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

The invention discloses a railway wagon small part bearing stop key nut looseness and loss fault detection method, and belongs to the technical field of freight train detection. The objective of the invention is to solve the problems of high cost, low efficiency and the like of a current detection mode depending on manual image viewing and the problem of low accuracy of detection by an existing image automatic processing technology. The method comprises the following steps: acquiring an image, extracting a bearing blocking key region image, constructing a sample data set, and training a deep learning network by using the sample data set to obtain a trained deep learning network; extracting a bearing blocking key area image for a real vehicle passing image, and obtaining a three-value segmentation image by using the trained deep learning network; carrying out fault detection by utilizing the three-value segmentation image, and if no nut exists in a bolt area, indicating that the stop key nut is lost; if partial area of the bolt is above the nut, indicating that the stop key nut is loosened; and if the situation is detected, carrying out fault alarming. The method is mainly used forbearing stop key nut looseness and loss fault detection.

The invention discloses a reinforcement learning controller of a bearingless permanent-magnet synchronous motor and a construction method of the reinforcement learning controller. Input ends of a suspension winding current control module and a torque winding current control module are connected with an actor-critic module, the actor-critic module comprises an actor neural network, a critic neural network, a reinforcement signal module and an instantaneous difference module, output ends of the critic neural network and the reinforcement signal module are connected with an input end of the instantaneous difference module, an output end of the instantaneous difference module is connected with input ends of the actor neural network and the critic neural network, a displacement error and a rotational speed error are common inputs of the actor neural network, the reinforcement signal module and the critic neural network, outputs of the critic neural network are a suspension winding given current and a torque winding given current under a d-q coordinate system, rotational speed and displacement feedback information is evaluated by an actor, the actor is guided to control the suspension winding current and the torque winding current, and stable tracking control of a bearingless permanent-magnet synchronous motor system is achieved.

The invention provides an image recognition system and method based on a blockchain and federated learning, and belongs to the technical field of industrial Internet of Things, and the system comprises a model construction module and an image recognition module. In the Internet of Things image recognition process, an image collaborative recognition mechanism based on a blockchain and federated learning is provided, firstly, a set of credible collaborative mechanism is established between distributed Internet of Things devices which are not credible mutually by introducing the blockchain, and distributed image collaborative recognition is achieved; secondly, by utilizing a federated learning algorithm, an image recognition model is jointly learned among distributed Internet of Things equipment, so that centralized transmission of original data is avoided; and finally, by using the global model trained by federal learning, the Internet of Things equipment can quickly and efficiently complete an image recognition task locally, and accurate image recognition is realized while the data privacy is ensured.

The invention discloses a reinforcement learning controller of a bearingless permanent-magnet synchronous motor and a construction method of the reinforcement learning controller. Input ends of a suspension winding current control module and a torque winding current control module are connected with an actor-critic module, the actor-critic module comprises an actor neural network, a critic neural network, a reinforcement signal module and an instantaneous difference module, output ends of the critic neural network and the reinforcement signal module are connected with an input end of the instantaneous difference module, an output end of the instantaneous difference module is connected with input ends of the actor neural network and the critic neural network, a displacement error and a rotational speed error are common inputs of the actor neural network, the reinforcement signal module and the critic neural network, outputs of the critic neural network are a suspension winding given current and a torque winding given current under a d-q coordinate system, rotational speed and displacement feedback information is evaluated by an actor, the actor is guided to control the suspension winding current and the torque winding current, and stable tracking control of a bearingless permanent-magnet synchronous motor system is achieved.

The invention discloses a method for measuring coal humidity at a coal humidity control outlet, which includes data collection, data screening, data processing and modeling, model checking, and humidity detection. The invention discloses a method for measuring the coal humidity at the outlet of coal humidity adjustment. By constructing the model of the temperature value of the steam outlet of the dryer and the humidity value of the outlet coal, the humidity value of the outlet coal can be obtained in real time by directly measuring the temperature value of the steam outlet of the dryer. It is impossible to achieve accurate online measurement of the export coal humidity value, and there is a problem of time lag in laboratory measurement.

The present invention discloses a CO-OFDM system, its receiving end, and a nonlinear equalization method and device. The method includes: clustering received OFDM symbols; determining the OFDM symbols as edge points or Non-edge points; input the OFDM symbols determined as edge points into the radial basis function RBF neural network; output the nonlinear equalization result of the OFDM symbols according to the output of the RBF neural network of the edge points or the clustering results of the non-edge points. The complexity of the nonlinear equalization scheme of the present invention is not high, and the nonlinear equalization performance is good.

The invention discloses a method for dynamic collaborative dispatching of double yard bridges in a container yard, which includes step 1, using container yard operation simulation to simulate and learn the designed Q value table, and obtaining the Q value table after the simulation study; step 2, using The learned Q value table and action selection strategy dynamically generate the action instructions of the yard bridge, and the yard bridge selects tasks to carry out operations according to the action instructions. The method improves the operation efficiency of stacking and retrieving boxes in the storage yard, and reduces the interference of double yard bridges and the waiting time of vehicles.

The invention discloses an intelligent ship tracking method based on composite orthogonal neural network predictive control, comprising the steps of: obtaining a predetermined trajectory during the movement of the ship, and calculating the predetermined trajectory and the predicted output through an optimization algorithm to calculate the position of each propeller The optimization algorithm predicts the thrust; predicts the thrust through the neural network, and outputs the thrust that each thruster should produce by weighting the optimization algorithm predicted thrust and the neural network predicted thrust; predicts the position, heading, and speed of the ship through the prediction model; Correct the predicted values ​​of position, heading, and speed, and use the corrected predicted values ​​as the aforementioned predicted output. The present invention combines a compound orthogonal neural network to propose a new model prediction strategy. The neural network algorithm is simple, the learning convergence speed is fast, and it has excellent characteristics such as linear and nonlinear approximation precision, and the learning algorithm of the neural network can be completed offline. The online computing time is greatly reduced.

The present invention proposes a method for predicting user quality of experience (QoE) of an IPTV unbalanced data set based on a PNN-PSO algorithm. The method comprises the steps of: first extracting the influencing factors associated with the user's quality of experience from the set-top box record of IPTV, obtaining the characteristic input of the PNN-PSO model, and carrying out the definition of relevant concepts and the extraction and preprocessing of data; then analyzing The association between each feature and the user quality of experience (QoE); then based on the PNN-PSO neural network, the user quality of experience (QoE) prediction model is established, and the IPTV data set is used to realize the prediction of user quality of experience (QoE). This method overcomes the local minimum problem in the PNN algorithm and the problems caused by data imbalance, the learning convergence process is faster, and the prediction is more accurate and efficient; making full use of the IPTV set-top box data set can help IPTV operation Providers can make adjustments in a timely manner to improve the quality of user experience.

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 proposes a reinforcement learning variable-duration signal light control method based on the Internet of Things equipment, which mainly includes the following aspects: a reinforcement learning method based on the concept of "intensity" of the intersection is designed, and the data collected by the Internet of Things equipment is collected. Various real-time traffic information (such as vehicle position, speed, etc.) to control the phase selection of signal lights. At the same time, a most reasonable green light duration can be selected according to the number of vehicles in each lane. The invention can quickly converge to an excellent signal light control strategy under the condition of dynamic traffic changes, greatly shorten the learning time of the strategy and improve the control quality of the strategy.

The invention discloses a fault detection method for loose and missing key nuts of small parts bearings of railway freight cars, belonging to the technical field of freight train detection. The present invention aims to solve the problems of high cost and low efficiency in the current detection method relying on manual viewing of images, and the problem of low detection accuracy in the existing automatic image processing technology. The invention collects images and extracts the image of the bearing block key area, constructs a sample data set, uses the sample data set to train the deep learning network, and obtains a trained deep learning network; extracts the image of the bearing block key area from the real passing car image and uses the trained depth The learning network obtains the ternary segmentation image; using the ternary segmentation image for fault detection, if there is no nut in the bolt area, it means that the key nut is missing; if part of the bolt area is above the nut, it means that the key nut is loose; if the above If so, a fault alarm will be issued. It is mainly used for the detection of loose and missing faults of bearing retaining key nuts.