26results about How to "Control complexity" patented technology

The invention relates to a method, a device and equipment for constructing an elastic network regression model of a battery retention rate. The method comprises the following steps: acquiring sample data of a sample battery and an initial elastic network regression model; inputting the sample data into the initial elastic network regression model, and outputting a prediction result; determining the target loss of the initial elastic network regression model according to the prediction result; the target loss comprises an error term and a regularization term; and training the initial elastic network regression model according to the target loss to obtain a trained elastic network regression model. According to the method, penalty for the regression coefficient is introduced into the regularization item, the complexity of the elastic network regression model is controlled, and the elastic network regression model is prevented from being over-fitted, so that the training efficiency of the initial elastic network regression model is improved, and due to the fact that the number of sample data needed for training the regression model is small, the training efficiency of the regression model is improved. And the complexity of training the initial elastic network regression model is reduced.

The invention discloses a CNN-XGBoost power load prediction method based on an attention mechanism, and the method comprises the steps: obtaining related historical load data and feature data, carrying out the data correlation analysis, employing the CNN to construct a feature extraction module based on the advantages of the CNN in feature extraction, and carrying out the calculation of the data correlation. Aiming at the problem that a traditional convolutional neural network cannot distinguish whether the extracted features are effective for prediction, Attention is introduced, different weights are given to the features, the features with different output contribution degrees are distinguished, the extracted features are emphasized consciously, the relation between the extracted features and labels is explored, and an XGBoost algorithm is combined to determine whether the extracted features are effective for prediction or not. According to the method, the advantages of the CNN algorithm in the aspect of feature extraction and the advantages of the XGBoost algorithm of high nonlinear load prediction precision and high prediction convergence speed can be integrated, and the accuracy and stability of short-term load prediction of power consumption of the whole network are improved.

The invention discloses a generalizable deep forgery detection method based on multi-classification guidance, and belongs to the technical field of image processing. According to the technical scheme, data enhancement is carried out on an image, and then the image is input into a binary classification network and a multi-classification network of a double-flow neural network; combining the features of the two branches through a fusion module; the fused features are input into a binary classifier to determine whether the input is authentic or counterfeited. The method has the beneficial effects that the generalizable deep counterfeiting detection method based on multi-classification guidance shows remarkable beneficial effects in the aspects of improving the model generalization ability, enhancing the feature expression ability, optimizing the loss function design, preventing over-fitting, improving the calculation efficiency, enhancing the label consistency and the like; the method is beneficial to suppressing the propagation of false information, maintaining the right to know and security of the public, enhancing the vigilance of people to the authenticity of information, laying a solid foundation for the applications of identity authentication, digital forensics and the like in multiple fields, and making an important contribution to the construction of a safer and more credible information society.

The invention relates to the technical field of climate event simulation, in particular to an optimization method of climate mode simulation composite extreme wind and rain event frequency based on deep learning, and the method comprises the following steps: obtaining observation data of rainfall and wind speed of a target area and simulation data of a plurality of climate modes; calculating rainfall and wind speed thresholds by adopting a percentile threshold method, and determining the occurrence frequency of a composite extreme wind and rain event; evaluating a climate mode simulation result by using the observation data, selecting the frequency as a characteristic value of the deep learning model, and using the observation frequency as a true value; constructing a multi-layer perceptron neural network, and optimizing model training by fusing a mean square error, a standard deviation ratio and a weighted loss function constrained by a minimum value; and inputting the simulation result of the climate mode into the trained model, and generating the correction result of the composite extreme wind and rain event frequency, thereby improving the simulation precision of the composite extreme wind and rain event frequency, effectively reducing the deviation, and providing support for disaster estimation and decision-making under the climate change background.

The invention relates to the technical field of network security, in particular to a depth-first search attack graph generation method and system based on node asset weight, which introduces a deep learning priority algorithm of dynamic adjustment of node weight and intelligent node weight calculation, reduces the scale and complexity of an attack graph as much as possible while ensuring the accuracy of the attack graph, and improves the security of the attack graph. The attack graph generation efficiency is improved; comprising the following steps: calculating a comprehensive importance score of each node by comprehensively evaluating characteristics of each node in a network, and obtaining an initial weight value according to the comprehensive importance score; the method comprises the following steps: dynamically adjusting an initial weight value of each node in a network through real-time flow monitoring and regular polling of a network state to obtain a corrected weight value; traversing the nodes in a descending order according to the corrected weight values, and recording accessed nodes and paths; the current node has no adjacent unaccessed node or has reached the preset search depth limit; and stopping after all possible paths are searched, and generating an attack graph.

The invention relates to the field of artificial intelligence, in particular to a mineral resource recognition model training method based on data collected by an unmanned aerial vehicle. The training method comprises the steps of collecting real geological data, training a data generation model, generating simulated geological data by adopting the trained data generation model, and training a feature extraction model and a classification model by taking the simulated geological data and the real geological data as training samples together. An existing mineral resource identification model training method based on data acquired by an unmanned aerial vehicle has the problems of low model precision and low calculation speed of a mineral resource identification model. According to the mineral resource identification model training method based on the data acquired by the unmanned aerial vehicle provided by the invention, the model precision of the mineral resource identification model is relatively high, and the calculation speed is relatively high.

This invention discloses a method, system, and storage medium for multivariate time-series feature extraction and grade prediction in flotation processes. The method includes the following steps: Step S1: Raw data input and encoding; encoding and structured input of time-series data composed of various process variables; Step S2: Extracting dynamic features; Step S3: Condition-guided encoding modeling; Step S4: Enhancing the saliency of key variables and important time segments; employing a multi-head attention mechanism to match key-value pairs generated from target-guided query vectors and multi-scale features; Step S5: Outputting a prediction module; performing feature fusion and nonlinear mapping on the attention mechanism output to output the predicted concentrate grade and recovery rate for future times. The system and storage medium are both based on the above method. This invention has advantages such as higher intelligence, better controllability, and improved prediction accuracy and model adaptability for key indicators in the flotation process.

The invention provides an oil-water separation effect prediction method and device based on a Lonion model. The method comprises the following steps: defining an input independent variable vector f (x) = [x1, x2,..., x8] T; performing feature engineering extension on the input independent variable vector to generate an extended feature vector f (z); the extended feature vector f (z) comprises an original feature vector, a polynomial feature vector and a physical interaction feature vector; inputting the extended feature vector f (z) into a pre-trained prediction model to obtain a predicted value, output by the pre-trained prediction model, of the oil content of the effluent of the air flotation synergistic device; inputting the predicted value into an integrated prediction framework to obtain a final predicted value and uncertainty estimation; displaying the final predicted value and the uncertainty estimation to a user through a display interface; according to the technical scheme, the accuracy, robustness and physical consistency of oil content prediction of the effluent can be improved.

The embodiment of the application provides a hydraulic fracturing fracture width profile experiment device and method, the method can reconstruct the continuous distribution of the fracture width profile along the height direction of the fracture by combining a plurality of optical fiber discrete measuring points and vertical interpolation, ensures the spatial resolution, controls the complexity and layout cost of the experiment system, and guarantees the physical rationality of the result by non-negative constraint, and meets the monitoring requirement of the complex expansion scene of the multiple cluster fractures.

A winding structure of a motor stator comprises an iron core, a plurality of slots and a plurality of hairpin-shaped leads. The plurality of slots are located on the iron core and allow the plurality of hairpin-shaped wires to be placed in from the insertion side and to penetrate out from the extension side. The grooves surround the rotor accommodating space in the circumferential direction. The first-phase winding is configured from a first-phase wire inlet slot position and comprises a plurality of first-phase return connecting wires. The second phase winding is configured from the wire inlet slot position of the second phase and comprises a plurality of second phase return connecting wires. The third-phase winding is configured from the wire inlet slot position of the third phase and comprises a plurality of third-phase return connecting wires. And the wire inlet slot position of the first phase, the wire inlet slot position of the second phase and the wire inlet slot position of the third phase are respectively positioned in three different polar regions.

The application discloses a design method of gas passage shunting of a gas shunting device, and belongs to the technical field of gas shunting devices. The gas shunting device comprises a plug gas distribution seat, at least one middle layer shunting plate and a bottom layer shunting plate which are sequentially connected and can form multiple shunting layers; at least two plug seats are arranged on the plug gas distribution seat and are in communication with the shunting layers; connectors are arranged on the middle layer shunting plate and the bottom layer shunting plate and extend to the outside of the bottom layer shunting plate; the design method is that the number of connectors is set according to the number of interfaces on one plug seat, and the number of shunting layers is determined according to the total number of interfaces on all plug seats. Through the shunting layout design of multiple layers, the number of shunting layers can be increased to reduce the complexity of gas passage layout as the number of device plug increases, the problem of large size of the device can be effectively controlled, the problems of the prior art, such as the complex single-layer gas passage layout, are overcome, the assembly process is simplified, the connection convenience is improved, the assembly efficiency is improved, and the assembly cost is saved.

The invention provides a distributed photovoltaic low-voltage distribution network voltage estimation method and system, and belongs to the field of distribution network voltage assessment, and the method comprises the steps: 1, inputting the voltage data, time information and space information measured by a measurement device into a learning supervision regression model, the learning supervised regression model maps voltage data measured by the measuring device to adjacent low-voltage power distribution network nodes through a random forest algorithm to obtain voltage data of the adjacent low-voltage power distribution network nodes; step 2, training the learning supervised regression model, and when a regression index is minimum, stopping training to obtain a trained learning supervised regression model; 3, performing voltage estimation on unmonitored nodes in the low-voltage power distribution network by using the trained learning supervision regression model; and based on the measurement data of the limited node, mapping the measurement data of the limited node to the adjacent node without the measurement device, and further carrying out voltage estimation on the unmonitored node in the low-voltage power distribution network containing the distributed photovoltaic.

The invention provides a self-adaptive quantization LDPC decoding method and device, and relates to the technical field of optical communication, and the method comprises the steps: obtaining original information data, carrying out the preprocessing of the original information data, and obtaining standard information data; data adjustment is carried out on the standard information data, adjustment information data is obtained, and frame data information in the adjustment information data is in Gaussian distribution; the adjustment information data is input to an LDPC decoding initial quantization unit, decoding initialization processing is carried out on the adjustment information data, quantization information data is obtained, and the LDPC decoding initial quantization unit iteratively generates variable node information according to the adjustment information data; inputting the quantized information data into a self-adaptive quantized decoding processing unit to perform main decoding processing on the quantized information data to obtain main decoded data, and performing self-adaptive quantized processing on variable node information when variable nodes of the self-adaptive quantized decoding processing unit are updated to obtain updated node information; and integrating the main decoding data and the updated node information to enable the adaptive quantization decoding processing unit to output a target decoding result. The method is beneficial for improving the sensitivity of LDPC decoding.

The invention discloses a user question-oriented rapid feedback method and related equipment, belongs to the technical field of artificial intelligence, and aims at effectively improving the shopping experience by utilizing large-scale question and answer data generated by users to assist the learning of questions and related comment tasks. The rapid feedback algorithm based on the attention model has a remarkable effect in solving the problem positioning in the shopping process, and is superior to similar algorithms. According to the method, the user can be helped to quickly find the answer, the shopping efficiency is improved, and the shopping experience of the user can be further improved. In addition, the purchase demand contained in the questions of the user also provides beneficial information for commodity recommendation, so that the comprehensiveness and individuation degree of shopping experience can be further improved by performing commodity recommendation according to the questions of the user.

The invention discloses a surface acoustic wave signal processing method and system based on an NQAM demodulation algorithm, and the method comprises the steps: obtaining a surface acoustic wave signal, carrying out the interference removal of the surface acoustic wave signal based on an IMS adaptive filtering strategy, and obtaining a target surface acoustic wave signal; and demodulating the target surface acoustic wave signal according to a preset NQAM demodulation algorithm to obtain phase information and amplitude information corresponding to the target surface acoustic wave signal. The error is reduced, the model complexity is controlled at the same time, and stable filtering performance can be ensured to be provided under the limiting condition.

The invention relates to the technical field of big data processing, in particular to a data model modeling method and system and a storage medium. The method comprises the following steps of obtaining an original big data set and performing standardization processing to obtain a standard big data set; performing feature automatic generation processing and regularization recursion feature screening processing on the standard big data set to obtain large data volume related optimization features; performing preliminary modeling processing and model performance optimal screening processing on the large data volume related optimization features through a decision tree, a support vector machine and a deep neural network basic algorithm to obtain a big data performance optimal model; performing hyper-parameter combination optimization selection and model optimization adjustment on the big data performance optimal model to obtain a big data optimization adjustment model; and performing feature update real-time monitoring and adaptive update adjustment optimization on the large data volume related optimization features to generate a large data adaptive adjustment model. The generalization ability of the data model can be improved.

The invention discloses a highland barley field carbon-ammonia source sink imbalance risk early warning system, and belongs to the field of gas intelligent early warning, an optical system module comprises a wide temperature range wavelength locking double DFB laser set, a first electric control gradient refractive index liquid crystal lens, a second electric control gradient refractive index liquid crystal lens, an off-axis integral cavity, a convergent lens and a detector, and is used for achieving optical detection of CO2, CH4 and NH3 in a highland barley field; the self-adaptive power supply module provides a stable power supply for the system by adopting a magnetic flux-temperature double feedback and time division multiplexing mechanism aiming at the low-voltage and low-temperature extreme environment of the Tibet Plateau. The temperature control module provides high-precision temperature stability control for the laser based on a bimodal dynamic gain-thermal disturbance prediction composite sliding mode control algorithm; and the main control module is used for generating a laser driving signal and performing early warning on the imbalance risk of the highland barley field by using the rough set-deep belief network hybrid early warning module. The problems of poor system stability, low detection precision and slow early warning response in a plateau extreme environment can be solved.

This invention relates to the field of bearing fault identification. To address the problems of existing technologies where anomaly suppression relies on fixed thresholds and multi-domain feature selection lacks operational stability and redundancy constraints, this invention provides a bearing fault identification method and system based on anomaly suppression and multi-domain features. The bearing fault identification method based on anomaly suppression and multi-domain features includes: cleaning the sample set; extracting candidate features from the cleaned sample set; constructing a multi-objective feature scoring function; forming a multi-domain core feature subset from candidate features with scores not lower than the feature selection threshold; training a fault diagnosis model; preprocessing and feature extraction of the bearing vibration signal under the diagnostic operating condition to obtain multi-domain core features; and then processing these features through the trained fault diagnosis model to obtain the corresponding fault type. This method can still stably output bearing fault diagnosis results with high accuracy and low false alarm rate even under noise and fluctuating operating conditions.

The invention relates to the field of deep learning, in particular to a method and a device for generating individual prejudice samples in a deep neural network. Comprising the following steps: selecting clustering centers in different areas of a data set, clustering original data in the data set according to the clustering centers, and taking clustered samples as clustering results; according to global protection attributes of the samples, judging whether the samples in the clustering result are prejudice samples or not, performing global disturbance on the samples which are judged to be negative, and taking the samples which are judged to be positive and the samples subjected to global disturbance as global prejudice samples; and selecting local protection attributes under gradient guidance, performing local disturbance on the local protection attributes in the global prejudice samples to obtain local prejudice samples, and taking the global prejudice samples and the local prejudice samples as generated individual prejudice samples. According to the method, the problem of overfitting possibly caused by accelerating the iteration process by introducing momentum optimization is solved, the complexity of the generation process is controlled, and the generalization of the method is improved.

The application provides a data processing method of a laser radar, an electronic device and a storage medium. The method is applied to the technical field of laser radars. The laser radar comprises a transmitting unit and a receiving unit. One transmitting unit corresponds to N receiving units, and N is a natural number greater than 1. The method comprises the following steps: selecting echo signals of M receiving units, wherein M is less than or equal to N; and outputting detection information corresponding to a target object according to the echo signals of the M receiving units. The method is beneficial to reducing the process complexity of the laser radar and improving the detection accuracy and flexibility of the laser radar.

The embodiment of the specification discloses a training method and device of a time series prediction model, and a computer system. The method comprises the following steps: first, performing a target operation on a first input by using a differential attention module in the time series prediction model, wherein the first input is determined according to a first number of index sequences corresponding to a first entity, and each index sequence comprises a historical sub-sequence and a target sub-sequence to be predicted. The target operation comprises the following steps: first, processing a query matrix corresponding to the first input by using a learnable first prototype key matrix and a second prototype key matrix respectively to obtain first and second attention scores; then, determining a comprehensive attention score, which is positively correlated with the first attention score and negatively correlated with the second attention score; next, processing a value matrix corresponding to the first input by using the comprehensive attention score to obtain a first output corresponding to the first entity; then, determining a prediction sequence corresponding to the target sub-sequence based on the first output; and finally, training the time series prediction model according to the prediction sequence.

The invention discloses a thermal power unit NO X Emission modeling method and system, the method comprising: based on the influence of NO X The basic parameters of emissions are obtained by obtaining the historical operation data of the thermal power unit and establishing the training set D0; the density clustering algorithm and the sparse greedy matrix are used to simplify the training set D0 to obtain the sample representative set D2; a semi-parametric support vector regression machine model is constructed based on the sample representative set D2, and the weights of the semi-parametric support vector regression machine model are solved based on the iterative reweighted least squares method; based on the training set D0, the Apache Spark parallel computing framework is used for parallel design to obtain the final weights of the semi-parametric support vector regression machine model; based on the final weights, the semi-parametric support vector regression machine model is parallelized using Apache Spark to obtain the thermal power unit NO X Emission model. When processing large-scale thermal power plant data, the present invention can shorten the modeling time and meet industrial requirements.

The application discloses a power grid similar operation section sample set construction method and system, and belongs to the technical field of power system dispatching control. The method comprises the following steps: defining initial electrical characteristics of a power grid operation section; building a feature extraction network comprising a graph neural network, a fusion layer and a time sequence encoding layer; adopting a time sequence comparison learning and node mask reconstruction double-task joint training strategy to realize space-time feature decoupling; preprocessing historical sections and building a feature library; matching the first K similar sections through similarity retrieval and building a labeled sample set. The system comprises corresponding function modules. The application solves the problems of high feature dimension, poor time sequence consistency and dependence on artificial experience in the prior art by deeply fusing graph structure learning and time sequence modeling, realizes automatic and high-precision construction of a power grid historical operation section sample set, and can provide reliable data support for power grid safety analysis and dispatching decision.

The invention provides an underwater small target detection system capable of reserving fine-grained information, and the system comprises a fine-grained down-sampling module which carries out the down-sampling through the linear combination between channels and employing an SPD-Conv layer, and obtains a series of sub-feature maps; the C2fSCConv module is introduced into a backbone network of YOLOv8, that is, space channel reconstruction convolution SCConv is fused into a Bottleneck bottleneck module in the C2fSCConv module to replace a second CBS module in the Bottleneck bottleneck module, and effective information of a sub-feature graph is extracted through stacking of the bottleneck module; and the deconvolution channel feature fusion module is arranged between the YOLOv8 backbone and the neck and is used for carrying out reinforced fusion on the sub-feature maps based on the effective information of the sub-feature maps. According to the method, challenges in underwater small target detection can be processed more effectively, the detection precision is improved, the feature redundancy and the calculation amount are reduced, and therefore the requirements of practical application are met.

The invention provides a visual language model zero sample classification method based on Lasso regularization dynamic integration. The visual language model zero sample classification method specifically comprises the following steps: S1, constructing a model pool comprising a plurality of pre-trained visual encoders; step S2, in a training stage, adaptively screening out a model subset which contributes significantly to a current task from the model pool by using the sparse characteristic of Lasso regularization, and learning the weight of the model subset; s3, introducing a dynamic regularization coefficient lambda adjustment mechanism based on verification loss to balance the complexity and generalization ability of the model; and S4, in a reasoning stage, performing weighted fusion on the prediction result of the selected model by using the sparse weight obtained by training to obtain a final classification result. According to the method, the accuracy and robustness of the visual language model in the zero sample classification task can be effectively improved.

The invention discloses a road disease detection method and device based on a YOLO-DMD model, and the method comprises the steps: constructing the YOLO-DMD model through introducing a C2f-Deform module, an MSDA mechanism and a Dyhead dynamic detection head based on a YOLOv8 algorithm model; acquiring a road streetscape image through streetscape shooting equipment or a vehicle-mounted automobile data recorder; inputting the image into a YOLO-DMD model for size adjustment and normalization processing to obtain a standardized image; performing feature extraction on the standardized image through a backbone network embedded with a C2f-Deform module to obtain an initial disease feature map; based on the initial disease feature map, feature fusion is carried out through an MSDA multi-scale deformable attention mechanism in a neck network, and a fused feature map is generated; and based on the fused feature map, carrying out disease classification and positioning through a Dyhead dynamic detection head, and outputting disease category, confidence and bounding box information to complete road disease detection. According to the method, the problems of insufficient extraction, poor multi-scale fusion suitability, poor detection precision and robustness caused by classification positioning coupling and the like in the prior art are solved.