270 results about "Transfer model" patented technology

A method and apparatus is provided for delivering a content file to a client over a packet-switched network. The method begins by determining a suitable throughput required to deliver the content file to the client. Next, the throughput available in a peer-to-peer network for delivering the content file to the client is determined. The required throughput is compared to the available throughput. If the available throughput is less than the required throughput, the available throughput is supplemented with additional throughput. The content is then delivered to the client over the packet-switched network using the available throughput of the peer-to-peer network and the additional throughput.

The invention provides a posting / subscribing system for adding message queue models and a working method of the posting / subscribing system. The system comprises a distributed message-oriented middleware server-side cluster, a client-side cluster and a distributed coordinate server cluster. Service nodes in the message-oriented middleware server-side cluster are composed of corresponding modules of an original posting / subscribing system, and the function of the message-oriented middleware server-side cluster is the same as the information transferring relation; a client side is simultaneously provided with a calling interface of a posting / subscribing model and a calling interface of a message queue model which are used for achieving the purpose that a user can select and use the two different message transferring models directly without transforming the bottom layer middleware, so that a coordinate service module and a message processing module are additionally arranged on the client side based on an original framework. The system supports the posting / subscribing message transferring model and the message queue message transferring model simultaneously and guarantees that the availability, flexibility and other performance of the two models are all within the acceptable range, and the problem that it is difficult to transform the bottom layer message transferring model for the user is better solved.

Methods, systems, and apparatuses are provided for controlling an environmental maintenance system that includes a plurality of sensors and a plurality of actuators. The operation levels of the actuators can be determined by optimizing a penalty function. As part of the penalty function, the sensor values can be compared to reference values. The optimized values of the operation levels can account for energy use of actuators at various operation levels and predicted differences of the sensor values relative to the reference values at various operation levels. The predicted difference can be determined using a transfer model. An accuracy of the transfer model can be determined by comparing predicted values to measured values. This accuracy can be used in determining new operational levels from an output of the transfer model (e.g., attenuating the output of the transfer model based on the accuracy).

A first input is received from a plurality of sensors. A first state including a first location based on the first input is determined. The first state is associated with a first probability. A second input is received from the plurality of sensors. A second state including a second location is determined based on the second input associated with a second probability. It is determined that the second state corresponds to an actual state based on a transition model and the second probability. The transition model associates the first state with the second state and indicates a likelihood of a transition from the first state to the second state. A rule to change a state of at least one network connected device is triggered based on the second state.

The invention relates to image process, motion analysis and the like. In order to implement cross-camera tracking of moving targets, the invention uses the technical scheme that the cross-camera tracking method for the multiple moving targets comprises the following steps of: 1, motion filtering for a single camera, specifically, decomposing complex motions of human body into a finite number of combinations of relatively simple motion filtering models, wherein each single motion filtering model is represented by a combination of a linear motion state transition model and a gaussian noise, and the motion filtering method comprises the two following core parts: a motion model: an observation model; zk = Hsk + wk2, cross-camera oriented motion association: (1) creation of a panorama; (2) target similarity measurement; (3) motion trail association; and transforming the problem of cross-camera association of multiple motion trails into the problem of matching of bipartite graphs, and solving the problem through the integer programming under the following constraint conditions. The cross-camera tracking method for the multiple moving targets is mainly used in the image processing, the motion analysis and the like.

Methods, systems, and apparatuses are provided for controlling an environmental maintenance system that includes a plurality of sensors and a plurality of actuators. The operation levels of the actuators can be determined by optimizing a penalty function. As part of the penalty function, the sensor values can be compared to reference values. The optimized values of the operation levels can account for energy use of actuators at various operation levels and predicted differences of the sensor values relative to the reference values at various operation levels. The predicted difference can be determined using a transfer model. An accuracy of the transfer model can be determined by comparing predicted values to measured values. This accuracy can be used in determining new operational levels from an output of the transfer model (e.g., attenuating the output of the transfer model based on the accuracy).

The invention provides an underwater autonomous robot fixed depth control method based on reinforcement learning and belongs to the underwater robot control field. The method comprises steps that firstly, a Markoff decision process model for underwater autonomous robot fixed depth control is constructed, and expressions of a state variable, a control variable, a transfer model and a one-step lossfunction for underwater autonomous robot fixed depth control are respectively acquired; the decision network and the evaluation network are respectively established; through reinforcement learning, the decision network and the evaluation network can be continuously updated whenever an underwater autonomous robot moves forwards for each step during fixed depth control training till convergence; thefinal decision network for fixed depth control is acquired. The method is advantaged in that fixed depth control on the underwater autonomous robot under the condition of a completely-unknown underwater autonomous robot dynamic model is realized, and the practical value is high.

A data processing apparatus includes an obtaining unit configured to obtain time-series data from a wearable sensor, an activity model learning unit configured to learn an activity model representing a user activity state as a stochastic state transition model from the obtained time-series data, a recognition unit configured to recognize a current user activity state by using the activity model of the user obtained by the activity model learning unit, and a prediction unit configured to predict a user activity state after a predetermined time elapses from a current time from the current user activity state recognized by the recognition unit.

The invention provides a state self-checking method of a combined speed measuring and positioning system for a train, which comprises the following steps: judging the state of a speed sensor by using the respective multi-path information of the speed sensor for carrying out the redundant check; judging the measuring state of the speed sensor by taking radar measuring data as reference; judging the measuring state of a radar sensor according to the measuring data of the speed sensor and the output state word of a radar; and determining a system structure reconstruction scheme based on a system state transfer model according to the state judging results of the speed sensor and the radar, outputting speed measuring and positioning information which does not contain invalid or fault sensor or radar measuring results, and giving system state information representation or giving state representation of invalid measuring output when all sensors are invalid. The method of the invention improves the state detection performance of the sensors, reduces the complexity of the combined system, and can further save the engineering expenses and the maintenance cost.

The invention discloses a breast structure disorder identification method based on two transfer and a convolutional neural network. The method is characterized by transferring model parameters trainedin an ImageNet data set to a target convolutional neural network to carry out initialization on model parameters thereof ( first transfer); then, carrying out first fine-tuning training (second transfer) on a target convolutional neural network model through malignant tumor images and normal breast tissue images; carrying out second fine-tuning training through class-information-known structure disorder images and normal breast tissue images in a training set; and finally, carrying out identification on class-information-unknown structure disorder and normal tissues. The method realizes identification based on the convolutional neural network and transfer learning; and compared with the existing method, the method achieves better classification effect and performance.

The invention discloses a short-term wind speed forecasting method based on a deep neural network transfer model. The method comprises the following steps that (1) normalization preprocessing and division of sample sets are carried out on data of two or more wind power plants, (2) the deep neural network transfer model is established, (3) layered training is started from bottom to top in an unsupervised learning mode, (4) supervised learning is carried out from top to bottom on the basis of the third step, (5) weight parameters of connection of a top layer and hidden layers are finely adjusted so as to obtain an output layer, corresponding to the wind power plants, in a deep neural network, and (6) inverse normalization is carried out on the result output by a deep neural network so as to obtain the predicted value of wind speed. Transfer learning is introduced to the wind speed forecasting field, knowledge of other wind power plants rich in data is transferred to target wind power plants, and the problem that the newly built wind power plants have few data is solved effectively. By means of the effective transfer scheme based on the deep neural network, the wind speed prediction accuracy of the target wind power plants is greatly improved.

A font style transfer and transformation method based on Manchu character picture belongs to the field of image processing, which solves the problems of greatly reducing the time and labor cost spentin processing a large number of Manchu character images. The main points are that the manchu font style transfer model is trained step by step: the manchu character training data set is used for training the model to realize the manchu character learning; a mandarin font style migration data set is use to train that model again to realize the mandarin font style migration learning, and the effectis to save a large amount of time and labor cost.

A model to predict a transition of a user's usage for a terminal device is generated from an operation record. A usage cluster generating unit (111) generates multiple user groups (131) consisting of users whose characteristic values representing a usage characteristic calculated from operation record information are similar according to the operation record information of the multiple first users of the terminal device. A usage transition model generating unit (112) analyzes that the characteristic value representing the usage characteristic calculated from the operation record information is similar to which of the usage characteristics of the plurality of the user groups (131) for each operation record information of each divided section, each divided section being obtained by dividing the operation record information of a plurality of second users of the terminal device, and generates a transition model (141) representing a transition relationship between the user groups (131) according to the analysis result.

The invention discloses a matching error calculation method for error transfer modeling of a precision mechanical system. The method comprises the following steps of: measuring shape errors D1 and D2 of two matching surfaces by applying a three-coordinate measuring machine to obtain data of a difference surface, and determining a contact point according to the data of the difference surface; calculating deformation errors delta 1 and delta 2 of the two matching surfaces of a part according to the contact point and assembly force applied to the two matching surfaces, thereby obtaining actual matching surface data D1+delta 1 and data D2+delta 2 considering the shape errors and the deformation errors of the two matching surfaces of the part; and calculating matching error component of the two actual matching surface to obtain matching error so as to be used for error transfer modeling of the precision mechanical system. In the method, the shape errors of the matching surfaces and the part deformation errors produced under the action of the assembly force are considered, a more accurate error transfer model can be established for the precision mechanical system on the basis, and the accuracy for the manufacturing quality prediction and control is improved.

The invention provides an existing motion data-based realistic motion transfer and generation method and system. The method includes the following specific steps that: a data glove is adopted to clearly and continuously shoot the motions of a measured object; a target transfer model is built; a skeleton sequence is reconstructed by means of inverse kinematics method; skeleton size is transferred by means of a forward kinematics method; skinning and drawing weighting operation is performed on a target skeleton and the target model; the real value of the three-dimensional position of the skeleton is calculated; realistic material texture is drawn on the model; and rendering is carried out, and a result is outputted. The method and system of the invention have the advantages of high automation degree, high precision, simple process and easiness in operation. With the method and system of the invention adopted, a lossless skeleton sequence can be obtained, and professional or popular motion transfer requirements can be satisfied.

The invention provides a design method for a threshold value and an envelop wire of a near-earth alarm system, and aims at improving limitation in design of the threshold value and the envelop wire of the existing near-earth alarm systems. Probability of generation of earth collision danger in the flight process is calculated via comparison of landform height and flight track height on the basis of establishment of a landform height statistic model and a state transfer model; and the design method for the threshold value and the envelop wire of the near-earth alarm system is provided in combination with performance analysis of the near-earth alarm system so that timely and effective alarm information with low false alarm rate is provided for airplanes.

The invention discloses a bearing capacity assessment calculation method for a complex region of an existing concrete bridge. The method comprises the steps of determining load and boundary conditions of the complex region for diseases such as reinforcement corrosion, concrete cracking, carbonization and the like appearing in the complex region; establishing a pull rod model and a press rod model of the complex region; according to a detection result, further determining reduction coefficients of a pull rod, a press rod and a joint after the diseases are considered; making a calculation according to a pull rod strength control rule, a press rod strength control rule and a joint strength control rule to obtain a minimum bearing capacity value; and finally, according to a bearing capacity assessment result, determining a reinforcement load transfer model, continuously adjusting the positions of the reinforced pull rod, press rod and joint, or continuously increasing the reinforcement amounts of the pull rod and the press rod until the usage requirements are met. According to the method, the problems of inapplicability of a section method in calculating the complex region of concrete, complex calculation for a finite element method and difficulty in arrangement of reinforcement are solved and the bearing capacity of a complex region of an old bridge can be accurately estimated.

The invention relates to an energy storage optimal configuration method considering system gas and thermal power units. The method comprises the following steps: firstly, according to historical data,forecasting the output of a new energy generator set, constructing a typical scene set of the output of the new energy generator set, and combining the load fluctuation characteristics, constructinga load scene set of a power system; This paper analyzes the operation characteristics of gas-fired units and thermal power units in the start-up and stop stages, establishes the state transfer equations, defines the state transfer conditions, and establishes the state transfer models of gas-fired units and thermal power units in the start-up and stop stages, so as to realize the transfer and switching between different states of gas-fired units and thermal power units in the start-up and stop stages. The paper also analyzes the operation characteristics of gas-fired units and thermal power units in the start-up and stop stages. According to the system and operation parameters, considering the wind power consumption objective, the optimal energy storage allocation model considering the climbing ability and multi-stage state transition of gas turbine and thermal turbine is constructed, with the objective of minimizing the relevant investment and total operating cost. Solve the above power system energy storage optimal allocation problem, and obtain the power system energy storage optimal allocation scheme.

The invention relates to an aircraft route tracking method based on deep reinforcement learning, and belongs to the field of robot intelligent control. The aircraft route tracking method comprises thesteps of: constructing a Markov decision process model for aircraft trajectory tracking control, and respectively obtaining expressions of a state variable, a control variable, a transfer model and aone-step loss function for the aircraft route tracking control; respectively establishing a strategy network and an evaluation network; through reinforcement learning, enabling an aircraft to continuously update the strategy network and the evaluation network when making a step forward every time in route tracking control training until convergence; and obtaining a final strategy network for theroute tracking control. The aircraft route tracking method has strong expansibility, is not only suitable for the problem of route tracking control of the aircraft, but also can be expanded to other complicated problems of aircraft control by only resetting the state variable, a control input and a loss function and adjusting the structure and parameters of a neural network.

The invention relates to an optical transfer model measuring method in a coding structured light system. The optical transfer model measuring method in the coding structured light system achieves optical transfer modeling of an optoelectronic system for a strip coding structured light system. Firstly, optical transfer modeling of the structured light system is finished theoretically, corresponding measurement legends are designed according to the characteristics of a strip coding light image, measurement of a line spread function of the system is finished through a knife-edge method by using a projected knife edge target, and then Fourier transform is carried out on the line spread function to obtain the optical transfer model of the coding structured light system. The optical transfer model is used for guiding the design of coding structured light, improving the measurement accuracy and expanding the measurement applicability of different reflection surface objects, improves the performance of the coding structured light system, and widens the application range of the structured light technology.

The invention discloses a navigation method and device based on extended Kalman filter. By building a model, including a state model, a state transition model and an observation model, for an actual navigation system, the filtering process in the navigation method is subjected to stagewise optimization and matrix operation is improved. Relationship and independency among eleventh order parameters are analyzed, matrix operation is studied and the system is split, so as to achieve short operation period, reduce calculation amount, save memory space, improve efficiency and lower requirements for hardware.

An image matching apparatus includes a numerical analyzing unit that solves an equation of motion numerically. The equation of motion is formulated as an image matching state transition model representing a mapping relation between a source image and a destination image. The equation of motion represents a force applied on each lattice point of the destination image including potential energy force. The numerical analyzing unit calculates the potential energy force, and sets a convergence value of a solution of the equation of motion as a mapping relation between the lattice point of the source image and the lattice point of the destination image based on the numerical analysis.

A simulation method of simulating a behavior or a mechanism using a hybrid model including a state transition model and a continuous system model. The hybrid model is analyzed to extract a first description data of the state transition model and a second description data of the continuous system model. The method includes generating a table representing a relationship between continuous system equations and switching conditions thereof, based on the extracted first description data. The method also includes generating a plurality of internal data expressions of the continuous system equations, based on the extracted second description data.

The invention provides a multi-person abnormal behavior detection method based on security monitoring video data. According to the method, a standard AV output signal of a monitoring camera is acquired; pedestrian characteristics are extracted, and coarse detectors, coarse pedestrian ROIs and precise ROIs are respectively acquired; pedestrian behavior tracking is carried out, a particle filtering method is employed to respectively surround each tracking target of a video into a rectangular frame, a multi-order autoregression process mathematics model is established for state transferring of each tracking target, and a state transferring model for describing actual motion situations of motion targets is acquired; under the particle filtering framework, a particle filtering human body tracking method integrated with color and shape characteristics is acquired; abnormal-pedestrian classification is carried out, and optical flow characteristics of the precise ROIs are calculated; each frame of gray-scale image in the monitoring video flow is set to be a Markov random field ; characteristics of pedestrians determined to have abnormal traffic behaviors in video monitoring scenes are extracted, a continuous hidden Markov model is established, and the abnormal behaviors are identified.

The invention discloses a subway passenger flow organization method based on major stop parking and multi-station synergistic flow-limiting. The method comprises the following steps: counting the passenger flow characteristic of each station on a subway line; setting a train stopping scheme of the subway line according to the passenger flow characteristic; constructing a space-time transfer model of the train and the passenger according to the passenger flow characteristic of each subway station and the train stopping scheme of the subway line; computing a weighted value of a line multi-station transportation system according to the space-time transfer model of the train and the passenger; dynamically adjusting a decision variable, optimizing integral weighted value of all passengers to obtain an optimal train stopping scheme and the corresponding flow-limiting scheme. Through the adoption of the technical scheme disclosed by the invention, the limitation of isolatedly and statically considering the passenger flow limiting in single station is overcome, the influence of the passenger flow change in each station to the adjacent station is considered, the stations and the line are regarded as an integer so that the passenger flow changes of the whole line is dynamically linked, the method has obvious improvement in comparison with the existing single-station flow-limiting scheme.

An information processing apparatus includes: model learning means for self-organizing, on the basis of a state transition model having a state and state transition to be learned by using time series data as data in time series, an internal state from an observation signal obtained by a sensor; and controller learning means for performing learning for allocating a controller, which outputs an action, to each of transitions of a state or each of transition destination states in the state transition model indicating the internal state self-organized by the model learning means.

The invention discloses a reinforcement learning reward self-learning method in a discrete manufacturing scene. The method comprises the following steps: 1, refining the process of the current production line, wherein g belongs to G = {g1, g2,..., gN}, and the intelligent agent reaches a preset target g and is recorded as an interaction sequence episode; according to the initial parameters, obtaining multiple sections of episodes corresponding to the g1 as a target, taking state actions in the episodes and a state difference value delta as a training data set to be input into a GPR module, andobtaining a system state transition model based on state difference; enabling the intelligent Agent to continue to interact with the environment to obtain a new state st, and enabling a Reward network to output r (st), enabling an Actor network to output a (st), enabling a Critic network to output V (st) and enabling a GPR module to output value function Vg as the updating direction of the whole;when the absolute value of Vg-V (st) is smaller than epsilon, considering that award function learning under the current procedure is completed, and carrying out parameter storage of the Reward network; continuously carrying out interaction, and generating the following sub-target g < n + 1 > as the episodes of the updating direction for updating the GPR; and when the set target G = {g1, g2,...,gN} is all realized in sequence, finishing the process learning of the production line.

The invention discloses a multi-target detecting and tracking method based on RGB-D data. The method includes the steps of effectively fusing colored image information and depth information to detect moving targets, substituting the results of multi-target detection into an observation likelihood model of the RJMCMC particle filtering algorithm, establishing a reasonable state transfer model at the same time, and tracking multiple targets by defining reasonable reversible jumping actions and corresponding suggestion distribution. Multiple targets are detected by fusing information of an RGB image and a depth image, and therefore the detection missing rate and the false detection rate are decreased; multiple targets are tracked through the RJMCMC particle filtering algorithm, and by setting the reasonable reversible jumping actions and selecting the corresponding suggestion distribution, shielding and interacting between multiple targets can be eliminated, and the problem that the number of multiple targets changes can be solved.