90results about How to "Increase the speed of learning" patented technology

The invention discloses a short-term load prediction method based on similar day segmentation and an LM-BP network. According to the method, through quantitative calculation of comprehensive correlative coefficients between meteorological factors and a historical load curve corresponding to a to-be-predicted day, the load curve of the to-be-predicted day is segmented, and according to prediction load curves within different time intervals, corresponding similar days are separately figured out; selection of the similar days is conducted through comprehensive consideration of a multi-feature similarity judgement standard of tendency similarity and shape similarity which are based on historical day meteorological similarity and historical load data. A similar day sample with the highest similarity is selected form historical data of the same kind; different neural network models are constructed through different training samples within different load prediction time intervals, and therefore the prediction precision of the neural network models is further improved. By means of the method, the calculation speed and convergence speed of a prediction algorithm are increased.

The present invention provides a system and method for building fast and efficient support vector classifiers for large data classification problems which is useful for classifying pages from the World Wide Web and other problems with sparse matrices and large numbers of documents. The method takes advantage of the least squares nature of such problems, employs exact line search in its iterative process and makes use of a conjugate gradient method appropriate to the problem. In one embodiment a support vector classifier useful for classifying a plurality of documents, including textual documents, is built by selecting a plurality of training documents, each training document having suitable numeric attributes which are associated with a training document vector, then initializing a classifier weight vector and a classifier intercept for a classifier boundary, the classifier boundary separating at least two document classes, then determining which training document vectors are suitable support vectors, and then re-computing the classifier weight vector and the classifier intercept for the classifier boundary using the suitable support vectors together with an iteratively reindexed least squares method and a conjugate gradient method with a stopping criterion.

The invention provides an under-actuation unmanned light board track tracking control method of ICA-CMAC neural network based on RBF identification. The under-actuation unmanned light boat track tracking control method of the ICA-CMAC neural network based on RBF identification uses a position reference system and a posture reference system to measure USV position information and heading posture information, performing filtering and space-time alignment on obtained USV posture and position signals to obtain a current USV accurate position and a posture, and adopting parallel control of an ICA-CMAC neural network and an integration divided type PID. The ICA-CMAC neural network realize feedforward control; credibility distribution is performed through introducing a balance learning constant; an USV inverse model is identified according to an adjustment index and a sigma learning rule; a generated output is used part of an USV input; and then a controller master control output including a PID controller and the ICA-CMAC neural network. The under-actuation unmanned light board track tracking control method of the ICA-CMAC neural network based on RBF identification solves a problem of USV track tracking under a condition that external interference is not determined, reduces dependence on an accurate mathematic model, enhances an adaptive adjustment capability and an interference-resistance capability of a system and improves on-line learning speed of an algorithm and track tracking accuracy.

The present invention relates to an echo canceler for training an echo path estimation without providing a hindrance to a speech. The echo canceler of the present invention includes a pseudo noise generator for generating a certain pseudo noise. This pseudo noise is forcibly supplied to a transmission line for transmitting a far-end talker's voice. Here, in the case where the far-end talker's transmitting speech level is faint or the far-end talker is in a speechless condition, a certain correlation is established between the pseudo noise and a signal of the transmission line for transmitting the near-end talker's voice. Based on such a correlation as just mentioned, a coefficient for generating an echo replica is calculated. Therefore, a training for estimating an echo path is performed based on the pseudo noise and without depending on the far-end talker's voice, thereby generating an appropriate echo replica.

A learning classification device includes a data memory, a learning unit, and a classification unit. The data memory is configured to store training data for learning a decision tree. The learning unit is configured to read a plurality of feature quantities included in the training data from the data memory by single access and derive data of a node based on the plurality of feature quantities, to learn the decision tree. The classification unit is configured to determine where the training data read from the data memory is to be split to from the node, based on the data of the node derived by the learning unit.

A temperature adjusting device is included with a sand and dust environmental testing system. The sand and dust environmental testing system includes: a circulating air duct, including a contraction section, a test section and a separation section etc.; a main blower; U-type separators and guide vanes, respectively configured at four corners of the circulating air duct. The temperature adjusting device includes: heat exchange pipes distributed in the U-type separators; four groups of guide vanes. Three groups of the guide vanes are configured with heat exchange medium holes and the other group of the guide vanes is equipped with heating device. An electric heater is accommodated in the second contraction section. An intelligent temperature control includes: building a neural network system structure; blended learning training of neural network parameters; PI control; and amplitude limitation processing.

The invention relates to the technical field of space-air-ground-sea integrated communication, in particular to a multi-beam low-orbit satellite intelligent dynamic channel resource allocation method.The method comprises: using a low-orbit satellite as an Agent; mapping the number of beam service request users at the current service request moment and the initial state of channel occupation by beams into an environment; at each service request moment, sending a service request, according to the environment where the low-orbit satellite is located, using a Q-learning algorithm to observe an environment state and acquire an award signal of the environment; learning a state-action value function, changing the channel resource allocation strategy step by step, to complete dynamic channel resource allocation of the low-orbit satellite mobile communication system. It can be avoided that the same channel resource is allocated to users or beams within the multiplexing distance, and a channelresource allocation scheme capable of enabling the overall performance of the low-orbit satellite mobile communication system to be optimal is achieved.

The present invention provides a system and method for building fast and efficient support vector classifiers for large data classification problems which is useful for classifying pages from the World Wide Web and other problems with sparse matrices and large numbers of documents. The method takes advantage of the least squares nature of such problems, employs exact line search in its iterative process and makes use of a conjugate gradient method appropriate to the problem. In one embodiment a support vector classifier useful for classifying a plurality of documents, including textual documents, is built by selecting a plurality of training documents, each training document having suitable numeric attributes which are associated with a training document vector, then initializing a classifier weight vector and a classifier intercept for a classifier boundary, the classifier boundary separating at least two document classes, then determining which training document vectors are suitable support vectors, and then re-computing the classifier weight vector and the classifier intercept for the classifier boundary using the suitable support vectors together with an iteratively reindexed least squares method and a conjugate gradient method with a stopping criterion.

The invention discloses a model-transfer-based large-sized new compressor performance prediction rapid-modeling method, which comprises the following steps: determining a rated value of each parameter and a stable running interval on the basis of a performance prediction model for an existing similar compressor by utilizing the prior experience knowledge of a new / old compressor; designing an experiment to acquire a small number of experimental data samples, performing normalization processing on the acquired samples according to rated running parameters of a new compressor, establishing a performance prediction model for the new compressor by utilizing an ELM (Extreme Learning Machine) neural network, performing transfer learning, and performing model transfer training by using experimental sample input data and a predicted output value of the basic model as input variables of the new model and using experimental sample output data as the output of the new model; testing the effectiveness of the new model by using the experimental samples. According to the method, the performance prediction model for the new compressor can be rapidly developed under the condition of less experimental data information by virtue of the performance prediction model for the existing similar compressor and the prior knowledge of the new compressor, so that the modeling efficiency and accuracy are improved.

The invention discloses an axial split-phase magnetic suspension flywheel motor suspension force numerical modeling method. The method comprises the steps of simulation and experiment design, sample collection and processing, model offline training and model online optimization. On one hand, the extreme learning machine based on principal component analysis is used for improving the adaptability and robustness of the model to parameter changes, rapid and accurate modeling of small sample data is achieved, and the precision and speed of the suspension force model are improved. On the other hand, a differential evolution algorithm is introduced to optimize the network structure, so that the built model can meet the precision requirement and the real-time control requirement, the problems oftoo many hidden layer neurons and huge network structure are avoided, the calculation speed of the model is increased, and the model is more suitable for modeling of such motors.

A fuel injection controller includes an oxygen sensor that responds to an oxygen concentration inside an exhaust passage, and an injection amount control unit programmed to control a fuel injection amount based on the output of the oxygen sensor. The injection amount control unit includes an injection amount correction value computing unit that determines an injection amount correction value based on the output of the oxygen sensor, a short-time learning value computing unit that determines a short-time learning value based on the injection amount correction value, a long-time learning value computing unit that determines a long-time learning value based on the short-time learning value; a feedback correction amount computing unit that computes a feedback correction amount, an injection amount control value computing unit that computes a control value of the fuel injection amount, and a long-time learning value holding unit that holds the long-time learning value.

A plant control system includes: a numerical calculation execution part which calculates the operation characteristic of the plant; a model for simulating the plant control characteristic according to information on the numerical calculation result; a learning part which learns the plant operation method by using the model; a learning information database which stores learning information data on the learning part; a pattern generation part which generates pattern data expressing a state input based on the learning information data in the learning part with a smaller input number than the model input dimension; a pattern database which stores the pattern data generated in the pattern generation part; and a learning result determination part which selects a learning result having a preferable control effect from the learning result obtained by using a plurality of patterns.

The invention discloses a reinforcement learning method in combination with knowledge transfer. The reinforcement learning method comprises the following steps: S1, designing a mapping relation between BP neural network autonomous tasks; S2, performing case storage on source task learning experience, and constructing a linear perceptron to learn an action mapping relation between a source domain and a target domain; S3, applying a case-based reasoning mechanism; S4, carrying out similarity calculation and case retrieval, and accelerating learning of related but different tasks by using the learnt experience in the case library as a heuristic expression; the method is applied to the learning method of the autonomous skills of the unmanned vehicle. According to the method, the advantages ofreinforcement learning and transfer learning are combined, and experience obtained by the robot from a simple domain or a source domain can be applied to a complex domain or a target domain through transfer acceleration; the learning speed is high, and the dimensionality disaster can be avoided; and the autonomous skill learning speed and efficiency of the unmanned vehicle are remarkably improved.

The invention belongs to the technical field of seismic and logging joint inversion, and discloses a CNN well-seismic joint inversion method and system, a storage medium, equipment and application. The method comprises the steps: searching an inversion mapping operator f1: y-> x from seismic data y to logging data x, i.e. X = f1 (y), with the seismic data y as the input and the logging data x as the output; reconstructing a logging curve in the forward direction; and reversely updating the weight and the bias. A four-layer network structure containing two hidden layers comprises an input layer, a first convolution layer, a second convolution layer and an output layer, and the two hidden layers are convolution layers. Some virtual logging curves are interpolated by using a Kriging interpolation technology, and virtual logging data and real logging data are used as training data for convolutional neural network learning. Under the condition that a real well is not additionally added, the number of learning samples can be increased through virtual well logging, an inversion mapping operator is searched for in a wider range, and over-fitting of local training data is prevented.

The time taken to learn a model from training examples is often unacceptable. For instance, training language understanding models with Adaboost or SVMs can take weeks or longer based on numerous training examples. Parallelization thought the use of multiple processors may improve learning speed. The invention describes effective methods to distributed multiclass classification learning on several processors. These methods are applicable to multiclass models where the training process may be split into training of independent binary classifiers.

The invention discloses an improved PMADDPG multi-unmanned aerial vehicle task decision-making method based on transfer learning. The method comprises the steps of firstly, performing reasonable mathematical modeling for threats such as air defense missiles and radars in an environment under the background of a multi-unmanned aerial vehicle actual combat environment, then creating a plurality of different two-dimensional combat environment models, designing constraint conditions, and learning and training the plurality of combat environments in sequence, thereby obtaining a final multi-unmanned aerial vehicle task allocation model. According to the method, the defect that in the prior art, task decision making can only be carried out in a known or static combat environment is overcome, decision making can also be carried out efficiently in an unknown combat environment, tasks of unmanned aerial vehicles are achieved, and the viability of an unmanned aerial vehicle group in the unknowncombat environment is greatly guaranteed.

Systems, methods, and apparatus for equalization preference learning are provided. An example method includes receiving a first label for a first audio concept for a media object and applying active learning to select a first example not yet rated by a first current user. The example method includes collecting a first user rating, by the first current user, of the first example compared to the first audio concept and applying transfer learning to combine the first user rating with ratings from prior users of examples not yet rated by the first current user to build a model of the first audio concept. The example method includes creating a tool operable by the first user to generate examples close to and far from the first label to modify the media object.

The present invention relates to a filter bank approach to adaptive filtering method using independent component analysis. More particularly, the invention relates to a method of improving the performance of adaptive filtering method by applying independent component analysis that is capable of reflecting the secondary or even higher order statistical characteristics to adaptive filtering algorithm using the filter bank approach.In order to implement the conventional adaptive filter algorithm using independent component analysis to the real world problem, a large number of filter training coefficients are required and also a large amount of calculation is required when a training is undertaken. This results in a very slow learning speed and the deterioration in the quality of result signals.The adaptive filtering method using independent component analysis according to the present invention provides a method of reducing the large amount of calculation required for filter training, improving the learning speed and the quality of result signals by utilizing a filter bank approach.Hence, the filter bank approach to adaptive filtering method using independent component analysis according to the present invention is capable of improving the performance over the conventional adaptive filtering method using independent component analysis.

Disclosed is a decision-making agent having a hierarchical structure. The present invention allows a user without knowledge about reinforcement learning to learn by easily setting and applying core factors of the reinforcement learning to business problems.

The invention belongs to the field of combining artificial intelligence with medical image analysis, and relates to a method for identifying pulmonary nodules by using a convolution neural network based on particle swarm algorithm optimization. The method comprises the following steps: obtaining CT slice images of patients' lungs; preprocessing CT slice images, and labeling the obtained CT slice images to form a data set; constructing convolution neural network; optimizing super parameters of convolution neural networks using particle swarm optimization; The optimized convolution neural network is trained by the labeled dataset. The features of pulmonary nodules were extracted by using the trained convolution neural network. The invention optimizes the convolution neural network through the particle swarm algorithm, solves the problem that the optimal super parameter of the convolution neural network is difficult to be selected manually, and has very good identification effect for lungnodules.

Disclosed is an epsilon-greed based online sequential perceiving and opportunity accessing method. In each time slot, users perceive channels sequentially and get access opportunistically to realize transmission. The method includes a step of initializing relative parameters and making access decisions based on online learning in each time slot, and has capabilities of learning environments actively and adapting to dynamic changes of environments. In addition, the method is an online decision implementing method, and the next decision is adjusted in real time according to each decision and feedback by systems, so that long-term accumulated throughput gain of the systems is maximized.

The invention relates to a wireless communication technology, in particular to an anti-interference wireless communication method based on deep reinforcement learning. The method comprises the following steps: using two convolutional neural networks: one convolutional neural network calculates a value function, and the other convolutional neural network performs action selection based on a calculation result of the value function; adopting priority experience sampling in an experience playback stage, so that experience samples with higher priorities are sampled preferentially, updating parameters of the convolutional neural network based on the experience samples, and updating the priorities of all the experience samples through calculation of the updated convolutional neural network; adopting a forward action reservation strategy, designing a Gaussian-like function to judge the value of the current action, and dynamically adjusting and controlling the probability that the current action is continuously executed. According to the method, the optimal sending power and the optimal communication frequency band can be intelligently selected, the learning speed of the whole system is improved, and the optimal sending mode can be learned under the condition that a third-party attacker model is unknown.

The present invention relates to an improved adaptive weighted average image denoising method based on an extreme learning machine. The method comprises the following steps: 1) inputting a training image set and a target image to be subjected to denoising processing; 2) employing the extreme learning machine to combine the pixel value and the rank logical difference of the pixel for performing training of the training image set to obtain the extreme learning machine; 3) detecting of the target image through the extreme learning machine model, and obtaining the image noise position; 4) employing an adaptive weighted average algorithm for image denoising processing; and 5) outputting the target image after the denoising processing, and performing assessment of the denoising effect. The improved adaptive weighted average image denoising method based on the extreme learning machine is high in applicability, high in feasibility, fast in calculation speed, high in effectiveness and high in practical value and can reach good image denoising effect.

An object is to provide a technique that a current state of a fuel cell may be detected more accurately. A fuel cell system includes a controller, a fuel cell, and an impedance measurer that may measure an impedance of the fuel cell. The controller obtains a first impedance value that expresses the impedance of the fuel cell in a predetermined state, acquires a second impedance value that expresses the impedance of the fuel cell that is measured by the impedance measurer during operation control of the fuel cell, and performs operation control of the fuel cell using the first impedance value and the second impedance value.

The invention discloses a frequency hopping sequence prediction method based on an optimized wavelet neural network, belonging to the frequency hopping sequence prediction method field. 1, performingtime domain analysis on frequency hopping signal to obtain the frequency hopping sequence at the current time; 2, preprocessing frequency hopping sequence to obtain a training sample and a test sample; 3, inputting training sample into the initialized neural network to carry out DBSCAN clustering calculation and weight optimization sequentially to complete the training; 4, inputting test sample into a trained neural network for prediction, and obtaining a frequency hopping sequence at the next time; The invention solves the problem that when the wavelet neural network is used for predicting different frequency hopping sequences, there is no universal and effective algorithm in the network training process, which leads to the problem that the number of hidden layer nodes and the initial value of wavelet translation factor can not be determined adaptively. The prediction accuracy of the same hidden layer node network is improved, the subsequent learning speed of the network is accelerated, and the running time of the program is shortened.

When the variation of the KCS learning value is large, an ISC learning control is performed in a learning acceleration mode in which the amount of the ISC learning update is large and the ISC learning is performed at high speed. By thus performing the ISC learning control in the learning acceleration mode in which the ISC learning is performed at high speed, the ISC learning control can quickly follow up a great variation of the KCS learning value, and the time required for the ISC control value to converge to the most suitable one can be shortened. With this, the ISC learning control performed during idling of a hybrid car can be completed in a short time, and the fuel consumption enhancement effect by idling stop can be heightened.

The invention belongs to the technical field of mathematical teaching, and particularly relates to an intelligent mathematical media teaching system. The intelligent mathematical media teaching systemcomprises a student side, a teacher side, a server side and an interaction device, wherein the student side and the teacher side can be installed and run on various intelligent devices such as mobilephones, tablet computers and computers; the server side is installed and run on a server in a school network center; the interaction device is installed on a desk in a multimedia classroom; and the student side, the teacher side, the server side and the interaction device are wirelessly connected through network signals. The intelligent mathematical media teaching system can objectively reflect the real learning level of each student, thereby being convenient for teachers to fully understand the students. Meanwhile, the intelligent mathematical media teaching system can increase the interaction between the teachers and the students in the teaching process, improves the teaching quality and the teaching effect, can bear most of the management work for the students, and reduces the burden of the teachers.

The invention relates to a telecommunication package recommendation method. The method comprises the following steps: determining a user feature vector of a user and a package feature vector of each package in a package library; connecting each package feature vector with the user feature vector, and inputting the package feature vectors and the user feature vectors into a factorization machine model to obtain an implicit vector matrix; and inputting the implicit vector matrix into a trained multi-layer perceptron model to obtain the probability that the user orders each package, and determining a package recommendation list of the user according to the probability. According to the invention, the deep features of the user and the package can be learned, and the accuracy of recommending the package to the user is improved. The invention also relates to a telecommunication package recommendation device, a storage medium and equipment.

The present invention generally relates to the development of cognitive function in infants. More particularly, the present invention provides the use of lactoferrin for improving memory and / or learning speed, and / or for promoting brain maturation in infants under physiological, i.e. non-pathological conditions. In one aspect, the present invention shows the utility of lactoferrin for improving long-term memory, e.g. long-term location memory in a healthy infant.