33results about How to "Fast learning convergence" patented technology

The invention discloses a reinforcement learning controller for a five-degree-of-freedom bearingless permanent magnet synchronous motor and a construction method thereof. The reinforcement learning controller is composed of a current control module, six differentiators and three actor-critic modules, wherein an output of the three actor-critic modules is connected with a bearingless permanent magnet synchronous motor system via the current control module. According to the controller and the construction method thereof, the actors and critics in reinforcement learning in the field of artificial intelligence is combined with a traditional vector control technology, the critic evaluates rotation speed and displacement feedback information of a five-degree-of-freedom motor system and guides the actuator to output each current of the motor, a controller parameter can be online updated without an accurate motor model, stable operation of the motor can be maintained, strong capacities of resisting motor parameter variation and load disturbance are provided, meanwhile influences of uncertain factors, such as system parameter variation and load sudden change, on the system performance are overcome, and better robustness is provided.

The invention discloses an ultrasonic detection defect qualitative recognition method based on a neural network, and the method comprises the steps: carrying out the preprocessing of a damage signal through employing a wavelet packet threshold noise reduction algorithm in a wavelet analysis algorithm, reserving a useful signal in a first intrinsic mode component as much as possible, and carrying out the mode decomposition of the signal through employing a complementary set empirical mode decomposition algorithm; carrying out soft threshold noise reduction and rigrsure noise reduction, finally, carrying out superposition reconstruction on two parts of processed intrinsic mode components to obtain a final signal, and secondly, extracting feature vectors of different damage conditions to form a learning sample of a multivariable interpolation radial basis function. According to the method, noise reduction processing can be carried out on the collected signals, the convergence speed is high, the method is simple and effective, the radial basis function neural network after learning training has the capacity of ultrasonic detection defect qualitative recognition, device damage and the damage degree can be accurately recognized, and the damage positioning can be achieved.

The invention discloses an image processing method and device, a readable storage medium and computer equipment. The method comprises the following steps of acquiring a standard blurred image and a standard clear image corresponding to the standard blurred image; carrying out full convolution network training on the standard blurred image and the standard clear image, wherein the full convolutionnetwork comprises at least one training network layer, and each training network layer comprises a convolution layer, a mask activation layer and an ReLU activation layer; acquiring a to-be-processedblurred image; processing the to-be-processed blurred image by utilizing a test full convolution network in order to acquire a clear image corresponding to the to-be-processed blurred image, wherein the test full convolution network comprises at least one test network layer, and each test network layer comprises the convolution layer and the ReLU activation layer. By means of the image processingmethod and device, the readable storage medium and the computer equipment, a problem that the clearness effect of characters during image processing is poor can be solved.

The invention provides a multi-modal medical image multi-organ positioning method based on a one-to-one target query Transform, and belongs to the technical field of medical image processing. According to the method, the correlation between positions and sizes of organs is simulated by utilizing a conditional Gaussian model and a self-attention mechanism of Transform. The one-to-one target query architecture compulsorily executes a unique target query on each target organ, and the query sequence is the predicted category, so that classification is not needed, the network structure is simplified, redundant calculation is reduced, and the learning convergence speed is higher. According to the method, a 3D multi-modal image is projected to two orthogonal 2D planes before organ detection is executed, then complementary information from the multi-modal image is combined through a multi-modal fusion method, and finally, an obtained 2D bounding box is subjected to back projection to obtain a 3D bounding box, so that the calculation burden is reduced, and a more stable organ positioning result is obtained.

The invention discloses a tenant priority management method for a multi-tenant big data platform, and the method comprises the following steps: a tenant priority manager which consists of a resource request waiting queue and a radial basis neural network is initialized; When the cluster load is light, the resource request of each tenant can be met, and at the moment, the tenant priority manager isnot started; When the resource request of the tenant cannot be met, the resource request of the tenant is collected through a tenant priority manager and corresponding information is put into a resource request waiting queue; If the tenant needs to improve the resource request priority, an application is sent to the tenant priority manager, the tenant priority manager calls the radial basis neural network to calculate the tenant resource request priority, and then a judgment document is generated and sent to the system administrator for auxiliary judgment. Compared with a traditional resourceallocation strategy of the multi-tenant big data platform, the method can effectively overcome the defect of tenant priority processing of the multi-tenant big data platform.

The invention discloses a federal learning poisoning detection method and device based on feature confrontation, and the method comprises the steps: dividing all clients of each round of parameter training into benign clients and defense clients, and configuring a defense patch data set for the defense clients; in each round of training, enabling the benign client to optimize the benign model by using the local data set, enabling the defense client to optimize the defense model by using the defense patch data set and the local data, and enabling the server to aggregate all the benign models and the defense models to obtain a federated learning model; after multiple rounds of training are finished, using the federated learning model of the last round for detecting a poisoning sample, and during detection, according to a prediction result of a target label of a test sample in the federated learning model, and judging whether the test sample is poisoned or not by judging whether the prediction result of the defense target label in the federated learning model meets the label mapping relation after the optimal defense patch data is added into the test sample, namely realizing federated learning poisoning detection.

The invention discloses a container yard turnover falling optimization method under incomplete container picking information. The method comprises the following steps of performing simulation learningon a designed Q value table by utilizing container yard container picking operation simulation to obtain a Q value table after simulation learning; and step 2, dynamically generating an action instruction of the container turnover falling by utilizing the Q value table after learning and an action selection strategy, selecting a container falling position of a container to be turned over according to the action instruction, and adaptively updating the Q value table according to execution feedback of the action instruction in the yard container picking operation process. By adopting the method, self-adaptive adjustment of a container turnover falling optimization instruction in the yard container overturning operation environment can be realized through Q value learning in the container picking process; and the learning and convergence speed of a Q algorithm is increased, and the container turnover rate of a multi-layer stacking container yard and the secondary container overturning rate of the container yard are lowered.

The invention provides a method and a device for main and standby protection of packet transport networking equipment. The method comprises the following steps: router state synchronization is carried out to a first service processing unit in a main state and a second service processing unit in a standby state; and after the first service processing unit and the second service processing unit are subjected to main and standby switching, the second service processing unit carries out service flow data forwarding to the synchronized router information. According to the invention, a timing processing mechanism is adopted to execute router state synchronization, so that after the main and standby switching, the convergence of an MAC (Media Access Control) address and a router address of the standby service processing unit is quicker, thereby improving the main and standby switching performance of the service processing unit.

The invention discloses a CO-OFDM system and a receiving end thereof, and a nonlinear equalization method and device. The method comprises the following steps: clustering received OFDM symbols; judging the OFDM symbol as an edge point or a non-edge point according to a clustering result; inputting the OFDM symbols judged as the edge points into a radial basis function RBF neural network; and outputting a nonlinear equalization result of the OFDM symbol according to the output of the RBF neural network of the edge points or the clustering result of the non-edge points. The nonlinear equalization scheme provided by the invention is low in complexity and good in nonlinear equalization performance.

The invention discloses a measurement method of outlet coal moisture of coal moisture control. The method includes data collection, data screening, data processing and modeling, model inspection and moisture detection. According to the measurement method of the outlet coal moisture of the coal moisture control disclosed by the invention, an outlet coal moisture value can be obtained in real time through constructing a dryer steam outlet temperature value and outlet coal moisture value model and directly measuring a dryer steam outlet temperature value, and problems that accurate online measurement of the outlet coal moisture value cannot be realized, and time delaying of measurement in laboratories exists are solved.

The invention discloses a variance-covariance combination-based taxi passenger flow volume prediction method. The method comprises the steps of S1, predicting taxi demand passenger flow volume in a certain day in the future through a radial basis function neural network prediction method; s2, predicting the passenger flow volume required by the taxi in the same day in the future through a wavelet neural network prediction method; and S3, combining the prediction result obtained in the step S1 and the prediction result obtained in the step S2, and adopting a variance-covariance combined prediction method to obtain a final taxi passenger flow prediction result. According to the method, two combined prediction methods of the radial basis function neural network and the wavelet neural network are combined, the advantages of the two methods are complemented, the disadvantages are avoided, and thus the prediction precision and stability of the taxi passenger flow volume are greatly improved.