51 results about "Autologistic regression" patented technology

A method of emotion recognition in cross-subject EEG signals, belonging to technical field of deep learning, includes the following steps: S1, constructing the extracted DE features into positive and negative samples by using a positive and negative sample generator; S2, sending the DE features of an anchor and the positive and negative samples into the encoder for coding, mapping the DE features to a latent space, performing regression prediction on the encoded anchor samples in the latent space by using an autoregressive model, training the encoder by using a probability supervision contrastive loss function; and S3, connecting the trained encoder to the classifier for fine tuning, and training the classifier through the cross entropy loss function; in this process, the encoder does not perform gradient propagation to complete cross-subject emotion recognition.

The invention discloses a first view angle online action segmentation method based on uncertainty, and relates to the technical field of video action analysis. According to the method, a strategy of dynamically changing the network according to uncertainty is adopted, and efficient and accurate online action segmentation under the condition of information loss in the future is realized. According to the invention, a method for scenario context according to an uncertainty principle is adopted. The method comprises two key parts: 1, a context aggregation converter which combines a joint auto-encoder and an auto-regression design to enhance contextualization of frame-level features and effectively overcome limitations caused by lack of subsequent contexts in a streaming video; and 2, action contextualization based on evidence: quantifying uncertainty in a model decision by using evidence deep learning, and correcting fuzzy prediction based on a previous context. According to the mode, the context scene information is supplemented, and the problems in the prior art can be effectively solved. The method can be widely applied to scenes related to first person view angle video action analysis, such as human body behavior monitoring based on wearable equipment, character action understanding in augmented reality and virtual reality scenes and the like.

Systems, devices, and methods are provided for training and / or inferencing capacity planning using a machine learning model. A first time series may be provided as an input to a machine learning model, which may be an Auto Regressive Integrated Moving Average (ARIMA)-based forecasting model. The machine learning model may be trained solve a conditional maximum likelihood problem by performing quantile regression. The machine learning model may forecast one or more innovations using Monte-Carlo simulations. The machine learning model may generate, as an output, a value that corresponds to an amount of computing resources that is predicted, over a second time series, to be sufficient to satisfy a threshold level of availability or quality.

The invention relates to a multi-source meteorological data fusion technology, and discloses an air temperature reconstruction method fusing data of a micro-meteorological device, which improves the temporal-spatial resolution and precision of ground temperature under a complex terrain. The method comprises the following steps: densely deploying micrometeorological devices in a complex terrain area to obtain high-frequency observation data, carrying out quality control, and carrying out hierarchical processing in two dimensions of space and time by taking pattern forecast grid point data as an initial background field: in the spatial dimension, dynamically updating a fusion weight for grid points with observation stations by using geographical weighted regression, and carrying out data fusion; a residual machine learning model is combined with multi-topographic feature correction for grid points without observation stations, and a high-precision space fusion background field is generated; in the time dimension, errors after space fusion are decoupled into a trend term and a periodic term, an autoregressive integral moving average model is used for predicting a trend, a Fourier algorithm is used for correcting a periodic phase, and then time dimension machine learning correction is carried out on a grid point of an observation-free station. And finally, complete-process automatic, high-temporal-spatial-resolution and low-error complex terrain area air temperature reconstruction is realized.

The invention is suitable for the technical field of geographic element data analysis, and provides a geographic spatiotemporal weighted vector autoregression-based multi-element collaborative change modeling method, which comprises the following steps of: constructing a GTWR-VAR model, and combining a multivariable feedback mechanism of vector autoregression with locality of geographic spatiotemporal weighted regression to obtain a GTWR-VAR model; the method comprises the following steps of: quantifying spatial-temporal heterogeneity of mutual influence among geographic elements, introducing a Bayesian optimization mechanism, adaptively determining an optimal lag order, a spatial bandwidth and a time bandwidth by taking verification set mean square error minimization as a target, and finally adopting Monte Carlo block permutation test to determine the optimal lag order, the spatial bandwidth and the time bandwidth on the premise of retaining inherent time autocorrelation of data. And evaluating and eliminating non-significant items in the local coefficient to ensure the statistical effectiveness of the result. According to the method, the continuous change rule of the influence mechanism between the geographic elements in time and space is disclosed, and a more accurate and reliable basis is provided for regional response research under fine management, prediction and climate change.

The invention discloses a river runoff prediction method, system, equipment and medium under a rainfall condition, and the method comprises the steps: collecting data information of a cascade hydropower station drainage basin and an adjacent region, and carrying out the data preprocessing of the collected data; performing sample construction on the preprocessed data; a meteorological prediction model in the cascade hydropower station drainage basin is constructed based on an auto-encoder, and meteorological auto-regression prediction is carried out; and based on a meteorological autoregression prediction result, using a neural network model to automatically weight and predict the rainfall, using the predicted rainfall, and calling software to predict the river runoff. Through a physical model, the interpretability of the method model is enhanced, and complementation of a physical mechanism and data driving is realized.

The application discloses a kind of intelligent verification and deviation correction method and system of power selling day clearing result.The method collects power selling day clearing data, which is decomposed into metering, price, loss and settlement rules and other business components and evaluated uncertainty, according to tail thickness, divided into light tail type or near Gaussian type component and heavy tail type component;Through first-order Taylor expansion to obtain sensitivity coefficient, saddle point approximation second-order semi-analytic expansion is used for heavy tail component to obtain second-order cumulant curvature;Spearman dependence matrix is constructed, and positive definite correlation matrix is generated by ridge parameter disturbance and normalization, and combined with standard deviation to form covariance matrix, calculate the combined standard uncertainty and the corrected component degree of freedom, generate asymmetric verification threshold;When day clearing result is over threshold, based on contribution degree vector, deviation is counter-distributed, and kalman filter with residual auto-regression is used to update component, and output correction result.

The invention relates to the technical field, in particular to an intelligent optimization algorithm for the temperature distance in the future 15 days, and the algorithm comprises the steps: carrying out the preprocessing operation of the obtained actual meteorological data, building a linear regression model based on the prediction result of numerical prediction, calculating the average temperature distance between the meteorological data of numerical prediction and the actual meteorological data, and carrying out the calculation of the average temperature distance between the meteorological data of numerical prediction and the actual meteorological data. Training a linear regression model, and inputting future temperature anomaly for correction; an autoregression extrapolation model is constructed according to the early-stage air temperature anomaly, and the future average air temperature anomaly is obtained through the autoregression extrapolation model; and comprehensively considering the priority of the prediction result, outputting an optimal prediction result, and performing accuracy evaluation on the prediction data. According to the intelligent optimization algorithm for the temperature distance and flatness in the future 15 days, the advantages of the two models are fully combined, the result of a linear regression correction model is taken as a main part, an autoregression extrapolation model is taken as a constraint condition, and whether linear regression extrapolation is in an over-correction state is judged, so that the situation of over-correction of the linear regression model is reduced.

The present invention discloses a method, device, medium, and equipment for predicting the load of an air-conditioning system, and relates to the technical field of load prediction. The method first obtains the historical load data of the air-conditioning system and the load influencing factors of the same period as the historical load data. A preliminary load prediction model is then constructed based on the historical load data using a seasonal difference autoregressive sliding average model, and the prediction residuals of the preliminary load prediction model for the historical load data are determined. A generalized regression neural network is then trained based on the load influencing factors and the prediction residuals of the preliminary load prediction model for the historical load data to obtain a residual prediction model. Finally, the load data to be predicted can be input into the preliminary load prediction model for preliminary load prediction, and the residual prediction model is used to correct the load data, thereby determining the predicted load of the air-conditioning system at the target time based on the preliminary load and the corrections. The present invention improves convergence speed, prediction efficiency, and prediction accuracy.

An extensible network configuration synthesis method based on an autoregressive model can effectively relieve the problems of too high complexity and slow solution in the configuration synthesis process by dividing a network into a plurality of small sub-regions and adopting a decomposition-combination strategy; compared with a traditional method, the efficiency of configuration synthesis can be remarkably improved when a large-scale network is processed, and the expandability of the large-scale network is improved; through an incremental configuration network and an auto-regression-based configuration generation strategy, dynamic adjustment can be performed under continuously changing network conditions, so that the robustness and flexibility of the system are enhanced.

The invention belongs to the technical field of computers, and discloses a GUI (Graphical User Interface) positioning method and system based on a routing prediction framework. According to the method, a route prediction mechanism based on lexical item type judgment is introduced in the autoregression decoding process of the multi-modal large language model, and an interface semantic generation task and a spatial positioning task are dynamically shunted, so that a user instruction with a target element can jump out of a discrete text generation path; continuous space coordinate regression is directly carried out based on a hidden state and structure enhanced interface image features, and a null positioning result is returned through a special refusal response lexical item when a target element does not exist, so that the coordinate quantization error and illusion positioning problems are effectively avoided under an end-to-end unified framework, the positioning reasoning delay is remarkably reduced, and the positioning accuracy is improved. Precision, real-time performance and robustness of graphical user interface element positioning are improved, and the overall understanding ability of the model for interface structure semantics and spatial layout relation is enhanced.

The application provides a source-sink landscape-based soil heavy metal remediation method and system, and relates to the field of ecological environment; the method generates target region small watershed space data by analyzing digital elevation model through GIS; on the other hand, the source-sink landscape type of the target region is determined according to satellite images and remote sensing images, and the pollution source intensity is determined by combining rainfall runoff and soil environment sampling monitoring to establish a source-sink landscape geographic information database; soil heavy metal monitoring and spatial evaluation are carried out by selecting typical landscape types, and a self-regression model is established by spatial autocorrelation analysis of soil heavy metal spatial distribution information and source-sink landscape types, and the best scheme of landscape space optimization is proposed based on the model; based on the optimal scheme, the landscape space pattern most conducive to the soil heavy metal self-repairing process is obtained; finally, physical, chemical and biological treatments are further carried out for different landscape types, and the source-sink landscape-based soil heavy metal remediation application is completed.

A cross-subject electroencephalogram-based emotion recognition method, belonging to the technical field of deep learning. The method comprises the following steps: S1, using a positive and negative sample generator to construct extracted differential entropy features into positive and negative samples; S2, feeding differential entropy features of an anchor and the positive and negative samples into an encoder for encoding, mapping the encoded differential entropy features to a latent space, using an autoregressive model to perform regression prediction on an encoded anchor sample in the latent space, using a probabilistic supervised contrastive loss function to train the encoder, training, by means of reducing the distance between positive sample pairs and increasing the distance between negative sample pairs, the encoder to complete representation learning, and discarding the regression model after the representation learning is completed; and S3, connecting the trained encoder to a classifier for fine tuning, and training the classifier by means of a cross-entropy loss function, during which the encoder does not perform gradient propagation, so as to complete cross-subject emotion recognition. The present invention has a higher recognition accuracy and a smaller standard deviation.

A method for estimating a covariance with respect to a plurality of bonds is provided. The method includes: receiving historical bond market returns data; using a first algorithm based on an Auto-Regressive-Moving-Average (ARMA) model to calculate ARMA model regression errors based on the historical bond market data; using a second algorithm based on a logarithmic Generalized AutoRegressive Conditional Heteroskedasticity (GARCH) model to calculate an estimated volatility vector based on the ARMA model regression errors; using the ARMA model regression errors and the calculated volatility vector to estimate a time-varying covariance matrix of the ARMA model regression errors with respect to the historical bond market data; using the estimated time-varying covariance matrix of the ARMA model regression errors and the calculated volatility vector to estimate a time-varying covariance matrix of the bond returns; and using the estimated time-varying covariance matrix to calculate a set of predicted hedge ratios.

The invention relates to an abnormal prediction method and system for dissolved gas in oil based on multi-expert learning, and the method comprises the steps: collecting historical concentration data of each dissolved characteristic gas in transformer oil, arranging the historical concentration data into a dissolved characteristic gas time sequence data set according to a time sequence, and cleaning the data set; training an autoregressive moving average model, a gray scale prediction model and a deep learning regression model through the cleaned dissolved characteristic gas time series data set, and fusing a long-short-term memory network as a backbone to obtain a multi-expert model; and inputting real-time concentration parameters based on the characteristic gas dissolved in the oil into the multi-expert model, predicting a concentration signal of the characteristic gas dissolved in the oil at the next moment, and judging that the concentration of the characteristic gas is abnormal if an error between the concentration signal and the actual concentration signal at the next moment exceeds a set threshold value. According to the method provided by the invention, regression prediction of the dissolved gas in the oil and abnormal state detection of the sensor can be rapidly completed, and a reliable basis is provided for a multistage early warning task of the dissolved gas in the downstream oil.

A method for differential diagnosis of gallbladder adenoma and cholesterol polyps based on the fusion of deep learning and clinical features in ultrasound images includes: performing deep learning on ultrasound images using a deep learning network to output an adenoma / polyp risk score; the deep learning network is selected from one of VGG, ResNet, DenseNet, GoogLeNet, Vision Transformer, and Inception; the adenoma / polyp risk score and clinical features are input into a trained machine learning model for learning to obtain the final probability of identifying the polyp as a gallbladder adenoma rather than a cholesterol polyp; the clinical features include: echo intensity, maximum polyp diameter, and age; the machine learning model is selected from one of Logistic Regression, Gradient Boosting, K-nearest neighbors, Gaussian naïve Bayes, AdaBoost, Random Forest, ExtraTrees, and SVM.

The invention belongs to the technical field of power systems, and particularly relates to a new energy inertia active support capability evaluation method based on artificial intelligence. The method comprises the following steps: acquiring real-time frequency measurement data of wind power and photovoltaic units; detecting and separating pulse and Gaussian noise, and processing by using an improved adaptive median filter and an improved weighted mean filter to obtain high-precision data; constructing an active auto-regression ARX model of a steady-state / dynamic scene to improve a recursive least square FF-RLS algorithm identification parameter with a forgetting factor, and obtaining a real-time inertia support evaluation value; and inputting the evaluation value and the minimum inertia demand threshold value into a fuzzy logic controller to generate an adequacy index, and if the adequacy index is lower than a preset threshold value, sending an insufficient support early warning. Accurate real-time evaluation is realized, and the frequency safety of the power grid is guaranteed.

The application discloses a sequence data timespectrum radiation unification processing method and system under reference data constraints, and the method comprises the following steps: acquiring a remote sensing image sequence and generating a multi-source radiation normalization image database to form a multi-source reference sequence data pair; a reconstruction self-similarity normalization pre-training model is constructed, the reconstruction self-similarity normalization pre-training model comprises a decomposition and reconstruction module, a self-similarity weight matrix module and a variational normalization module, the decomposition and reconstruction module obtains a final timespectrum feature sequence, and the self-similarity weight matrix module obtains a timespectrum self-similarity weight matrix; the variational normalization module processes to form timespectrum radiation consistent sequence data; and corresponding timespectrum radiation consistent sequence data is obtained based on the reconstruction self-similarity normalization model. The application constructs a variational model framework of timespectrum normalization, cooperates self-similarity weight, timespectrum self-regression and similarity, realizes timespectrum radiation normalization of a timespectrum sequence cube, and solves timeserial remote sensing data which is accurate in trend and faithful in spectrum.

The present invention relates to the field of artificial intelligence and provides an enterprise strategic intelligent early warning method. The method performs NLP recognition processing on external operating data to extract external feature data; performs standardization processing and coding processing on the enterprise's internal evaluation data to obtain internal reference data; obtains primary reference information through a multiple regression model, obtains auxiliary reference information through a temporal differential autoregressive sliding average model, obtains a predicted customer throughput trend based on the primary and auxiliary reference information, and obtains current enterprise risk warning information based on the customer throughput trend and preset trend risk comparison information. In this way, intelligent early warning of customer risks is achieved, and possible losses discovered afterwards are reduced. In other words, intelligent early warning of customer risks is achieved, and possible losses discovered afterwards are reduced. The accuracy and practicality of the model are evaluated to improve the throughput prediction accuracy.

The invention relates to an autoregression image generation method based on semantic detail decoupling and local information enhancement. The method comprises a VQ-GAN feature learning process based on semantic detail decoupling and an autoregression generation process based on local context lexical elements. In the feature learning process, a stage type and alternating type double training strategy is adopted, firstly, a small-size semantic codebook is trained to learn an image core semantic structure and construct the basic reconstruction capability, then a large-size detail codebook is introduced, and through the layering process of semantic quantification, residual calculation and detail quantification and alternating training of three iteration rounds and one period, the image core semantic structure can be learned. And semantic-detail hierarchical double-codebook functional differentiation and collaborative optimization are realized. In the autoregression generation process, a local window and a dynamic cache mechanism are introduced to extract preorder local context lexical elements, and conditional guidance of local information is enhanced; and finally, combining an adaptive progressive classifier-free guidance strategy to dynamically adjust the guidance intensity of the reasoning stage, and balancing the semantic consistency and the degree of freedom of details.

The invention discloses a prediction method, apparatus and device, and a computer readable storage medium. The method comprises the steps of obtaining historical personnel distribution data corresponding to a to-be-predicted personnel group type; performing autoregression, difference and moving average calculation on the historical personnel distribution data to obtain first predicted personnel distribution data; obtaining historical personnel distribution prediction related data corresponding to the to-be-predicted personnel group type; capturing a dependency relationship of historical personnel distribution prediction related data to carry out personnel distribution prediction to obtain second predicted personnel distribution data; performing dimension reduction processing on the historical personnel distribution prediction related data to obtain prediction related data after dimension reduction; and performing personnel management service by using a support vector machine regression model according to third prediction personnel distribution data obtained by performing personnel distribution prediction on the first prediction personnel distribution data, the second prediction personnel distribution data and the prediction related data after dimension reduction. According to the invention, accurate prediction of personnel distribution data is realized, and personnel service management is facilitated.

The invention discloses a ground surface settlement monitoring method based on synthetic aperture radar images, and relates to the technical field of ground surface deformation monitoring. Comprising the following steps: acquiring a real-time rail lifting value of a to-be-processed synthetic aperture radar image obtained by rail lifting imaging; inputting the real-time rail lifting value into the trained ridge regression model to obtain a real-time rail lifting fusion preliminary prediction value; carrying out modeling on a residual error generated by the trained ridge regression model in a prediction process through the trained autoregression integrated moving average model to obtain a real-time prediction residual error; and superposing the real-time lifting rail fusion preliminary prediction value and the real-time prediction residual error to obtain a real-time lifting rail fusion final prediction value, and using the real-time lifting rail fusion final prediction value to extract ground surface settlement deformation information. According to the method, the rail lifting fusion value can be obtained only by processing the rail lifting value, the dependency of traditional InSAR double-rail fusion is broken through, and a new normal form is provided for single-rail data deformation monitoring.

The invention discloses a base station package type solar panel fault detection method, and relates to the technical field of fault detection, and the method comprises the steps: obtaining the power data and environmental parameters of a solar panel, employing high-frequency sampling and sliding window filtering processing, carrying out the modeling through a time sequence, combining the characteristics of autoregression and moving average, recognizing the short-time power fluctuation characteristics, and carrying out the fault detection of the solar panel. Calculating a power change rate, and detecting abnormal power fluctuation; the method comprises the following steps: a power abnormity screening stage: comparing with historical power data based on time sequence analysis, introducing an illumination power mapping model, establishing a mapping relation between illumination and power through regression analysis, and eliminating misjudgment caused by short-time illumination change, and a fault identification stage: analyzing temperature distribution of a solar panel in combination with infrared thermal imaging, and calculating a temperature mean value and a variance, judging whether a local overheating phenomenon exists, and judging whether power reduction is caused by short-time shielding or panel damage through cross comparison of illumination data and thermal imaging data.

The present invention relates to an edge prediction method based on a deep regression recurrent neural network, comprising the following steps: Step S1: Obtain historical edge load data and construct a data set; Step S2: Preprocess the data set; Step S3: Take the users of the edge server as covariates; Step S4: Input the preprocessed data set and covariates into a prediction model to predict future edge loads; Step S5: Based on the edge load prediction results, assist the edge computing service in formulating a resource scheduling plan. The present invention realizes scalable and effective workload prediction, and effectively improves the resource allocation efficiency in cloud computing.

The invention discloses a financial market mobility toughness network deconstruction and vulnerability traceability method, system and device and a computer readable storage medium, and relates to the technical field of metering economy and financial risk early warning. The method comprises the following steps: acquiring standardized fluidity indexes and external fluidity impact indexes of multiple sub-markets; constructing a quantile time-varying parameter vector autoregression model, and estimating quantile impulse response under multiple quantiles; based on the pulse response, respectively calculating two kinds of toughness indexes, namely, the flowability impact absorption strength and the absorption duration of the financial market; further, quantile regression is combined with skewed t distribution mapping to obtain conditional distribution and in-danger growth of mobility and toughness acceleration; and realizing vulnerability traceability by adopting marginal contribution decomposition of orthogonalization residual errors, and outputting time-varying risk source weights and alarms. The technology can be stably operated under the extreme descending state and the prosperous state, the state dependence of toughness and cross-market asymmetric overflow can be depicted, and quantitative early warning and source tracking of the mobility toughness lower tail risk are achieved.

The invention provides a line icing thickness prediction fusion method based on a two-way parallel network. The method comprises the steps that a multi-dimensional data sequence is acquired, and a neural network prediction model predicts an empirical thickness prediction curve of icing thickness change conditions in a future time period according to the multi-dimensional data sequence; a thickness observation sequence is obtained, the regression prediction model carries out iterative calculation according to the thickness observation sequence to obtain the current potential state of the current moment, the regression prediction model carries out multiple times of autoregression iterative calculation based on the current potential state, and a mechanism thickness prediction curve of a future time period is predicted; a plurality of prediction features are obtained, the nonlinear element model generates nonlinear weights based on the prediction features, and the empirical thickness prediction curve and the mechanism thickness prediction curve are multiplied by the corresponding nonlinear weights and are added so as to be fused into a final prediction curve. According to the method, the prediction results of the two branches can be complementarily corrected, the disturbance of complex environmental factors to the prediction process is reduced, and the accuracy of the final prediction curve is improved.

The invention discloses a multi-scale adaptive time sequence prediction method and system based on multi-expert cooperation, and belongs to the technical field of time sequence data analysis. The method comprises the following steps: firstly, dividing a time sequence data sample into a plurality of normalized time sequence blocks with equal length; secondly, obtaining a time sequence prediction result for the normalized time sequence block by using a hybrid multi-expert self-adaptive predictor; and finally, reconstructing the hybrid multi-expert adaptive predictor as an auto-regression iterative residual network architecture, so as to extract complementary multi-scale information in the time sequence representation to improve the time sequence prediction performance. The method is realized through a multi-scale adaptive time sequence prediction system, and comprises a normalized time sequence partitioning module, a mixed multi-expert adaptive prediction module and an iterative residual error autoregression learning module. According to the method, the limitation of a fixed-length prediction head can be broken through through iterative residual error autoregression learning, the problem that a shared linear prediction layer is insufficient in capturing a heterogeneous time sequence mode is solved through hybrid expert prediction, and a high-precision and high-flexibility cold load adaptive time sequence prediction solution is provided for realizing intelligent environmental control and energy-saving operation.

The application discloses a water quality prediction method based on ARIMA and an improved chicken swarm algorithm, and comprises the following steps: acquiring water quality data from an initial period to a t period; performing stationarity test on the water quality data by using a unit root test method to determine a difference order d; determining an autoregressive order p and a moving regression order q by using an Akaike information criterion AIC or a minimum Bayesian information criterion BIC; establishing an ARIMA water quality prediction model according to the difference order d, the autoregressive order p and the moving regression order q, obtaining water quality prediction data of the t period through the ARIMA water quality prediction model; introducing an optimization operator, re-encoding the chicken swarm algorithm in combination with the water quality prediction data of the t period; initializing a population, defining a fitness function and stipulating a chicken swarm swimming strategy; performing chicken swarm identity attribution on the water quality prediction data according to the size of the fitness, and performing position updating through different identities to obtain optimal water quality prediction data of the t period. The application can realize more accurate prediction of water quality data.

The invention discloses a VAE-IAF-based spectrum self-supervised learning model construction method and application thereof, and the method comprises the steps: building a self-supervised learning model framework which comprises a VAE-IAF model and a spectrum feature adaptive regression network; the VAE-IAF model comprises a variational automatic encoder and an inverse autoregressive flow, and an encoder of the variational automatic encoder performs feature extraction on input high-dimensional spectral data to obtain spectral features; performing distribution estimation on the spectral features extracted by the encoder by the inverse autoregression flow to obtain an optimized potential vector; and the spectral feature adaptive regression network receives the optimized potential vector output by the inverse autoregression flow, carries out regression prediction and outputs a prediction result. Nondestructive detection of the zinc content of the crop leaves is achieved based on the spectrum self-supervised learning model, and the method is high in detection precision and does not depend on large-scale labeled data and complex preprocessing steps.