42 results about "Heteroscedasticity" patented technology

The invention discloses a power grid frequency balance intelligent control system based on a distributed architecture, relates to the technical field of power grid control, and is used for solving the problems that the existing WLS / UKF and other methods are generally assumed to be good in synchronization, are insensitive to heteroscedasticity time delay and missing measurement, and are difficult to ensure robustness under real communication and equipment conditions. Through a through link, under the real clock misalignment, time delay jitter and packet loss conditions, measurement of a unified time axis and quality quantification are realized, and frequency estimation bias and uncertain transmission are significantly reduced. Reconstruction, filtering and triggering are penetrated with the same confidence degree, and abnormal measurement suppression and effective information utilization are improved; the fusion efficiency and timeliness of asynchronous arrival data are improved by utilizing graph prior and information domain incremental updating; and regional measurement abnormity and real power imbalance are discriminated through double-model causality, and error event reporting is reduced.

The invention discloses a rainfall downscaling method and system based on a deep learning network model fusing rainfall priori knowledge, and the method comprises the steps: firstly collecting the topographic data and low-resolution day-by-day rainfall data of a target region, and taking the data as input data; a short-term high-resolution precipitation field generated in a mesoscale weather forecast WRF mode is used as training truth value data; according to the method, the function of accurately downscaling the rainfall data in combination with the convolutional neural network and the long and short term memory network is realized, the spatial-temporal correlation of rainfall is fully considered in the downscaling process, and meanwhile, a likelihood function combined with coupled censored data, Box-Cox conversion and time variation variance Gaussian distribution is adopted as a rainfall loss function; the method not only can represent zero expansibility, skewness and heterovariance characteristics of rainfall, but also can improve the rainfall downscaling precision and quantify the uncertainty of rainfall downscaling, and is suitable for wide popularization and use.

The invention discloses a bridge temperature-induced strain prediction method under the influence of non-constant noise, and the method comprises the steps: building a training set containing main beam temperature and temperature-induced strain according to historical monitoring data, and assuming a test set; under a heterovariance Gaussian process framework, constructing conditional distribution of temperature-induced strain of a main beam of the test set as an undetermined heterovariance Gaussian regression model; a variational Bayesian inference method is adopted to construct a variational free energy boundary based on variational posteriori distribution, and approximation is performed on the variational free energy boundary to obtain an edge variational boundary; assuming that the variational posteriori distribution of the noise logarithmic variance function obeys multivariate normal distribution, simplifying an edge variational boundary into a new edge variational boundary, and solving to obtain a hyper-parameter and a variational parameter; and substituting the obtained result into the to-be-determined heterovariance Gaussian regression model, obtaining new main beam temperature data, and predicting the temperature-induced strain of the main beam in the test set. The bridge temperature-induced strain prediction method under the influence of the non-constant noise is established, the prediction precision can be ensured, and the calculation efficiency can be improved.

The invention belongs to the technical field of power system carbon emission analysis and intelligent prediction, and provides a cross-regional power carbon emission factor prediction and robust optimization method and system. The representation of the power grid cross-regional node i at the moment t is obtained, wherein the representation integrates multi-scale space representation, time sequence representation and physical consistency constraint; based on the Bayesian thought, hybrid modeling with heterovariance likelihood combined with posterior weight uncertainty is adopted, and a carbon emission factor prediction mean value and a prediction covariance of the node i are obtained; according to the method, uncertainty prediction of carbon emission factors is integrated into a decision model, and a feasible and stable decision can still be obtained when a prediction error and a model deviation exist under a given carbon budget or carbon cost through probability constraint. Through the method, accurate prediction and robust application of the cross-regional electric power carbon emission factor are realized, so that refined accounting and low-carbon operation and maintenance decision are supported.

The invention discloses a public transport passenger flow prediction method based on IC card data, which relates to the technical field of intelligent transportation, and comprises the following steps: respectively carrying out feature inspection on a week time sequence, a day time sequence, a time interval sequence and a holiday effect sequence, selecting an adaptive model, carrying out parameter and noise variance feature analysis on the adaptive model by using an RLS algorithm, and carrying out prediction on the public transport passenger flow. Outputting four groups of prediction result sequences; models corresponding to the four groups of prediction result sequences are converted into a state space form, a noise variance estimation value is estimated online through an RLS algorithm, and four groups of time-varying fusion weights are generated; and constructing an observation noise covariance matrix based on the four groups of time-varying fusion weights, performing recursive optimization through a Kalman filter, and outputting a hybrid prediction result by weighting and fusing observation prediction values of the four groups of sequences. According to the method, accurate fitting of passenger flow periodicity, seasonality and heterovariance is realized, and the single-model prediction precision is improved.

The invention belongs to the technical field of automobile intelligent manufacturing and data processing, and particularly relates to a three-dimensional assembly tolerance prediction method for automobile assembly. The method comprises the steps of obtaining ideal three-dimensional coordinates and actual measurement coordinates of key measurement points under standard working condition characteristics on an automobile final assembly production line, and constructing a historical data set of input characteristic vectors and prediction target vectors so as to train a joint Gaussian process regression model; predicting a deviation correction amount and a heterovariance under the new working condition characteristics to obtain a dynamic correction coefficient, and further updating input parameters in the simulation model; the updated input parameters are substituted into the simulation model, a new tolerance distribution diagram of the key measuring points is generated, a comprehensive risk assessment index is calculated, and tolerance prediction and risk early warning of the automobile assembly process are achieved according to an assessment result. According to the method, the problem that a static simulation model cannot reflect a dynamic production process is solved, and the tolerance prediction accuracy is remarkably improved.

A method for automatically abstracting and selecting an optimal set of variance-related functions that are an indicator of an individual outcome in healthcare, wherein the method comprises: generating, by one or more processors, an abstracted set of variance-related candidate-patient functions, wherein the abstracted set of variance-related candidate-patient functions are temporally heteroscedastic functions; optimizing, by one or more processors, each patient function from the abstracted set of variance-related candidate-patient functions by identifying a time period in which the variances and heteroscedasticity of each patient function are maximized, wherein this optimization produces an optimal abstracted set of variance-related patient functions from the time period in which the variances and heteroscedasticity of each patient function are maximized;by one or more processors comparing the optimal abstracted set of variance-related patient functions with a historical dataset for a patient population to create a predictive set of variance-related patient functions, wherein the predictive set of variance-related patient functions predicts a health-related target outcome of the patient population; by one or more processors generating a current optimal patient set of variance-related patient functions for a current patient; by one or more processors comparing the optimal set of variance-related patient functions for the patient population with the current optimal patient set of variance-related patient functions for the current patient;In response to the optimal set of variance-related patient functions for the patient population, which matches the current optimal patient set of variance-related patient functions for the current patient within a predefined limit, one or more processors determine whether the health-related target outcome matches a predefined health-related target outcome for the current patient; and in response to the health-related target outcome matching the predefined health-related outcome for the current patient, one or more processors issue an alert regarding the predefined health-related outcome for the current patient.

The application discloses a chromium-nickel high geological background area agricultural product risk assessment method and system, relates to the chromium-nickel pollution monitoring technical field, first, laser echoes, binocular parallax and illumination changes are synchronously acquired in the same time window and are triplely aligned, and multi-modal original ranging signatures are generated; then, according to echo integrity and texture complexity, real-time reliability weights are calculated, and nuclear decay weighting is implemented on depth candidates; subsequently, target short-time motion coherent information is fused, and with the aid of residual learning network and Lyapunov criterion iterative correction, steady-state depth sequences are output; finally, based on Bayesian residual and heteroscedastic Gaussian process, pixel-level uncertainty boundaries are generated, and reliability weights are used to online adjust laser power, exposure time and frame rate, realizing a ranging, evaluation and acquisition parameter self-adaptive closed loop, and providing data support for fast, fine and reliable chromium-nickel pollution risk determination.

The invention relates to a pollutant detection method and system based on machine learning, and relates to the technical field of pollutant detection.The method comprises the steps that original spectral data are collected and denoised through a Savitz-golay smoothing filter, parameters are dynamically adjusted, and stable spectral data with pollutant core characteristics reserved are obtained; the method can effectively solve the technical problem that low-concentration interval heterovariance noise masks weak features of left censored data, and avoids the defect of excessive smoothness or insufficient denoising of a traditional denoising method. Secondly, in combination with a blank spectrum noise optimization detection limit and a quantification limit, effective features are extracted by utilizing machine learning, and incomplete observation characteristics of left-censored data can be adapted, so that the problem of systematic bias caused by data distribution hypothesis distortion of an existing algorithm is solved, and the defect that a low-concentration risk early warning window is neglected in a traditional method is overcome; finally, filter parameters are analyzed and optimized through a loss function, a detection result is finally output, and detection limit heterogeneity caused by instrument drifting and working condition fluctuation can be dealt with.

This invention discloses a DDoS attack detection method based on heteroscedastic unscented Kalman filtering. Addressing the issues of limited resources in edge networks and low anomaly detection accuracy and high latency in low signal-to-noise ratio environments, this invention utilizes Sketch for full-scale, unsampled traffic feature extraction in the programmable data plane and reports observations to the control plane through differential processing. In the control plane, a third-order state-space model incorporating instantaneous traffic rate, rate of change, and acceleration is innovatively constructed. An additive heteroscedastic noise model is proposed to adaptively handle Sketch hash collisions and traffic shot noise, and unscented Kalman filtering is employed for state estimation. Simultaneously, an intelligent hybrid control mechanism is introduced to overcome filtering lag, and finally, NIS and CUSUM are combined for dual-modal anomaly detection, outputting the system state. This invention enables high-precision real-time detection of DDoS attacks with extremely low resource overhead, effectively solving the trade-off between measurement and detection in existing technologies.

The application belongs to the technical field of underwater acoustic array signal processing, and discloses a covariance fitting azimuth estimation method under non-Euclidean metric based on matrix reconstruction, which constructs a noise standard deviation matrix, judges the noise type, extracts the heteroscedastic characteristic from the noise standard deviation matrix, uses the extracted noise power to whiten the array received signal, calculates a sample covariance matrix SCM, reconstructs a model covariance matrix MCM matched with the whitening processing, uses the symmetric Kullback-Leibler divergence as the non-Euclidean metric to calculate the difference between the SCM and the MCM, and realizes the DOA estimation by traversing the azimuth grid points and minimizing the symmetric Kullback-Leibler divergence. The application uses the symmetric SKL divergence to solve the matrix-matrix comparison problem, and focuses on the low SNR condition. The more complete modeling of the noise interference parameter can reduce the DOA estimation failure threshold by about 20 dB.

The invention relates to the technical field of global navigation satellite system data processing, and discloses a GARCH-based GNSS cycle slip detection threshold real-time determination method, which comprises the following steps: acquiring GNSS double-frequency observation data, constructing a Morse-Wiberner and geometric distance-free combination observation value, and carrying out differential processing to obtain a noise time sequence. Secondly, traversal calculation is carried out based on an akaike information criterion, and the optimal order and the sliding window length of the GARCH model are determined; then, a current epoch condition standard deviation is predicted using the model, and a multiple factor is adjusted in combination with an environmental state to construct a dynamic detection threshold. And finally, comparing an absolute value of the noise sequence with a threshold value to judge a cycle slip, and calculating an integer solution by constructing an observation equation to complete repair. According to the method, the heterovariance characteristic of the observation noise is modeled by using the GARCH model, so that the adaptive adjustment of the detection threshold along with the environmental noise level is realized, the misjudgment rate in a complex environment is effectively reduced, and the detection accuracy is improved.

This invention discloses a method, apparatus, medium, and device for obtaining atmospheric boundary layer height. The method includes: accessing multi-source observation data to generate observation records with quality indicators and initial observation uncertainties; performing spatiotemporal mapping to generate a registered profile sequence with an uncertainty field; generating a set of physical candidate heights in parallel based on at least one physical mechanism among thermal gradient, dynamic shear, and material distribution, and recording the variance of the source observations after interpolation for each physical candidate as the candidate uncertainty; constructing feature vectors for the pixels to be estimated; inputting the feature vectors into a first-layer candidate scorer to output a confidence score for each physical candidate, and inputting the feature vectors and confidence scores together into a second-layer heteroscedasticity regressor to output a point estimate and pixel-level uncertainty of the boundary layer height; using the point estimate and pixel-level uncertainty as observation terms, performing Kalman filtering assimilation with the background field, and outputting the atmospheric boundary layer height.

The invention relates to the technical field of equipment predictive maintenance, in particular to an industrial equipment residual life prediction method based on an SRenbPI algorithm, which comprises the following steps: preprocessing industrial equipment source domain data, including constructing a heteroscedasticity noise simulation data set and introducing real industrial sensor data, and setting standardized input through normalization, sliding window construction and label; in the training stage, a bootstrap sampling strategy is adopted to construct an integrated regression model, noise influence is quantized through a scaling residual formula, and a residual set is formed; in the prediction stage, noise scale estimation and residual quantile are combined to generate an unequal-width prediction interval, meanwhile, a dynamic updating mechanism is introduced, a residual set is updated in real time to adapt to data distribution changes after a plurality of test samples are processed every time, and the model does not need to be trained again. According to the method, the adaptive defect of a traditional static residual set in industrial equipment full-life-cycle monitoring is effectively overcome, and the prediction interval precision and real-time performance are remarkably improved.

The invention relates to an agricultural productivity evaluation method based on Newey-West standard error. Relates to the technical field of agricultural economic data processing and metering analysis, in particular to an agricultural productivity evaluation method based on Newey-West standard errors. By correcting a traditional standard error, synchronous processing of time sequence data autocorrelation and heterovariance is realized. The method comprises the following steps: obtaining crop yield data over the years of each province and agricultural input data over the years of each province; constructing a least square regression model by taking sum as a dependent variable and an independent variable; calculating a vector estimation value of a regression coefficient; constructing a robust covariance matrix, and standardizing error SE; the constructed estimation covariance matrix is corrected and obtained; and calculating a t statistic and a P value according to the SE, and evaluating the agricultural productivity according to the t statistic and the P value.

The invention discloses an autonomous underwater vehicle (AUV) causal reinforcement learning decision-making method for bistatic sonar underwater target tracking, and belongs to the field of autonomous underwater vehicle (AUV) intelligent decision-making and underwater target tracking. The method comprises the following steps: firstly, constructing a causal model for describing a target tracking problem, and then establishing a measurement generation model based on a heterovariance neural network and a binary classification network; on the basis, a real-time maneuvering strategy of the AUV is trained by adopting a deep reinforcement learning framework, and high-fidelity training data is generated by utilizing an anti-factual reasoning mechanism so as to optimize a learning process. According to the method, the problems of insufficient tracking precision and low sample efficiency of a traditional decision method are effectively solved, the tracking precision and the tracking holding time of the AUV to the underwater target can be remarkably improved, and meanwhile, the data requirement for a real training sample is greatly reduced.

The invention discloses a natural gas online water analyzer field calibration method, equipment and a medium, and relates to the technical field of natural gas analysis and testing. According to the method, the uncertainty of calibration gas is introduced into the calibration model based on the least square method, the target function is established by weighting the sum of squares of residual errors, and higher weights are given to data points with lower uncertainty, so that the accuracy and the reliability of model fitting are improved, the problem of heterovariance of measurement errors is effectively solved, and the accuracy and the reliability of measurement are improved. And the accuracy of parameter estimation is improved.

The invention discloses an abnormal power fluctuation detection method and system fused with time sequence modeling. The method comprises the following steps: acquiring daily electric quantity year-on-year data of each industry and an industry power consumption data set; the method comprises the following steps: constructing a time sequence model based on year-on-year data of daily electric quantity of each industry, calculating posterior distribution of a predicted value of the time sequence model through an L-BFGS quasi-Newton method, and calculating a data basic confidence interval of a next time point; inputting each characteristic value of the time sequence model one day before the date to be predicted, the mean absolute error of the predicted value of the short-term historical window time sequence model, the residual standard deviation and the industry power consumption data set into a full-connection network to obtain a heteroscedasticity confidence interval; and calculating a final threshold interval based on the two confidence intervals to carry out industry electricity consumption abnormity judgment. According to the method, the problem that the phenomena of false alarm and missing report occur frequently due to the fact that the power utilization characteristic difference between different industries is not considered and the change of the electric quantity rule along with time is difficult to adapt in the prior art is solved.

The invention relates to a bridge data reconstruction method based on multi-scale space-time fusion and uncertainty perception, belongs to the technical field of civil engineering structure health monitoring and artificial intelligence data processing, and aims to solve the problems that space-time feature mining lacks dynamic adaptability, multi-scale feature fusion is insufficient and the like in the prior art. The method comprises the following steps: collecting time sequence data of a multi-dimensional sensor and generating a structured data set; extracting multi-scale time sequence features by using a parallel causal convolutional network; constructing a dual-path time sequence coding architecture comprising an LSTM main path and a GRU auxiliary path; after hierarchical attention and adaptive time sequence fusion, sensor spatial correlation dynamic modeling is introduced to generate high-dimensional enhanced features; a dual-branch network is utilized to predict the uncertainty of the mean value and the heterovariance, end-to-end optimization training is carried out through a mixed loss function, and a trained model is used for outputting reconstructed bridge data with an uncertainty confidence interval. According to the method, the precision and reliability of bridge monitoring data reconstruction can be remarkably improved.

The invention belongs to the technical field of underwater acoustic array signal processing, and discloses a non-European metric covariance fitting orientation estimation method based on matrix reconstruction, and the method comprises the steps: constructing a noise standard deviation matrix, and judging the type of noise; extracting heterovariance characteristics from the noise standard deviation matrix; whitening an array receiving signal by using the extracted noise power, and calculating a sample covariance matrix SCM; a model covariance matrix MCM matched with the whitening processing is reconstructed; a symmetric Kullback-Leibler divergence is used as a non-Euclidean measure, and the difference between the SCM and the MCM is calculated; and the DOA estimation is realized by traversing the azimuth grid points and minimizing the symmetric Kullback-Leibler divergence. According to the method, the problem of matrix-matrix comparison is solved by applying symmetric SKL divergence, the situation of low SNR is concerned, and the DOA estimation failure threshold can be reduced by about 20dB through more complete modeling of noise interference parameters.

The invention relates to the technical field of electronic data processing, and discloses a semiconductor part defect diagnosis method based on multi-modal data fusion, which comprises the following steps: acquiring visual full-field scanning data and targeted sampling inspection data of the surface of a semiconductor part, and unifying the visual full-field scanning data and the targeted sampling inspection data into the same mechanical coordinate system; calculating a process potential energy heterovariance index of each micropore position, wherein the process potential energy heterovariance index is jointly determined by the radial distance and the local visual variance of the micropores; constructing a potential energy regulation factor based on a process potential energy heterovariance index, correcting a weight and an error in the collaborative Kriging interpolation model by using the potential energy regulation factor, and calculating a prediction true value of each micropore in the whole field; and comparing the predicted truth value with the design tolerance of the micropores, marking the micropores exceeding the tolerance range as defect holes, and diagnosing a processing technology fault source causing defects according to the spatial distribution morphological characteristics of the defect holes. According to the method, the detection rate of the tiny spiral defects in the edge area is increased.

The industrial robot absolute positioning error compensation method solves the problems of weak normal observation and unstable pose estimation under long distance or inclined angle in the existing robot absolute positioning, and belongs to the technical field of industrial robot precision control. The present application comprises: constructing a multi-layer parallel ArUco marker field in the workspace and optimizing the layout; the robot end monocular camera collects the marker field image along the planned path, and the enhanced image is obtained through edge-ROI guided super-resolution reconstruction; corner detection and pose solution are carried out based on the enhanced image, the actual pose is obtained and compared with the theoretical pose to obtain the pose error, and the re-projection error is recorded, and the data set is composed of the robot state characteristics; the data set is used to train the heteroscedastic Gaussian process regression model, wherein the observation noise variance of each sample is proportional to the square of the re-projection error; for the target task path point, the state characteristics are input into the model to predict the pose error and compensate to the theoretical instruction to generate the modified control instruction.

The invention discloses a forward calculation-based gear tooth surface waviness Fourier analysis error interval evaluation method, which comprises the following steps of: firstly, constructing an isomorphic evaluation data set, and statistically analyzing the distribution characteristics of actual sampling intervals to determine a frequency domain credible range; the core of the method is that Monte Carlo forward modeling is introduced, and systematic deviation and random fluctuation caused by sampling interval fluctuation to a frequency spectrum are quantified. Based on this, a heterovariance normal distribution model of the measured amplitude is established, the maximum likelihood estimation is utilized to reversely deduce the true value of the amplitude, and the pivot vector method is adopted to calculate the error interval of the amplitude of the key frequency point under the predetermined confidence level. Furthermore, the optimal sampling length and the resampling interval are optimized and determined through forward modeling calculation. According to the method, the limitation that only a single spectrum result can be given in traditional measurement is broken through, quantitative calibration of errors introduced into a measurement system is achieved, a reliability evaluation result with a confidence interval is provided, and a closed-loop optimization decision basis can be provided for measurement instrument selection and parameter setting.

The application discloses a power grid frequency balance intelligent control system based on a distributed architecture and relates to the technical field of power grid control, and is used for solving the problem that the existing WLS / UKF method generally assumes good synchronization, is not sensitive to heteroscedastic time delay and missing measurement, and is difficult to ensure robustness under real communication and equipment conditions; through a through link, under the conditions of real clock misalignment, time delay jitter and packet loss, measurement unified time axis and quality quantization are realized, frequency estimation bias and uncertain transmission are significantly reduced; the same confidence is used to penetrate reconstruction, filtering and triggering, abnormal measurement suppression and effective information utilization are improved; the fusion efficiency and timeliness of asynchronous arriving data are improved by using graph prior and information domain incremental updating; measurement abnormality and real power imbalance are distinguished by double model causal discrimination, and false event reporting is reduced.

The application provides a non-uniformity correction method for a non-refrigeration long-wave infrared defocus plane polarization detector, and belongs to the technical field of infrared polarization imaging. In view of the problems that the existing correction method ignores noise heteroscedasticity and joint solving coupled iteration leads to poor stability, the application establishes a super-pixel linear response model, constructs a correction function containing a calibration matrix and an offset vector, weights each sample time-domain noise variance reciprocal, constructs a weighted least square objective function, realizes differentiated weighting of parameter estimation, adopts a step-by-step optimization strategy to decouple the bivariate joint problem into two independent sub-problems, derives a closed-form analytical solution, and finally completes correction through linear transformation of the correction function. The application can adaptively suppress heteroscedastic noise interference, avoid numerical instability of iteration, and improve correction accuracy and robustness.

The invention discloses a distributed optimization sampling method for different nodes under a heterovariance situation, and the method comprises the steps: calculating an optimal target sampling number corresponding to each node in a distributed random optimization problem, and enabling the sampling number of each node to depend on a local variance, thereby achieving the distributed optimization sampling of different nodes under the heterovariance situation, and improving the sampling efficiency. The limitation that the sampling number of each node is equal in the prior art is broken, the sampling cost under the heterovariance condition is greatly reduced, and the calculation efficiency of the distributed random optimization method under the heterovariance condition is greatly improved; meanwhile, the convergence of the provided method is theoretically guaranteed, and the reliability of the distributed optimization sampling method is improved.

The invention discloses a mixed subspace adaptive Riemannian gradient descent target detection method, and belongs to the technical field of underwater sonar signal processing and intelligent sensing. The method comprises the following steps: firstly, constructing a mixed low-rank subspace and heterovariance noise model, and modeling a background into a union set of a plurality of subspaces; then, calculating the posterior responsibility degree of observation data belonging to each subspace mode by using a soft distribution mechanism of temperature control; thirdly, based on a noise variance contour of online learning, a sparse foreground signal is accurately recovered by adopting a weighted soft threshold operator; and finally, introducing an adaptive gain control strategy, updating the subspace basis matrix on line through Riemannian gradient descent, and dynamically adjusting noise parameters. The method has the advantages of being high in heterovariance noise robustness, capable of automatically adapting to abrupt change of the background environment, high in detection precision and the like, and is suitable for real-time detection and tracking of underwater sonar small targets.

Aiming at the problems of high volatility, nonlinearity and long-term memorability of coffee future prices and difficulty in capturing implicit modes in residual errors by a traditional model, the invention provides a method for estimating coffee future prices based on VMD (variational mode decomposition) and GARCH (generalized autoregressive conditional hetero-variance). In the first stage, noise reduction is performed through VMD decomposition, a trend component, a periodic component and a random component are extracted, fluctuation aggregation is described by using GARCH, and then primary multi-day prediction is obtained through LSTM construction; in the second stage, the residual error in the first stage is input into LLMs according to a controllable format for fitting correction so as to mine covered small-amplitude structural information. Through rolling window verification and ablation experiments, the method is remarkably superior to various baselines in RMSE, MAE, MAPE and direction accuracy, the prediction precision, robustness and economic feasibility are improved, and the method can be used for providing a scientific basis for multi-day price prediction of coffee futures and can also be used as a new mode innovation reference idea for price prediction of agricultural and sideline products.

This application discloses an integrated assessment method and device for multivariate seismic vulnerability of tunnels that incorporates physical priors, relating to the technical field of seismic analysis and seismic risk probability assessment of underground engineering structures. The method includes: generating predicted samples based on a logarithmic domain joint distribution description; inputting each predicted sample into a heteroscedastic moment estimation substitution model to obtain predicted values ​​of the conditional mean and conditional variance of the logarithmic domain demand response; reconstructing and dynamically anchoring the conditional marginal distributions of each response based on the predicted values ​​and residual information, and coupling each anchored conditional marginal distribution to obtain a multi-response joint probability model; calculating the point-state failure probability of each projected sample under a preset damage state based on the multi-response joint probability model, constructing discrete sample pairs; and fitting each discrete sample pair to obtain seismic vulnerability curves under each preset damage state. This application improves the accuracy, robustness, and engineering usability of tunnel seismic vulnerability assessment.

The invention discloses a stepwise regression analysis method-based college student physique test score prediction model, and belongs to the technical field of sports and metering economics. The method comprises the following steps: collecting and preprocessing college student physique test multi-dimensional index data; constructing a multiple linear regression model between the total physical test score and a plurality of explanatory variables; screening significant variables by adopting a stepwise regression analysis method, and optimizing a model structure; performing multi-collinearity, heteroscedasticity and sequence correlation test and correction on the model; and finally verifying the validity of the variable set by using a random forest model. According to the method, key influence factors can be automatically identified from numerous physical indexes, a prediction model with high goodness of fit and high interpretation is constructed, accurate prediction of student physical scores is realized, and a scientific basis is provided for physical health management of colleges and universities.