67 results about "Extreme gradient boosting" patented technology

A method for predicting air quality with the aid of machine learning models includes: (A) providing air pollution data to perform an eXtreme Gradient Boosting (XGBoost) regression algorithm for obtaining a XGBoost prediction value; (B) providing the air pollution data to perform a Long Short-Term Memory (LSTM) algorithm for obtaining an LSTM prediction value; (C) combining the air pollution data, the XGBoost prediction value and the LSTM prediction value to generate air pollution combination data; (D) performing an XGBoost classification algorithm to obtain a suggestion for whether to issue an air pollution alert; and (E) performing the XGBoost regression algorithm on the air pollution combination data to obtain an air pollution prediction value. Two layers of machine learning models are built, and a situation where prediction results are too conservative when a single model does not have enough data can be improved.

The invention provides a blast furnace molten iron silicon content change trend prediction method, a blast furnace molten iron silicon content change trend prediction device and a storage medium. Themethod includes the step that characteristic parameters during an iron making process which are related to silicon content change in blast furnace molten iron are acquired, and a silicon content change trend is predicted on the basis of trained blast furnace molten iron silicon content prediction models, wherein the blast furnace molten iron silicon content prediction models include a first layerprediction model and a second layer prediction model, wherein the first layer prediction model is used for preliminarily predicting the silicon content change trend according to an original sample, and the second layer prediction model predicts the silicon content change trend for the second time according to the prediction result of the first layer prediction model so as to obtain the silicon content change trend. According to the blast furnace molten iron silicon content change trend prediction method, the blast furnace molten iron silicon content change trend prediction device and the storage medium of the invention, a limit gradient enhancement and long-term and short-term memory network fusion model is established to predict the trend of the silicon content in the molten iron, a reference basis is provided for a blast furnace operator to judge the change trend of the condition of the blast furnace and control the amplitude of the blast furnace in advance, and therefore, the smoothness of the iron making process can be ensured, and the quality of the molten iron can be maintained within a normal range.

The invention discloses a phishing website detection method based on multi-feature fusion. The phishing website on the internet can be detected in real time by the method. The URL features, the HTML features and the text vectors based on TF-IDF are firstly extracted according to the URL of the webpage; then the text vectors are classified by using Logistic regression, Logistic regression featuresare constructed and multi-feature fusion of the Logistic regression features and the webpage URL features and the HTML features is performed; and finally the XGBoost (eXtreme gradient boosting) modelis trained and classified prediction is performed on the phishing website to be detected. Multiple key features are omnidirectionally extracted from multiple dimensions based on the URL of the website, the problem of high feature dimension of the text vectors can be effectively solved by using the Logistic regression feature fusion method and the operation efficiency can be greatly enhanced in comparison with that of the existing feature fusion method; besides, the XGBoost classification model can further enhance the phishing website detection accuracy in comparison with the conventional classification model so as to reduce the missing report rate of phishing website detection.

The application discloses an invoice trade name classification method, system and equipment and a computer readable storage medium. The method comprises the following steps: determining the trade namevector of a to-be-classified invoice trade name based on preset rules; inputting the trade name vector to a pre-created target trained model to obtain an invoice trade name type output by the targettrained model, for judging if the type is a target type, wherein the target trained model is obtained by pre-utilizing target training samples to train a to-be-trained model created by the Extreme Gradient Boosting (XGBoost) model based on the gradient boosting decision tree, and the target training samples include training samples marked with target invoice trade name types and training samples obtained based on Rocchio algorithm and not marked with the trade name types. By using the invoice trade name classification method provided by the invention, the labour cost can be reduced and the accuracy of classification results can be obviously improved.

The invention discloses a detection method based on fusion of a simple neural network and an extreme gradient promotion model. The method comprises steps of obtaining a data set, carrying out the preprocessing of the data set, and dividing the data set into a training set and a test set; performing data increment operation on minority class samples in the data set to balance the data set; trainingthe fusion detection model by using the training set; wherein the fusion detection model comprises a simple neural network and an extreme gradient boosting model, and supervised learning training isperformed on the data set till the model converges; and performing intrusion detection on to-be-detected data by utilizing the converged fusion detection model to obtain an intrusion detection result.According to the method, the limitation of insufficient generalization ability of a single machine learning model in different scenes is improved, and the defect of poor association rule mining ability of machine learning for deep information is overcome, compared with a traditional method, cost of manually mining association rules is saved, data features are more effectively utilized, and the intrusion detection rate is increased.

The invention discloses a system for early detection and risk prediction of schizophrenia. The technical problem that currently, it is hard to accurately and objectively carry out early detection andrisk prediction on schizophrenia can be solved. The system comprises a test module, a collecting and preprocessing module, a data analysis module, a learning classification module and a classificationperformance evaluation method. According to the system for early detection and risk prediction of the schizophrenia, high-risk individuals, ultra-high-risk individuals and initially-attacked people of the schizophrenia are subjected to quantitative and objective detection recognition and risk prediction by comprehensively utilizing cognitive tests, clinical detection, P50 sensory gating electrophysiological characteristics and an XGBoost(the Extreme Gradient Boosting) machine learning method; specifically, by combinedly using cognition, clinic and P50 sensory gating experiments, and adoptingsource location, brain networks, machine learning and other analysis methods, initially-attacked people, high-risk people, ultra-high-risk people and normal contrast people are classified, and the classification accuracy, sensitivity and specificity are improved.

The invention discloses a wind turbine generator bearing early warning method based on feature fusion. The method comprises the following steps: preprocessing CMS data; obtaining a time domain characteristic index; obtaining trend characteristic indexes; obtaining a frequency domain characteristic index; obtaining envelope characteristic indexes; carrying out feature fusion; training an extreme gradient lifting model; training a function set in the classification model, and constructing a learning objective function of XGBoost; and calculating the learning objective function of the XGBoost. According to the technical scheme, the vibration data of the CMS state monitoring system is utilized, the generator bearing fault mechanism is analyzed from the time domain feature, the trend feature, the frequency domain feature and the envelope feature, the four features are fused, the feature vector representing the running state of the generator is effectively extracted, the method greatly improves the recognition sensitivity of the algorithm for the generator set bearing fault, achieves the fault early warning through employing a limit gradient lifting method, and is higher in accuracy.

The invention belongs to the technical field of runoff prediction, and discloses a runoff probability prediction method and system based on deep learning, and the method comprises the steps: employinga maximum information coefficient to analyze the linear and nonlinear correlation between variables, so as to screen a runoff correlation factor; building an extreme gradient boosting tree model on the basis of correlation analysis, and inputting runoff correlation factors into a trained XGB model to complete runoff point prediction; inputting a point prediction result obtained by the XGB model into a GPR model, and performing secondary prediction to obtain a runoff probability prediction result; selecting confidence and acquiring a runoff interval prediction result under the corresponding confidence through Gaussian distribution; and optimizing hyper-parameters in the XGB model and the GPR model by adopting a Bayesian optimization algorithm. A high-precision runoff point prediction result, an appropriate runoff prediction interval and reliable runoff probability prediction distribution can be obtained, and the prediction method plays a crucial role in utilization of water resourcesand reservoir scheduling.

The invention discloses a credit score integrated classification system and method based on deep learning, and the system comprises a data obtaining and preprocessing unit which is used for obtaininga credit data set, carrying out the data preprocessing to obtain a sample data set, and dividing the sample data set into a sample training set and a sample testing set; an integrated classification training unit which is used for training the sample training set through the RNN sub-unit, the LR sub-unit and the XGBoost sub-unit to obtain predicted credit probability of the sample training set through each sub-unit; and a voting unit which is used for carrying out majority voting on the three predicted credit probabilities obtained by the integrated classification training unit, if two or morepredicted credit probabilities are higher than 0.5, the customer credit is good, and otherwise, the customer credit is poor. According to the method, the deep learning algorithm recurrent neural network RNN is applied to the credit scoring problem, the logistic regression LR, the extreme gradient boosting tree XGBoost and the recurrent neural network RNN are integrated in parallel, the diversityand accuracy of the model are considered, and the performance of the model is improved.

The invention discloses a radar radiation source individual identification system based on a radar pulse sequence. The radar radiation source individual identification system comprises a radar individual pulse sequence database, a data preprocessing module, a different individual weight calculation module, a weighted extreme gradient lifting radar individual modeling module, a radar individual identification module and a radar individual identification final result calculation module. According to the invention, individual identification of the radar radiation source is realized based on the radar pulse sequence. In the radar radiation source individual identification system based on the radar pulse sequence, an original pulse sequence is used as input, wavelet decomposition is performed on the original radar pulse sequence, so that multi-scale input can be obtained. More distinguishable subtle features are mined, a weighted ensemble learning algorithm is adopted to establish a radar radiation source individual recognition model, the problem that the number of radar individual samples in a database is unbalanced can be solved, and meanwhile the method has the advantages of being high in feature extraction capacity, high in accuracy, high in modeling speed and the like.

The invention discloses a high-proportion photovoltaic power distribution network voltage prediction method based on a time convolution neural network. The high-proportion photovoltaic power distribution network voltage prediction method comprises the step of: 1, carrying out data preprocessing on original load data, namely carrying out normalization on voltage time sequence data through employinga maximum and minimum interval scaling method based on multiple time scales to obtain a complete voltage sequence; 2, constructing an input feature vector set, namely performing feature screening based on an extreme gradient boosting tree algorithm of a decision tree, constructing a training sample set, outputting each feature weight, and screening out different feature subsets in combination with the weights and the voltage prediction model condition; and 3, establishing a voltage prediction framework based on the high-proportion photovoltaic power distribution network, and training a time convolution network prediction model to obtain a voltage prediction result. According to the high-proportion photovoltaic power distribution network voltage prediction method, the extracted features are combined with the time and input into different channels of the time convolutional neural network model to obtain the prediction result, so that the purpose of remarkably improving the voltage prediction precision of the power distribution network is achieved.

The invention relates to a cloud workflow task execution time prediction method based on limit gradient improvement, and belongs to the technical field of cloud computing. According to the method, influence factors of task execution time are classified from three levels of workflow task composition, resources on which task operation depends and a physical execution environment of the resources, and comprehensive modeling of the influence factors of the task execution time is achieved. Secondly, aiming at the condition that the sample data set has a data missing value, the data set with the missing value is complemented by adopting a machine learning method; and finally, by means of the multi-type data processing capability of the extreme gradient lifting algorithm, the parameter design isrelatively simple, the calculated amount is small, the advantages of a serial learner and a parallel learner are combined, and a cloud workflow task execution time prediction model is trained by adopting the extreme gradient lifting algorithm. Compared with an existing prediction model, limitation on the sample data type is reduced, prediction errors are reduced, and the generalization ability ofthe model is further improved.

The invention relates to a method for adjusting process parameters in a polyester fiber polymerization process, which comprises the following steps of: acquiring a predicted value of a polyester fiberperformance index, comparing the predicted value with a set value, and adjusting the process parameters in the polyester fiber polymerization process according to a comparison result; the collectionprocess of the predicted value comprises the following steps: firstly, collecting m parts of characteristic data from all characteristic data acquired by a sensor; performing feature correlation sorting on the m parts of feature data based on an extreme gradient boosting tree algorithm of grid search; selecting the first n parts of feature data with relatively high feature correlation from the m parts of sorted feature data; normalization processing is carried out on all samples of the n parts of feature data; and finally, processing the n normalized characteristic data by using a weighted double-flow bidirectional long-short-term memory attention network, and respectively outputting predicted values of the four polyester fiber performance indexes. According to the method, the predicted value of the polyester fiber performance index can be obtained with high precision, so that industrial production can be better guided.

The invention relates to an establishment method of a disease risk prediction model and a recommendation method of a disease insurance product. The establishment method comprises the steps: obtaininghistorical diagnosis and treatment data of medical insurance insured personnel in a preset region; performing classified sampling processing on the historical diagnosis and treatment data to obtain asample data set, wherein each piece of sample data comprises historical disease diagnosis coding information of each sample in a preset time range; preprocessing the sample data set to remove invaliddata; clustering historical disease diagnosis coding information of all samples in the preprocessed sample data set according to disease attributes and focus parts to obtain disease clustering featuretags; screening the disease clustering feature tags by adopting a preset feature selection algorithm to obtain serious disease clustering feature tags, and establishing a disease risk prediction model corresponding to the preset serious disease according to the serious disease clustering feature label, the gender, the age and the doctor-seeing behavior information in combination with an extreme gradient boosting algorithm, thereby laying a foundation for the accuracy of disease insurance product recommendation.

The invention discloses a disk fault prediction method and system, and the method comprises the steps: carrying out the sampling and marking of a disk data set through employing the SMART technology,and obtaining a positive sample corresponding to a fault disk and a negative sample corresponding to a normal disk; extracting SMART features of each positive sample and each negative sample accordingto a preset time sequence to obtain time sequence features of each positive sample and each negative sample; importing a self-defined loss function into the extreme gradient boosting XGBoost algorithm to obtain an improved XGBoost algorithm; wherein the loss caused by positive sample misclassification in the user-defined loss function is greater than that caused by negative sample misclassification; and importing the time sequence characteristics as input and the positive sample and the negative sample as output into an improved XGBoost algorithm, so that the improved XGBoost algorithm performs machine learning on the disk data set to obtain a disk fault prediction model. According to the technical scheme, the problems that in the prior art, a positive sample corresponding to the faulty disk is difficult to predict, and the prediction accuracy of the faulty disk is not high can be solved.

The invention relates to a psychosis high-risk identification model based on an extreme gradient boosting algorithm. A method comprises the steps: establishing a psychosis high-risk identification model and identifying psychosis high-risk condition by utilizing the established model, wherein the step of establishing the psychosis high-risk identification model comprises the substeps: obtaining thedata of a screening tool as training features, normalizing the training features, training an XGBoost model, screening the features and simplifying the features; the step of identifying the psychosishigh-risk condition by using the established model comprises the following substeps: acquiring screening data of a subject, extracting specified features, and sending the specified features to the established identification model for identification. The method has the advantages that the schizophrenia is a severe mental disease, and the social function of a patient suffering from the schizophrenia is severely reduced, and even severe patients often have impulsion, self-injury, injury and other behaviors, so the establishment of the schizophrenia high-risk identification model improves the identification rate of the schizophrenia high-risk identification model, facilitates the early diagnosis and intervention of the schizophrenia patient, and reduces family and social burdens.

The invention discloses a dynamic safety assessment method based on an extreme gradient boosting decision tree. The method comprises the steps: 1, establishing an original sample set; 2, forming an intermediate sample set; 3, forming an efficient sample set; 4, performing offline training on the extreme gradient boosting decision tree by utilizing the efficient sample set, and constructing a dynamic safety evaluation model based on the XGBoost decision tree; 5, due to the influence of various operation factors of the power system, updating the efficient sample set through time domain simulation, and updating aDSA model by using the updated sample set; 6, based on the real-time measurement data of the synchronous phasor measurement unit, performing online DSA on the power system by using the DSA model. The invention aims to provide a dynamic safety assessment method based on an extreme gradient boosting decision tree. According to the method, online DSA can be efficiently and accuratelycarried out on the power system, the robustness of the evaluation model is good, and the method can adapt to the change of a network topology structure.

The invention discloses a hydrogen fuel cell vehicle endurance mileage prediction method based on an XGBoost algorithm, the core of the method is that an algorithm model is preset, the algorithm model is an XGBoost (eXtremeGradient Boosting, extreme gradient boosting) algorithm, the method comprises the following steps: firstly, obtaining various current related data of a vehicle, analyzing and judging the current vehicle working condition, and also obtaining vehicle fuel cell data, hydrogen system data and other vehicle data, performing proportional combination on the current data and historical data corresponding to the vehicle working condition and then inputting into the algorithm model, and predicting a future consumption curve of hydrogen under the current working condition. The algorithm has the advantages that the algorithm model required by the current operation for predicting the endurance mileage is optimized and adjusted by continuously utilizing the residual error of the previous algorithm model so that the algorithm model required by the prediction operation is more and more convergent along with the increasing and increasing of the data volume and the increasing and increasing of the number of iterations, and the prediction precision is continuously improved.

The invention provides a critical patient prognosis prediction method, provides a critical patient prognosis prediction method based on an ensemble learning method, and belongs to the field of critical patient prognosis evaluation. The method comprises the steps that related data of a patient and a hospital are included through a structured language; feature variables are screened based on a random forest, and a prediction model is constructed through regularization logistic regression, K nearest neighbor, a support vector machine, the random forest, limit gradient lifting and a deep neural network algorithm; and through prediction probabilities of different methods, ensemble learning is carried out through a limit gradient lifting algorithm to finally predict whether the critical patient is dead or not, and the occurrence probability of the critical patient is further calculated. According to the method, patient data and hospital feature information are utilized as much as possible, the prognosis of the critical patient is predicted in an individualized mode, prediction bias caused by a certain model is weakened, and prediction accuracy is improved.

The invention discloses an online unit load prediction method based on ensemble learning. The method comprises the following steps: firstly, reading historical unit load data of a past month from a database, secondly, preprocessing the data, detecting a gross value according to a Pauta criterion, and processing the detected gross value in a backward filling manner; further performing feature extraction on the preprocessed data, including time features, correlation features and aggregation features, and constructing a feature data set; training a limit gradient lifting model and a lightweight efficient gradient lifting model through the feature data set, and training a long-short-term memory neural network model through a prediction result of an LGBM model and original feature data; and finally, taking output results of the three models as three inputs to train a linear regression model to output a final load prediction result. In a model training stage, network parameters are adjusted by taking a minimum prediction average error as a principle, so that an optimal unit load prediction model is established.

The invention discloses a chaos theory and integrated learning-based coal mining machine cutting height prediction system. The system is used for predicting the cutting height of the coal mining machine. The system comprises a data preprocessing module, a Chaotic characteristic calculation module, a Phase space reconstruction module, a coal mining machine cutting height extreme gradient boosting (eXtrome Grade Boosting) modeling module, and a coal mining machine cutting height extreme gradient boosting (eXtrome Grade Boosting, XGBoost) model prediction module. The chaotic characteristics of the cutting height sequence of the coal mining machine can be mined, and the cutting height of the coal mining machine can be quickly and accurately predicted.

The invention provides a sewage treatment process soft measurement method based on an extreme gradient lifting algorithm, and belongs to the field of industrial sewage treatment process soft measurement modeling and application. Firstly, for the problem that missing values exist in data collected in the industrial process, a proximity algorithm model is adopted for missing value filling; secondly,soft measurement modeling is performed by using an extreme gradient boosting algorithm; and finally, seven parameters of the model are adjusted by using a grid search method, so that a more accuratesoft measurement model is obtained. The method can improve the prediction precision of the effluent ammonia nitrogen concentration in the sewage treatment process.

The invention provides a method for predicting delirium through artificial intelligence, and belongs to the field of critical patient evaluation. The method comprises the steps: related data of a patient and a hospital are included through a structured language; feature variables are screened based on a random forest, and a prediction model is constructed through regularization logistic regression, K nearest neighbor, a support vector machine, the random forest, limit gradient lifting and a deep neural network algorithm; and through prediction probabilities of different methods, ensemble learning is carried out through a limit gradient lifting algorithm to finally predict whether a critical patient has delirium or not, and the occurrence probability of the delirium is further calculated. According to the method, patient data and hospital feature information are utilized as much as possible, and individualized evaluation for predicting the critical patient is carried out, so that prediction bias caused by a certain model is weakened, and prediction accuracy is improved.

The invention is suitable for the technical field of energy, and provides an electric power price prediction method and device. The method comprises the steps: obtaining historical data according to electric power transaction information, and the historical data comprises historical price data and influence factor data; processing the historical data to obtain a price prediction result and an influence factor prediction result; and processing a price prediction result and the influence factor prediction result by adopting an extreme gradient boosting method to obtain a prediction result of theelectric power price. By correcting the price prediction result, the prediction efficiency and accuracy are remarkably improved, the prediction calculation time is shortened, and meanwhile the methodhas the advantages of being high in data utilization rate and low in data usage amount. The method is simple in overall operation, guarantees the prediction accuracy, convenience and quickity, and saves the calculation resources.

The invention provides a credit card fraud detection method and equipment based on a clustering sample and a limit gradient. The method comprises steps 1 to 8. According to the credit card fraud detection method, the clustering algorithm, the sample adaptive weight calculation and the oversampling algorithm are combined, the inter-class imbalance problem is solved, the intra-class imbalance problem is effectively avoided, the data quality of artificially synthesized samples is improved, the limit gradient boosting tree serves as a classifier in a credit card fraud detection model, and the classification efficiency of the credit card fraud detection model is improved. A better classification effect can be obtained, and after an oversampling algorithm based on clustering and adaptive weight is combined, the finally generated credit card fraud detection model has good detection accuracy.

The invention relates to the field of neural networks, and discloses an abnormal object recognition method and device, a medium and electronic equipment. The method comprises the steps of obtaining object data and a label corresponding to the object data and representing whether the object data is abnormal or not; dividing the object data into a training set and a test set; inputting the object data in the training set and the corresponding labels into a plurality of to-be-trained deep neural network models for training to obtain a plurality of models; inputting the object data of the test setinto a deep neural network model to obtain an abnormal probability output by the model; determining a target deep neural network model according to the abnormal probability output by each model; cascading the target deep neural network model with the extreme gradient boosting model to obtain a cascading model, and training the cascading model by using the training set to obtain a trained cascading model; and inputting the to-be-identified object data into the trained cascade model for prediction. According to the method, the accuracy of identifying the abnormal object is improved, and the missing identification rate of the abnormal object is reduced.

The invention discloses an air conditioner cooling load prediction method and system considering data characteristics after frequency domain decomposition. The method comprises the following steps: taking factors of which the load sequence importance is greater than a set threshold value as input variables; decomposing an original load sequence into two intrinsic mode functions IMF1 and IMF2 and an error sequence ERR by using a variational mode decomposition technology; establishing a least square support vector machine prediction model for the intrinsic mode function IMF1 to obtain a prediction component of an IMF1 subsequence; establishing an extreme gradient boosting decision tree prediction model for the intrinsic mode function IMF2 to obtain a prediction component of an IMF2 sub-sequence; performing normal fitting on the probability distribution of the error sequence ERR showing the Gaussian noise part to obtain a prediction component of a sub-sequence of the error sequence ERR; and superposing the prediction component of the IMF1 sub-sequence, the prediction component of the IMF2 sub-sequence and the prediction component of the error sequence ERR sub-sequence, and outputting to obtain a final cold load prediction value. According to the method, the prediction precision under the environment noise condition is improved, and the method has the practical engineering application background and important practical significance.

The invention discloses a soft measurement method for on-line detection of components in a special rectification process, which comprises the following steps of: 1, selecting a soft measurement model input variable by adopting a principal component analysis random forest combined variable, 2, introducing a generalized robust loss function into a limit gradient algorithm, and optimizing a loss function hyper-parameter by adopting a Bayesian method, 3, inputting variables through a soft measurement model, and training in combination with a limit gradient algorithm to obtain a trained ARXGBoost model; 4, substituting real-time data into the ARXGBoost model for online detection; the adaptive updating robust limit gradient lifting algorithm has excellent prediction performance, realizes automatic updating of soft measurement, enhances the adaptive capability of the algorithm, improves the prediction reliability, performs online measurement and prediction on components, realizes direct control of the components, adopts adaptive adjustment robust parameters, and has the advantages of high robustness and high reliability. The prediction effect of the dynamic stage is good.

The invention relates to an OCPX adaptive learning method and system based on fast moving consumer arrival, and the method comprises a data extraction step: enabling an algorithm platform to extract advertisement putting data and conversion data based on a behavior monitoring data log; a feature extraction step: selecting a corresponding sample from the conversion data according to the advertisement putting data, and performing feature extraction according to the sample; a model training step of training a first-second classification model according to an extreme gradient boosting tree principle based on the features; a result uploading step: sequentially scoring the user packets according to the binary classification model, selecting a part of users in the user packets as prediction results according to scoring results, and uploading the prediction results to a front-end processor; and a result query step, in which the advertisement service party completes query of the prediction result and corresponding operation through the front-end processor. The delivery effect and user experience are optimized through a self-adaptive state formed by a process closed loop from extraction of self-owned data to training and pushing to a front-end processor for advertisement query to carry out data query.