44 results about "Predictive regression" patented technology

Embodiments of the invention utilize advanced statistical data analytics to predict and provide recommendations for root-cause analysis for service access QoE issues in networks, such as 3G/4G networks. Using FCAPS data as predictor variables, embodiments are configured to set up the problem as a predictive regression or classification problem to estimate service access QoE related indicators. Some embodiments perform training and tuning of various non-linear statistical modelling algorithms, based for example on tree and ensemble methods, using network deregistration information from RAN logs.

The invention provides a method and a system for predicting transaction per second (TPS) transaction events of a bank background. The method includes acquiring transaction data of the bank background, and extracting TPS data and TPS data features from the transaction data of the bank background, wherein the TPS data features refer to features of each moment, and the features are formed by linking various relevant information extracted from the transaction data of the bank background; selecting a random forest model, and training the random forest model according to the TPS data to obtain a trained TPS transaction event trend prediction regression model; inputting test set data into the TPS transaction event trend prediction regression model to perform TPS transaction event trend prediction; displaying TPS transaction event trend prediction results in a graphic manner. According to the method and the system for predicting TPS transaction events of the bank background, a reference can be provided for improvement of bank background service, and advice can be given for decision of bank fault elimination methods.

The invention provides a power grid long-term load characteristic predication method based on variation of an electricity consumption structure. The power grid long-term load characteristic predication method comprises the following steps: predicating a typical daily average load of all electricity consumption departments in each season in a target year; calculating a typical daily average load of the whole society in each season in the target year; predicating a typical daily load rate of all the electricity consumption departments in each season in the target year; calculating the maximum typical daily load of all the electricity consumption departments in each season in the target year; predicating the concurrence coincidence factor of the maximum typical daily load of all the electricity consumption departments in each season in the target year; calculating a typical daily load rate of the whole society in each season in the target year; constructing a predication regression model of a typical daily peak valley rate of the whole society in each season in the target year; and predicating the typical daily peak valley rate of the whole society in each season in the target year. The power grid long-term load characteristic predication method can reasonably predicate the typical daily average load rate and typical daily peak valley rate of a regional power grid in the whole society in each season so as to provide reference evidences for electricity market analysis and power grid planning workers to meet the requirements of a reasonable foresight plan of the power grid according to a regional long-term load characteristic variation principle.

The invention discloses a prediction method for the net photosynthetic rate of a population and aims to overcome the problems that it is difficult to establish a prediction regression equation for the net photosynthetic rate of a population, considerable related data are needed in establishment of a prediction model for the net photosynthetic rate of the population and prediction accuracy is not high. The method comprises the following steps: 1, acquiring proportioning data of spectral radiance of every wave band of visible light in a region, wherein a portable multispectral radiometer is installed at a height H higher than the canopy of a plant in the test region with an area of S, the proportioning data M_D of spectral radiance of every wave band of visible light in the test region in different periods is obtained, and [0,1] normalization processing is carried out; 2, acquiring data of the net photosynthetic rate of the population; 3, constructing a bionic kernel function; 4, establishing an SVM training set and an SVM prediction set; 5, carrying out tool selection and parameter optimization on a prediction model; and 6, predicting the net photosynthetic rate of the population, which comprises obtainment of the prediction model Model, obtainment of a predicted value Predict of the net photosynthetic rate of the population and determination of reliability of the prediction model Model.

This patent concerns novel technology for detecting backdoors in neural network, particularly deep neural network (DNN) classification or prediction/regression models. The backdoors are planted by suitably poisoning the training dataset, i.e., a data-poisoning attack. Once added to an input sample from a source class of the attack, the backdoor pattern causes the decision of the neural network to change to the attacker's target class in the case of classification, or causes the output of the network to significantly change in the case of prediction or regression. The backdoors under consideration are small in norm so as to be imperceptible to a human or otherwise innocuous/evasive, but this does not limit their location, support or manner of incorporation. There may not be components (edges, nodes) of the DNN which are specifically dedicated to achieving the backdoor function. Moreover, the training dataset used to learn the classifier or predictor/regressor may not be available. In one embodiment of the present invention, which addresses such challenges, if the classifier or predictor/regressor is poisoned then the backdoor pattern is determined through a feasible optimization process, followed by an inference process, so that both the backdoor pattern itself and the associated source class(es) and target class are determined based only on the classifier or predictor/regressor parameters and using a set of clean (unpoisoned) samples, from the different classes (none of which may be training samples).

The invention discloses a brain age prediction method based on a 3D convolutional neural network. The method comprises the following steps: performing stratified sampling on brain nuclear magnetic resonance imaging data; inputting the brain nuclear magnetic resonance imaging data subjected to stratified sampling into a 3D convolutional neural network through multiple threads for training, and extracting feature data; constructing a brain age prediction regression model according to the extracted feature data; and outputting a brain age prediction result according to the brain age prediction regression model. According to the method and the device, the technical problem of inaccurate children brain age prediction caused by information loss in the aspect of feature selection of a traditionalmachine learning model in related technologies is solved, and the technical effect of improving the accuracy of children brain age prediction is achieved.

The invention provides a pipeline detection system and method based on deep learning and an unmanned aerial vehicle, and belongs to the field of industrial robots. A ground station part comprises a data management module and a first wireless communication module. An airborne part comprises a second wireless communication module, a visible light camera, an infrared camera, a detection system airborne control part and a memory. The method comprises steps of adopting a bilateral filter to denoise an image; carrying out edge detection on the image by adopting a Canny operator and then mapping theimage back to the original image to carry out sharpening operation; simplifying the convolution and pooling operation into a feature map which can be recognized by a feature extraction network; constructing an RPN network to perform prediction regression on a target box in the feature map; carrying out standard post-processing through SoftNMS, and reserving a prediction box with the highest prediction score as detection output; and generating a target mask image, namely a final pipeline detection image. The oil pipeline electric leakage identification accuracy and efficiency are high; automatic detection of the oil pipeline is realized, the labor cost is saved, and the working efficiency is improved.

The invention provides a method and a device for predicting regression test failure on the basis of the repair deficiency change, and relates to the technical field of source code change analysis. The method and the device are used for solving the problem that in the prior art, corresponding test cases capable of causing the regression test failure cannot be found. The method comprises the following steps of: obtaining influence factors through the analysis on the repair deficiency change history; selecting feature vectors from the influence factors to build a feature model; performing machine learning on the built feature model by a Logistic regression model to obtain a prediction model; using the prediction model for predicting whether the repair deficiency change can cause the regression test failure or not; and recommending the test cases capable of causing the regression test failure through the analysis on a static call graph. The method and the device provided by the invention are suitable to be used for providing concrete test cases causing regression test failure during the repair of deficiencies in software source code.

The invention discloses a water quality abnormal event identification and early warning method based on pipe network multivariate water quality time series data, and belongs to the technical field ofwater supply pipe network water treatment. Firstly, water quality data, collected by SCADA, of monitoring points are preprocessed, and simulated water quality abnormal event data are simulated; and secondly, a prediction regression model is established for a plurality of water quality indexes in the normal operation state, well-trained regression prediction models of each water quality index are integrated to construct a final regression prediction model. Thirdly, a standard deviation of residual distribution of a predicted value and a true value of each water quality index is determined, theregression prediction model is evaluated, and an optimal arithmetic multiplier is determined. Finally, the probability of the water quality abnormal event is updated by utilizing a time sequence Bayesian principle, an event alarm is given, and an alarm signal of a final model, the occurrence probability of the water quality abnormal event and an abnormal water quality index are given. The method has the advantages of low operation cost, simple operation, good effect and the like, and can greatly reduce the false alarm rate and the missing report rate.

The invention discloses a learning model for predictive regression and an implementation method. The model comprises a similarity calculation module used for carrying out similarity calculation on target enterprise data input into the model to obtain similar enterprises; a scoring module used for scoring the similar enterprises and screening out similar enterprises in preset interval scores; and an output module used for outputting a preset processing result of the similar enterprises in the preset interval score. The technical problem that experts or consulting institutions are poor in effectof processing mass enterprise data is solved. The health diagnosis and development prediction result of the unknown target enterprise can be generated according to the similar enterprise track. In addition, the method is suitable for a scene of predicting and diagnosing enterprise development in the industry through a machine learning model.

The invention discloses an image color difference detection method based on feature perception and multi-channel learning, and the method comprises the steps: 1), constructing a training set for training a color difference detection network which is composed of a multi-channel learning module, a feature perception module, a region suggestion network and a prediction regression network; 2) inputting the image into a multi-channel learning module to obtain a comprehensive feature map of the image; 3) inputting the image comprehensive feature map into a feature sensing module to obtain a sensing weighted feature map; 4) inputting the perceptual weighted feature map into a region suggestion network to obtain a block feature map; 5) inputting the block feature map into the prediction regression network to obtain a chromatic aberration offset and a position, calculating loss with a true value, and performing back propagation to adjust parameters; 6) iteratively training to a preset value, and determining a color difference detection network; and 7) inputting the to-be-detected image into the chromatic aberration detection network to obtain chromatic aberration offset and position. According to the invention, high-speed and high-precision chromatic aberration detection of images with complex textures and patterns can be realized.

The invention discloses an intelligent equipment early warning method based on a multi-input multi-output ResNet network. A deep residual network ResNet is used to process the data prediction regression problem. Through the reasonable construction of the training data set and the ResNet network, one network can perform prediction regression on multiple features at the same time. The number of training parameters is reduced, and the training speed and efficiency are improved. The method comprises the following steps: selecting equipment-related historical data, and preprocessing the equipment-related historical data; constructing a multi-input and multi-output ResNet network according to the multi-input and multi-output ResNet; training the multi-input multi-output ResNet network by using the preprocessed equipment related historical data to obtain an intelligent early warning model, and analyzing a training result to obtain a residual threshold; collecting relevant real-time data of the equipment, calculating a predicted value by utilizing the intelligent early warning model, and judging the running state of the equipment according to a residual error threshold value. The method isused for intelligent early warning of equipment.

Predictive regression models are widely used in different domains such as life sciences, healthcare, pharma etc. and variable selection, is employed as one of the key steps. Variable selection can be performed using random or exhaustive search techniques. Unlike a random approach, the exhaustive search approach, evaluates each possible combination and consequently, is a computationally hard problem, thus limiting its applications. The embodiments of the present disclosure perform i) parallelization and optimization of critical time consuming steps of the technique, Variable Selection and Modeling based on the Prediction (VSMP) ii) its applications for the generation of the best possible predictive models using input dataset (e.g., Blood Brain Barrier Permeation data) and iii) business impact of predictive models that are requires the selection of larger number of variables.

The invention discloses a partial least square method-based pipeline corrosion prediction method under metro stray current. The method comprises the steps of 1, selecting a test point, and performing finite element simulation; 2, preprocessing data ; 3, obtaining a training sample; 4, selecting a first principal component pair; 5, carrying out regression of dependent variables and independent variables to principal components; 6, carrying out residual matrix replacement; 7, establishing a partial least square regression equation; 8, verifying cross validity; 9, removing redundant position variables, and selecting optimal sensor arrangement; and 10, testing the test sample. According to the method, the effective pipeline corrosion prediction current density is obtained through earth surface potential data obtained in the interval in combination with finite element simulation, the prediction regression equation is obtained according to the partial least square method, the corrosion condition of a pipeline is judged, the number of sensors is reduced, meanwhile, the prediction precision can be kept within the acceptable range, and the method has great practical significance for intuitively monitoring the corrosion condition of the pipeline and saving the cost.

The invention relates to a certificate bill positioning detection method based on a numerical prediction regression model. The method comprises the following steps: (1) obtaining a training sample; (2) constructing a numerical prediction regression model, wherein the numerical prediction regression model comprises a lightweight neural network and a spatial transformation network which are connected in series, the input of the lightweight neural network is a to-be-positioned image, the output of the lightweight neural network is a feature convolution graph, the input of the spatial transformation network is the feature convolution graph, and the output of the spatial transformation network is coordinates of four key points of a certificate bill in the to-be-detected image; (3) designing a loss function; (4) training a numerical prediction regression model by using the training sample in the step (1); (5) inputting a to-be-positioned image into the trained numerical prediction regressionmodel, and obtaining coordinates of four key points of the certificate bill in the to-be-detected image; and (6) selecting a certificate bill image according to the coordinate circles of the four keypoints of the certificate bill. Compared with the prior art, the method is accurate and reliable in result.

The embodiment of the invention provides a station advertisement media resource value and income prediction regression method and a prediction model. According to the station advertisement media resource value and income prediction regression method and the prediction model of the invention, the production and operation data of a high-speed rail station are acquired from different data sources totrain a regression prediction model; training and testing are combined in a training process; an optimal regression algorithm is searched in a cross validation mode; parameter optimization is carriedout on the regression algorithm; and therefore, a scientific and feasible railway operation data analysis and evaluation prediction model can be established, more effective support is provided for theoperation and development of railway media advertisements, and the problem that station advertisement media resource value and income prediction lacks a scientific method system in the prior art is solved.

The invention discloses a regression model construction optimization method and device, a medium and a computer program product, and the method comprises the steps: obtaining feature data from each preset data source, and predicting the feature data of each preset data source based on each preset regression parameter prediction model and the data source missing probability of each preset data source, respectively generating prediction weight parameters corresponding to the feature data; generating a prediction regression value based on each prediction weight parameter and each feature data; and constructing a target regression model by optimizing each preset regression parameter prediction model based on the real regression value and the prediction regression value corresponding to each feature data. The technical problem of low regression model construction efficiency caused by data source missing is solved.

The invention discloses an ADHD pathogenic subcutaneous nucleus prediction method, system and device and a storage medium. The method comprises the following steps: acquiring a first magnetic resonance image data set of each subtype ADHD case, a second magnetic resonance image data set of a normal child subcutaneous nucleus and first scale data of each subtype case; carrying out subcutaneous nucleus structure covariant analysis on the first magnetic resonance image data set and the second magnetic resonance image data set to obtain a first abnormal subcutaneous nucleus; obtaining a first contribution score according to the two subcutaneous brain area volumes of the first abnormal subcutaneous nuclei; obtaining a trained prediction regression model according to the first contribution score and the first scale data; and inputting a first magnetic resonance image of an ADHD case to be predicted into the prediction regression model to obtain a prediction result of the subcutaneous nuclei. The method can provide effective scientific basis for clinical treatment and rehabilitation decision. The method can be widely applied to the technical field of medicine.

The embodiment of the invention discloses a method and a device for detecting an iron receiving state. The method comprises the following steps: acquiring multi-channel iron notch video stream data; decoding the multi-channel iron notch video stream data to obtain a plurality of frames of iron notch images, each iron notch corresponding to at least one frame of iron notch image; the multiple frames of iron notch images are input into a target recognition and detection network, iron flow detection information of at least part of iron notches is obtained through the target recognition and detection network, and the target recognition and detection network comprises a multi-scale feature extraction module, a channel attention module and a prediction regression module; the multi-scale feature extraction module is configured to obtain feature maps of different scales of the same iron notch based on an input iron notch image, and the channel attention module is used for enhancing channel attention of the feature maps. And the prediction regression module is configured to obtain the iron flow detection information of the at least part of iron notches according to the feature map output by the channel attention module.