104results about How to "Improve learning accuracy" patented technology

The invention discloses a video rain removing and snow removing method based on multi-scale convolution sparse coding. Under the assumption of a low-rank background, the rain and snow components and the moving prospect in the video are estimated at the same time. Firstly, video data containing rain and snow noise is acquired, and a model is initialized; a generation model of the rain and snow graph is built according to the characteristics of the rain and snow and the video prospect; according to the structural characteristic that the rain and snow are imaged in the video, the moving rain andsnow are repeatable and multi-scale rain strip local blocks on the image, a multi-scale convolution sparse coding model is established related to the rain and snow; a moving object detection model isestablished according to the characteristics of the video foreground sparsity; the model is integrated into rain and snow under the maximum likelihood estimation framework model; a rain-snow video anda rain-removing snow model are applied, so that rain and snow videos and other statistical variables are obtained, and rain and snow videos are output. The invention aims to establish a high-qualityvideo rain-removing snow model based on the rain and snow generation principle and the rain and snow noise structure characteristics, so that the snow removing and snow removing technology can be widely applied in more complicated practical scenes.

The invention discloses an intelligent screening system based on the numerical control processing technology for difficult-to-machine metal, which comprises the following subsystems: a parameter database subsystem, a fixed data source subsystem, an online detection and feedback subsystem, a technology intelligent comprehensive screening optimized scheme system, a data mining and supplementing subsystem, a simulation verification subsystem and an application operating system. The system has the characteristic of recognizing the reasonableness, the advancement and the high efficiency property of each technology scheme in the database, and is used for collecting the processing information of difficult-to-machine metal materials, the machine tool and cutter selection experience and cutting technological parameters accumulated in production practices and experiments. The roughness test data of the optimized cutting technological parameters is selected for processing, so that a reasonable and mature technological scheme is recommended for manufacturing enterprises, and the numerical control processing precision of the difficult-to-machine metal materials is controlled. The purposes of increasing the processing efficiency of the difficult-to-machine materials, reducing processing cost and acquiring high quality products are achieved.

The invention discloses an image noisy point detection method based on a convolution neural network. The method includes the steps that sample images are collected, manual labeling and classification is conducted according to the types of noisy points, the sample images are input in the convolution neural network for training of classification models, and sample image blocks which are wrongly classified are collected in the classification process for secondary learning classification. The noisy points are labeled and classified in the manual and machine combination mode, supervised learning is achieved, learning accuracy of the convolution neural network is improved, and therefore the noisy points can be directly classified through the best trained classification model in the image noisy point detection process, and detection results are more accurate.

The invention discloses a building load forecasting method and device based on an improved IHCMAC (Hyperball Cerebellar Model Articulation Controller) neural network model. The method comprises the steps of: simulating the actual operation of a building to obtain building cold / heat load data and influencing factor data; determining input variables of the model according to the degree of correlation between the influencing factors and the building cold / heat load; clustering the input variables according to a particle swarm-K mean clustering algorithm to obtain values of L clustering centers, i.e., model node values, and defining a Gaussian kernel function for each node; and updating the weights of the nodes via a weight training algorithm to obtain a building load forecasting value of the model. The method has the advantages of fast convergence, high learning precision and strong generalization ability, and can provide a decision basis for energy-saving optimization control of a building system.

An air fuel ratio control apparatus for an internal combustion engine can improve learning accuracy in the air fuel ratio control even if the air fuel ratio of a mixture detected by an air fuel ratio detection device shifts from an actual air fuel ratio thereof. The apparatus controls the air fuel ratio of an exhaust gas flowing into an exhaust gas purification device based on an air fuel ratio feedback value and an air fuel ratio learning value. A temperature detection device detects the temperature of the exhaust gas purification device. A determination device determines, based on a difference between a detection value of the temperature detection device and a target temperature, that the air fuel ratio detection device shifts to a rich or lean side. The update of the air fuel ratio learning value is inhibited when the air fuel ratio detection device shifts to a rich or lean side.

A method for removing rain in a video based on noise modeling is disclosed. Under the assumption of a low-rank background, the rain bar noise component and the moving foreground in the video are simultaneously estimated. First, video data containing rain noise is acquired and a model is initialized; a rain map generation model is created according to the characteristics of the rain noise and the video foreground; the structural characteristics of the rain imaging in the video-a rain bar formed by moving rain droplets on each small block in an image is identical in the direction, the small block prior distribution of the rain bar is established; a moving object detection model is established according to the characteristics of the video foreground sparsity; the model is converted into a rain removal model under the maximum likelihood estimation framework; a rain-containing video and the rain removal model are applied to get a rain-removed video and other statistical variables, and the rain-removed video is output. The method aims to build a high-quality video rain removal model based on a rain map generation principle and rain bar noise structure characteristics, thereby more accurately allowing the video rain removal technology to be widely applied to complex raining scenes with the moving foreground.

A system for analyzing language using supervised learning method. With system extracts portions matching with structures of problem expressions from a raw corpus that is not supplemented with analysis information, then converts the extracted portions corresponding to the problem expressions into supervised data including problems and solutions and stores in the data storage. The system extracts sets of solutions and features from the supervised data stored in the data storage, carries out machine learning processing using the sets and stores learned results as to what kind of solution is the most straightforward for which feature in the learning results database. The system then extracts sets of features from the inputted object data, extrapolates analysis information showing the most optimum for a certain feature, from the sets of features based on the learning results database.

The invention discloses a multitask machine learning method and a multitask machine learning device both used for image classification. The method and the device are characterized in that low rank approximation of a residual structure and a covariance matrix of a regression matrix are utilized simultaneously, probability modeling is performed on the residual structure, the regression matrix, a low rank decomposition of the regression matrix and the covariance matrix of the regression matrix, learning of parameters of a probability model is performed through a variational deduction method or a sampling method, and a regression matrix high in accuracy is acquired finally and used for image classification. By the scheme, on one side, correlativity information among multitasks in the residual structure is utilized, so that parameter learning accuracy can be improved to improve classification accuracy; on the other side, by performing low rank approximation on the covariance matrix of the regression matrix, calculating complexity of an algorithm can be effectively lowered.

A method (and structure) for processing an inductive learning model for a dataset of examples, includes dividing the dataset into N subsets of data and developing an estimated learning model for the dataset by developing a learning model for a first subset of the N subsets.

The invention discloses a twist drill abrasion monitoring method. Firstly, four parameters such as drilling force, torque, electric current of a spindle motor and electric current of a feed motor in the different abrasion states are collected under the same cutting parameter; the four parameters are subjected to wavelet packet decomposition to obtain energy spectrums of eight frequency bands; and the energy spectrums of the second frequency band, the third frequency band, the fourth frequency band and the fifth frequency band are used as characteristic values serving as conditional attributes, the three abrasion states of a drill bit of a twist drill are used as decision attributes, a decision table is established, and the characteristic values are used as input neurons of a BP neural network for training and learning. The abrasion state recognition of the twist drill is performed through the established BP neural network, and the recognition rate is high.

Certain embodiments of the invention may include systems and methods for identifying drug targets using biological networks. According to an example embodiment of the invention, a method is provided for predicting the effects of drug targets on treating a disease. The method can include constructing a structure of a Bayesian network based at least in part on knowledge of drug inhibiting effects on a disease; associating a set of parameters with the constructed Bayesian network; determining values of a joint probability distribution of the Bayesian network via an automatic procedure; deriving a mean Bayesian network with one or more averaged parameters based at least in part on the joint probability values; and calculating a quantitative prediction based at least in part on the mean Bayesian network.

The invention discloses an electric automobile power battery SOC (State of Charge) detection system. The characteristics lie in that the detection system comprises a battery parameter acquisition platform and a battery SOC estimation system, the battery parameter acquisition platform is responsible for real-time parameter acquisition for the voltage, current and temperature of an automobile power battery pack, and the battery SOC estimation system can accurately estimate a battery SOC value through the acquired parameters; and the battery SOC is a non-linear, delayed, multivariable coupling and complex highly demanding real-time system. The detection system effectively solves a problem that the traditional automobile battery SOC estimation method is difficult to achieve an ideal effect.

A first intake air amount an engine is calculated based on a detected value of an intake air flow rate of an air flowmeter. A second intake air amount is calculated based on any one of a detected value of an intake pipe pressure and a throttle opening degree instead of the detected value of the intake air flow rate. When it is determined that the intake pulsation is not large, a difference amount of the second intake air amount from the first intake air amount is calculated. A corrected second intake air amount, which is a sum of the second intake air amount and the difference amount, is set as an intake air amount calculated value when it is determined that the intake pulsation is large.

An active learning system samples known data, and learns the known data independently in a plurality of learning machines, and selects data to be next learned for unknown data. The active learning system comprises a sampling weighting device for weighting the known data when they are sampled, a prediction weighting device for weighting the learning results of the learning machines when they are integrated, and a data weighting device for weighting the learning results when the data to be next learned is selected. When an extreme deviation is present in the count of data, each of the weighting devices more heavily weights a larger count of data.

The invention discloses a meta learning-based combined prediction method for a time-varying nonlinear load of an electrical power system. The method has the beneficial effects that (1) meta learning is a final result obtained by learning for a plurality of times on the basis of the learning result, and an output result of the previous layer of model and a characteristic attribute of a prediction sequence are utilized as input information of the next layer of learning, so that the previous learning can be fully applied to the later conclusion process, so that system deviation in the used learning algorithm can be found out and corrected and the learning accuracy is improved; (2) the optimal weight is obtained by setting the mean square error to the minimum and adjusting a gating network parameter through a decision condition in meta learning. Thus, important reference basis is provided for optimization and determination of the weight of the load combined prediction model.

A method includes estimating a model parameter in a case where operating states of plural electric devices are modeled by a probability model by using a total value of power consumption of the plural electric devices connected with a panel board. In the estimating, the model parameter in which likelihood calculated by a likelihood function becomes a maximum is estimated based on characteristics of power data that may be predetermined as prior knowledge from an operation tendency of each of the plural electric devices, the probability model is a factorial hidden Markov model (FHMM), and the likelihood is a value that indicates certainty of a pattern of a total value of the power consumption, which is modeled by the FHMM, of the plural electric devices with respect to a total value of the power consumption that is actually measured.

In a focus control for following a focus position of light beam to be radiated to an optical disk to an optical disk surface, the technique is for shorten the memory length in the learning controller. The learning controller is provided to a focus feedback control system and updates the memory comprised of N numbers of memory cells with a shorter sample period than a time obtained by dividing the time required for one rotation of the disk by N numbers and outputs the learning results in the memory with the shorter sample period.

The invention discloses an intelligent SOC (State of Charge) prediction device for an electric vehicle power battery, which is characterized by comprising a battery parameter acquisition platform and a battery SOC prediction system, wherein the battery parameter acquisition platform is used to acquire real-time parameters of voltage, current, temperature, and ambient temperature of the vehicle power battery pack; and the battery SOC prediction system is used to predict the battery SOC value through the acquired real-time parameters. The battery SOC is a nonlinear, delayed, multivariable-coupling, and complex real-time system with extremely high real-time performance requirements. The problem that the conventional prediction device can not achieve ideal battery SOC prediction precision effects can be effectively solved.

The invention discloses a data security sharing method based on a hash map and federated learning. Detection of a federated learning local model is added into a hashgraph consensus algorithm of a block chain 3.0 technology; dishonest nodes are prevented from providing an error model; meanwhile, the federated learning data model is realized through a method of carrying out weighted aggregation on alocal model. The method comprises the following steps: 1) adding detection on the federated learning local model into a block chain 3.0 technology hash graph consensus algorithm to prevent dishonestnodes from providing an error model; and 2) the dishgraph node detection process mainly comprising the following steps of: generating an event, performing Gossip communication, performing consensus byadopting a virtual voting algorithm, and realizing successful detection of dishonest nodes in a federal learning process based on a data security sharing model of hash and federal learning.

A service providing apparatus in which a learning control unit performs control as to whether or not to perform learning of a first estimation unit based on an output of an environment decision unit for outputting information about environment of a user and outputs of the first estimation unit and a second estimation unit for making stochastic decisions on service targeted for supply.

The invention discloses a radar high-resolution range profile target recognition method based on a state space model, mainly used for solving the problem of a large demand on training samples and poor recognition performance in the traditional radar high-resolution range profile target recognition technology. The realization process of the radar high-resolution range profile target recognition method comprises the steps of: extracting frequency spectrum amplitude signals after training sample normalization to be used as recognition characteristics of the training samples; modeling the recognition characteristics of the training samples by using a state space model; estimating all parameters of the state space model of the training samples by using an expectation maximization method, storing all the parameters in a recognition system template base; and extracting frequency spectrum amplitude signals after training sample normalization to be used as recognition characteristics of the training samples, and recognizing the recognition characteristics of the training samples. The invention has the advantages of a small demand on the training samples and high recognition performance, and can be used for recognizing radar targets.

The present invention involves a method, a system, and software for semantic analysis of disparate data in an environment having a plurality of datasets having distinct information fields. A candidate generation module involves creating graphs with information fields from the plurality of datasets as nodes, then creating smaller graphs containing source and sink vertices based on heuristic values. An electrical network computation module involves representing graphs as an electrical circuit to calculate the voltage of each node and the current of each edge by solving a system of linear equations. A diverse subgraph generation module involves selecting paths that carry the larger amount of current and have more new nodes in an iterative process. With each iteration, the path that scores the highest marginal current per number of existing types of nodes is selected and added to the diverse subgraph.

The invention provides an intelligent psychological pressure assessment and early warning system for multiple groups under an epidemic disease condition, and belongs to the field of artificial intelligence pattern recognition. The system comprises a data acquisition module, a psychological assessment module and an alarm module, wherein the data acquisition module is configured to acquire at leastone physiological signal of a tested individual and perform preprocessing; the psychological assessment module is configured to input the preprocessed physiological signal into a preset neural networkmodel to obtain the probability that the tested individual is in different psychological states, and then determine the current psychological state level of the tested individual; and the alarm module is configured to send out alarm information when the psychological state level of the tested individual exceeds a safety level. According to the system, only the related physiological signals of thetested individual need to be collected, and a psychological level classification result can be efficiently and accurately obtained without any inquiry, so that the individual with the psychological abnormality can be intervened and treated in advance, and the physiological and psychological health of the tested personnel is guaranteed.

A control apparatus controlling a motor for driving an object to switch a shift range of a transmission includes: switching devices for energization to windings of the motor; a controller for the switching devices; a current detecting circuit for a current in the windings and the switching device; a current limit circuit keeping an average of the current within a predetermined range; a standard position learning device of the motor at a learning range such that the current limit circuit limits the current, and the motor rotates until the object stops at a limit position of a movable range; a shift range determination device; and an error determination device determining that the current detecting circuit or the current limit circuit malfunctions when the object does not reach the limit position, the shift range is in a non-learning range, and the standard position learning device starts to learn the standard position.

An alternator configured to acquire a motive force from an engine to generate electricity and a capacitor capable of storing therein the electricity generated by the alternator are provided. An electricity consumption amount required during learning is calculated, and a capacitor electricity amount is compared with the calculated electricity consumption amount, and when the capacitor electricity amount is equal to or less than the calculated electricity consumption amount, the alternator is operated to generate electricity until the capacitor electricity amount is increased beyond the calculated electricity consumption amount. On the other hand, when the capacitor electricity amount is greater than the calculated electricity consumption amount, or when the capacitor electricity amount is increased beyond the calculated electricity consumption amount according to electricity generation by the alternator, the learning is performed while using the capacitor electricity without performing electricity generation by the alternator.

Engine control unit enabling to learn a change of relationship between a throttle opening degree and an intake air quantity (opening degree—air quantity characteristic) while restricting lowering of fuel efficiency to a minimum and to improve accuracy of engine stall prevention, torque control, including: a learning means which learns a characteristic change amount of the opening degree—air quantity characteristic; a learning requirement determining means; and a learning moving means causing executing learning in a stable driving state. The learning requirement determining means derives a separation amount between an intake air quantity corresponding to a throttle valve opening degree at present time and an actual intake air quantity detected in the stable driving state and determines requirement of learning with use of the separation amount and a threshold value.