151results about How to "Good prediction accuracy" patented technology

A moving picture coding apparatus divides a picture into basic blocks and generates a prediction image of a block to be predicted in a basic block by using adjacent pixels in reference pixel blocks adjacent to the block to be predicted as reference pixels to perform predictive coding of a moving picture. When some of the reference pixels are not available, pixel values of the reference pixels that are not available are calculated based on pixels in the reference pixel blocks. The prediction image of the block to be predicted is generated by using the calculated pixel values instead of the reference pixels that are not available.

The invention provides a method and system of the prediction of time series data. The method comprises the steps that wavelet decomposition is conducted on the sequence formed by t-1 moment data, n subsequences are obtained; stationary detection is conducted on n subsequences respectively; for non-stationary time series, an advance learning LSTM model is built using the t-1 moment data, and the values of t moment are predicted respectively, and forecasts of the non-stationary part are obtained by summing; similarly, for stationary sequences, ARMA models are respectively built and the values of t moment are predicted, and the forecasts on stationary part are obtained by summing; finally the prediction values of the non-stationary part and the stationary part at t moment are summed, to obtain the final forecast value. By the method and system of the prediction of time series data, by wavelet decomposition, the advantages of LSTM and ARMA are combined, in comparison with traditional methods, better effects are provided in dealing with non-stationary time series. In addition, by benefiting from the unique LSTM structure in the model, the forecasting and the generalization ability of the method is better, and the method is suitable for time series prediction in various fields.

The invention discloses a super-short-term prediction method of photovoltaic power station irradiance. The method includes the steps that irradiance data are extracted from a history database, data of a night time quantum are removed, corresponding extraterrestrial theoretical irradiance is calculated, data abnormal detection is carried out based on the preceding operations, and the data are normalized in the difference value ratio method of an extraterrestrial irradiance theoretical value and practical irradiance; a training sample set is extracted according to input and output dimensionality of a model; a model of an irradiance time sequence is built through an ANFIS, a the rule quantity and an initial parameter of the ANFIS model are determined in a subtractive clustering method, and a fuzzy model parameter is optimized in a counter propagation algorithm and a least square method; a prediction sample is input, and a prediction value is obtained through calculation; the prediction value is added to form a new sample set, and multiple steps of prediction are achieved in a cycling mode; counter normalization processing is carried out on the prediction value. Super-short-term prediction of the irradiance can be achieved only by means of a history irradiance time sequence, prediction accuracy is good and the method is easy to carry out.

A method of finding defects in sealing material formed as a frame line on a glass plate includes irradiating the frame line of sealing material. A temperature of the irradiated sealing material is measured and a change of the temperature caused by a nonuniformity in sealing material is detected. Another aspect features a method of hermetically sealing a thin film device between glass plates. Sealing material is dispensed on a cover glass plate in the form of a frame line cell. The sealing material is pre-sintered onto the cover glass plate and cooled. A laser beam is moved around the frame line on the sealing material. A temperature of the sealing material contacted with the laser beam is measured. A change in the temperature (ΔT) caused by a nonuniformity in the sealing material is measured. Further aspects include a feedback process, infrared imaging and use of delta temperature data to increase sensitivity of temperature measurement data.

The invention discloses a gray generalized regression neural network-based small sample software reliability prediction method. The method comprises the following steps of: first, respectively emulating and expanding failure time data and test coverage rate data in collected small sample software reliability test data by using an improved Bootstrap method to form expanded reliability data which has the same failure statistical rule as small sample reliability data; then, obtaining a three-dimensional curve of the failure time, the test coverage rate and the unreliability of the expanded reliability data; next, establishing a gray generalized regression neural network; later on, training the gray generalized regression neural network by adopting the expanded reliability data and establishing a small sample software reliability prediction model; and finally, predicting by using the model to obtain software reliability prediction information. The method avoids solving a complex multivariate likelihood equation, and solves the problem that an available prediction model can be obtained only by training a large number of models in artificial neural network modeling in software reliability prediction.

A method of building a set of experimental prediction model that requires fewer experimental frequency, shorter prediction time and higher prediction accuracy by using the advantages of combining the experimental data of Taguchi method and neural network learning is disclosed. The error between the experimentally measured result of photolithography and the simulated result of the theoretical model of near field photolithography is set as an objective function of an inverse method for back calculating fiber probe aperture size, which is adopted in the following Taguchi experiment. The analytical result of Taguchi neural network model of the present invention proves that the Taguchi neural network model can provide more accurate prediction result than the conventional Taguchi network model, and at the same time, improve the demerit of requiring massive training examples of the conventional neural network.

A regeneration water factory effluent residual chorine risk prediction method comprises: data monitoring and collection are performed so that an index monitored and collected is obtained; the data is transmitted to a server through a communication system; effluent residual chorine risk prediction, analysis and decision are performed by the server according to the collected data: the corresponding monitoring data stored in a data base of the server is read, the monitoring data is input in a water quality model and the effluent residual chorine risk prediction of technology is performed; the index influencing effluent residual chorine risk probability is judged according to a potential of hydrogen (pH) value, the water temperature, ammonia concentration and chemical oxygen demand distribution situation monitored under different risk probabilities; and solutions under different effluent residual chorine risk probabilities are made. The regeneration water factory effluent residual chorine risk prediction method is good in model prediction accuracy (the accuracy is higher than 95%) so that when facing water quality change, a researcher can predict effluent residual chlorine risk accurately in time. The regeneration water factory effluent residual chorine risk prediction method offers a certain reference to operation and management of an actual water factory.

The invention provides a three-dimensional wake numerical simulation method based on a 2D_k Jensen model, which can obtain a novel three-dimensional wake model. The method is characterized in that: step 1, calculating an inflow wind shear curve u0 (z) and the turbulence intensity distribution I0 (z) in the vertical direction; 2, calculating that expansion coefficients kx, z of the wake; 3, predicting that wake wind speed based on the original Jensen model to obtain the initial wake wind speed u* (x, z); 4, calculating that expansion radius rx, r of the wake; 5, utilizing that wake expansion radius rx, r obtained in the step 4, based on the cosine-type velocity distribution proposed by the 2D_kJensen model, three-dimensionally modifying the initial wake wind speed u* (x, z) obtained in thestep 3 to obtain the corrected wake wind speed distribution u (x, y, z). The three-dimensional wake model obtained by the invention inherits the advantages of the engineering model, and has good prediction accuracy for the wake velocity deficit in the direction of flow direction, the cross wind direction and the vertical direction, and can effectively reflect the asymmetric characteristics of thevelocity deficit in the vertical direction, and the calculation accuracy is even better than the numerical simulation result based on the CFD method.

The invention provides an automobile intelligent steering automatic prompting device and method, belongs to the field of intelligent traffic, and aims at solving problems that a driver forgets to turn on a steering lamp when changes lanes in driving or turns a corner so that driving safety is enhanced. The device comprises four parts including a GPS positioning unit, a video recognition unit, a central processor unit and an execution unit. The video recognition unit includes a sight line residing time calculation unit, a lane guidance arrow recognition module and a vehicle both-side lane line recognition module. The execution unit comprises a steering prompting module and a lane changing prompting module. According to the method, the current real-time position of the vehicle is solved via the GPS positioning unit, and the video recognition unit calculates time of the sight line residing at left and right rearview mirrors, recognizes the type of a line ground guidance arrow, and calculates the dynamic distance of the both-side wheels to the lane lines. The central processor unit receives information transmitted by the GPS positioning unit and the video recognition unit, judges the relative position of the lane in which the vehicle is positioned, predicts the behaviors of the driver and issues instructions to the execution unit so that steering and lane changing prompting is realized.

The invention relates to a recommendation prediction method based on attribute information reference self-leaning. Attribution information is utilized for slowing down cold start, the advantages of user and commodity collaborative filtering and model decomposition based on a matrix are combined, the training speed is high, and the interpretability is achieved; meanwhile, on the condition that rating data is sparse, the prediction precision is superior to that of user and commodity collaborative filtering and the matrix-based decomposition method.

Embodiments relate to using a branch target buffer preload table. An aspect includes receiving a search request to locate branch prediction information associated with a branch instruction. Searching is performed for an entry corresponding to the search request in a branch target buffer and a branch target buffer preload table in parallel. Based on locating a matching entry in the branch target buffer preload table corresponding to the search request and failing to locate the matching entry in the branch target buffer, a victim entry is selected to overwrite in the branch target buffer. Branch prediction information of the matching entry is received from the branch target buffer preload table at the branch target buffer. The victim entry in the branch target buffer is overwritten with the branch prediction information of the matching entry.

The invention provides a pseudo-acoustic curve rebuilding and sparse pulse joint inversion method, and belongs to the field of reservoir prediction research. The method comprises the following steps of: 1, performing cross analysis on an acoustic curve and a natural gamma-ray curve; performing statistics on a mathematical relationship of an acoustic parameter and a natural gamma-ray parameter according to the cross analysis; if the two curves have good consistency, regarding that the oil and gas reservoir prediction by the natural gamma-ray curve is effective; and entering the second step; 2, using a wavelet multi-scale transform and information fusion technology to rebuild a DT curve (acoustic-time difference curve) to obtain a pseudo-acoustic curve, wherein the curve includes both natural gamma-ray features and acoustic features; and 3, performing sparse pulse seismic inversion under the DT curve constraints to obtain a final inversion result.

The invention discloses a model modification method with damp considered based on basic stimulation response data and relates to model modification methods with damp considered. The model modification method with the damp considered based on the basic stimulation response data aims to solve the problem that damp modification is not considered in the prior art and comprises the steps that (1) a finite element model is established, and partitioning is conducted on a physic matrix of the model; (2) a transfer function based on basic stimulation is obtained by means of matrix transformation; (3) test data are introduced, and a deviation function of the test data and simulation data is established; (4) a model modification equation is established according to the relation between the deviation function and physic parameter deviation, and required modification can be obtained by solving the equation; (5) a modification parameter obtained in the step (4) is plugged into the finite element simulation model, and then a modified finite element simulation model is obtained. The model modification method with the damp considered based on the basic stimulation response data is applied to the field of finite element simulation of structures as well as the relevant fields such as the test field and the design field.

The invention relates to an electric system wide area dynamical control method and system for compensating distributive communication time delay, and belongs to the field of electric system wide area dynamical control. The electric system wide area dynamical control method provided by the invention comprises the following steps of: after a wide area dynamic controller obtains measurement data, respectively predicting and extrapolating the measurement data according to a pre-set prediction method and maximum communication delay, so as to obtain controller input data which are sufficient to compensate distribution characteristics of the communication time delay; respectively calculating according to the controller input data to obtain a series of control data; uniformly sending the control data to a network control unit; and selecting the corresponding control data as output according to current time by the network control unit for electric system wide area dynamical control. With the adoption of the technical scheme provided by the invention, different prediction algorithms can be selected aiming at different measurements, so that the prediction precision and the instantaneity can be improved; the prediction method is used for covering most part of the communication time delay, so that the distribution characteristics of the communication time delay can be effectively compensated; and a used wide area dynamical controller is mounted in a dispatching center, so that the coordination control of the plurality of wide area dynamical controllers can be conveniently carried out.

The invention discloses an unsupervised monocular depth estimation algorithm based on deep learning. Through comparing the difference between the optical flow generated by the motion of the camera andthe all-optical flow, the detection of the moving target in the scene is realized; finally, the depth estimation effect of the algorithm is improved. According to the method, under the condition thata training label is not needed, for a moving monocular camera video, unsupervised estimation of a depth image, a camera pose and a moving optical flow can be achieved at the same time, the predictionprecision of the three tasks is excellent, and the precision and robustness of the algorithm are effectively enhanced by detecting a dynamic target in a scene.

The invention provides a method for adaptively adjusting operation modes of a process industry on the basis of working conditions, and belongs to the field of information processing. The method includes preprocessing complex industrial field data, and combining the complex industrial field data with technical analysis to select a plurality of variables so as to describe the working conditions; building operation index prediction models to predict indexes, and comparing prediction results to actual measurement results and judging deviation of the prediction results and the actual measurement results; setting threshold values to judge whether production is in a normal state or not, and adaptively adjusting parameters if the production is in an abnormal state; combining samples which generate misjudgment results with an improved support vector machine process to secondarily predict the indexes, secondarily matching prediction results with an operation mode library, searching the optimal operation modes to optimize operation and continuously adjusting and updating matching of the operation modes along with continuous change of the working conditions so as to support operation optimization. The method has the advantages that the parameters can be dynamically and adaptively adjusted on the basis of the working conditions, so that the operation modes can be matched, and the mode matching accuracy and the mode optimization accuracy can be improved.

The invention discloses a feature extraction multi-objective optimization method for a wear condition of a milling tool. The multi-objective optimization method comprises the following steps of collecting multiple physical field signals by multiple sensor channels; calculating a plurality of statistical feature parameters and wavelet energy in a time domain and a frequency domain of a signal of each sensor channel to form a feature parameter candidate set; constructing a multi-objective optimization model by taking tool condition prediction accuracy and a feature parameter number as optimization objectives; performing global optimization on the optimization model by employing an intelligent optimization algorithm; and making a sensor feature parameter set corresponding to an optimal solution of the optimization model serve as feature parameters required by tool condition monitoring. The multi-objective optimization method disclosed by the invention has the following advantages and effects that tool wear loss prediction accuracy corresponding to each feature parameter set is inspected from prediction accuracy by taking the prediction accuracy and the feature parameter number as theoptimization objectives, so that the phenomenon of high relativity but low prediction accuracy is avoided.

A method for video encoding based on an image super-resolution, the method including: 1) performing super-resolution interpolation on a video image to be encoded using a pre-trained texture dictionary database to yield a reference image; in which the texture dictionary database includes: one or multiple dictionary bases, and each dictionary basis includes a mapping group formed by a relatively high resolution image block of a training image and a relatively low resolution image block corresponding to the relatively high resolution image block; 2) performing motion estimation and motion compensation of image blocks of the video image on the reference image to acquire prediction blocks corresponding to the image blocks of the video image; 3) performing subtraction between the image blocks of the video image and the corresponding prediction blocks to yield prediction residual blocks, respectively; and 4) encoding the prediction residual blocks.

The invention relates to an intelligent diagnosis diagnosis method and device based on an ensemble learning framework, equipment and a medium, and the method comprises the steps of collecting multi-source operation data of a target metrological verification automatic system, and carrying out the preprocessing of the multi-source operation data to obtain a corresponding fault data sample set; constructing a readable target data set of a preset fault diagnosis model according to the fault data sample set, wherein the preset fault diagnosis model adopts an XGBoost model, training the XGBoost model according to the target data set, and taking the corresponding XGBoost model after training as a target fault diagnosis model; and acquiring online operation data of the target metrological verification automatic system in real time, and analyzing the online operation data according to the target fault diagnosis model to predict the fault condition of the target metrological verification automatic system. According to the invention, the fault condition of the metrological verification automatic system can be accurately predicted on line in real time, and the fault diagnosis efficiency and accuracy are improved.

In a neural network, hidden layers are modified by supplying input data, an output label, and internal teaching labels to the neural network; causing the neural network to process the input data through the hidden layers and outputting a result of the processing for comparison with the output label; supplying the internal teaching labels to the hidden layers and calculating scores for the hidden layers based on the internal teaching labels; and modifying the hidden layers or hidden nodes based on the calculated scores and the comparison of the processing result with the output label. The modifications to the hidden layers or hidden nodes may involve pruning hidden nodes by dropping lower scoring nodes; reducing a number of bits in computations and outputs; reducing a number of bits in selected nodes; bypassing lower scoring nodes; modifying activation functions of the hidden nodes based on the calculated scores; and/or adding hidden layers or hidden nodes.

The invention discloses a digestive tract endoscope image processing method and device, a storage medium, equipment and a system, belongs to the technical field of artificial intelligence, and relatesto a computer vision technology and a machine learning technology. The digestive tract endoscope image processing method comprises the following steps: acquiring a to-be-detected digestive tract endoscope image; classifying the digestive tract endoscopic images to be detected based on a first model, the first model being obtained by training based on a first training data set under the constraintof a second model, the first training data set comprising a pure data set and a noise data set, and the second model being obtained by training based on a second training data set before training thefirst model; wherein the pure data set comprises sample images with consistent annotations, and the noise data set comprises sample images with inconsistent annotations, and the second training dataset is a subset of the first training data set and comprises a pure data set, and the sample images are digestive tract endoscope images. According to the digestive tract endoscope image processing method, the training data volume is increased, and meanwhile, the influence of label labeling errors on the model prediction accuracy can be reduced.

The invention discloses an image quality evaluation method based on importance between deep network features. According to the method, a module for judging the importance relationship between the feature maps is added to end-to-end neural network model training, the image quality can be predicted more accurately, and a high generalization capability is shown on each image quality assessment (IQA)data set. The method specifically comprises the following steps of 1) preparing an image quality assessment data set for training a test network model, and randomly dividing the image quality assessment data set into a training set and a test set according to image contents; 2) adding a SeNet module into the vgg-16 network to build a neural network model VGG*-SE in multiple different combination mechanisms for image quality assessment, using the training data set to train the network models respectively, selecting the model as a final model when the trained model reaches expected precision onthe test data set, and storing parameters after training of the network model; and 3) calculating the prediction precision of the test set by using the selected final model.