56results about How to "Improve forecast quality" patented technology

The invention discloses a method for short-term prediction of salt tide at a water intake in a tidal river region. The method comprises the following steps of: analyzing the following influencing factors: the flow of upstream hydrologic monitoring stations and the tide level difference, water unavailable hours and average chloride concentration of the water intake in a day so as to establish a BP-RAGA coupled neural network salt tide forecasting model and predict the water unavailable hours and average chloride concentration of the water intake in the next day; and determining the water available hours and amount of each water intake on the basis of the prediction so as to determine the time and space distribution of raw water intaking amount of the tidal river in the next day and finally draw up a dispatching predetermined plan of a raw water system by combining an optimal dispatching module of the raw water system, so that the safe operation of the raw water system and safe urban water supply are ensured. Compared with the conventional prediction systems, the method can dynamically determine the weight of each influencing factor in the prediction module by fully using existing information, improve the sensitivity and precision of the prediction and have lower rate of missing report and false report.

In a method for increasing the spatial resolution of a weather forecast, in which forecast predicted meteorological parameters are available for grid points of a first grid and the predicted meteorological parameters are interpolated to grid points of a second grid so as to obtain interpolated meteorological parameters, wherein the grid points of the second grid have a higher spatial resolution than the grid points of the first grid, a statistical postprocessing method is used to predict at least one output quantity of an analysis method by which forecast meteorological parameters can be interpolated to the grid points of the second grid.

The invention discloses a high-risk pollution source classification forecasting method based on principal component analysis and random forest. The method includes the steps of collecting and integrating environmental pollution source behavior data of enterprises into primary selection indexes, and screening out illegal pollution source behavior indexes influencing pollution sources to serve as a high-risk pollution source index system; conducting data cleaning and data normalization processing on the environmental pollution source behavior data; finding out a function relationship indicating whether or not the high-risk pollution source index system and the pollution sources are illegal, and building a random forest model; conducting model training and evaluating the precision of the random forest model after training is finished; sorting importance degrees of the pollution source behavior indexes; conducting the principal component analysis to obtain principal components, utilizing the principal components to conduct weighting and work out comprehensive scores; according to the comprehensive scores, judging the risk score coefficient of each enterprise, automatically ranking the risk core coefficients and generating a TOP enterprise list, wherein the risk score coefficients indicate the occurrence probability of illegal behaviors of the corresponding enterprises. The high-risk pollution source classification forecasting method based on the principal component analysis and the random forest can reduce complexity of operations and improve forecasting precision and the quality of results.

The present disclosure describes a system (100) for reducing background noise from a speech audio signal generated by a user. The system (100) includes a user device (102) receiving the speech audio signal, a noise reduction device (118) in communication with a stored data repository (208), where the noise reduction device is configured to convert the speech audio signal to text; generate synthetic speech based on the converted text; optionally determine the user as an actual subscriber based on a comparison between the speech audio signal with the synthetic speech; and selectively transmit the speech audio signal or the synthetic speech based on comparison between the predicted subjective quality of the recorded speech and the synthetic speech.

A process modelling technique uses a single statistical model, such as a PLS, PRC, MLR, etc. model, developed from historical data for a typical process and uses this model to perform quality prediction or fault detection for various different process states of a process. Training data sets of various states of the process are stored and the training data divided into time slices. Mean and / or standard deviation values are determined for both the time slice parameters and variables and the training data. A set of deviations from the mean are determined for the time slice data and the model generated based on the set of deviations. The modeling technique determines means (and possibly standard deviations) of process parameters for each of a set of product grades, throughputs, etc., preferably compares on-line process parameter measurements to these means and use these comparisons in a single process model to perform quality prediction or fault detection across the various states of the process. Because only the means and standard deviations of the process parameters of the process model are updated, a single process model can be used to perform quality prediction or fault detection while the process is operating in any of the defined process stages or states. Moreover, the sensitivity (robustness) of the process model may be manually or automatically adjust each process parameter to tune or adapt the model over time. An alternative aspect is a method of displaying process alert information using a user interface having multiple screens.

The invention relates to a method and to an arrangement for the forecast of wind-resources of a wind-farm. The forecast is done by a numerical weather-prediction-tool, where the weather-prediction-tool uses a long-term data-set of meteorological data. This data are related to the location of the wind-farm. A wind-speed measurement is done by at least one wind-turbine of the wind-farm to do a parameterization of an atmospheric turbulence. The wind-speed measurement is used to generate a data-stream, which is combined with the data-set of the meteorological data to do the forecast.

The invention discloses a method for quick predicting total bacterial count of a potable water network based on BP (Back Propagation) neural network, comprising the following steps of (1) acquiring the total bacterial count of the tested potable water network and other related water quality index values influencing the count as detection data; (2) establishing an error-based backpropagation neural network; (3) training and testing the neural network; and (4) predict the total bacterial count of the potable water network by utilizing the neural network passed the test. The invention can establish a prediction model of the total bacterial count of the potable water network by 6 water quality indexes having high relativity with the total bacterial count and being quick tested and obtained by instruments only through limited tests, accurately and quickly predict the total bacterial count in the potable water network through computer simulation tests and scientific prediction, provide reliable information for water supply enterprises and guarantee bacteriological water quality security of the water supply network.

The invention discloses a prediction algorithm for recognizing tyrosine posttranslational modification sites. The algorithm comprises the steps of data collection, data processing, feature coding, feature optimization and model training and evaluation. The invention furthermore discloses application of the prediction algorithm. According to the algorithm, features of the tyrosine posttranslational modification sites are extracted comprehensively from the perspectives of protein sequence information, evolutional information and physical and chemical properties, Elastic Net is used as an optimization means to automatically select variables to screen multidimensional features, redundant information is removed, a prediction model of tyrosine nitration, sulfuration and phosphorylation sites is constructed in combination with an SVM, the prediction capability of the prediction model is improved, and the prediction quality of the tyrosine posttranslational modification sites is remarkably improved. Through a developed prediction software platform TyrPred, predictive analysis of nitration modification sites, sulfuration modification sites and phosphorylation modification sites of tyrosine on intact protein is realized, and a convenient, economical and rapid research tool and important reference are provided for research of tyrosine posttranslational modification.

The invention relates to a prediction method of pharmacokinetic property and toxicity of a drug molecule based on a support vector machine, which belongs to the molecule design field assisted by computers. The method fully takes advantage of the statistical learning modeling of the support vector machine, adopts an integrated method and simultaneously carries out the selection of a drug molecule descriptor and the optimization of SVM parameter. The method thereof comprises the following implementation steps: the descriptor is calculated and pre-treated, a descriptor data set is re-scaled, and the integrated method is adopted to carry out the selection of the descriptor and the optimization of the SVM parameter simultaneously. The optimization of the SVM parameter uses a conjugate gradient method to optimize penalty function C and kernel function Gamma. Genetic algorithm is used for selecting the descriptor and the individual fitness degree function adopts the fitness function which can comprehensively reflect prediction accuracy and the number of descriptors. In the integration of the selection of the descriptor and the optimization of SVM parameter, fitness degree function of each individual is calculated by SVM optimized parameter to complete the data integration of roulette, hybridization and mutation. The method fully takes two processing advantages of SVM and computer and significantly improves prediction result and efficiency.

The invention relates to an adapter device (10) and a method for charging a vehicle (20), having an interface (11) for detecting internal vehicle operation data (30) containing factors which report driving habits according to lifestyle, and an interface (12) for detecting details related to the fluctuation of energy prices (31), having a device (13) for detecting and planning requirements, said device (13) being designed for deducing an energy requirement profile (40) using the vehicle operation data (30) and for producing a future requirement plan based on at least one of the named factors, said device (13) further being suitable for deducing the duration and frequency of vehicle down times (41, 41') by incorporating the requirement plan, having a charging optimizing device (14) which isdesigned for comparing the vehicle down times (41, 41') with the energy price fluctuation (31) details and for producing a vehicle charging plan (42) which is optimized for time and / or price and is based on the results of the comparison, and having a charging control unit (15) which is designed for charging the vehicle (20) from an energy storage (21) in a manner controlled by the charging plan.

The present disclosure describes a system (100) for reducing background noise from a speech audio signal generated by a user. The system (100) includes a user device (102) receiving the speech audio signal, a noise reduction device (118) in communication with a stored data repository (208), where the noise reduction device is configured to convert the speech audio signal to text; generate synthetic speech based on the converted text; optionally determine the user as an actual subscriber based on a comparison between the speech audio signal with the synthetic speech; and selectively transmit the speech audio signal or the synthetic speech based on comparison between the predicted subjective quality of the recorded speech and the synthetic speech.

The embodiment of the invention provides an image parallel processing method and device and electronic equipment, and relates to the field of video encoding and decoding. The method comprises the steps that firstly, an obtained original image is divided into a plurality of prediction units, each prediction unit comprises a plurality of to-be-processed pixel groups, and each to-be-processed pixel group comprises to-be-processed pixel points; intra-frame prediction is performed on to-be-processed pixel points in a plurality of to-be-processed pixel groups according to a plurality of preset parallel processing units, the plurality of parallel processing units corresponding to the plurality of to-be-processed pixel groups. According to the method and the device, the plurality of parallel processing units are superposed to perform intra-frame prediction on one prediction unit in parallel, so that the whole intra-frame prediction process is accelerated, and the overall parallel granularity,processing efficiency and prediction quality are remarkably improved.

The invention discloses a training method and a using method of an image semantic segmentation model. The training method comprises the steps that training images marked with semantic segmentation information in advance are input into a feature extraction module of a network; the feature extraction module combines the two advantages of the high processing speed of a low-resolution picture and the high deduction quality of a high-resolution picture, and a feature map obtained through calculation is output; then the feature map is sent into a segmentation module for deconvolution, and the feature map is restored to 1 / 4 size of the original map; the type weight of each pixel is marked to obtain a prediction result; and finally, the parameters of the network are corrected according to the prediction information of the trained image and the information marked in advance. The use method is similar to the training method, and the last 1 / 4 size of image is upsampled and recovered to the original image size. According to the segmentation method, the calculated amount and the consumed time are greatly reduced, and the segmentation method can run at the speed of 30 frames under the high resolution of 1024 * 2048, and meanwhile the high-quality inference effect is achieved.

The invention discloses a group recommendation method based on a bidirectional tensor decomposition model. The method comprises the following steps: 1) defining an interactive relationship DS representing a group G, a user U and a product I; 2) establishing a tensor decomposition model; 3) performing transform solving on the tensor decomposition model by utilizing a Bayesian personalized ranking method so as to obtain each parameter value in the tensor decomposition model; 4) obtaining group preference rg,i of the gth group on the ith product, and traversing all commodities to obtain the group preference of the gth group on all the products; and 5) performing descending sort on the group preference of the gth group on all the products, and selecting the previous N products to serve as a recommended product list to be pushed to the gth group. According to the method disclosed by the invention, the personal preference is modeled into a bidirectional process, the real formation process of the personal preference can be effectively reflected, the group recommendation precision is improved, and the method has better robustness.

The invention discloses a method for predicting inhibition concentration of a cytochrome P450 enzyme CYP1A2 inhibitor by utilizing simplified partial least squares. The method comprises the steps of: 1) collecting, processing and optimizing sample sets; 2) constructing molecular descriptors of the inhibitor; 3) preliminarily screening molecular descriptors of the inhibitor; 4) scaling data sets of molecular descriptors of the inhibitor again; 5), dividing data sets of molecular descriptors of the inhibitor; 6), setting up a QSAR model; and 7) predicting inhibition concentration of the cytochrome P450 enzyme CYP1A2 inhibitor.The method utilizes the method of partial least squares (PLS) to further estimate and simplify constituents of partial least squares (PLS) in the presence of serious multiple correlations of independent variables on the basis of regression modeling and uses the simplified method of partial least squares (SIMPLS) to precisely predict inhibition concentration of the cytochrome P450 inhibitor.

The invention provides a periodic prediction neural network based industrial data fault prediction method. The method includes steps: (1) designing mode exploration branch rings, and establishing an equipment operating state mode for storage and representation of various operating states and time-series relationships of equipment; (2) designing a periodic prediction neural network for predicting anext state and fault probability according to time series of a current period in a cloud environment; (3) in the cloud environment, performing dynamic node selection and data duality in an operatingprocess to iteratively explore new faults which are not occurred, and establishing a data experiment mechanism to realize fault prediction in a real-time cloud environment. The periodic neural networkprediction based industrial data fault prediction method has the advantage that by combination of industrial data fault prediction and the neural network and design of a periodic prediction neural network model, algorithm prediction precision is improved while the algorithm execution rate is increased.

The invention discloses reference frame generation method and device of a B-frame image based on bidirectional prediction, and belongs to the technical field of video coding. The method comprises thefollowing steps: acquiring a first reconstructed image and a second reconstructed image which are equal in time distance and reverse in direction to the B-frame image to be coded; generating the predicted image of the B-frame image to be coded based on the first reconstructed image and the second reconstructed image; and performing quality enhancement on the predicted image so as to obtain the reference frame of the B-frame image to be coded. The generated reference frame and the B-frame image to be coded are enabled to be at the same moment by bidirectional prediction so as to improve the prediction quality of the B-frame image to be coded.

The invention provides a carbonatite reservoir prediction method based on a pre-stack multi-attribute and ancient landform fusion technology. The method comprises the following steps: 1) performing rock physical analysis on an area where a layer section to be predicted is located based on drilled well data, and determining pre-stack elastic parameters sensitive to a carbonate reservoir; 2) for theinterval to be predicted, determining a relational expression between the longitudinal wave velocity and the transverse wave velocity based on the drilled well data, determining a relational expression between the density and the longitudinal wave velocity, and then performing pre-stack AVO attribute inversion calculation to obtain specific numerical values of the determined pre-stack elastic parameters sensitive to the carbonate reservoir; 3) respectively determining a current structural form and an ancient landform for the interval to be predicted; and 4) for the interval to be predicted, performing data normalization and fusion processing on each pre-stack elastic parameter obtained by inversion of the determined current structural form, ancient landform and the determined pre-stack AVO attribute, and performing carbonate reservoir prediction by using a fused result.

The invention discloses a CNN-LSTM-based wind speed prediction method. The CNN-LSTM-based wind speed prediction method comprises the steps of: cleaning original record data of meteorological elements;taking the data of the F meteorological elements of N stations as input, and performing standardization processing on the data through adopting a Z-score method to enable the data to meet (0, 1) standard normal distribution; linearly combining the original meteorological elements by utilizing a PCA technology, and converting the original meteorological elements into a group of linearly irrelevantvariables; extracting a meteorological element feature set influencing the wind speed change through adopting an LASSO algorithm, and taking the meteorological element feature set as the input of a prediction model; extracting a potential spatial relationship between a target station and adjacent stations through adopting a spatial feature extraction algorithm to obtain Tspatial feature vectors on forecast times, and analyzing and checking the spatial relationship of wind speed change in combination with a Moran index; extracting a time feature relationship on the T spatial feature vectors through employing a time feature extraction algorithm, and continuously optimizing the time feature relationship by adopting an Adam algorithm; and taking an MAPE as an evaluation index, and verifying the accuracy rate of wind speed prediction on the test set.

The invention discloses a three-dimensional dose prediction method and system in a personalized precise radiotherapy plan. The method comprises the following steps of: 1, acquiring radiotherapy related information such as an electronic computed tomography image, a dangerous organ structure mask image and a three-dimensional dose distribution image; 2, performing data preprocessing operation on the images in the step 1; 3, inputting the acquired image data into a two-stage generator network to generate a three-dimensional dose distribution image and a confidence map; 4, confronting the three-dimensional dose distribution prediction image and the three-dimensional dose distribution real image by adopting a Markov discriminator; 5, jointly optimizing a prediction model through a reconstruction loss function, a reconstruction loss function with the confidence coefficient weight and a confrontation loss function; and 6, generating three-dimensional dose distribution by using the trained prediction model. Through the three-dimensional dose prediction method and system in the personalized precise radiotherapy plan provided by the technical schemes of the invention, manual intervention in the radiotherapy plan can be reduced, the dose prediction precision is improved, and personalized precise radiotherapy is realized.

The invention relates to an air quality prediction method based on a deep neural network. The method comprises steps: S1, the concentration of each air pollutant at each time point at different locations is acquired; S2, the concentration of each air pollutant is ranked to acquire a concentration vector; S3, the concentration vectors at the same time point at different locations are used as a datagroup, and all data groups in a time sequence are used as a data set; S4, a known set containing the concentration values of all pollutants is selected from multiple data sets, a pollutant concentration data group which is missing in a to-be-detected set is found out, and a corresponding time point is determined; S5, a data group corresponding to the time point is selected from the known set as input data, the rest data groups are used as output data, and input vectors are acquired; and S6, a deep neural network model is built, other data groups in the to-be-detected set are used as input values to be inputted to the model to acquire output values as prediction values. Compared with the prior art, the method has the advantage of improving the prediction precision.

The invention discloses a ship trajectory prediction method based on a CNN and a BILSTM. The method comprises the following steps: 1, preprocessing information data; the information data comprises longitude, latitude, course and navigational speed acquired by a ship AIS (Automatic Identification System); 2, dividing the information data preprocessed in the step 1 into a training set and a test set; 3, importing the training set in the step 2 into a convolutional neural network (CNN) for feature extraction; 4, combining the features extracted in the step 3 with data of the training set to form input data of trajectory prediction; 5, importing the input data in the step 4 into a BILSTM neural network model for learning, and obtaining a ship motion rule model implied in the trajectory data; and 6, predicting the trajectory of the ship by using the model in the step 5. The method is used for solving the problem that the ship trajectory prediction accuracy is low in a complex water traffic environment.

Provided is a method for computer-implemented determination of a data-driven prediction model. The method processes digital input data having input and output variables and being semantically annotated based on a digital semantic representation having a hierarchical tree structure where each tree in the structure represents an input variable of the data, the leaf nodes of the respective tree being the discrete values of the input variable. The method of the embodiment provides a recoding of those discrete values by cutting off hierarchical levels of the respective trees. Based on this recoding, a plurality of data modifications is determined for the input data. Those data modifications are trained by a machine learning method where the trained machine learning method with the highest prediction quality is derived from the trained machine learning methods.

The invention relates to a formation pressure prediction method based on stochastic simulation. The method comprises the steps that according to an information entropy theory, the optimum size of a data template is determined, and the data template is utilized to scan a training image of formation pressure and extract pattern features of the training image; dimensionality reduction is carried outon the pattern through principal component analysis or multidimensional transformation, a k-means algorithm is utilized for classification, a mean value of categorical data is acquired and is taken asa representative mean value of the category, the known formation pressure data is compared with the above mean value to obtain differences, and formation pressure data in a category with the minimumdifference is taken as a predicted value. Compared with the prior art, the method can effectively achieve the formation pressure data prediction, and the predication speed and quality can be improved.

The invention relates to a density forecast method of the chlorophyll a stemmed from a water body of the BP neural network. The density forecast method comprises the following steps: (1) the chlorophyll a in the tested water body and the value of other correlative water quality index which influences the chlorophyll a are acquired as the examination data. (2) The neural network of an error back propagation is established. (3) The neural network is trained and tested. (4) The neural network which passes the test is utilized to forecast the chlorophyll a in the water body. Other water qualitieswhich influence the chlorophyll a are: Ammonia nitrogen, total nitrogen, total phosphorus, orthophosphate, permanganate index, temperature, dissolved oxygen, pH, suspension, five-day biochemical oxygen demand. The step (1) also comprises a normalization process. The data of the chlorophyll a and other ten water quality indexes are between -1 and +1 after the data of the chlorophyll a and other ten water quality indexes are normalized. The neural network comprises an input layer, an intermediate layer and an output layer. The invention can establish a forecast model related to the chlorophyll a, just needing the experiment which has the limited times. The chlorophyll in the river can be accurately and quickly forecasted through the computer simulation experiment and the science forecast.

The invention discloses a power load probability prediction method based on a chaotic population algorithm and a Bayesian network, and the method comprises the steps: 1, obtaining the actual data of an air temperature, relative humidity, wind power and power load time sequence, carrying out the preprocessing of each column of data, and dividing a training set and test set data; 2, performing wavelet threshold de-noising processing on the original data of the power load, and restoring real information of a time sequence of the power load; 3, constructing a Bayesian network model to obtain an initial prediction interval; 4, calculating an interval change amplitude range, and obtaining an optimal interval change amplitude by applying a chaotic population algorithm; 5, chaotic search is adopted in the neighborhood of the optimal interval change amplitude, and a final prediction interval is obtained. The uncertainty of the power load can be measured by constructing the prediction interval,so that an effective reference can be provided for the optimized operation of the power system.

The invention discloses a wind power probability prediction method based on a chaotic firefly algorithm and a Bayesian network, and the method comprises the steps of 1, obtaining the wind speed, the wind direction, the air temperature and the wind power actual power data, preprocessing the data, and selecting the data of a training set and a test set; 2, performing empirical mode decomposition on the original data of the wind power to enable a wind power time sequence to be more stable; 3, constructing a Bayesian network model to obtain an initial prediction interval; 4, calculating an interval change amplitude range, and obtaining an optimal interval change amplitude by using a chaotic firefly algorithm; and 5, performing chaotic search near the optimal interval change amplitude to obtain a final prediction interval. According to the method, the uncertainty of measuring the wind power by constructing the prediction interval can be constructed, so that an effective reference can be provided for a power dispatching decision.

The embodiment of the invention provides a motion compensation method for a double-forward prediction unit. The method mainly comprises following steps of obtaining two initial double-forward prediction image blocks for predicting a current image block; selecting pixels on the two initial double-forward prediction image blocks; utilizing a gradient calculation formula to calculate x direction and y direction derivatives of the pixels on the forward prediction image blocks; through adoption of a training window, calculating the offset values of the pixels on the forward prediction image blocks; and adjusting the positions of the pixels on the forward prediction image blocks according to the x direction and y direction derivatives and the offset values of the pixels on the forward prediction image blocks. According to the motion compensation method for the double-forward prediction unit provided by the embodiment of the invention, the prediction effects of the prediction image block can be further improved on the basis of not increasing a code rate; the accuracies of prediction image blocks are improved; therefore, the prediction quality of the current image block is improved; and the coding efficiency of the double-forward prediction unit is improved.

The invention discloses a text abstract generation method based on input sharing. The method comprises the following steps: calculating a text to obtain sentences, and combining the sentences to obtain a plurality of candidate abstract texts to obtain a candidate abstract data set; obtaining a model input sequence, and calculating an attention mask matrix of the model input sequence; constructing a text abstract generation model, and obtaining a final candidate abstract score corresponding to the candidate abstract text; the candidate abstract scores are used, training of a text abstract generation model is supervised according to the real scores of the candidate abstract texts, and a trained text abstract generation model is obtained; and obtaining a to-be-reasoned text, preprocessing the to-be-reasoned text, inputting the preprocessed to-be-reasoned text into the trained text abstract generation model, and selecting a candidate abstract text with the candidate abstract score as an abstract text of the to-be-reasoned text. According to the method, the number of words input into the model can be reduced, and the training speed and the reasoning speed of the model are greatly improved on the premise that most generation effects are reserved.

The invention discloses an optimization design method for a microwave circuit, and the method comprises the following steps: obtaining a sample model design parameter through LHS, and obtaining a corresponding sample response through a Matlab-HFSS joint simulation technology; calculating correlation coefficients of all sample responses and target responses, selecting the sample with the maximum correlation coefficient as an optimization sample, and other samples serving as training samples; training the ELM by using the training sample, predicting design parameters responding to the optimized sample, and optimizing an input weight and a threshold value of the ELM by using BSO; and establishing a mapping relationship between the microwave circuit model design parameters and the response by using the ELM after optimizing the input weight and the threshold, training by using all the training samples in the training process, and predicting the model design parameters corresponding to the target response in the prediction process. According to the method, the training and prediction quality of the neural network is improved, the number of required training samples is reduced, the time required for optimization design of the microwave circuit is shortened, the automation of the optimization design of the microwave circuit is realized, and the optimization design efficiency is improved.