64 results about "Prediction rate" patented technology

The invention discloses an elevator failure recognition method based on a convolutional neural network. The elevator failure recognition method comprises the steps that 1, elevator motion data are collected and are converted into a time-spectrogram as a sample set through wavelet transform; 2, the time-spectrogram in the sample set is divided into a training set and a test set, failure types and failure degrees of samples in the training set are marked as known labels of data samples; 3, the convolutional neural network is built, the time-spectrogram in the training set is input to the convolutional neural network, and the characteristics of the previous layer are extracted and classified; 4, according to the labels given in the step 2 and the characteristics extracted in the step 3, a multi-class SVM classifier is trained; 5, after training, the prediction rates, for all types of failures, obtained through the SVM classifier are obtained; and 6, detection and recognition are conducted. The elevator failure recognition method which is novel in angle, consistent to actual situation and high in accuracy is achieved, the requirement for hardware is low, and portability is high.

The present invention provides a method of using the machine learning to predict a hard disk fault, and belongs to the cloud storage safety technology field. The method of the present invention uses the mass smart data sets provided by the blackblaze, according to the condition that the smart data of the hard disks of different brands is distributed unevenly, uses a random forest algorithm to train and model the historical data, generates a fault prediction model, and enables the fault prediction rate to be improved.

The invention relates to an abnormal-data alarming method. The method includes: acquiring historical operation data and current operation data in a preset time period; building a trend model corresponding to the historical operation data; calculating a periodic factor corresponding to the historical operation data; utilizing the trend model and the periodic factor to calculate an alarming threshold value of abnormal data in the current operation data; utilizing the trend model and the periodic factor to calculate a prediction value of the current operation value, and calculating a corresponding prediction rate according to the prediction value and the current operation data; when the prediction rate is greater than the alarming threshold value, generating alarming information for the abnormal data. By adopting the method to pre-warn the abnormal data, operation time can be saved effectively, and working efficiency can be improved. In addition, the invention further provides an abnormal-data alarming device.

The invention relates to a deep-learning-based weather forecasting method and system, so that a problem that weather forecasting is carried out by manual data screening based on priori knowledge in the prior art can be solved. The method comprises: S1, storing collected historical weather data and real-time weather data; S2, establishing a layer-by-layer deep learning model based on the weather data and continuously correcting the deep learning model; S3, synchronously collecting the real-time weather data into the deep learning model; and S4, obtaining an output result based on the deep learning model with the real-time weather data inputted. Therefore, data can be stored in different databases; output results of different needs are obtained by constructing different deep learning models; and a model with high prediction rate can be obtained by weather forecasting deep learning.

The invention relates to a breakout prediction method for continuous casting, belongs to the technical field of metallurgical continuous casting, and in particular relates to a breakout prediction method in plate blank continuous casting process. In terms of temperature monitoring-based breakout prediction algorithm, the self-adapting genetic algorithm is introduced into the BP (Back Propagation) neural network to achieve automatic optimization of the network structure and improve identification accuracy of a temperature monitoring model. In terms of friction monitoring-based breakout prediction algorithm, a friction monitoring model based on logical decision and neural network is established. Based on the prediction results of the two models, the method achieves combination of temperature monitoring accuracy and friction monitoring sensitivity, provides the prediction mechanism based on temperature monitoring in addition with friction monitoring and couples the temperature monitoring model to the friction monitoring model, so as to ensure the accuracy of final prediction and effectively reduce the false prediction rate.

The invention discloses a page pushing method and device. One specific mode of execution of the method comprises the steps that the user click rates of pages loaded by a client side are collected; a plurality of models used for predicting click rates of pages are located, and the prediction click rates of the pages are calculated through the models; a standby model is selected from the models based a comparative result of the prediction click rates and the user click rates; the prediction rates of multiple pages to be pushed are calculated through the standby model; a page is selected from the pages to be pushed based on the prediction click rates of the pages to be pushed. By means of the mode of execution, targeted page pushing is achieved.

The present invention discloses a streaming media service rate prediction method and device. According to the scheme of the present invention, the service rate influence data of the previous data sampling time period of the current moment can be obtained, the obtained service rate influence data is inputted to a set least squares support vector machine (LS-SVM) rate prediction model to obtain the service prediction rate of the next data sampling time period of the current moment, and the corresponding streaming media data is generated according to the service prediction rate, wherein the LS-SVM rate prediction model is obtained by training multiple sets of sample data within the previous set time period of the current moment according to a set LS-SVM rate prediction algorithm, so that the service rate prediction accuracy is improved, the code rate of the generated streaming media matches a current network rate more accurately, and the problems, such as the video lag, the network bandwidth waste, etc., are avoided.

The invention relates to a method for predicting lower limb joint angles on the basis of electromyography wavelet correlation dimensions. The method includes acquiring surface electromyography signalsfrom related muscle groups of the lower limbs of human bodies and determining action signal sections of the surface electromyography signals by the aid of energy thresholds; carrying out wavelet denoising on the surface electromyography signals of the action signal sections to obtain effective surface electromyography signals; carrying out wavelet multi-scale decomposition on the effective surface electromyography signals, extracting low-frequency coefficients of each layer, and further computing correlation dimensions of the low-frequency coefficients of each layer; combining the low-frequency coefficients and correlation dimension numbers with one another, computing wavelet correlation dimension coefficient features of the effective surface electromyography signals, and inputting the features into prediction networks; dividing extracted electromyography signals into training sets and test sets and extracting features according to processes; training networks by the training sets, and then verifying the prediction accuracy by the test sets. The method has the advantages that as shown by experimental results, the method is high in human body lower limb movement knee joint angle prediction rate, and prediction results of the method are superior to prediction results of other prediction methods.

The invention relates to the technical field of intelligent questions and answers, in particular to a Bert model-based intention recognition and slot value filling joint prediction method, which comprises the following steps of: inputting a target text to obtain a word vector, a segment vector and a position vector of the target text, splicing the word vector, the segment vector and the position vector as an input vector of a Bert model, and performing prediction on the input vector of the Bert model; inputting a trained Bert model, outputting an intention representation vector and a slot value sequence representation vector by the trained Bert model, performing weight calculation on the intention representation vector and the slot value sequence representation vector in a Gate layer to calculate a joint action factor, acting the joint action factor on the slot value sequence representation vector, and finally outputting predicted intention classification and a slot value sequence. According to the method, a Gate mechanism is used on a Bert layer, the internal relation between intention recognition and slot value filling is fully utilized, and the task error prediction rate is reduced.

The invention provides a streaming media adaptive transmission method, a terminal and a system. The method comprises the following steps: receiving a streaming media data packet, predicting a networkcondition according to the streaming media data packet, and obtaining a network transmission prediction rate; Monitoring the service QoE of the streaming media according to the streaming media data packet, and obtaining a predicted QoE acceptable transmission rate according to the service QoE; Obtaining a player buffer margin and a terminal CPU utilization rate, obtaining a suggested transmissionrate according to the network transmission prediction rate, the QoE acceptable transmission rate, the player buffer margin and the terminal CPU utilization rate, and sending the suggested transmissionrate to a media server; Wherein the media server changes the encoding rate of the encoder in real time based on the suggested transmission rate, so that the output rate of the streaming media data packet does not exceed the suggested transmission rate.

The invention relates to a determination method of organic porosity of a shale gas reservoir based on well logging data. The method comprises following steps: collecting shale gas reservoir organic porosity data and shale gas reservoir well logging data obtained through work area parameter well core experimental analysis; calculating inspected organic matter maturity Roa1 of the parameter well shale gas reservoir; calculating organic porosity [phi]toca1 of the parameter well inspected shale gas reservoir; obtaining organic porosity coefficient K of the work area shale gas reservoir by means of the cross-plot technology; collecting well logging data of the shale gas reservoir to be processed; calculating the reservoir inspected organic matter maturity Roa; calculating the organic porosity [phi]toca of the inspected shale gas reservoir of the well to be processed; calculating the organic porosity [phi] toc of the gas reservoir; outputting the calculation results of the organic porosity [phi]toc of the shale gas reservoir of the well to be processed. By means of the method, reliable data is provided for shale gas reservoir evaluation and prediction rate is increased. The method is simple and has wide application scope. The method is applied in 182 wells in Fuling shale gas field, and the prediction rate is high.

The invention discloses a high-dynamic rapid star tracking method. The method comprises 1, designing a weighted value according to multiframe attitude information output by a star sensor at the T1-Tn moment and a proximity relationship of the multiframe attitude information, forecasting attitude of the star sensor at the next T<n+1> moment, searching a star point corresponding to the attitude from a star database by the forecasted attitude, and producing a virtual forecasted large view field star image for matching by mapping, and 2, carrying out matching identification on a star position A and a star mass center position B at t+delta t moment to realize star tracking. The high-dynamic rapid star tracking method utilizes multi-frame attitude information prediction, has a fast prediction rate, improves prediction accuracy, realizes tracking and matching based on a forecasted virtual wide field star image without novel star identification and improves a star tracking algorithmic speed.

The invention provides a deep-learning-based early screen apparatus for the lung cancer. The apparatus is composed of an image processing module, an image analysis module, an image analysis result processing module. The image processing module is used for preprocessing an image to obtain an image meeting a deep learning standard. The image analysis module is used for inputting the image into a neural network after deep learning to detect a lung nodule in the image, so that the neural network outputs a suspicious lung nodule and a corresponding confidence value. The image analysis result processing module is used for selecting N highest values, extracting a last convolution layer for each highest value, introducing extraction results into a pooling layer and an all-connection layer, and thus calculating the probability of the lung cancer. According to the deep-learning-based early screen apparatus provided by the invention, the blank of the intelligent device for early screening of thelung cancer is filled and an automatic low-cost high-confidence apparatus is provided for intelligent medical imaging diagnosis. The operation has characteristics of full automation and manual intervention prevention, so that the precious time of the medical staff is saved; and the lung cancer prediction rate is consistent.

The invention discloses a rapid prediction method for heat conduction-radiation coupling heat transfer characteristics of a periodic structure composite material at a high temperature, and the method comprises the steps: decomposing a to-be-solved temperature field into a macroscopic average field and a mesoscopic temperature fluctuation through a multi-scale progressive analysis method, respectively carrying out the calculation, and finally reconstructing the macroscopic average field and the mesoscopic temperature fluctuation into a complete temperature field. The calculation process comprises the steps of carrying out grid division on a macroscopic prediction model and a characterization unit under a microscopic scale, solving a periodic vector function in the characterization unit under the microscopic scale, calculating macroscopic equivalent physical property parameters, solving a macroscopic scale heat conduction-radiation coupling heat transfer equation, and finally reconstructing a multi-scale temperature field. The multi-scale model established by the method can accurately calculate the temperature field of the periodic structure composite material, and can significantly improve the prediction rate of the high-temperature heat transfer characteristic of the composite material.

The invention discloses a brushless direct current motor sensor fault detection method based on a convolutional neural network. The method specifically comprises the following steps: acquiring original data of the brushless direct current motor during operation; converting the original data into a time-frequency spectrogram as a sample set through wavelet transform; marking fault types and fault degree of samples in the training set as known labels of the data samples; establishing a convolutional neural network, inputting the time-frequency spectrogram in the training set into the convolutional neural network, and extracting and classifying features of a previous layer; training a multi-class SVM classifier according to the given label and the extracted features; after training is completed, acquiring the prediction rate of the SVM classifier for each type of faults; and finally, analyzing the system state of the brushless direct current motor, and predicting possible faults. The invention can qualitatively and quantitatively evaluate the operation state of the monitored brushless direct current motor sensor and predict the development trend of the monitored brushless direct current motor sensor; therefore, the fault diagnosis process is more intelligent, and the detection accuracy is higher.

A digital image automatic labeling method based on uncertainty analysis, including the steps of extracting image features based on a deep convolutional neural network, constructing an image automaticlabeling system based on a variable precision neighborhood rough set, and labeling unlabeled images. The method includes the following steps: collecting the image data and labeling to obtain a training set, and extracting a feature vector of the image through the deep convolutional neural network; obtaining a classification model based on the neighborhood estimation class conditional probability density; in prediction, extracting image features, and estimating the position of the image to be classified by using upper and lower approximation concepts of the rough set; directly judging the membership of the labels for the images located in positive and negative domains, and judging the images in the boundary domain by using a Bayesian decision rule. According to the digital image automatic labeling method based on uncertainty analysis, the position of images to be labeled in the sample space are estimated by introducing upper and lower approximation concepts of the rough set, the error prediction rate of the irrelevant labels is reduced, and the problem of uncertainty existing between the underlying image feature and the high level semantics in image automatic labeling is solved.

The invention discloses a mulberry pyralid larva based on visible and near-infrared hyperspectral imaging and a rapid recognition method for damage of mulberry pyralid larva to mulberry leaves. The method comprises the following steps: picking three kinds of mulberry leaves which are healthy, have larva damage and have larvae, acquiring visible and near-infrared hyperspectral imaging data, recognizing ROI (Return on Investment) of samples after image correction, and respectively obtaining five types of ROIs of leaf veins, healthy mesophyll, slightly damaged mesophyll, seriously damaged mesophyll and larvae; establishing partial least square discriminant analysis and least square support vector machine models. The successive projections algorithm, informative variable elimination, UVE-SPA (Uninformative Variable Elimination-Successive Projections Algorithm) and competitive adaptive re-weighted sampling are used for variable selection; the selected optimum model is a UVE-SPA-LS-SVM (Uninformative Variable Elimination-Successive Projections Algorithm-Least Square-Support Vector Machine) model based on visible range data and has a correct prediction rate value of 97.30%. The method disclosed by the invention can achieve effects of rapidly and nondestructively distinguishing mulberry pyralid larvae and damage degrees thereof to the mulberry leaves, providing high-quality mulberry leaves for silkworm raisers and improving the yield of silkworm and the quality of silk, and has an important popularization value on agriculture detection of plant diseases and insect pests.

The invention provides a protein structure prediction method, a protein structure prediction device and a medium. The protein structure prediction method is applied to the computer equipment, the computer equipment comprises a CPU and at least one GPU, and the method comprises the following steps: obtaining a target protein sequence of a to-be-predicted protein structure. And in the CPU, according to the sequence length of the target protein sequence, determining an alignment quantity threshold value of a matching sequence corresponding to the target protein sequence. And comparing the target protein sequence with a plurality of protein sequences in a preset protein sequence library according to the comparison quantity threshold, and determining a matching sequence corresponding to the target protein sequence. And determining a matching structure corresponding to the matching sequence in a preset protein structure database. And inputting the matching sequence and the matching structure into a protein structure prediction model preset in a GPU for protein structure prediction, and obtaining a protein prediction structure corresponding to the target protein sequence. The memory occupation of the GPU can be reduced, the operation speed of the GPU is improved, and the prediction rate is accelerated.

A dynamic analysis apparatus includes a hardware processor. The hardware processor is configured to perform the following, calculate a prediction rate multiplied by a respiratory function value of the subject in predicting the respiratory function value when an exclusion target portion is excluded; obtain input of the exclusion target portion in an anatomical unit from the input unit, based on the anatomical unit, specify a partial region of the lung field in which a characteristic amount relating to a respiratory function in the plurality of frame images is calculated, calculate the characteristic amount related to the respiratory function in the partial region of the lung field specified from the plurality of frame images and the characteristic amount related to the respiratory function of an entire lung field, and calculate the prediction rate based on a characteristic amount ratio which is a ratio of the two calculated characteristic amounts.

A method for protecting a machine learning (ML) model is provided. During inference operation of the ML model, a plurality of input samples is provided to the ML model. A distribution of a plurality of output predictions from a predetermined node in the ML model is measured. If the distribution of the plurality of output predictions indicates correct output category prediction with low confidence, then the machine learning model is slowed to reduce a prediction rate of subsequent output predictions. If the distribution of the plurality of categories indicates correct output category prediction with a high confidence, then the machine learning model is not slowed to reduce the prediction rate of subsequent output predictions of the machine learning model. A moving average of the distribution may be used to determine the speed reduction. This makes a cloning attack on the ML model take longer with minimal impact to a legitimate user.

The invention discloses a streaming media transmission optimization method and system, and the method comprises the following steps: S1, receiving streaming media data, carrying out the statistics of a streaming media code rate, and carrying out the network bandwidth prediction through a Kalman filtering model; S2, carrying out code rate estimation and prediction in a non-packet-loss input mode by using a streaming media code rate and a network bandwidth prediction result; S3, performing bidirectional predictive code rate verification on the round-trip time, the packet loss rate and the code rate estimation predicted value to obtain a sending end output code rate; and S4, according to the given code rate, outputting a corresponding streaming media code rate to carry out flow control. Network condition prediction suitable for wired and wireless scenes is provided, real-time and rapid prediction of the current network rate is met, smooth transition of the code rate is guaranteed, streaming media playing is smoother, TCP friendliness is guaranteed in combination with the TFRC algorithm, meanwhile, the video code rate under low bandwidth is guaranteed, the situation that the video is stuck and paused due to the fact that the prediction rate is too low is reduced. The method is a congestion control solution based on streaming media transmission.

The invention relates to the field of computers, and provides an information recommendation method and device to solve the problem that prediction accuracy is low. The method comprises the steps that in response to an information recommendation instruction, a recommendation information set is obtained; based on an information click rate prediction sub-model, a purchase behavior prediction sub-model and an information conversion rate prediction sub-model in the recommendation prediction model, an information click rate, a purchase behavior prediction rate and an information conversion rate corresponding to each piece of recommendation information in the recommendation information set are obtained; and a corresponding first recommendation probability is obtained based on the information click rate, the purchase behavior prediction rate and the information conversion rate corresponding to each piece of recommendation information, and the recommendation information of which the first recommendation probability exceeds a set threshold value is pushed to the target object. The model for predicting the purchase intention of the target object is added in the recommendation prediction model, so that the prediction accuracy of the whole model can be effectively improved.

An embodiment of the invention provides a health condition predicting method based on data mining and a device thereof. The method comprises the steps of obtaining a physiological parameter which corresponds with the current time of a to-be-tested user; performing characteristic extraction on the physiological parameter for obtaining a random item or combination of corresponding offset degree, instantaneous variance and drift degree; constructing a long-term change trend according to the offset degree, constructing a short-time change violence according to the instantaneous variance, and constructing long-term drift degree according to the drift degree; constructing a danger early-warning factor according to the long-term change trend, the short-time change violence and the long-term driftdegree, and predicating the health condition of the to-be-tested user through the constructed danger early-warning factor and the predicating model for obtaining a predicating result. The device is used for executing the method. According to the health condition predicating method, the corresponding characteristic parameter is obtained through performing characteristic extracting on the physiological parameter; the danger early-warning factor is constructed according to the characteristic parameter; and a predication result is obtained by means of the prediction model, thereby effectively reducing an error prediction rate in a monitoring process.

The invention discloses a tin-bismuth alloy performance prediction method based on transfer learning. According to the invention, the method comprises the steps: constructing a deep belief network, transferring the learned priori knowledge to the feature learning of the new tin-bismuth alloy through the model, and predicting the performances of the new tin-bismuth alloy at different proportions. According to the method, the existing tin-bismuth alloy performance data and prior knowledge of a prediction model are utilized, and under the experiment condition based on a small amount of data, theproblem that data characteristics cannot be effectively learned due to the fact that the sample amount of a deep learning algorithm is too small can be effectively solved; according to the method, thecorresponding performance of the new tin-bismuth alloy can be predicted through transfer learning, and the vacancy of the tin-bismuth alloy performance prediction field to a certain degree is made up; the prediction rate of the performance prediction model achieved through the method is greatly increased compared with the prediction rate of a performance prediction model without transfer learning.

The invention discloses a photovoltaic power generation power prediction method based on deep learning. The photovoltaic power generation power prediction method comprises the following steps: A, acquiring photovoltaic power generation data and sending the photovoltaic power generation data to a memory for storage; B, performing feature extraction on the stored photovoltaic power generation data;C, encrypting the data subjected to feature extraction; D, using the encrypted data as input of a BP neural network, wherein the output of the BP neural network is to-be-predicted photovoltaic power generation power; And E, performing deep training on the BP neural network to obtain the photovoltaic power generation prediction power. The prediction method is high in precision and prediction rate,and the adopted data preprocessing method can realize data sorting, noise reduction and data filtering, so that the subsequent data processing efficiency is improved; According to the adopted featureextraction method, the first keyword and the second keyword are searched, so that the extraction difficulty can be reduced, and the feature extraction precision is improved; The adopted encryption method can perform multiple encryption on the photovoltaic data, so that the security and confidentiality of the data are improved.