the adaptability and generalization ability of autonomous driving decision making are improved. In addition, a random network exploration (RND) module is also combined, the diversity and effectiveness of exploration are further enhanced, and a near-end strategy optimization algorithm used by the reinforcement learning decision module is perfected. By combining the distributed architecture and the RND module, the training process is remarkably accelerated, the exploration efficiency is improved, the decision-making quality of the automatic driving system is comprehensively improved, and the reliability and efficiency of automatic driving decision-making are improved.

The invention relates to the technical field of farmland identification, and discloses a remote sensing image classification and farmland identification method based on deep learning, comprising the following steps: S1, data collection: acquiring remote sensing image data covering a target farmland area; s2, data preprocessing and labeling: preprocessing the remote sensing image data, and labeling the preprocessed data; s3, model construction: constructing a convolutional neural network model, and training the model by using the preprocessed remote sensing image data and annotation data; and S4, remote sensing image classification and farmland identification: inputting a to-be-identified remote sensing image into the trained convolutional neural network model, and outputting a classification result by the convolutional neural network model. According to the invention, transfer learning and fine tuning technologies are utilized, and a pre-trained deep learning model is applied to a remote sensing image farmland identification task, so that the remote sensing image farmland identification task can adapt to farmland identification tasks of different regions, different seasons and different crop types.

The invention discloses an in-structure sound field bounded uncertainty analysis method based on a physical information neural network. The method comprises the following steps: establishing a physical information loss function and establishing a physical information driven neural network; obtaining an input uncertain parameter value range of the internal sound field, obtaining a plurality of parameter levels, and training a sound pressure prediction model under the parameter levels; acquiring sound pressure of a sample point and establishing a data set; establishing a data driving residual error based on the data set, establishing an overall loss function by using the data driving residual error and a physical information loss function, and training a parameter-containing physical information driving neural network to obtain a parameter-containing internal sound field sound pressure prediction model; and establishing an interval model, and obtaining an internal sound field sound pressure uncertainty distribution range by using a global search algorithm and the parameter-containing internal sound field sound pressure prediction model. The method can solve the problem of prediction of the sound field in the structure containing uncertain parameters, and can effectively improve the calculation efficiency on the premise of ensuring that the calculation precision meets the engineering requirements.

The invention belongs to the technical field of reliability-based design optimization, and particularly relates to a reliability design optimization method combining transfer learning and a physical information neural network. Comprising the following steps: constructing a physical information neural network model by adopting training data of a source domain, and evaluating the reliability of the source domain. A physical information neural network model constructed by adopting training data of a source domain is updated by utilizing a layer freezing and fine tuning technology in transfer learning, so that the reliability of a target domain with different statistical attributes is evaluated, the calculation overhead is saved, and the calculation efficiency is improved. A first-order scoring function is utilized to carry out random sensitivity analysis on the failure probability so as to ensure normal operation of design optimization. And finally, providing estimation reliability, an objective function value and corresponding sensitivity information for an optimizer to obtain a corresponding updated design point. And repeating the iterative design process until the optimal design is realized. The training speed of the physical information neural network model is increased, and the precision of the physical information neural network model is improved.

The invention discloses a TEC space-time prediction method based on fusion of feature analysis and a self-attention mechanism, and the method comprises the steps: screening strong effective features through analyzing the correlation between TEC historical data and each feature parameter, calculating the attention weight between data points on each time sequence through employing a self-attention coding mechanism, and carrying out the weighted summation to obtain the attention output. In an ionosphere total electron content prediction task, firstly, for sun and geomagnetic activity characteristic parameters, the correlation between each parameter and a TEC sequence in the same time period and a TEC sequence after a day interval is evaluated, strong correlation characteristics are screened, TEC sequences of adjacent grid points are used as characteristic quantities, multiple spatio-temporal characteristic matrixes of different time scales in a global range are established, and the total electron content of the ionosphere is predicted. And encoding the input feature information according to a self-attention encoding mechanism to obtain an attention output result, and decoding in a decoding layer through a linear model to obtain a TEC predicted value at a future moment. According to the method, efficient extraction of spatio-temporal information can be completed, and higher-precision prediction of the TEC is realized through a self-attention mechanism.

The invention aims to provide a named entity recognition method and device based on a large language model, and the method comprises the steps: carrying out the data cleaning of a to-be-trained text, carrying out the data enhancement of the cleaned data, and obtaining target data; the lora structure is fused into a large language model LLaMA model, and a new large language model is obtained; training the new large language model by using target data to obtain a target model; and extracting a target entity from the text by using the target model to complete identification of the named entity. Therefore, the problem of insufficient effect caused by less parameter quantity and knowledge in the traditional method is solved, the training data is increased, the effect of the model is improved, the training speed of the model is improved by adopting lora, and the problem of insufficient training resources is solved.

The invention discloses a model training method and device, equipment and a medium, and the method comprises the steps: obtaining model demand information, and creating a to-be-trained model, an initial parameter auxiliary table and an initial training state table based on the model demand information; training the to-be-trained model according to a preset data set, the initial parameter auxiliary table and the initial training state table, and determining a current loss function and a deviation parameter of the training; updating the initial training state table based on the deviation parameter to obtain a next training state table; according to the current loss function, determining dynamic gradient information and combining with a preset fluctuation threshold to update the initial parameter auxiliary table; and performing iterative training on the to-be-trained model according to the next training state table and the next parameter auxiliary table until a training ending condition is met, and obtaining a final training model. The training process is recorded in the table, the parameter dynamic gradient information and the change condition of the loss function are determined for dynamic adjustment, the parameters are adjusted in a self-adaptive mode, and the training speed and the model accuracy are improved.

The invention relates to the technical field of artificial intelligence, and discloses a multi-modal model training fine-tuning optimization method based on artificial intelligence, which comprises the following specific steps: preparing and preprocessing data, collecting a large-scale multi-modal data set, constructing an image text data set containing one million high-quality images and text description accurately corresponding to the images, and performing fine-tuning optimization on the image text data set. According to the method, data are subjected to comprehensive and meticulous cleaning operation, noise data and mistakenly labeled data in the data are strictly removed, so that the accuracy and reliability of the data are ensured, and the effective utilization rate of the data can be increased by at least 20% through data cleaning; wherein the image data is screened, blurred, low-resolution and unclear-content images are removed, it is ensured that each image can clearly convey specific visual information, and text description, grammar error checking, spelling error and semantic ambiguity parts are corrected or deleted.

The invention provides a large model training and evaluation method, and belongs to the technical field of artificial intelligence, and the method comprises the steps: obtaining intelligent question and answer data and a training data set, and the intelligent question and answer data is generated by a user side in response to the interaction operation between a user and a pre-trained large model in a dynamic changeable environment; performing online annotation processing on the intelligent question and answer data to obtain online annotation data; performing data integration processing on the online annotation data and the training data set to obtain an integrated training data set; performing model training on the pre-trained large model by adopting the integrated training data set to obtain an intermediate large model; performing dynamic simulation evaluation processing on the intermediate large model to obtain a first model performance evaluation result, and performing model parameter adjustment processing according to the model performance evaluation result to obtain a target large model; wherein the target large model is used for outputting a question-answer result according to the input user. According to the method, the dynamic adaptability and the accuracy of the output result are improved, so that the questioning experience of the user is improved.

The invention discloses a regional school site selection optimization method and system based on an improved black plinuan search algorithm, and the method comprises the steps: building a regional school site selection optimization model through an XGBoost algorithm, and enabling the regional school site selection optimization model to achieve the optimization of a multi-target optimization function composed of school bearing pressure, student school accessibility and school service coverage rate as a target; according to the method, a regional school site selection optimization model is solved through an improved Black-South optimization algorithm, and the improved Black-South optimization algorithm is obtained through a Sine chaotic mapping initialization population, a golden sine search strategy and the search performance of a Gaussian disturbance improvement algorithm, so that the diversity of the population is increased, and the global search and local optimization capability of the algorithm is improved. According to the method, the regional school site selection optimization model can be established through the XGBoost algorithm, the optimization speed can be effectively improved, and the global search and local optimization capability of the algorithm can be effectively improved by improving the Black-Youlinuan optimization algorithm and increasing the diversity of populations.

The invention relates to an artificial intelligence model building method and device, equipment and a storage medium. The method comprises the following steps: preprocessing input original data to obtain a privacy protection data set; performing feature extraction through a multi-scale convolutional neural network connected in parallel in the target model to obtain a target multi-level local feature; dividing the multi-level local feature into two parts to obtain a first part of local feature and a second part of local feature, and inputting the second part of local feature into a bidirectional long-short-term memory network and a bidirectional gating loop unit network which are connected in parallel in a target model for feature extraction to obtain a global semantic feature; splicing and weighting the first part of local features and the global semantic features to obtain fusion feature representation; and performing model parameter optimization on the target model based on the fusion feature representation to obtain a trained artificial intelligence model. According to the invention, the training speed and convergence stability of the artificial intelligence model are improved.

The invention relates to a remote sensing target detection method and device based on multiple scales and an efficient attention mechanism, and the method comprises the steps: obtaining a target remote sensing image, inputting the target remote sensing image into a target detection model, and obtaining a target detection result; wherein the target detection model comprises the following steps: replacing a 3 * 3 common convolution in the sixth network layer ELANBlock of the YOLOv7 by using a 3 * 3 depth separable convolution; replacing a 3 * 3 common convolution in the ELANBlock of the eighth network layer of the YOLOv7 by using a 5 * 5 depth separable convolution; using a 7 * 7 depth separable convolution to replace a 3 * 3 common convolution in the ELANBlock of the tenth network layer of the YOLOv7; an MIA-M module based on multi-scale efficient attention is used for replacing the 17th network layer ELANBlock and the 23rd network layer ELANBlock of the YOLOv7; an MIA-S module based on multi-scale efficient attention is used for replacing the ELANBlock of the 21st network layer of the YOLOv7; and replacing the ELANBlock of the 25th network layer of the YOLOv7 with a multi-scale efficient attention MIA-L module to obtain a target detection model.

The invention relates to a tropical cyclone strength intelligent forecasting method and device based on feature adaptive optimization. The method comprises the following steps: constructing an initial data set; the initial data set is divided into a model training sample and an independent test sample; selecting a pre-training model; inputting the model training sample into a pre-training model for feature adaptive optimization, and obtaining a high impact factor set influencing tropical cyclone strength prediction accuracy; performing model optimization training; and carrying out model robustness test and tropical cyclone strength intelligent forecasting. According to the method, the high impact factor set which is most critical to tropical cyclone strength forecasting can be rapidly extracted, and the tropical cyclone strength forecasting accuracy is remarkably improved.

The invention discloses an intelligent building decoration design method based on deep learning, and the method comprises the following steps: S1), employing a monocular camera to shoot pictures of a building at all angles, and obtaining an original picture; s2) extracting keywords, and making cue words; s3) building decoration data collection; s4) performing data screening and preprocessing; s5) customizing a loss function, and training the diffusion model to obtain a trained diffusion model; s6) inputting the original image into the trained diffusion model to generate a preliminary image; inputting the preliminary image and the cue word into a pre-trained conditional generative adversarial network to obtain a preliminary design drawing; and S7) inputting the preliminary design drawing into a pre-trained variational auto-encoder to obtain an architectural decoration design reference draw.According to the intelligent architectural decoration design method based on deep learning, through the design of the self-defined loss function and the dynamic weight, in combination with the deep learning model, the image is automatically processed, and the design reference drawing is generated, so that the working efficiency is improved.

The invention provides a humanoid robot control method and system, a storage medium and a program product, and belongs to the field of computer vision. The method comprises the following steps: preprocessing expert action data, and processing the expert action data into expert data equivalent to a skeleton architecture of a target robot; in a simulation environment, a humanoid structure is adopted to build a robot, joint parameters of the robot are configured, and the degree of freedom of each joint is controlled by an independent physical control module; constructing a strategy representation method of the robot, wherein the strategy representation method comprises a state space, an action space, a reward function and a multi-frame control mode; the robot is initialized; and the difference between the robot action and the expert action on each frame is minimized, the reward function is maximized, and the robot is driven to learn. The method can assist the learning process of the humanoid robot, so that the robot can be personified while completing tasks, and the training speed is improved.

The invention discloses a business processing method, a test system, a medium and a product, and relates to the technical field of data processing. Before actual service demand processing is carried out, the performance parameters of the calculation accelerators are processed through the preset topological structure, the weight parameters and the set communication strategies, the target topological structure corresponding to the benchmark test is obtained, and resources of the calculation accelerators are made full use of. And in the target model of the actual current business demand, training the target model according to the parallel training strategy corresponding to each calculation accelerator and the target sample number of the target model so as to determine model parameters in the operation process. A conventional model does not need to be adjusted after the whole training process is finished and then return to iteration, time consumption is reduced, and meanwhile, by determining the parallel training strategy of each calculation accelerator, the operation efficiency of the training process and the evaluation effect of the model training parameters corresponding to the user requirements are improved.

The invention discloses a vision and language fused multi-modal large model system, which comprises a vision encoder fused with DINOv2 and SigLIP and used for collecting low-level spatial attributes and semantic attributes; and the multi-layer perceptron projector is used for mapping the visual features to a language embedding space and a Mama backbone model network based on a state space model. Compared with a multi-modal large language model which depends on a Transform network as a basic model, the large model system disclosed by the invention has the advantage that indexes such as reasoning speed and accuracy are improved.

The invention discloses a hierarchical object modeling and rapid reconstruction method based on NeRF, and the method comprises the following steps: image preprocessing: initially obtaining an image video frame through a camera, carrying out the related processing of an input image, including semantic segmentation, depth estimation, normal estimation and internal and external parameter calibration, and generating a corresponding semantic graph, a depth graph, a normal graph and camera parameters; space partitioning: partitioning the scene according to the semantic graph information and defining a bounding box; adaptive sampling: performing adaptive sampling by combining the depth map, the normal map and the camera parameters, and determining a sampling point set; and performing hierarchical vectorization SDF reconstruction, performing hierarchical vectorization SDF reconstruction by using the sampling points to obtain hierarchical SDF representation describing the shape of the object, converting the hierarchical SDF into a triangular mesh model, and completing three-dimensional object modeling and rapid reconstruction. The invention provides a hierarchical object modeling and rapid reconstruction method based on NeRF, objects in a scene are rapidly reconstructed, and downstream editing tasks, interaction and other applications are supported.