200 results about "Support vector regression model" patented technology

The invention relates to a real-time predicting method for arrival time of a bus. The real-time predicting method comprises the following steps: S1, respectively extracting historical floating bus data and historical running time data of a site section between every two adjacent sites as a training set, screening factors influencing the running time of the bus at one site section according to the training set, and respectively constructing support vector regression models of corresponding site sections according to the screened factors; S2, determining site sections contained in a running route between a current site and a target site of the bus, acquiring respective real-time floating bus data and real-time running time data of the site sections as a predicting set, predicting the running time of the bus at the site section by using the support vector regression model according to the predicting set, and obtaining the arrival time of the bus from the current site to the target site according to the predicted running time of the bus at each site on the running route. According to the method provided by the invention, the prediction accuracy of the arrival time of the bus can be improved.

The invention discloses a new short-term wind speed forecasting method of a wind farm, which comprises the following steps: collecting wind speed data of the wind farm, forming the time sequence of the historical wind speed and carrying out normalization treatment; applying the chaos analysis method for analyzing the time sequence of the historical wind speed after the normalization treatment for obtaining phase space reconstruction parameters of a wind power system in the area located by the wind farm, wherein the parameters are delay time and embedding dimension of the time sequence; utilizing the parameters for carrying out treatment on the time sequence of the historical wind speed after the normalization treatment, and obtaining a training sample set required by a support vector regression model for wind speed forecasting; adopting the training sample set for training the support vector regression model; utilizing the support vector regression model after training for carrying out short-term wind speed forecasting on the wind farm, and obtaining the normalized result of the short-term wind speed forecasting of the wind farm; and carrying out anti-normalization treatment on the obtained normalized result of the short-term wind speed forecasting of the wind farm, and obtaining the short-term wind speed forecasting result of the wind farm.

The invention provides an offshore crane gearbox fault diagnosis device and method based on multivariate data. The device comprises a temperature sensor, an acceleration sensor, an embedded monitoring unit and a remote monitoring and maintenance center. Temperature and acceleration sensors are arranged. A GPRS module transmits collected data to an upper computer. Bearing temperature trend prediction based on a gray model and support vector regression model residual compensation is carried out according to a collected temperature signal. Vibration fault feature extraction is carried out on a vibration signal by using the combination of empirical mode decomposition and envelope spectrum analysis. Gearbox lubricating oil samples are regularly extracted for convention physical and chemical property analyzing and emission spectrographic analyzing. Wear trend analysis and fault early warning are carried out according to the content of metal wear in oil samples. Fusion comparing is carried out on three analysis results to give a gearbox fault diagnosis result. According to the invention, the fault diagnosis accuracy of an offshore crane gearbox can be effectively improved, and the diagnosis result is accurate and reliable.

The invention provides a deep convolutional neural network-based human face image quality prediction method. The method comprises the main steps of building a human face image quality prediction model; creating a human face image quality tag; and automatically predicting human face image quality. According to the process, a support vector regression model is trained by using a deep convolutional neural network (ConvNet); facial features are extracted for predicting facial image quality; a large database of unrestricted facial images is established; image features are extracted by using the deep ConvNet; and the human face image quality is predicted from the features extracted by the ConvNet through adopting the support vector regression model. According to the method, clustering is performed by comparing paired facial images; through matrix solving, a complete quality level is derived; and by depending on a supervised learning technology, the facial image target quality tag is created,so that the human face image quality prediction accuracy is improved, and a further contribution is made for new design of engineering field and innovative solutions of monitoring field.

The invention discloses an indoor positioning method based on manifold learning and an improved support vector machine. The method comprises a step of determining a positioning area, dividing the positioning area according to an indoor structural characteristic and a layout characteristic, and obtaining a classification result, a step of obtaining offline training data, and collecting hotspot RSS signal values which can be received by the reference points in different classification area as a training data set, a step of using an isometric mapping algorithm to carry out training data characteristic extraction, a step of using the training data to carry out support vector machine classified training, using a taboo search algorithm to carry out support vector machine classification hyper parameter searching, and establishing the support vector regression model of each category at the same time, a step of carrying out online positioning, collecting the RSS signal value of each hotspot of a target, using a support vector machine classification model to carry out classification, and obtaining the rough positioning area of the target, and a step of carrying out the accurate positioning of the target by using the support vector regression model according to the classification result. According to the method, the time-varying characteristic of the wireless signal intensity is effectively suppressed, and the precision is obviously improved.

A digital watermarking based method for objectively evaluating quality of stereo image includes: at transmitting terminal, extracting characteristics of the left-view image and the right-view image of an undistorted stereo image in DCT domain and embedding digital watermarking obtained by processing quantization coding on the characteristics into the DCT domain; at a receiving terminal, detecting the digital watermarking embedded in the distorted stereo image and processing decoding and inverse quantization to extract the embedded characteristics of the left-view image and the right-view image of the stereo image, obtaining a stereo perception value and a view quality value of the distorted stereo image according to the embedded characteristics, and finally obtaining an objective quality score of the distorted stereo image utilizing a support vector regression model.

The invention discloses an iron steel electricity consumption forecasting method and device. The forecasting method comprises the following steps: acquiring iron steel historical data of an area to be forecasted within a preset period of time to serve as sample data, wherein the iron steel historical data comprises electricity consumption index and iron steel electricity consumption; performing dimensionless treatment on the sample data to obtain normalized electricity consumption index and normalized iron steel electricity consumption; adopting the normalized electricity consumption index as an input variable, and the normalized iron steel electricity consumption as an output variable to realize network training so as to build a nerve network model; adopting the normalized electricity consumption index as the input variable and the normalized iron steel electricity consumption as the output variable to build a support vector regression model; combining the nerve network model and the support vector regression model to obtain a combined model, and using the combined model to forecast the iron steel electricity consumption. According to the invention, the demand on the forecasting precision of the iron steel electricity consumption is satisfied to the utmost extent, and reference is provided for economical and reasonable planning of a power grid.

The invention discloses a fault prediction method in industrial production based on particle swarm optimization support vector regression, and the method comprises the steps: calculating an average deviation and a variance, carrying out the feature extraction of multi-dimensional data in an industrial production process, and obtaining the feature data of an original input sample set; Constructinga time sequence of feature data of the original input sample set, selecting previous h continuous feature data from the time sequence, and establishing a row number h-according to a set mapping dimension m; Wherein m + 1 is an input sample with the column number being m; And carrying out fault prediction on the industrial production process by using an input sample and a trained support vector regression model. According to the method, the particle swarm algorithm is adopted, three key parameters of the support vector regression model are optimized at the same time, a feasible and efficient method is provided for optimization of the parameters of the support vector regression model, and the accuracy of fault prediction in the industrial production process through the support vector regression algorithm is improved.

The invention discloses a visual line tracking method based on projection mapping correction and ocular fixation point compensation under natural lights; the method comprises the following steps: extracting double-eye iris center and eye inner and outer angle point and mouth corner point characteristics; calculating a projection mapping relation between a rectangle formed by the eye inner and outer angle points and mouth corner points and the rectangle information when the head is in a calibrated position, thus carrying out projection mapping correction for iris center positions and eye inner and outer angle point positions, and eliminating influences caused by head motions; using corrected left and right eye iris centers to respectively form 4 vectors with the left and right eye inner and outer angle points, and a polynomial mapping model is combined to obtain a real time ocular fixation point; finally using a support vector regression model to carry out ocular fixation point compensation. The visual line tracking method is high in precision, can reduce head motion influences, and can realize visual line tracking under natural lights.

The invention discloses a lithium battery SOH estimation method based on data driving. The method is characterized by comprising the following steps: 1) recording charging data of a lithium battery ina constant-current charging working mode in real time; 2) performing capacity increment curve calculation on a constant-current charging voltage curve by adopting a simplified dQ/dV processing mode through capacity increment analysis; determining the peak value intensity and the peak value position voltage of a No.2 peak of the capacity increment curve as characteristic vectors for estimating SOHthrough grey correlation analysis; 3) constructing a support vector regression model by taking the peak intensity and the peak position voltage as feature vectors as input and SOH as output; 4) fusing a differential evolution strategy with a grey wolf optimization algorithm to form an improved grey wolf optimization algorithm IGWO; and 5) performing three-parameter joint optimization on hyper-parameters in the support vector regression model through the IGWO. The method has the advantages that the defects in the prior art can be overcome, and the structural design is reasonable and novel.

The invention discloses a wind power plant ultrashort term wind speed prediction method based a combination kernel function. The method comprises the following steps: adopting a support vector machine based on the combination kernel function, combining a wavelet kernel function and a polynomial kernel function, combining a global characteristic of the polynomial kernel function with high precision and good learning capacity of the wavelet kernel function, effectively improving predictive capacity of a support vector machine model, and reducing predicting errors. Meanwhile, in building of the support vector machine model, choosing similar data, building models in a classified mode, enabling a training sample and a test sample to have a similar relation, and therefore strengthening nonlinear fitting capacity of a support vector regression model. According to combination of the combination kernel function and the similar data, the wind power plant ultrashort term wind speed prediction method based the combination kernel function improves precision of wind speed prediction.

The invention relates to a traffic accident rate prediction system based on on-line variational Bayesian support vector regression, which comprises a data preprocessing module, an online variational Bayesian support vector regression model building module, an online variational bayesian support vector regression model training module, and an online variational bayesian support vector regression model prediction module. This method effectively solves the problem that the traditional support vector regression model predicts the speed of traffic accident rate is slow, the prediction result is inaccurate, and it is difficult to solve the problem on line and show its practical value.

A method for predicting residual service life of gas turbine bearing based on support vector regression By collecting the health status data of gas turbine bearing and extracting the time-domain and frequency-domain features after preprocessing, Then the time-domain and frequency-domain features are fused by the data fusion tool through the principal component analysis method, and the low-dimensional feature indices which characterize the bearing degradation are obtained and used to train the residual service life prediction model based on support vector regression. Finally, the residual service life prediction model is used to predict the bearing life in real time. The prediction error of the invention is lower than the prediction result of the ordinary support vector regression model andthe neural network model, and the prediction result precision is high.

The invention discloses a non-reference stereoscopic image quality evaluation method based on a convolution neural network. Firstly, according to the stereoscopic perception characteristics of human visual system, the left and right views of stereoscopic image are fused into the middle view to simulate the process of stereoscopic perception of human eyes. Then the convolution neural network (CNN)is used as the feature extraction tool, the concept of transfer learning is used to adjust a convolution neural network to extract quality-related features from images adaptively, and the strong feature expression ability of neural network is utilized to avoid the incompleteness and high complexity in the process of manual modeling. Finally, the features extracted from the convolution neural network are sent to the support vector regression model, which is used to establish a mapping between the subjective scoring value and the features, so as to obtain the final quality evaluation score. Theimage quality objective evaluation score based on the method provided by the invention has high consistency with the subjective quality score provided by the database, and has good accuracy and robustness.

The invention discloses a method for predicting the daily activity-travel time of a commuter and aims at solving the problems of only consideration of certain period of activity-travel, certain travels or activities and the like in one day in the prior art. The method comprises the following steps of investigating travel data of the commuter; dividing the daily activities and the travels of the commuter into five activity-travel periods in sequence; constructing an integral framework of a model system for predicting the daily activity-traveling time of the commuter; screening an influence variable for the model system; setting selection branches of a departure time selection model and a stop/start time selection model; applying an Ordered Probit model to construct a departure time selection sub-model and a top/start time selection sub-model of the model system; applying a support vector regression model to construct a travel time-consuming prediction sub-model and an activity time-consuming prediction sub-model in the model system; predicting and calculating daily activity-travel time elements of the commuter; and recognizing and removing overlapped time periods to form the final daily activity-traveling time arrangement of the commuter.

The invention provides a blood glucose prediction method and device. The method includes the following steps: cleaning original measured blood glucose data; smoothing and de-noising the cleaned data by using Kalman filtering; applying an autoregression model, a support vector regression model and a long-term and short-term memory network model to perform prediction according to the filtered data,and obtaining the blood glucose prediction results of the models; calculating the weights of the models separately; performing linear combination on the blood glucose prediction results of the respective models and the weights of the respective models to acquire the combined prediction model prediction value. Compared with the prior art, the method combines various model advantages, and realizes various model weights to predict blood sugar, so that the prediction result is more accurate, and the application range is wider.

A camera calibration error compensation method based on multi-dimensional characteristics includes the following steps: (1) preparing data: collecting p images of standard targets, obtaining p images with errors, selecting q key points through each images, obtaining p*q key points; (2) extracting the characteristics of the key points: extracting the characteristics of each key point; (3) calculating the actual errors of the p*q key points (delta x, delta y) p*q; (4) conducting simulated training: conducting a support vector regression model training by using support vector machine (SVM) light tools; and (5) estimating errors: obtaining the actual position (x, y) q of q key points, then extracting the characteristics of the q key points according to step (2), storing the characteristics of the q key points in a to-be-regressed characteristic file, and calculating the compensation value (delta x, delta y) of each key point. According to the camera calibration error compensation method based on the multi-dimensional characteristics, association characteristics of a scene image are used, the compensation value of each collected image estimated in support vector regression is adopted, and by means of the camera calibration error compensation method based on the multi-dimensional characteristics, compensated light target center is close to an ideal light target center.

The invention provides a photovoltaic power prediction method based on seasonal regionalization. The photovoltaic power prediction method comprises the steps: S1: collecting the historical information of a photoelectric power station; S2: classifying the solar irradiance data and the power data collected in the step S1 according to seasons; S3: preprocessing the data of every season after classification in the step S2, and using the preprocessed data to establish a support vector regression model to obtain the corresponding relationship between the power and the solar irradiance; S4: according to the season type belonging to a predicted day, determining a regression model which is demanded by power prediction; S5: aiming at the data of the time quantum before the predicted day, performing data preprocessing, and establishing a least squares model to obtain the relationship between the power and the solar irradiance within the recent time quantum; and S6: calculating the prediction power y. The model established by means of the photovoltaic power prediction method based on seasonal regionalization is high in versatility and generalization, and can improve the prediction accuracy of a photovoltaic power station to a certain degree.

The invention discloses a method for performing non-reference image quality prediction by using multilayer depth characterization. The method comprises the following steps: step (1) data preprocessing: zooming all images to a unified size, subtracting a draw value, and converting binary data into a data format that can be identified by a deep neural network; step (2) feature extracting and processing: performing feature extraction by using a 37-layer VGGnet model that is trained on ImageNet, extracting each layer of features, and processing the same to obtain a column vector; and step (3) score predicting: inputting the column vector obtained by fusing each layer of features in a support vector regression model to obtain a predicted score of each layer of features, and using an average value of the scores of the layers as a quality evaluation score of the whole image. The invention provides a simple and efficient new method for image quality evaluation. The best effect in the existingimage quality evaluation field is obtained.

The invention discloses a depth neural network model modeling method for high-speed railway train delay time prediction, Combined with the characteristics of the obvious interaction between adjacent trains and the time series and non-time series influencing factors of train delays, a deep neural network model including circulating neural network and all-connected neural network is proposed in thetechnical field of rail transit. In this model, the non-time-series factors of delay are input into the fully connected neural network, and the time-series factors are input into the cyclic neural network to learn the interaction relationship between adjacent trains by its feedback mechanism. In order to identify the influence of the interaction between trains on the train delay, the prediction accuracy is high and the practical application ability is good. The absolute error and relative error of the prediction are lower than the support vector regression model, the ordinary neural network model and the Markov model.

The invention discloses an order production progress real-time prediction method based on double-layer transfer learning, and the method specifically comprises the steps: dividing a data set added with real-time production state data into a historical order data set and a current order data set, and carrying out the normalization processing; Establishing a feature extraction model of the historical order data set by adopting a deep sparse autoencoder, and constructing a support vector regression model mh by taking the historical order data subjected to feature extraction as a secondary training set; Finely adjusting the deep sparse autoencoder based on the current order data set, and establishing a first layer of transfer learning model; And establishing a support vector regression model mc by using the current order data set after feature extraction, and establishing a second-layer transfer learning model as a production progress prediction model of the current order based on the similarity of mc and mh. According to the invention, the problems of low prediction precision and poor generalization performance caused by a small amount of current order data are solved.

The invention discloses an equipment fault prediction method based on particle swarm optimization support vector regression, and the method comprises the steps: carrying out the feature extraction ofvibration signal data in industrial production key equipment based on a wavelet decomposition method, and obtaining feature data; Secondly, constructing a time sequence of the feature data, selectingfirst n continuous feature data from the time sequence, and establishing a row number n-according to a set mapping dimension m; Wherein m + 1 is an input sample with the column number being m; And finally, carrying out fault prediction on the equipment by using the trained support vector regression model by using the input sample. According to the method, the particle swarm algorithm is adopted, and three key parameters of the support vector regression model are optimized at the same time, so that a feasible and efficient method is provided for optimizing the parameters of the support vector regression model, and the accuracy of predicting the equipment fault by using the support vector regression algorithm is improved.

A method for measuring a water holding ratio of a horizontal well by fusing total flow and a conductivity probe array signal belongs to the technical field of multiphase flow detection. The method comprises: firstly, establishing a calculation relationship between an arc length of an oil phase sector of the horizontal well and the water holding ratio of the horizontal well; then acquiring the conductivity probe array signal, and obtaining a corresponding water holding ratio value as a reference value by using the arc length of the oil phase sector, which is recorded by a high speed camera; next, establishing a nonlinear support vector regression model from the total flow and the conductivity probe array signal to the reference value of the water holding ratio; and finally, performing high precise prediction on the water holding ratio of the horizontal well by using the established nonlinear support vector regression model. The experimental result shows that the method provided by present invention is feasible and effective; and not only is the method excellent in generalization, but also measurement precision of the water holding ratio can be greatly improved.

The invention discloses a gas load combination prediction method based on support vector regression and relates to gas load prediction methods. According to the combination prediction method, a data preprocessing technology, an improved genetic algorithm and support vector regression are combined, and the method is mainly used for solving the problems that in the prior art, urban gas load prediction is low in precision and poor in applicability. The method comprises the steps that first, a correlation coefficient method is adopted to analyze the correlation between different influence factorsand gas loads, and singular spectrum analysis is adopted to perform de-noising processing on the obtained main influence factors; second, processed data is adopted to train a support vector regressionmodel, nuclear parameters and penalty factors are optimized in combination with the improved genetic algorithm, and finally a support vector regression model with an optimal training result is obtained; and last, the trained support vector regression model is utilized to predict gas load indexes in a future period of time. Through the combination prediction method, a short-term gas load prediction error can be substantially lowered, and prediction precision can be improved.