42 results about "Quartile" patented technology

The invention provides a model and data driven low-orbit navigation enhanced satellite clock error forecasting method and equipment, and the method comprises the steps: carrying out the preprocessing of original clock error data of a low-orbit satellite: converting clock error time domain data into frequency domain data through time-frequency conversion, and employing a quartile method to recognize and eliminate frequency domain outliers and corresponding time domain abnormal values; constructing a frequency domain energy model and extracting periodic terms, including performing power spectral density analysis on the preprocessed clock error frequency domain data, and adaptively extracting significant periodic components of the low earth orbit satellite clock error through a threshold value; establishing a polynomial low earth orbit satellite clock error forecasting model considering periodic term correction and forecasting a clock error sequence to obtain a corresponding fitting residual error sequence and a clock error forecasting value; and normalizing the fitting residual error sequence, inputting the fitting residual error sequence into the gating circulation unit neural network, training and forecasting by adopting a window sliding input mode, and outputting a residual error prediction result. According to the method, the interpretability of the forecasting result can be effectively improved while the low earth orbit satellite clock error forecasting precision and stability are improved.

The invention discloses a relay protection method and device based on time sequence data prediction, and the method comprises the steps: collecting the operation sequence data of a main network power transformation relay protection device in real time through a wireless sensing node, and transmitting the operation sequence data to a data processing layer through a Zigbee wireless network protocol; historical data features are extracted based on a sliding window method, abnormal data are recognized and removed through a quartile method, and missing data are filled up through a linear interpolation algorithm; constructing a CNN-GRU prediction model fused with the attention mechanism, dynamically weighting key features by the attention mechanism, inputting the preprocessed data into the model for training, and generating an operation state evaluation value based on a prediction result, and when the evaluation value exceeds a preset threshold, triggering a multi-level early warning mechanism. And the prediction precision is verified by using a decision coefficient R2, the R2 is required to be greater than or equal to 0.95, and model parameters are dynamically updated to adapt to the working condition change of the power grid.

The invention discloses an interpretable load prediction method based on knowledge distillation and a time self-attention additive neural network, and the method comprises the steps: collecting the historical load and meteorological data of a power grid as the input characteristics of a model, carrying out the detection of a data quartile abnormal value, dividing the data quartile abnormal value into a training set, a test set and a verification set, and carrying out the detection of the data quartile abnormal value; standardization and abnormal value filling are carried out through Z-shaped orthogonalization and linear filling, and finally, a tensor form meeting the model input requirement is converted through a sliding window; designing a knowledge distillation'teacher-student 'framework based on multiple scales and multiple cycles; constructing a time self-attention additive neural network TSA-NAM as a student model; calculating a shape function representing the contribution degree and the characteristic value in the sub-network to obtain the interpretability of the characteristic dimension; exporting the attention weight of the time self-attention module to obtain the interpretability of the time dimension; performing simulation verification; according to the method, high reliability and high precision are guaranteed, and meanwhile, multi-dimensional interpretability is brought to power load prediction.

The invention discloses a multi-stage wind power abnormal data combination cleaning method, and belongs to the field of new energy power generation data processing. Aiming at the problems of various types of abnormal values of wind power original data, serious interference and low cleaning precision, and single anomaly detection means, easy missing detection and misjudgment and the like in the prior art, the invention provides a staged combined cleaning strategy, which comprises the following steps of: firstly, dividing equal interval sections of wind speed and power, and eliminating isolated point type anomalies in distribution by using double quartile analysis; a CFSFDP density peak value clustering algorithm is introduced, low-density anomaly clusters are mined according to a density-distance joint criterion, and the recognition capability of structural aggregation anomaly is improved through two rounds of clustering; performing segmentation modeling on a wind speed-power relation by using upper and lower envelope line fitting based on a function, and removing envelope outer drift type noise; and finally, complementing edge missing data by using an interpolation algorithm. According to the method, the systematicness, precision and adaptability of the wind power data cleaning process are remarkably enhanced, high-quality data are provided for subsequent power prediction and energy optimization scheduling, and the application prospect is wide.

The invention relates to a false comment detection method, belongs to the technical field of Internet information security, and mainly solves the problems of incomplete data acquisition, single feature modeling, slow model updating and rigid threshold judgment in the prior art. According to the method, comment text data are collected in real time through a distributed crawler system, non-language characters are filtered through a preprocessing module, the text length is standardized, and a three-dimensional feature set containing semantics, user behaviors and time-space association is extracted; a double-layer dynamic classification model is adopted for training; dividing a training set and a verification set based on a sliding time window, starting incremental learning and adjusting parameters of a full connection layer when the cosine similarity of new data and historical features is lower than 0.7, calculating weighted values of median and upper quartile in real time through a dynamic threshold module to judge abnormal comments, and generating a structured detection report with a hash check code. The method is suitable for real-time monitoring and risk management and control of online platform false comments, and the detection efficiency and accuracy are improved.

The invention discloses a traffic flow prediction method based on a multi-scale time window adaptive graph bias neural network, and belongs to the field of deep learning and intelligent traffic. The method comprises the following steps: collecting flow, speed and occupancy data of traffic network nodes, and constructing a historical data set; constructing a time embedding module, and extracting intra-day and intra-week time features; constructing a multi-scale time window trend sensing module, and capturing short-term fluctuation and long-term trend of the traffic flow; a time condition adaptive graph bias module is constructed, a graph bias matrix is dynamically generated according to the time context, and time-varying spatial dependence is modeled; a sparse space attention module is constructed, the calculation complexity is reduced, and spatial-temporal features are fused; and a self-adaptive double-end output module is constructed, a gating coefficient is generated based on the volatility and the quartile distance, and linear and nonlinear branches are fused to output a multi-step prediction result. According to the method, multi-scale spatial-temporal feature fusion and adaptive prediction of the traffic flow are realized, and the prediction precision and robustness are improved.

The invention discloses an electric energy meter metering misalignment judgment method based on multi-source data and consistency check, and belongs to the technical field of intelligent power grid metering monitoring. The method comprises the steps of collecting electric energy meter daily electric quantity, transformer area total electric quantity and user load characteristic data; identifying and correcting data exception by adopting a sliding window and a self-adaptive quartile distance method; constructing a multivariate linear fitting model of the total electric quantity of the transformer area and the electric quantity of each sub-meter, and calculating the average residual contribution degree of each electric energy meter through least square estimation and residual analysis; and calculating a metering deviation index in combination with electric quantity volatility, judging a suspected misalignment electric energy meter based on a confidence interval, and controlling a list length and effectiveness according to an accumulated contribution ratio threshold. Remote, real-time and automatic monitoring of mass electric energy meters is realized, the defects of high cost, long period and high false alarm rate of a threshold value method in traditional manual checking are overcome, and the accuracy, robustness and interpretability of metering misalignment judgment are remarkably improved through multi-source data fusion and statistical checking.

The present application relates to the technical field of civil structure monitoring, and discloses a setting method of dynamic threshold of civil structure monitoring data and a safety warning method. The setting method first collects all monitoring data in a previous preset time period of a sensor, fits the serial number corresponding to each monitoring data with the monitoring data according to the time sequence of data collection, and obtains a fitting curve. Then, the monitoring data is subtracted from the data of the corresponding serial number in the fitting curve in sequence, a group of data without trend items is obtained, the absolute value is taken, the normal distribution is calculated, the data in the upper and lower quartile ranges are selected and the mean value is taken as the accurate value, the data is fitted again through comparison and replacement, and the fitting curve under the influence of external load can be obtained. Finally, the monitoring data of a subsequent preset time period of the sensor is numbered in the same way, the monitoring data center curve of the future time period is obtained, and the dynamic threshold interval is generated accordingly. The present application can set appropriate dynamic thresholds for the civil structure monitoring data of different measuring points.

This application relates to the field of equipment performance testing and evaluation technology, and provides a method and system for equivalent reduction of equipment test sample size. By constructing a data fusion evaluation model based on actual and numerical test data, the fidelity of actual test response estimation under small sample conditions is improved. An equivalent measure is designed and calculated by fusing the mean square error of the evaluation model from a single data source with a preset test confidence level, and a box plot of the equivalent measure from a single data source is obtained. By connecting the lower quartile and upper quartile of each obtained box plot, equivalent composite confidence intervals for actual and numerical simulation tests are plotted. Based on these equivalent composite confidence intervals, the theoretically required actual test sample size is equivalently reduced to the numerical simulation test sample size, achieving an equivalent reduction of the actual test sample size by the numerical simulation test, thus solving the problem of equipment performance evaluation under insufficient actual test samples.

The invention discloses a multi-strategy large language model common disease prediction method based on heuristic global perception, and the method comprises the steps: selecting four classical graph theory evaluation indexes, analyzing the index values of all edges in a disease network, and setting the value ranges corresponding to an average value, a median, an upper quartile and a lower quartile as threshold values; and constructing a 12-dimensional binary value feature vector of each edge based on the threshold value as a global feature, inputting the global feature into an SGD selector for training, and fixing parameters. And then, by taking a global graph theory feature prediction result as guidance, adopting a large language model RAG framework and combining with multi-strategy literature retrieval, and realizing interpretable output through an improved COT cue word. And performing dynamic selection on prediction results of global and local features through a linear regression model by adopting a supervised training mode. A large number of experiments verify the effectiveness of the method: the performance of the LLMs in a disease link prediction task is significantly improved by fusing the global and local features of the prediction edge.

This application relates to the field of model training technology, specifically to a semi-real-time prediction model training method based on hierarchical triggering of data distribution drift. The method includes: extracting feature values ​​from transmission current signals to train a fault prediction model for real-time prediction of fault feature values ​​at all upcoming data collection times within a preset time period; calculating distribution drift evaluation values ​​and persistence evaluation values ​​to assess the significance of data distribution drift persistence within a preset adjacent time period before the current data collection time; statistically analyzing the quartiles of historical persistence evaluation values ​​to perform semi-real-time update training of the fault prediction model using hierarchical triggering; and using the real-time updated fault prediction model to predict transmission line faults. This application aims to improve the real-time accuracy of fault prediction models in predicting transmission line faults, thereby more effectively ensuring the safety and reliability of transmission lines.

The application relates to the technical field of wireless communication, and discloses a communication method and device for impulse noise suppression based on abnormal point detection, wherein at a receiving end, the method receives a transmission signal, divides the transmission signal into four parts according to a quartile algorithm, determines upper quartiles, a median and lower quartiles, then determines a normal value interval according to the upper quartiles, the median and the lower quartiles, detects abnormal signals disturbed by impulse noise in the transmission signal based on the normal value interval, repairs the abnormal signals, and obtains a de-noised signal.In the embodiment of the application, the transmission signal features and the quartile algorithm are comprehensively utilized to more accurately locate the impulse noise disturbance and improve the suppression effect.In addition, through the quartile-based algorithm, the calculation amount is small, the abnormal signals can be quickly and accurately detected and repaired, the operation is simple, the impulse noise disturbance can be effectively suppressed, the error rate of the system is reduced, and the reliability of data transmission is improved.

The invention belongs to the technical field of rail transit cost assessment, and particularly relates to a rail transit project cost intelligent assessment method based on big data, which comprises the steps of inputting a cost file of a project to be assessed, including sub-items, a list, a quota and labor, material and machine information; performing index physical examination early warning, matching historical sample data, rejecting abnormal values through quartile, calculating a confidence interval, and detecting cost deviation; checking the features, comparing with a standard feature configuration library, and verifying the integrity and normalization of the engineering features; through cooperation of index physical examination early warning and feature examination, the problem of engineering type recognition errors can be quickly responded, so that the defects of a static model are overcome, construction cost files with engineering type labeling errors can be quickly responded, the false alarm frequency can be reduced, and the defects of the construction cost files can be quickly analyzed based on a multi-stage examination mechanism. Compared with manual checking, the method and the device have the advantages that the speed and the efficiency are higher, and the checking is more comprehensive.

The application discloses an online fault diagnosis method for a hydro-generator based on a potential interval distance of a through screw, relates to the technical field of generators, and is suitable for detecting a stator winding ground fault. Voltage collection devices are additionally arranged on a plurality of through screws of the hydro-generator, real-time induced potential is acquired, a historical data matrix is established, and a data buffer matrix is updated at interval time; the data buffer matrix is divided into a data evaluation matrix, a fundamental wave and an odd harmonic phase are extracted through Fourier decomposition, a harmonic phase matrix is constructed, a phase difference matrix is obtained by calculating normalized phase differences of adjacent columns, a quartile distance is solved after being converted into a column vector array, a dynamic quartile distance array is formed, abnormal limit values are determined by combining a box plot principle and an error level of the voltage collection device, and faults are determined by traversing the array. The method does not need additional sensing devices and traditional threshold setting, is convenient to operate, has low implementation difficulty, and can accurately and timely identify faults.

The invention discloses an abnormal monitoring data identification and restoration method based on an improved box type method, and the method comprises the steps: S1, collecting a monitoring data segment with a fixed length, building a data label through a box type graph in combination with the analysis features of a scatter diagram, and obtaining the features of the box type graph; step S2, calculating four statistical parameters of each segment of data according to the box diagram characteristics and the maximum value, the minimum value, the median, the lower quartile and the upper quartile of the display data; s3, calculating a relative difference rate by taking a statistical index of fault-free data in the same structure monitoring time sequence as a standard; s4, identifying fault data segments according to the relative difference rate of the data segments, and classifying the fault data segments; s5, adopting different repair strategies according to different fault data segments; and S6, finally, verifying the repair effect of repairing the fault data segment by adopting the repair strategy. According to the invention, accurate identification and classification of various types of abnormal data are realized.

The invention discloses an adaptive integrated empirical mode decomposition method based on a quartile difference algorithm. The adaptive integrated empirical mode decomposition method comprises the following steps: inputting an initial decomposition signal; initializing a value range for adding white noise amplitude ratio coefficients and an amplitude ratio coefficient interval value, and gradually increasing the white noise amplitude ratio coefficients in the value range by taking the quartile difference of the initial decomposition signal as a reference to obtain white noise corresponding to each white noise amplitude ratio coefficient; sequentially adding white noise into the initial decomposition signal from small to large to obtain a noise-added decomposition signal; solving a local maximum value point position and a local minimum value point position of the noise-added decomposition signal to obtain a sequence formed by maximum value points and a sequence formed by minimum value points; calculating a quartile difference between the maximum value point sequence and the minimum value point sequence by adopting a quartile difference calculation method to obtain an extreme point distribution characteristic of the noise-added decomposition signal and a corresponding change relation curve between the extreme point distribution characteristic and a coefficient; obtaining the optimal value of the added white noise amplitude ratio coefficient according to the corresponding change relation curve; and performing integrated empirical mode decomposition on the initial decomposition signal by using the optimal value to obtain an n-order intrinsic mode component. By using the method, the purpose of adaptively optimizing decomposition parameters according to actual decomposition signals in integrated empirical mode decomposition can be achieved, signal extreme points are distributed more uniformly, the fluctuation degree is as small as possible, mode aliasing is reduced, and adaptive decomposition of nonlinear and non-stationary signals is achieved.

The present application relates to a shadow measurement method for building height inversion, belonging to the technical field of geospatial measurement. The present application realizes the optimal division of the shadow by combining the fishing net method and multiple constraint conditions; then the shadow length values of all the divided regions are counted, and the quartile method and the bidirectional approximation strategy are used to determine the optimal value of the shadow; finally, the optimal values of all the regions are comprehensively evaluated to determine the shadow length. The measurement method fully considers the optimization of the measurement region, the optimization of the statistical data, and the optimization of the optimal value, can maximize the reduction of the influence of the non-smooth shadow boundary and the shelter on the measurement results, improve the accuracy and robustness of the shadow method for calculating the building height, and thus realize the large-scale and large-area measurement of the urban building height, and provide basic research data for the inversion and expansion of the urban height.

The invention relates to the technical field of data processing, in particular to an intelligent measurement and control method and system for a high-voltage switch cabinet. The method comprises the following steps: calculating a deviation degree of an evaluation parameter of a node according to the evaluation parameter of the node and a quartile on historical data of the evaluation parameter of the node; obtaining the risk score of the task at the node according to the historical fault rate and the deviation degree of each evaluation parameter of the node and the priority of the task; calculating taboo duration of the task at the node according to the risk score of the task at the node and the importance degree of the node; in response to the fact that the risk score of the task at the node is larger than the risk threshold value, when the target path for transmitting the task is obtained based on the ant colony algorithm, the node in the communication network node graph is added into the taboo table based on the taboo duration, intelligent measurement and control of the high-voltage switch cabinet are achieved, and the measurement and control efficiency of the high-voltage switch cabinet is effectively improved.

The invention relates to an underground water environment risk assessment method based on big data, and relates to the technical field of underground water monitoring and pollution risk assessment. According to the method, water quality parameter data filling is carried out based on a big data pre-trained water quality parameter filling model, a Bayesian classifier or a K nearest neighbor mean value through adaptive missing rate and data type selection, and the calculation efficiency and prediction precision of different missing rates are balanced; a similarity function is defined by considering sampling sites and water quality parameters, repeated data screening is carried out, and data with similar water quality parameters and non-similar sites are prevented from being deleted. Clustering and quartering anomaly mixed detection is utilized to consider univariate statistical anomaly and multivariate structure anomaly, and true pollution data is prevented from being deleted by mistake. A groundwater monitoring data set is utilized, and a composite groundwater environment evaluation model is trained through a differential evolution optimization fusion weight mode to estimate a groundwater environment index value. Model complementation is utilized to reduce deviation; and the integration benefit of the model is maximized by optimizing the weight through differential evolution.

The invention discloses a meteorological science popularization work evaluation index system construction method. The method comprises the following steps: construction of an expert investigation method and an initial index library: adopting the expert investigation method to construct the initial index library; two-dimensional validity screening: introducing a consensus degree-concentration degree two-dimensional validity screening model, and establishing a two-dimensional screening matrix by calculating an average score M and a variable coefficient CV of each index; aHP-entropy weight method combined weighting: obtaining a subjective weight through an AHP analytic hierarchy process, obtaining an objective weight in combination with an entropy weight method, and then carrying out linear weighting combined weighting; nondimensionalization: performing annual data collection on the quantitative indexes, calculating quartile and range, and performing 10-score questionnaire and linear conversion on the qualitative indexes into 1-5 scores; a Pearson's correlation coefficient matrix is calculated through index correlation analysis, if the two indexes are highly correlated and reflect the same problem, one of the indexes is merged or deleted, and if different problems are reflected, the two indexes are reserved; and performing dynamic adjustment and case verification.

The invention discloses a data set construction method for post-training of a vertical large language model, and belongs to the technical field of vertical large language model training. The technical problem to be solved is that serious long-tail data distribution and noise pollution exist in a training data set after a vertical big language model in the field of hydropower construction. According to the technical scheme, the method is characterized in that a nonparametric outlier detection algorithm, namely a box plot, is adopted to regulate and control data distribution, quartile and quartile distance of a data set sample frequency are firstly calculated, outliers are identified, then quantitative data enhancement is performed on business key long-tail scarce outliers, outliers which cannot be enhanced, such as noise and acquisition errors, are directly cleared, and the accuracy of data enhancement is improved. And finally obtaining an optimized training data set.

The invention discloses a laying hen lossless data processing method based on improved SVR, relates to the technical field of data processing, and adopts an adaptive cap cleaning data based on quartile distance to dynamically adjust an abnormal value elimination boundary, avoid mistaken deletion of effective physiological signals, improve the robustness of data cleaning and improve the accuracy of data cleaning. The defect that data distribution is damaged by a traditional fixed threshold value and a simple filling method is overcome. Through a preprocessing assembly line of KNN topology filling and Yeo-Johnson manifold transformation, missing data are effectively repaired, data skewed distribution is corrected, and the problem of feature space measurement failure is solved. Based on self-adaptive feature screening of F statistics, redundant features can be accurately removed, data dimension reduction is realized, and dimensionality disaster and model overfitting are avoided. According to the established TTR-SVR model and parameter optimization strategy, optimization precision and calculation efficiency are both considered, the problem that traditional SVR hyper-parameter selection is unreasonable is solved, and the accuracy and efficiency of laying hen nondestructive monitoring data processing are integrally improved.

The application provides an unsupervised fault diagnosis method and system based on multi-scale feature enhancement and adaptive feature selection, and belongs to the field of industrial equipment fault diagnosis. In order to solve the problems that the existing feature extraction method has insufficient feature representation ability, single feature scale and serious redundant feature interference, the application extracts multi-dimensional statistical features such as quantile, quartile range and skewness, combines adaptive wavelet base selection of the information maximum redundancy minimum criterion, and significantly improves the representation ability of the multi-scale characteristics of the fault signal. The cosine similarity weighted Laplace score and variance double threshold mechanism are introduced, and the redundant features are effectively eliminated. Through the collaborative design of multi-scale interval enhancement and adaptive wavelet transform, the time-frequency domain features are effectively extracted, the time consumption of high-dimensional data processing is obviously reduced compared with the traditional elastic distance method, and the real-time demand of industrial online monitoring is met.

The invention discloses a credit overdue prediction analysis method and system based on a knowledge graph, and the method comprises the steps: extracting entity network structure and business association two-dimensional features based on the credit knowledge graph, forming a numerical feature set, matching overdue labels, and removing invalid samples; through double-index two-stage screening of core features, feature discretization is completed in combination with quartile and a credit business threshold value; effective association rules are mined by focusing on overdue labels to form a rule base, and a linkage judgment mechanism of'rule matching priority + GNN model understanding 'is adopted. According to the method, the problems of incomplete feature coverage, single screening, poor adaptability and the like in traditional prediction are solved, the prediction accuracy and the coverage are considered, the rule mining efficiency and the feature quality are improved, and reliable technical support is provided for credit decision-making of financial institutions.

The invention relates to the technical field of signal-to-noise ratio detection, in particular to a medical endoscope signal-to-noise ratio detection method, device and equipment and a storage medium. Obtaining brightness channel average quartile difference data, first color difference channel average quartile difference data and second color difference channel average quartile difference data from the image data according to a preset first percentile and a preset second percentile; calculating the average quartile difference data of the brightness channel, the average quartile difference data of the first color difference channel and the average quartile difference data of the second color difference channel according to a preset total noise synthesis formula to obtain total noise data; drawing a signal-to-noise ratio curve according to the total noise data; according to the scheme, multi-frame acquisition is combined with a clinical working distance, a uniform gray scale region and slight focusing blur, the noise data purity is guaranteed, the quartile difference can shield an abnormal value, non-Gaussian noise is adapted, three-channel noise is fused to meet the requirement, the extracted signal noise is traceable, and the evaluation accuracy is improved.

The invention relates to the technical field of network security, in particular to an intrusion detection method based on density divergence clustering, which comprises the following steps: acquiring a data set of network intrusion detection, and performing density divergence clustering to acquire attack data; wherein the density divergence clustering comprises the following steps of: calculating Euclidean distances of all feature data points in a data set in pairs, and taking preset percentage quantiles of all the Euclidean distances as truncation distances; calculating the density of each feature data point according to the Euclidean distance and the truncation distance; according to the density, density divergence clustering is carried out on the feature data points, the feature data points are distributed to corresponding clusters, and a clustering result containing a plurality of clusters is obtained; and for a clustering result, constructing a judgment index through quartile, judging and extracting an abnormal cluster, and obtaining attack data. According to the scheme, abnormal behaviors can be identified more accurately in a complex environment, and the method has higher anti-interference capability so as to effectively detect network attacks.

The invention relates to the technical field of equipment performance test evaluation, and provides an equipment test sample size equivalent conversion method and system. By constructing a digital-real test data fusion evaluation model, the fidelity of real-installation test response estimation under the small sample condition of the real-installation test is improved. The equivalent measure is designed and calculated by fusing the mean square error of the single data source evaluation model and the preset test credibility, and a single data source equivalent measure box diagram is obtained. And drawing a real installation test equivalent composite confidence interval and a digital simulation test equivalent composite confidence interval by respectively connecting a lower quartile and an upper quartile for each obtained box type graph. On the basis of the equivalent composite confidence interval of the real installation test and the equivalent composite confidence interval of the digital simulation test, equivalently converting a real installation test sample size required by a theory into a digital simulation test sample size, so that the equivalent conversion of the digital simulation test to the real installation test sample size is realized; the problem of equipment performance evaluation under the condition of insufficient real installation test samples is solved.

The invention provides a statistical prior information-based target frame regression loss function optimization method and device, and relates to the technical field of computer vision. The method comprises the steps of performing aspect ratio extraction according to a training data set based on each target category to obtain a first aspect ratio data set, and performing outlier elimination by using a quartile method to obtain a second aspect ratio data set; performing prior information calculation according to the second aspect ratio data set to obtain first prior information, second prior information and third prior information; based on the second prior information and the third prior information, performing aspect ratio range verification by using a baseline model to obtain a verification result; based on the verification result, target frame regression loss is calculated according to the training data set, the prediction data set and the first prior information; and performing parameter optimization on the baseline model according to the target frame regression loss to obtain an optimized baseline model. The target frame regression loss function optimization method is an efficient and accurate target frame regression loss function optimization method based on statistical prior information.

The application discloses a kind of based on robust statistical characteristics of milk powder particle morphology dispersion prediction method, belong to milk powder quality detection and analysis technical field.The method is by using quartile distance and 5% tail cut mean to represent particle morphology, effectively reduce abnormal value interference, compared with traditional mean-standard deviation method can reflect the real characteristics of particle group, specifically, the present application is directed to the quartile distance and tail cut mean of 9 kinds of shape factor parameters of milk powder particle are calculated, and further based on dispersion correlation 14 key features are screened out, the reliability of feature is improved, and then the accuracy of milk powder dispersion prediction is improved possibly;Further, the present application scheme is directed to small sample application scene, using Bootstrap resampling, vertically splicing original data and 3 groups of resampling data, generate enhanced data set, and using CatBoost regression model further improves the prediction accuracy.

The application discloses a kind of dynamic residual threshold adaptive selection space-time filtering method, device, equipment and storage medium, the method includes obtaining the remote sensing time series image data of target area;By model training and predicting the possible change trend of remote sensing time series data obtains the prediction result of remote sensing time series data.Furthermore, the absolute residual between image predicted value and image true value is calculated, and the residual threshold determined based on the absolute residual median, quartile range, skewness and kurtosis is selected as the threshold of filtering method selection, specifically, for high residual area, spatiotemporal weighted filtering strategy combined with Euclidean distance is used;For low residual area, the filtering reconstruction method in time domain is used, so as to ensure the continuity of image while suppressing non-systematic noise.Compared with prior art, it has higher degree of automation and adaptability, and is especially suitable for remote sensing time series data processing in high-altitude, frequent cloud cover or complex terrain area.