32 results about "Maximum entropy probability distribution" patented technology

An apparatus generates a statistical class sequence model called A class bi-multigram model from input training strings of discrete-valued units, where bigram dependencies are assumed between adjacent variable length sequences of maximum length N units, and where class labels are assigned to the sequences. The number of times all sequences of units occur are counted, as well as the number of times all pairs of sequences of units co-occur in the input training strings. An initial bigram probability distribution of all the pairs of sequences is computed as the number of times the two sequences co-occur, divided by the number of times the first sequence occurs in the input training string. Then, the input sequences are classified into a pre-specified desired number of classes. Further, an estimate of the bigram probability distribution of the sequences is calculated by using an EM algorithm to maximize the likelihood of the input training string computed with the input probability distributions. The above processes are then iteratively performed to generate statistical class sequence model.

The invention provides a carbonate rock physical parameter seismic inversion method, which belongs to the field of petroleum geophysical exploration. The method comprises the following steps: (1) carrying out pre-stack AVO three-parameter inversion based on a pre-stack angle gather to obtain formation elasticity parameters M, wherein M includes P-wave velocity, S-wave velocity and density; (2) establishing a statistical rock physical model based on logging data; (3) randomly simulating physical conditions of a reservoir based on the rock physical model obtained from step (2) to obtain a random simulation result; and (4) carrying out Bayes classified simulation on the random simulation result obtained from step (3) to obtain a posterior probability distribution, and taking the M obtained from step (1) as the input of inversion to find a R with the maximum posteriori probability distribution, wherein R is the final inversion result.

Circuits that solve stochastic problems and techniques for operating them. These natively stochastic circuits may produce samples from probability distributions of interest for particular stochastic problems, and may be combined together in any suitable way to yield potential solutions to stochastic problems. In some implementations, the stochastic circuits may generate samples from conditional probability distributions conditioned on input data provided to the stochastic circuits. The circuits may be constructed from multiple interconnected stochastic subcircuits such that a circuit may produce a sample from a joint probability distribution, or from a marginal distribution of a joint distribution. These circuits may be used to implement stochastic sampling algorithms to solve stochastic processes, and may include stochastic subcircuits that operate concurrently.

Systems and Methods for multi-modal or multimedia image retrieval are provided. Automatic image annotation is achieved based on a probabilistic semantic model in which visual features and textual words are connected via a hidden layer comprising the semantic concepts to be discovered, to explicitly exploit the synergy between the two modalities. The association of visual features and textual words is determined in a Bayesian framework to provide confidence of the association. A hidden concept layer which connects the visual feature(s) and the words is discovered by fitting a generative model to the training image and annotation words. An Expectation-Maximization (EM) based iterative learning procedure determines the conditional probabilities of the visual features and the textual words given a hidden concept class. Based on the discovered hidden concept layer and the corresponding conditional probabilities, the image annotation and the text-to-image retrieval are performed using the Bayesian framework.

The invention provides an air quality fast traceability forecast method based on a Lagrangian transportation model. The method comprises the steps that a mesoscale weather forecasting mode is used to simulate the meteorological field of a simulated area; according to the meteorological field requirements of the Lagrangian transportation model, key elements, which are output by a meteorological field simulation system, of the meteorological field are extracted, and the data format is converted; basic parameters of the Lagrangian transportation model are set; the space-time probability distribution of released particles in the meteorological field is calculated through the Lagrangian transportation model, and a particle probability distribution database is built; the space-time probability distribution of the released particles and a pollution source list in the simulated area are combined to calculate the potential source contribution distribution of pollutants; and finally, statistical simulation is carried out on the calculated pollutant concentration and the contribution of each industry to the pollutant concentration in the pollutant list. According to the invention, the air pollution source of an area and a city can be analyzed and forecasted, and the identification of the key area of pollution contribution and quick response demonstration of and pollution emission reduction measures can be realized.

A parameter identification-based fault isolation technique is provided for improving fault code settings and isolating faults. DTCs and PIDs are retrieved from a current serviced vehicle. PID data is obtained from prior serviced vehicles having same vehicle specifications as and under substantially similar operating conditions as the current serviced vehicles. A plurality of clusters is formed using PID data of the prior serviced vehicles using mutual dependence to one another. A joint probability distribution is determined as a function of a mutual dependence of the PID data. PID data is selected of the current serviced vehicle having coherence to the PID of the prior service vehicles. A mutual dependence of the PID data of the current serviced vehicle is projected on onto the corresponding joint probability distribution. An anomaly in the PID data of the current serviced vehicle is detected in response to the probability distribution of parameter identification data being outside of a predetermined variance.

The present invention provides a deep migration learning method for cross-domain analysis of big data, by determining the value of the loss function of the deep network according to the joint distribution difference and the classification error rate, wherein the joint distribution difference is in all task-related layers corresponding to the source domain The distribution difference between the features of the feature and the joint probability distribution between the labels and the features in all task-related layers corresponding to the target domain and the joint probability distribution between the labels; and based on the value of the loss function, the parameters of the deep network are updated to Adapt the deep network to the target domain; thus, in the transfer learning process of the deep network, the joint distribution difference is used as a component of the value of the loss function of the deep network, and by updating the parameters of the deep network, while ensuring the accuracy of the source domain Realizing the matching of the joint distribution of the source domain and the target domain improves the migration ability of all task-related layers, thus bringing better results to the transfer learning of deep networks between different domains.

Circuits that solve stochastic problems and techniques for operating them. These natively stochastic circuits may produce samples from probability distributions of interest for particular stochastic problems, and may be combined together in any suitable way to yield potential solutions to stochastic problems. In some implementations, the stochastic circuits may generate samples from conditional probability distributions conditioned on input data provided to the stochastic circuits. The circuits may be constructed from multiple interconnected stochastic subcircuits such that a circuit may produce a sample from a joint probability distribution, or from a marginal distribution of a joint distribution. These circuits may be used to implement stochastic sampling algorithms to solve stochastic processes, and may include stochastic subcircuits that operate concurrently.

The invention discloses an electric power system static safety assessment method based on a probabilistic tide. Statistical information of the random quantity in a system is obtained according to historical data; mutual conversion between a relevant non-normal random variable and an independent normal distribution random variable is achieved by improving Nataf conversion. By means of the electric power system static safety assessment method based on the probabilistic tide, arbitrary probability distribution of many random factors in the electric power system can be coped with, the correlation among the random factors is considered, the processing difficulty of the correlation is reduced, and the electric power system static safety assessment method is suitable for assessing the influence of all kinds of uncertain factors in the electric power system on the static safety of the electric power system.

The invention provides an air ticket personalized recommendation method based on shared account passenger prediction. The method includes the steps of 1. counting all of the orders under each account and analyzing the preference of a user on different air ticket attributes, and calculating the weight of the user corresponding to each attribute based on the user historical orders; 2. calculating the probability distribution of all of the passengers under the account according to the account historical behaviors and current conversation context; and 3. ordering from high to low according to the similarity between the search result and the preference model combined with passenger prediction, recommending the first K results to the users, wherein K is an integer. The recommendation method based on the shared account user prediction takes the probability distribution of the passengers to obtain a targeted recommendation result. The method provided is high in universality, if the method is needed to be applied to a new filed, one only needs to redefine the attribute characteristics through the combination with the filed knowledge, and a composite preference model based on the user probability distribution is constructed, so that the recommendation effect can be effectively improved.

The invention discloses a deep network intelligent investment system data analysis method integrating an attention mechanism. The method includes the following steps that: step 1, financial fields called by a sufficient quantity of local devices are obtained from a financial website and a stock database, the financial fields are filtered and integrated, so that a field X can be obtained; step 2, the field X is inputted into an encoder module, wherein the encoder module is composed of a long-term and short-term memory network, and encodes the X; step 3, an encoded field X vector obtains an attention allocation probability distribution value within a probability distribution value interval through an attention allocation module; step 4, the long-term and short-term memory network in the decoder generates price predictions on the basis of a field code containing attention probability distribution and historical information that has been generated before; step 5, a trained deep network outputs the prediction result of a certain trading day, and compares the prediction result of the trading day with a set threshold value, and the risks of financial products are determined; and step 6, appropriate financial products are screened according to user funds, and an optimal investment portfolio is configured.

The invention discloses a music genre classification method based on a recurrent neural network and an attention mechanism. The method comprises: firstly, musical signal being transformed by a short-time Fourier transform to obtain a sonagraph, using a two-way recurrent neural network to learn features according to the from the sonagraph, to obtain higher-level abstract features, and using a parallel attention model to learn from the sonagraph to obtain attention probability distribution corresponding to feature representation, the attention probability distribution being used to set differentweights of musical feature representation; then performing weighted average on the features according to the feature weights to obtain fused features; finally, classifying music genres according to the fused musical features. In the method, the parallel recurrent neural network and the attention model are used, feature learning is carried out automatically according to the musical signals, and reasonable weights are set by using the attention probability distribution as the feature, and the features are weighted and averaged and then are classified, so that accuracy of music genre classification is improved, and complexity and limitation of manual feature extraction are prevented.

The invention relates to a fault detection method based on a correlation probability model. The method comprises the following steps: selecting a to-be-detected parameter from a plurality of telemetryparameters; dividing the to-be-detected parameters into a plurality of working units in time; determining a correlation coefficient between every two parameters in the plurality of telemetry parameters in each working unit so as to determine a correlation coefficient vector of the working unit; establishing a correlation probability model for the plurality of working units to determine multivariate probability distribution of the plurality of working units; carrying out dimension reduction on the multivariate probability distribution through principal component analysis (PCA) to obtain unaryprobability distribution; and performing anomaly detection on the data of the unary probability distribution to determine an abnormal to-be-detected parameter. According to the invention, parameter abnormity can be rapidly detected in the early stage of satellite faults, so that health state monitoring and system maintenance of satellites are facilitated.

The invention discloses a nano porous media modeling method and system with the relative density and the edge diameter controllable. The method includes the following steps that S100, grid division isconducted on a cube model in the three-dimensional direction, and a three-dimensional grid model is obtained; S200, a finite difference method is used for obtaining the three-dimensional grid model with a three-dimensional bicontinuous porous structure; S300, the jump function is used for calculating the relative density of the three-dimensional grid model at the moment; S400, a random cut line is generated in a grid point of each edge of the three-dimensional grid model, a binormal distribution function is used for fitting the length probability density distribution of the random cut line, and the length corresponding to an extreme point with the maximum length probability distribution is the average edge diameter; S500, the steps S100-S300 are iteratively executed until the relative density and the average edge diameter of the three-dimensional grid model meet the preset requirements. By adopting the method, the nano porous media modeling method and system with the relative densityand the edge diameter controllable can be obtained, and the obtained model can be used for unidirectional stretching molecular dynamics simulation.

Systems and methods are described for simultaneously deriving an effective x-sigma corner for multiple, different circuit and / or process metrics for a semiconductor device. The result is an effective sigma that is representative of design intent. Some implementations account for covariance, and use joint probability as the criteria for the effective x-sigma corner (e.g., as opposed to a unique sigma level of each individual metric). Analysis results for each metric can be transformed to metric distributions in a common distribution framework, and a correlation matrix can be calculated. The transformed metric distributions can be input to a joint probability distribution set to achieve a target joint sigma level. The joint probability distribution and correlation matrix values can be used to back-calculate scaled x-sigma corners for each metric distribution. Simulation of the device can be performed at one or more of the scaled x-sigma corners.

The invention discloses a soft sensing method of key performance index based on Markov random field and EM algorithm. The method comprises the following steps: firstly, establishing a Markov random field model according to auxiliary variables and key performance indicators, and then establishing a joint probability distribution function according to the Markov random field model, wherein, the joint probability distribution function is a joint probability distribution among auxiliary variables and / or a joint probability distribution among key performance indicators; According to the joint probability distribution function, the mean square deviation probability model is established, where the mean square deviation probability model is the relationship between the expectation of the key performance index and the target key performance index when the auxiliary variable is given, and the target key performance index is the key performance index when the mean square deviation probability model approaches zero; the auxiliary variables are inputted into the mean square deviation probability model to obtain the target key performance index, thus solving the technical problem of the prior art that the key performance index in the industrial process can not be accurately measured in real time.

The invention discloses an SAR image segmentation method based on a triple discrimination random field. The defect that in the prior art, when a TMRF model of a triple Markov random field is adopted for capturing statistical correlation of observation data, the operating distance must depend on an adopted neighbourhood system is overcome. The SAR image segmentation method includes the following steps of (1) inputting an SAR image to be segmented, (2) extracting features and initializing, (3) obtaining joint probability distribution, (4) obtaining posterior marginal probability distribution, (5) segmenting the SAR image, (6) estimating model parameters and (7) outputting the segmented image. Through introducing related information between non-stationary information and pixel point observation values of the SAR image, consistency of segmented areas is kept, and marginal accuracy of a segmentation result is improved.

The invention discloses a method for extracting reservoir dispatching function based on joint probability distribution. By collecting reservoir basic data and long series of measured runoff data, theoptimal dispatching process sample series are obtained by deterministic optimal dispatching of reservoirs. The marginal probability distribution functions of decision variables and state variables aredetermined, and the joint probability distribution functions of decision variables and state variables are constructed by using Copula function, then the conditional probability distribution functions of decision variables are deduced when the state variables are given, and the reservoir operation functions are extracted and the uncertainties are analyzed. The reservoir dispatching function extracted by the invention can accurately capture the inherent non-linear and heteroscedasticity correlation structure of the decision variable and the state variable, and the calculated confidence interval estimation value can quantitatively evaluate the uncertainty of the reservoir dispatching, thereby providing a beneficial reference basis for the decision risk analysis.

The present invention provides a disturbance source tracing method adapted for tracing a disturbance source of a system including a plurality of candidate nodes by a computing apparatus. In this method, the computing apparatus obtains a plurality of topology information by analyzing the system using a plurality of process analyzing methods respectively. The topology information records causalitiesbetween the candidate nodes of the system. The computing apparatus applies each causality of the topology information to a probability distribution algorithm, and calculates a stationary probabilitydistribution of each of the candidate nodes. The computing apparatus also synthesizes the stationary probability distributions calculated with respect to each process analyzing method to calculate a probability distribution of each of the candidate nodes being the disturbance source and find the disturbance source causing the disturbance in the system from lots of oscillation circuits.

The invention discloses a regional energy system optimization method based on type II fuzzy chance constraint, belonging to the field of energy system optimization dispatching. The method comprises the steps of: A, building an optimization model with the minimum regional energy system cost as the objective and with various kinds of constrains as constraint conditions; B, constructing type II fuzzyparameters according to the method for determining weights and the parameter possibility distribution, constructing random parameter according to the parameter probability distribution, and converting the type II fuzzy parameters and the random parameters into deterministic parameters through a type II fuzzy reduction method based on critical values and an opportunity constraint method, and solving the model; C, choosing different default probabilities of the energy system and substituting the probabilities into the model for solving, and obtaining the relationship between system default probability and system cost, energy purchase, energy supply and pollutant emission. The invention can optimize the regional energy system under the condition of multiple uncertainties, and provides an optimal energy purchase, energy supply and pollutant emission scheme.

The invention discloses an uncertain power flow calculation method based on hybrid random and interval variables, which relates to the technical field of power system voltage stabilization. The methodincludes the following steps: natural factors such as wind speed and light intensity are expressed as interval variables, and a wind power generation interval model, a photovoltaic power generation interval model and a node load probability model are constructed; the wind power generation interval model, the photovoltaic power generation interval model and the node load probability model are introduced into a power flow equation, and a power flow model based on hybrid random and interval variables is established; and calculation is performed on the power flow model based on hybrid random andinterval variables by adopting a double-layer Monte Carlo method to obtain the maximum and minimum cumulative probability distribution of power flow, and thus, the probability interval of power flow exceeding the limit is determined. By using the method of the invention, the maximum cumulative probability distribution and the minimum cumulative probability distribution of power flow can be obtained, and thus, the probability interval of power flow exceeding the limit can be determined.

The invention proposes a ''dessert'' probability direct estimation method and a ''dessert'' probability direct estimation system based on prestack inversion. The ''dessert'' probability direct estimation method comprises the steps of: utilizing well logging data to obtain a joint probability distribution among inversion parameters, that is, the joint probability distribution of a longitudinal wavemodulus and a density; using a probability estimation method to obtain an initial ''dessert'' phase probability; calculating a conditional probability distribution of a ''dessert'' sensitive parameter longitudinal wave modulus and a shear modulus, and sampling from the conditional probability distribution to obtain new longitudinal wave modulus and shear modulus models; calculating a conditionalprobability distribution of the density according to the new longitudinal wave modulus model through combining the longitudinal wave modulus with the density, and sampling from the conditional probability distribution to obtain a new density model; and calculating a new ''dessert'' phase probability, and utilizing an acceptance probability formula to determine whether to update the inversion parameters and the ''dessert'' phase probability model. The ''dessert'' probability direct estimation method and the ''dessert'' probability direct estimation system can obtain a series of samples of the related inversion parameters and the posterior probability distribution of the ''dessert'' phase probability through continuous iteration, and realize the direct estimation of the ''dessert'' phase probability.

The invention provides a prediction method and system for the acoustic noise probability of a DC power transmission line. The method comprises that S1) multiple groups of training data are used to train an artificial neural network model, and in the training process, line parameters are taken as input data, and acoustic noise measuring values are taken as output data; S2) the line parameters in the multiple groups of training data are input to the trained artificial neural network model again to obtain multiple groups of acoustic noise predicting values correspondingly; S3) according to the groups of acoustic noise measuring values and the groups of acoustic noise predicting values, groups of corresponding error values of the artificial neural network model are calculated; S4) the groups of acoustic noise predicting values are divided into multiple intervals, and probability distribution of the groups of error values in the intervals is determined; and S5) the artificial neural network model is used to predict data to be measured, and an acoustic noise probability predicting result is obtained according to a prediction result and the error-value probability distribution of the interval to which the prediction result belongs.

Systems and methods are described for simultaneously deriving an effective x-sigma corner for multiple, different circuit and / or process metrics for a semiconductor device. The result is an effective sigma that is representative of design intent. Some implementations account for covariance, and use joint probability as the criteria for the effective x-sigma corner (e.g., as opposed to a unique sigma level of each individual metric). Analysis results for each metric can be transformed to metric distributions in a common distribution framework, and a correlation matrix can be calculated. The transformed metric distributions can be input to a joint probability distribution set to achieve a target joint sigma level. The joint probability distribution and correlation matrix values can be used to back-calculate scaled x-sigma corners for each metric distribution. Simulation of the device can be performed at one or more of the scaled x-sigma corners.

Systems, methods and computer program products for analyzing stochastic characteristics, including one or more subsystem for forecasting financial statements and financial ratios' probability distributions and configured for at least one of: analyzing insurance contracts and portfolios without simplifying contract and product terms and conditions; studying and illustrating financial statement probability distributions; analyzing insurance portfolios by creating net asset value distributions thereof without deterministic assumptions; modeling to support asset and liability management, wherein both assets and liabilities are simulated simultaneously and decisions are based on joint probability distributions thereof; and modeling to study effects of model specification changes by implementing new model definitions and by rerunning the model with constant random number generator seed.

The invention relates to the field of automatic abstracting, and discloses a graph model-based automatic abstracting method. According to the technical scheme, an LDA probability topic model is applied to measurement of semantic correlation between sentences and improvement of the measurement effect of sentence correlation; and an idea of topic correlation and position sensitivity of the sentences is provided, so that abstract generation is relatively reasonable and effective. The method comprises the following steps: firstly, obtaining topic probability distribution of a document and word probability distribution of the topic through training the LDA topic model, determining the topic probability distribution of the sentences and effectively converting a semantic similarity measurement between the sentences into a similarity measurement problem of the topic probability distribution of the sentences; with the sentences as nodes, building edges by referring tothe cosine similarity and according to the semantic similarity between the sentences and generating a text graph representing the document; calculating the topic correlation between the sentences according to the topic probability distribution of the sentences and the topic probability distribution of the document; and calculating the position sensitivity and the like of the sentences according to the positions of the sentences in the document.

The invention discloses a data difference analysis method and system based on probability density estimation, and belongs to the field of data analysis. The method comprises the following steps of firstly, establishing a data set, wherein the data in the data set changes; estimating the joint probability of the data before and after the change by using a probability density estimation method; selecting an optimal window width according to a maximum likelihood method, for different window widths, taking any one point in the data set every time, constructing the joint probability distribution by using the rest points in the data set, calculating a joint probability density value of the any one point on the joint probability distribution, obtaining a product of a plurality of joint probability density values as a likelihood value, and enabling the window width with the maximum likelihood value to be the optimal window width; and obtaining the data joint probability density distribution before and after change through the probability density estimation method according to the optimal window width, and analyzing the data difference. According to the method, the significance degree of each piece of data can be obtained without being limited by the data distribution and is used for discovering the remarkably changed data.

A grinding chatter prediction method based on maximum information entropy and directional divergence focuses on the study of acceleration vibration signals of a bench during plane grinding, and grinding chatter is predicted by means of unsupervised data processing, namely the chatter prediction method based on maximum information entropy and directional divergence. The method includes: firstly, precisely estimating probability density distribution of bench vibration signals according to the principle of maximum information entropy in informatics; and secondly, quantizing the maximum entropy probability density distribution by directional divergence, observing plane grinding states according to changes in directional divergence value, and predicting plane grinding chatter. The method avoids chatter by the real-time monitoring of grinding states. Compared with other chatter prediction methods, the method has such advantages that training data is not needed, calculating speed is high, processing results are highly resistant to interference, and recognition rate is high.

The invention discloses a method and system for analyzing the uncertain power flow of the power grid, including: obtaining the probability description of input random variables such as wind speed, light intensity, load, etc. of the power system, establishing a model of the random factors of the power system, and determining the required Target output variables such as node voltage and line power flow; obtain the sampling values ​​of input random variables such as wind speed, light intensity, and load, and obtain the origin moments of each order of output variables such as node voltage and line power flow according to the deterministic evaluation method; based on the origin moment Calculate the results, and use the maximum entropy probability distribution algorithm to solve the probability distribution of target output variables such as node voltage and line power flow. The present invention can effectively overcome the characteristics of complicated calculation process in the traditional point estimation method, make up for the inaccurate fitting problem inherent in the traditional point estimation and series expansion method, and has certain application value for analyzing the uncertain state of the power system .