115results about How to "The predicted value is accurate" patented technology

The invention discloses a complex scene-based human body key point detection system and method. The method comprises the following steps of: inputting monitor video information to obtain a single-frame static map and a multi-frame optical flow graph; extracting features of the single-frame static map through a convolution operation so as to obtain a feature map, and in order to solve influences, on personnel target detection, of interference targets under complex scenes, judging a practical confidence coefficient and a preset confidence coefficient of the feature map by adoption of a personneltarget detection algorithm so as to obtain a discretized personnel target surrounding box; and carrying out optical flow overlapping on the multi-frame optical flow graph to form a two-dimensional vector field; extracting features in the discretized personnel target surrounding box to obtain a feature map, obtaining key points and association degrees of parts, generating a part confidence map foreach part of a human body by utilizing a predictor, and accurately detecting human body key points through the part confidence map and the two-dimensional vector field. The system and method are usedfor human body key point detection under complex scenes so as to realize accurate detection of personnel target key points.

The invention discloses a method and device for obtaining a predicted value of a motion vector. The technical scheme provided by the embodiment of the invention is as follows: the method comprises the following steps of: obtaining position coordinates and a motion vector of more than one encoded/decoded block neighboring a current block; obtaining association parameters of the motion vector and the position coordinates in accordance with the preset congruent relationship between the motion vector and the position coordinates, and the position coordinates and the motion vector of more than one encoded/decoded block; and obtaining a predicted value of the motion vector of the current block in accordance with the preset congruent relationship between the motion vector and the position coordinates, and the position coordinates of the current block and the association parameters of the motion vector and the motion vector. The embodiment disclosed by the invention can be applied to the video image encoding and decoding process, such as H. 264/AVC and the like.

The invention discloses a dynamic migrating method of a virtual machine based on performance prediction. The method comprises the following steps of obtaining a characteristic vector by using a singular value decomposing principle by extracting system performance data and analyzing the similarity of the singular value to obtain; then, obtaining the performance prediction value at the next moment by a reverse SVD (Singular Value Decomposition) algorithm so as to start a dynamic migrating mechanism of the virtual machine according to future performance prediction values. According to the method, the performance prediction mechanism can make accurate prediction value and the dynamic prediction mechanism can balance the system load. The system can effectively cope with sudden load by effectively analyzing current information of a physical machine and effectively predicting the future load, so that the performance loss caused by unnecessary migration is reduced.

The invention discloses a point cloud prediction method based on a space sequence, a point cloud layering method and equipment. The point cloud prediction method comprises the steps: generating two ormore spatial sequences of point clouds; carrying out forward and/or backward search of the current point by using the two or more space sequences to obtain neighbors of the current point; and determining an attribute prediction value of the current point according to the neighbor of the current point. The point cloud layering method comprises the steps: generating spatial sequence codes of all the points by utilizing coordinates of point cloud; and for the points with the same high order of the spatial sequence code, selecting part of the points, and dividing the point cloud into two or morelayers. According to the point cloud prediction method provided by the invention, a flexible neighbor search mode is realized, a more accurate prediction value is provided during point cloud attributeprediction, hierarchical division of the point cloud is realized through the point cloud layering method provided by the invention, and the geometric information and attribute information correlationof the point cloud can be further utilized to improve the compression performance of the point cloud attributes.

The invention provides a prediction data-based transformer early warning evaluation method and device. The method includes the following steps that: a prediction model of gas concentration is established on the basis of acquired training sample data; a fuzzy similarity matrix is established on the basis of prediction data obtained by the prediction model, so that an early warning threshold value can be obtained; and an early warning is reported or not reported according to the early warning threshold value and a national standard limit value. With the prediction data-based transformer early warning evaluation method and device provided by the technical schemes of the invention adopted, the early warning threshold having a plurality of standards is set according to the features of various types of early warning indexes, and with national standards adopted in combination, early warning results can satisfy actual application requirements to the greatest extent.

The invention discloses a method for detecting the flow abnormity of a wireless sensor network based on a GM model. The method employs the GM (1, 1) model, is small in amount of used historical data, is quick in building speed of a model, is accurate in prediction value, and is very suitable for the condition that the node energy and calculation capability of the wireless sensor network are limited. enabling a historical modeling data quantity to be fixed through employing a sliding window in a proper size, thereby guaranteeing the quickness of modeling and also guaranteeing the latest effectiveness of historical data; optimizing albinism differential equation solving initial conditions of the GM (1, 1) model, and enabling the prediction value to be more accurate; generating a flow prediction value, finally used for abnormal judgment, at the next moment through the exponential weighting mean of the former L predication values, thereby introducing certain inertia to the prediction of flow. When an abnormal flow happens, a normal flow prediction model cannot be changed easily, but a normal flow prediction value can be obtained better, and the flow abnormality can be detected more easily.

The invention discloses an industrial process soft measurement method based on layer-by-layer data expansion deep learning, and belongs to the soft measurement technology field. According to the technical scheme disclosed by the invention, the sample size of the process variable is expanded by adopting the data expansion auto-encoder; a plurality of data expansion auto-encoders are stacked to forma deep network model; a large number of samples from a low level to a high level are obtained layer by layer from industrial process data, enough sample size is provided for a deep learning model, accurate prediction of product quality is achieved, the method can be widely applied to product quality prediction of complex industrial processes such as a hydrocracking process and a steel sintering process, and the method has the advantages of being high in prediction precision, good in generalization and the like.

The invention discloses a rapid food and drug detection method based on a near infrared technology. The method comprises the following steps: collecting a modeling detection sample, establishing a near infrared technology analysis model, debugging the model data, performing rapid field detection, drawing a conclusion and preserving the conclusion. According to the method disclosed by the invention, the rapid field detection is performed by utilizing a near infrared spectrum, and the information is simple and rich; nondestructive sampling is adopted, direct sample introduction is realized, the operation is simple, pretreatment of the sample is not needed, the sample is green and environmental-friendly, and the performance is reliable; only the detection amount of 0.1mg is needed, related consumables and maintenance cost are avoided, the operating cost and environmental protection risk are reduced, lots of manpower, material resources and financial resources are saved, actual problems such as long-term hidden grass-roots criminal activities, wide supervision range, large quantity and high difficulty are effectively solved, and the aim of guaranteeing the quality safety of foods and drugs is achieved.

The invention discloses a high-speed railway sound barrier insertion loss prediction method of a five-sound-source mode. The method comprises the following steps of equivalently simplifying a high-speed railway noise source into a wheel-rail area noise source, a train body lower part noise source, a train body upper part noise source, a current collection system noise source and a bridge structure noise source according to the composition, position, frequency characteristics and attenuation law of the high speed railway noise source, and respectively calculating the insertion loss of a wing plate for the five sound sources by virtue of a single-sound-source mode; spreading the noise of a wheel-rail area, a train body lower part, a train body upper part, a current collection system and a bridge structure to be superimposed with noise on a sensitive point after a sound barrier is installed, obtaining a main noise grade after the sound barrier is installed, subtracting the main noise grade from the noise grade before the sound barrier is installed, introducing an insertion loss correction item of the bridge wing plate, and obtaining an insertion loss prediction value by adopting the five-sound-source calculation method. By adopting the method, the weaknesses in the traditional sound barrier insertion loss prediction method can be overcome. The high-speed railway sound barrier insertion loss prediction method is applied to the high-speed railway sound barrier engineering design and the ambient noise influence valuation and has the advantages of accurate prediction value and high engineering practicability.

The invention provides an oil refining process product prediction method based on variable weighting deep-learning and a system. The method includes: acquiring process variables in a debutanizer process, and using the process variables as input of a trained deep-learning model to obtain a product quality prediction value on the basis of the deep-learning model, wherein the deep-learning model includes at least three variable weighting autoencoders, and when the deep-learning model is trained, hidden-layer feature data of a variable weighting autoencoder which is in every two adjacent variableweighting autoencoders and is arranged at a front are used as input variables of a variable weighting autoencoder arranged at a rear, and the variable weighting autoencoder arranged at the rear is trained. The multiple weighting autoencoders are utilized to be stacked to form the deep network model, deep output-related features from a low level to a high level can be gradually obtained, features related to a quality index can be enhanced, the accurate prediction value can be provided for product quality, and the method has the advantages of high prediction precision, good generalization and the like.

The invention discloses a video lossless compression method based on adaptive model selection. The method comprises the following steps: judging the type of an input video, determining operand, performing video pre-processing operation; adopting the airspace redundancy-removing method, the time domain redundancy-removing method and the direct mode redundancy-removing method to perform redundancy-removing operation to the current sample and obtain the predictive value of the current sample; utilizing the redundant information of space, time and frequency domain to separately perform infraframe prediction and interframe prediction to the original YUV sequence; if the original sequence is RGB sequence, performing infraframe prediction, interframe prediction and direct prediction to the converted sequence; using an adaptive prediction model selector to judge the predictive value of the current sample and determine that the minimum is the optimal prediction error; and finally performing context-based arithmetic encoding to the map value of the final prediction error and outputting to obtain bit stream. The invention uses the adaptive prediction model selector to replace the additional bit prediction model, thus obviously increasing the compression efficiency of video; and the method can be used for video compression in the fields of aviation and navigation.

The invention discloses a data mining method and system based on time series. The method comprises following steps: obtaining original event records within a set period of time and determining a basic event type and M related event types according to the original event records; generating a first historic time series and M second historic time series corresponding to the basic event type; calculating correlation coefficients between a low-frequency time series of the first historic time series and a low-frequency time series of each second historic time series; determining a low-frequency time series of a kth second historic time series when the correlation coefficient is larger than a first threshold value and establishing a prediction model according to the low-frequency time series of the kth second historic time series and the low-frequency time series of the first historic time series, and obtaining a predicted value of the basic event type according to the prediction model, wherein M is a positive integer, and K is an integer no more than M. The method and system are used for solving the problem of inaccurate predicted results in the prior art.

A control system for a plant is provided. This control system can control the plant more stably, when the model parameters of the controlled object model which are obtained by modeling the plant, which is a controlled object, are identified and the sliding mode control is performed using the identified model parameters. The model parameter identifier (22) calculates a model parameter vector (θ) by adding an updating vector (d θ) to a reference vector (θ base) of the model parameter. The updating vector (d θ) is corrected by multiplying a past value of at least one element of the updating vector by a predetermined value which is greater than “0” and less than “1”. The model parameter vector (θ) is calculated by adding the corrected updating vector (d θ) to the reference vector (θ base).

The invention discloses a rotary hot spare dispatching method in the construction of an intelligent power grid on the basis of a relevance vector machine. The invention relates to the rotary hot spare dispatching method in the construction process of the intelligent power grid on the basis of the relevance vector machine and aims to solve the problem that a rotary hot spare is difficult to set for a new energy resource power system to stabilize the power fluctuation of a wind power connected grid, wherein the new energy resources comprise scale wind power and the like. An initialization setting result is transferred into a wind power relevance vector machine forecasting system; a wind power station wind-power acquisition module acquires a measurement value of a wind power of a wind power station in real time and transfers data into the wind power relevance vector machine forecasting system after carrying out data preprocessing; the wind power relevance vector machine forecasting system receives the data and carries out forecasting on the wind power at the moment in the future so as to obtain a forecasting result, i.e. an error band of the wind power value at the moment in the future and the wind power; and the obtained forecasting value and error band are sent into a dispatching controller, wherein the forecasting value is a power generation plan of the wind power station in the future and the power range represented by the error band is rotary hot spare distributed to the wind power station. The rotary hot spare dispatching method is used in the construction of the intelligent power grid.

The present invention discloses a folium apocyni veneti total flavonoid near infrared super rapid detection method including chemical value measurement, original spectrum acquisition, spectral data preprocessing, PLS quantitative calibration model establishment, quantitative calibration model stability prediction, folium apocyni veneti total flavonoid mode content model prediction and other steps. Compared with the prior art, waste time and energy, higher operating and maintenance cost and other problems of folium apocyni veneti total flavonoid detection methods in the prior art can be fundamentally solved, a PLS quantitative model is established by use of near infrared spectroscopy for super rapid detection without sample pretreatment, the folium apocyni veneti total flavonoid near infrared super rapid detection method is convenient, fast, free of side effects on the human body and the environment and small in testing result relative deviation, and if a test sample chemical measured value has high precision, a predicted value of the method may be close to a true value.

The invention relates to an electric power engineering fund prediction method and device, belongs to the technical field of electric power engineering, and solves the problem of inaccurate prediction result in the prior art. The electric power engineering fund prediction method comprises the steps: comprehensively analyzing a plurality of historical projects to determine influence factors of electric power engineering fund budget, wherein the influence factors comprise project periods, the number of participants, material usage and equipment usage; collecting influence factors of historical projects and corresponding electric power engineering funds; performing normalization processing on the influence factors, and then dividing the normalized influence factors and the corresponding electric power engineering funds into two sets including a training set and a verification set; constructing a neural network model based on Bayesian formula optimization, and training and verifying the neural network model through the training set and the verification set to obtain a prediction model; and inputting the influence factors of the to-be-predicted project into the prediction model to obtain a fund prediction value. The accuracy of prediction results is improved.