55 results about "Normalization algorithm" patented technology

The invention discloses a casting surface defect automatic detection method, which comprises the following steps of: synchronously acquiring multi-moment surface images and morphology data of a casting through multiple channels, automatically partitioning and extracting texture attributes, and realizing accurate space-time indexing; a denoising and brightness normalization algorithm is adopted to improve the consistency of basic data; in combination with a deep learning segmentation model and feature analysis, time sequence space defect matching, affine mapping, pseudo label generation and self-supervision consistency training are completed step by step, and fine-grained optimization is performed on dynamic change of a segmentation boundary. Therefore, the accuracy of defect detection is improved, and high-quality data support is provided for production process improvement and defect traceability.

A decision basis analysis method, apparatus and device for a high-impedance grounding fault identification model. The method comprises: on the basis of unlabeled zero-sequence current data, training an initial fault identification model to obtain a preset high-impedance grounding fault identification model, wherein the preset high-impedance grounding fault identification model comprises an encoder and a decoder; analyzing the degree of importance of each encoded feature, which is output by the encoder, to an identification result of the model, and calculating a corresponding global Shapley value; using an instance normalization algorithm to perform normalization processing on encoded vectors, and then inputting the normalized encoded vectors into the decoder for decoding analysis, so as to obtain a decoded waveform; and performing spectrum analysis by means of comparing the decoded waveform with an original fault waveform, and analyzing a decision basis of the model on the basis of the global Shapley values, so as to obtain an identification basis of the model. The present application can solve the technical problems of existing grounding fault identification models requiring a large amount of labelled data and the models lacking interpretability, thus leading to a lack of specificity and reliability in model training.

The invention discloses a radiation source identification confrontation sample construction method based on momentum acceleration, and belongs to the technical field of wireless communication and intelligent identification. The invention aims to solve the problems of low attack success rate and poor generalization of an adversarial sample generated by an existing method. Comprising the following steps: performing data enhancement on an original radiation source signal to obtain a noise-added sample signal; performing transformation by using a wavelet transformation method to obtain a two-dimensional time-frequency graph; a classification result is obtained through a deep neural network identification model; setting an initial confrontation sample and an initial gradient disturbance direction, and carrying out iterative updating on the confrontation sample; in each round of iteration, calculating a loss function based on a classification result, calculating the gradient of the loss function to the current adversarial sample, and updating the gradient disturbance direction based on a Nesterov momentum acceleration gradient normalization algorithm to obtain an adversarial sample in the next round of iteration; and performing wavelet inverse transformation on the final adversarial sample to obtain a time domain signal of the final adversarial sample. The attack success rate and generalization performance of the generated adversarial sample are improved.

The invention discloses a short video content accurate recommendation system based on artificial intelligence image recognition, particularly relates to the technical field of short video personalized recommendation, and is used for solving the problem of matching of user interest and visual bearing capacity. The method comprises the following steps: extracting short video key frame spatial-temporal characteristics through a multi-scale convolutional neural network, constructing a cognitive load threshold curve in combination with a user micro gesture sequence, representing the tolerance range of a user to visual complexity in different time periods, and generating a visual complexity vector; calculating a visual load matching index by using an adaptive dynamic warping algorithm, and adjusting a candidate video sequence through a load penalty factor to generate an initial recommendation probability; modeling a user dynamic interest vector based on a gated loop unit network in combination with an attention mechanism; and finally, user interests and video semantics are fused through multi-dimensional features, a final recommendation list is generated through a multi-objective optimization algorithm under the constraint of visual load, and personalized recommendation of interest matching degree maximization and visual comfort optimization is realized.

Disclosed in the present invention is a method for rapidly testing the authenticity of degradable plastic on the basis of near-infrared technology, comprising the following steps: step S1, using packaging bag samples of known materials to collect spectra, and using a kennard-Stone method to perform division to obtain a training set and a prediction set; step S2, by means of the spectra in the training set in combination with analysis based on a normalization algorithm, using a support vector machine algorithm to perform modeling, and using sample data in the training set to train a support vector machine model; and step S3, using the prediction set and inputting same into the trained support vector machine model to obtain a prediction result of the model, and performing analysis and adjustment to obtain an optimized support vector machine model. The method of the present invention features a simple testing process and high identification accuracy, and enables rapid determination of whether a material is a degradable material and identification of the specific type of the degradable material; additionally, the method saves a large amount of manpower, material resources and financial resources, achieving high economic benefits.

The invention discloses a hydraulic pump fault detection method and device based on information fusion and a medium, and belongs to the technical field of hydraulic device fault detection. The method comprises the steps of collecting a vibration signal and a pressure signal of the hydraulic pump; processing the vibration signal and the pressure signal based on a mathematical morphology filtering algorithm to generate a de-noised vibration signal and a de-noised pressure signal; processing the signals through a convolutional neural network to generate a vibration feature map and a pressure feature map; generating a weighted vibration feature map and a weighted pressure feature map based on an attention mechanism; integrating the weighted vibration feature map and the weighted pressure feature map, and generating a comprehensive fault feature vector; generating normalized comprehensive features based on a normalization algorithm; and processing the normalized comprehensive features through a full connection layer and a Softmax classification layer of the convolutional neural network, and outputting a fault category diagnosis result. By means of the method, the technical effect of improving the accuracy and reliability of fault detection of the hydraulic pump is achieved.

The invention provides an electric leakage fault positioning and monitoring method and system for a distributed photovoltaic access transformer area, which is optimized from the aspects of data preprocessing, electric leakage monitoring and electric leakage fault positioning and on-line monitoring of a low-voltage distribution transformer area, and comprises the following steps: firstly, optimizing a normalization algorithm and providing a denoising algorithm of a mixed sign function; a surrounding and hunting strategy of a whale optimization algorithm is introduced into a particle swarm algorithm, a Levy flight strategy is added into the particle swarm algorithm, and the particle swarm algorithm with double strategy improvement is provided. Then evaluating the random forest model, realizing electric leakage fault positioning verification by adopting a bisection method according to an evaluation result, acquiring new data in real time, importing the new data into the random forest model to acquire an electric leakage fault point corresponding to fault electric leakage current, and finally timely and accurately finding out an electric leakage position and an electric leakage circuit. And accurate positioning of the leakage fault of the low-voltage distribution area is ensured.

The invention discloses an electricity consumption information acquisition terminal transaction semantic monitoring and causal delimiting method, which comprises the following steps: acquiring an application protocol data unit by utilizing an eBPF probe, generating a cross-layer unique transaction fingerprint based on extracted sessions, objects, businesses and calling identifiers, and generating a transaction event record by utilizing a pairing state machine; constructing a transaction observation window containing front and back extensions by taking a transaction observation timestamp as a reference, and screening and aggregating kernel events associated with the transaction fingerprint to obtain a transaction evidence set; and constructing a directed evidence graph based on the set, and calculating the confidence coefficient of each root cause category by using a scoring rule and an index normalization algorithm. According to the invention, non-intrusive deep observation of power acquisition business affairs is realized, and the automation level and fault diagnosis precision of power terminal operation and maintenance are improved.

The invention discloses a three-level inverter open-circuit fault diagnosis method based on fault feature vector prediction, and the method comprises the steps: reconstructing a theoretical stator current through reference current based on a motor steady-state model, and introducing an amplitude per-unit algorithm to calculate a current residual error, thereby recognizing the fault, and locking a fault phase; collecting a system state at the moment of fault triggering, deducing a modulation switch sequence in a future detection period on line by using a reference voltage vector time sequence prediction model, and generating a dynamic reference value of theoretical action times of each switch state; performing real-time statistics on the actually-measured cutoff times of the fault phase in different switch states by using an edge triggering mechanism, and constructing an actually-measured feature vector; and determining a fault device by calculating the minimum distance between the actual measurement vector and each fault modal theoretical feature. The method can effectively eliminate the influence of the rotating speed and load change of the motor on the diagnosis threshold, solves the problem that the fault features of the inner tube and the clamping diode are overlapped and are difficult to distinguish, and improves the robustness and accuracy of fault diagnosis.

The application discloses a kind of internal resistance big data driven safe operation method and system, method includes: the historical and real-time internal resistance data of target equipment is collected, and the internal resistance time series data set after normalization is generated by dynamic time normalization algorithm;The internal resistance time series data set is carried out multi-scale feature extraction and high-dimensional space mapping, and the internal resistance abnormal mode cluster implied in data distribution is identified using adaptive density clustering algorithm;Based on internal resistance abnormal mode cluster, multi-modal fault correlation analysis is carried out, and a fault prediction atlas is generated;According to the fault prediction atlas, a set of differentiated operation and maintenance strategies is generated, and the operation and maintenance management platform is driven to execute corresponding operation and maintenance instructions. Using the embodiment of the application, the timeliness and accuracy of fault early warning can be improved, the operation and maintenance cost is reduced, and the safe and stable operation of equipment is guaranteed.

The invention discloses an urban environment dynamic monitoring system and method based on deep learning, and the method comprises the steps: collecting multi-modal image data, eliminating the modal difference through a self-adaptive normalization algorithm, generating multi-modal features, introducing an attention mechanism to carry out the weighted fusion of the multi-modal features, and obtaining the fusion feature data; an FPN multi-scale feature pyramid network is constructed to carry out pollution source identification on fused feature data, a DSAM dynamic space attention module is embedded to adaptively adjust feature channel weights, and pollution source categories and position coordinates are output; the category and position coordinates of the pollution source are predicted based on an MTL multi-task learning framework, a space-time diagram convolutional network is introduced to evaluate the environmental quality, and an environmental pollution thermodynamic diagram is output; and outputting an urban environment governance strategy through a dynamic decision tree model according to the environmental pollution thermodynamic diagram. The positioning error is reduced, and the accurate traceability requirement is met.

The present invention provides a track similarity assessment method, device, electronic device and storage medium, which can obtain various index data that characterize the degree of deviation between the preset track and the actual track. Then determine the number of qualified index data in each index data. When the number of qualified index data is not less than one, the preset track and the actual track are processed respectively based on the preset similarity algorithm and the preset distance algorithm to obtain the various deviation distances between the preset track and the actual track. Each deviation distance is processed based on the preset normalization algorithm to obtain the normalized result of each deviation distance. Each of the normalized results is processed based on the preset scoring algorithm to obtain the similarity score of the preset track and the actual track. When the number of qualified index data is not equal to the number of index data, the score of each index data needs to be subtracted. The present invention realizes the quantitative evaluation of the similarity between the planned track and the actual track of flying equipment such as drones.

The invention discloses a laying hen epidemic disease detection method based on DistilBERT-GATv2 and BiLSTM-TCN, and the method comprises the following steps: extracting related data from unstructured text data, and constructing a high-quality structured text database; a DistilBERT model is adopted to carry out fine tuning training on the labeled corpus, accurate recognition of a target entity is achieved, and standardized entity categories and texts are output; realizing extraction of a semantic relationship between entities, and constructing disease triple data; a cross-sentence anaphora resolution mechanism and an entity normalization algorithm are introduced, semantic references are unified, and consistency and uniqueness of node semantics in the knowledge graph are ensured; storing the constructed triple data in a graph database to complete the construction of the domain exclusive knowledge graph; visual presentation of entity nodes, relation edges and query paths is achieved by configuring a visual component. According to the method, the epidemic disease type of the laying hen can be efficiently detected and diagnosed, the accuracy of an intelligent monitoring system is improved, manual intervention is reduced, and the automation level of laying hen disease prevention and control is improved.

This invention provides an automated evaluation system for vibration monitoring signals of aerospace products, comprising: a raw signal parsing module, a data preprocessing module, a signal parameter configuration module, a signal interpretation and evaluation module, and an evaluation conclusion output module. The raw signal parsing module parses unreadable raw data packets into a program-readable data format; the data preprocessing module eliminates signal start-time errors and performs filtering; the signal parameter configuration module configures parameters for single and related signals, forming a parameter configuration library; the signal interpretation and evaluation module evaluates the waveform similarity of single signals based on a cross-correlation normalization algorithm and interprets the time sequence relationship of related signals; the evaluation conclusion output module displays the evaluation results on a software interface. This invention achieves automated evaluation of vibration monitoring signals of aerospace products, replacing manual interpretation, improving evaluation efficiency and reliability, and possessing good versatility and scalability.

The invention discloses a method for constructing an intelligent colorectal cancer prediction model based on mass spectrum serum proteomics, which comprises the following steps of: screening out candidate micropeptides with diagnostic potential by combining high-precision mass spectrum quantification with AI function prediction, and further constructing the model by adopting an ensemble learning algorithm. The method is rigorous in process, and by introducing an advanced normalization algorithm, integrating a feature selection strategy and targeted sample imbalance processing, the biomarker screening robustness and the prediction model accuracy are remarkably improved. Meanwhile, through model interpretive analysis, theoretical support is provided for clinical application of the marker.

The present invention belongs to the technical field of intelligent power data processing, and provides a normalization algorithm-based concentrator temperature compensation method and device, and a storage medium. A concentrator device is placed in a high-temperature or low-temperature environment for a certain time and energized for a certain time, then values of a phase voltage correction register of the concentrator device are read and subjected to normalization processing, and a temperature compensation algorithm is designed to obtain temperature compensation for a phase voltage channel. The present invention is applicable to current channel compensation correction, active phase compensation correction and reactive phase compensation correction. Different from previous solutions using the same temperature compensation parameter, the present invention can perform point-to-point precise compensation on any phase of three-phase voltages, currents, active power and reactive power. After the temperature compensation, errors of alternating-current sampling analog quantities at different temperatures can be controlled to be 0.1% or less. In addition, temperature compensation calculation is no longer required during device operation, thereby saving calculation resources; and there is no need to configure hardware temperature compensation circuits, thus reducing hardware costs.

% Disclosed is a non-probabilistic model-based intelligent reliability assessment method for corrosion damage to a ship hull structure. The method includes: collecting corrosion data of the ship hull structure, and performing interval processing on the data through an embedded data processing system to obtain a mean and deviation of a residual corrosion thickness; determining a resistance of the ship hull structure through hull structure resistance calculation software; measuring external load data such as wind and wave loads, sailing speed, and cargo's center of gravity, and determining external loads based on geometric characteristics of the ship hull structure; establishing a failure function and simulating a failure mode of the ship hull structure in a failure analysis module; and solving a non-probabilistic reliability index of the ship hull structure through an interval variable normalization algorithm, and displaying an assessment result and giving an early warning through a ship safety management system.

Provided herein are compositions, systems, and methods for ranking cancer drugs for treating a subject's cancer cells, where a plurality of gene signatures (each with a plurality of gene signature genes) with associated cancer drugs are processed with raw mRNA expression levels for genes in the sample. The processing (e.g., by computer) can comprise: i) applying a normalization algorithm to generate normalized mRNA expression values for signature genes, ii) applying a median finding algorithm to the normalized mRNA expression values in each of the plurality of drug gene signatures to generate a plurality of median values, and iii) applying a ranking algorithm such that the median values are ranked from highest value to lowest value (or vice versa), with the highest value being associated with the most effective cancer drug, or most effective combination of two cancer drugs.

The application discloses a kind of multilayer automatic picking methods based on sparse two-stage dynamic normalization algorithm, including the following steps in turn: S1.similarity and backtracking path of two sequences are obtained;S2.seismic trace clustering;S3.multilayer automatic picking.The application provides multilayer automatic picking method, utilizes sparse two-stage dynamic normalization algorithm to carry out layer position tracking work, on the basis of traditional dynamic normalization algorithm, join the similarity of data itself and join constraint to reduce search range and second-order distance formula, solve the problem of low efficiency and precision in the application of classic dynamic time normalization algorithm in layer position tracking.The application is suitable for multilayer automatic picking based on sparse two-stage dynamic normalization algorithm, can obtain accurate layer position tracking work in complex geological environment, improve the efficiency and accuracy of layer picking.

The application relates to the technical field of cardiovascular disease diagnosis and treatment and artificial intelligence, and discloses a cardiovascular disease diagnosis and treatment scheme optimization system based on architecture, which comprises an original data preprocessing module, which is used for collecting and standardizing time series monitoring data and static data of medical history texts, adopts an improved and normalized algorithm, utilizes model structured text data, and outputs standardized patient feature data; and an improved logic analysis module, which is used for receiving the standardized patient feature data, optimizing the architecture by introducing a sparse attention mechanism, and performing time series correlation analysis, pathological feature mapping and individual difference modeling. The improved logic analysis module is used for introducing the sparse attention mechanism, deeply mining time series correlation and implicit pathological logic coupling relationships in complex time series monitoring data, generating a high-dimensional patient state feature vector, and solving the problem that traditional methods are shallow in analysis and cannot accurately identify individualized pathological states.

The application discloses a breast cancer subtype classification method and system based on a graph convolutional neural network, which converts breast cancer gene expression data into a graphical representation and captures the correlation between genes using a graph convolution module. The data is preprocessed, including removing duplicate samples and samples without subtype labels, and filling in missing values. A graph representation dataset of breast cancer gene expression is constructed, combined with biological prior knowledge. The local features of the nodes in the graph are captured using the graph convolution method, and the data is normalized using the batch normalization algorithm. The self-attention pooling mechanism is introduced to learn the contribution of the input data to the output data, and the key features are extracted and hierarchical pooled. The local features and hierarchical features are spliced into the classification model to obtain the classification result of the breast cancer subtype. The application can effectively capture the correlation between genes and improve the accuracy of breast cancer subtype classification, and has potential biomedical application value.

The present invention discloses a method for improving the Kettle conversion efficiency based on an optimized k-means algorithm. The method preprocesses the execution time of Kettle conversion steps, uses the maximum-minimum normalization algorithm to standardize the data, and obtains a set of standardized data; groups are generated through the optimized k-means algorithm, the number of groups is generated, and the centroid is selected in a non-random manner. The present invention dynamically records the time of Kettle conversion steps, uses an improved k-means algorithm to group the steps, automatically manages parameters and expands threads, improves the conversion efficiency of Kettle, and reduces the time for manual parameter adjustment. Compared with not using this method, the conversion efficiency is increased by 20%.

The present application relates to big data processing and traffic layering control technical field, especially in a kind of global member whole life cycle operation and traffic layering control method and system, the method, including building global data fusion layer, collection multi-source heterogeneous member data and carry out standardization processing, and generate global unique member identification by identity normalization algorithm;Determine life cycle state and calculate residual life cycle value, obtain member value evaluation result;According to member value evaluation result calculation optimal traffic distribution scheme, generate the operation strategy of adaptation different member using optimal traffic distribution result, implement execution to operation strategy, realize data privacy protection based on differential privacy and federated learning algorithm, the present application realizes the dynamic optimal allocation of operating resources under the multiple constraint conditions of budget, frequency and system load, solves the resource waste and mismatch problem caused by artificial configuration fixed proportion.

The invention discloses a load data normalization algorithm based on a day-ahead load, and relates to the technical field of electric load data normalization processing. According to a smart city energy optimization management method, meteorological data, historical load data and special event load data are collected, influence factors of an electric load are separated, and the load data are subjected to normalization processing; a targeted adjustment factor data set is formed, and the data accuracy and comprehensiveness are improved. The factor data set is integrated and adjusted into a unified sequence, the calculation complexity is reduced, the factor relationship is embodied, and the learning ability and expandability of the model to the complex relationship are enhanced. An adjustment factor rule is mastered through a training model, accurate prediction of load change is realized, and a parameter adjustment strategy is formulated based on a prediction result, so that the model dynamically adapts to change. The parameter adjustment strategy optimizes the normalized algorithm parameters, enhances the adaptability and robustness, and improves the overall accuracy and stability of the prediction result.

The present invention relates to the field of measurement technology, and in particular to an integrated sensing module for a dynamic weighing system, including a physical domain, an environmental domain and a virtual domain, which realizes dynamic adjustment of parameters through cross-domain interaction, so as to cope with high-frequency vibrations and rapid temperature difference changes in complex environments. The physical domain is responsible for collecting multi-source data, the environmental domain perceives external interference in real time, and the virtual domain realizes time series feature extraction and dynamic model updating by combining the optimization capabilities of recursive neural networks and reinforcement learning. This technical solution significantly improves the real-time and adaptability of the system, can accurately predict environmental change trends, adjust the weighing strategy in advance, and control the error rate at an extremely low level. At the same time, by combining the negative feedback mechanism with the optimization of the objective function, the sampling frequency is dynamically adjusted to reduce energy consumption while ensuring data processing efficiency. The deep integration of wavelet decomposition, dynamic normalization algorithm and neural network model greatly enhances the accuracy and stability of data processing.

The application discloses a cable online defect positioning method and equipment, belongs to the technical field of cable defect positioning, and is used for solving the technical problems that the current cable online detection technology has a positioning blind area, signal attenuation causes end positioning difficulty, and positioning effect still needs to be improved. The cable online defect positioning method comprises the following steps: determining an original positioning curve of a test cable according to an incident signal input into the test cable and a reflected signal collected; performing normalization processing on the original positioning curve to obtain a first positioning curve; performing average energy operator optimization on the first positioning curve to obtain a second positioning curve; and determining a defect position of the test cable according to the second positioning curve. The application utilizes the average energy operator to reduce noise interference caused by the normalization algorithm, improve the blind area of the first end signal oscillation, and greatly improve the positioning amplitude of the original positioning curve.

This invention provides a method, system, device, and medium for compliance risk avoidance in generative artificial intelligence. The method includes: acquiring user input text; preprocessing the input text and calculating its compliance potential value, wherein the compliance potential value is positively correlated with the compliance relevance, discourse empowerment, and group influence of each character in the input text; based on the calculated compliance potential value, using a normalization algorithm to calculate an empirical threshold for the input text, and dividing the input text into three levels according to the empirical threshold; constructing a three-level text database; generating a final question-and-answer result by calling the three-level text database according to the level of the empirical threshold corresponding to the input text; and displaying the final question-and-answer result to the user. This addresses how to ensure that the generated content does not have content compliance issues while providing users with more authoritative, comprehensive, and reliable answers when applying generative artificial intelligence in areas involving compliance expression related to content compliance.

The invention relates to the technical field of biological information, in particular to a multi-task gene regulation network model training method based on knowledge graph embedding and feature fusion. Implicit important information in a gene sequence is captured from three aspects of context perception representation, physicochemical properties and pseudo dinucleotide composition of the sequence, and the richness of feature embedding is improved; a bidirectional long-short-term memory network is adopted to carry out splicing learning on the extracted features so as to reduce the coding difficulty of a high-throughput gene sequence; two learning task modules with different emphasis are designed and focus on node features and edge features respectively, and more comprehensive feature utilization is realized through a multi-task learning mode; the loss of the two tasks is dynamically balanced by adopting a gradient normalization algorithm, so that the overall learning efficiency and stability of the model are improved. The invention aims to solve the problem of how to improve the regulation relation prediction accuracy of the gene regulation network.

The invention provides a method for detecting the content of ergosterol in lentinus edodes based on near infrared spectroscopy and application. And rapidly detecting the content of the ergosterol in the shiitake mushrooms by adopting a near infrared spectrum technology, and inputting spectral data of a sample to be detected into the optimal model to obtain a predicted value of the ergosterol, so that the prediction of the content of the ergosterol in the shiitake mushrooms is realized. According to the method, the SG smoothing algorithm and the normalization algorithm are combined to preprocess the spectrum, noise generated when the sensor obtains the spectrum data can be effectively removed, and baseline and scattering correction is carried out; the variable combined population analysis-genetic algorithm is used for carrying out characteristic wavelength extraction on an original spectrum, so that required important variables can be effectively extracted, redundant information is efficiently removed, and the operation rate of the model is improved; the content of ergosterol in shiitake mushrooms is predicted by adopting the least square support vector machine model based on the crown porcupine optimization algorithm, and the model is high in adaptive capacity and high in prediction accuracy. And a new technical approach is provided for rapidly detecting the content of ergosterol.

The invention discloses a track prediction method based on a CNN-AM-BiLSTM neural network model, and the method comprises the steps: collecting the track data of an air target through radar equipment, and constructing a data set; reducing the track data to a range of [0, 1] by using a normalization algorithm, and dividing into a training set, a test set and a verification set; building a CNN-AM-BiLSTM neural network model, wherein the CNN-AM-BiLSTM neural network model comprises a convolution module, an attention mechanism module, a sequence extraction module and an output module combination; taking the data set as the input of a neural network model, starting to train the neural network model, continuously adjusting hyper-parameters in the training process, and obtaining an optimal hyper-parameter group according to the change of a loss function; and inputting air target track data detected by radar equipment into the neural network model, and outputting future track data of the air target after model checking calculation. According to the method, the flight paths of various aerial targets can be predicted, and the prediction precision is improved.