18results about How to "Effective training" patented technology

The invention provides a deep learning irradiance numerical mode prediction correction method and system introducing clustering, and the method comprises the steps: enabling information sequences under the same generalized weather type to be classified into the same cluster through introducing a clustering algorithm; the variational mode decomposition algorithm can decompose historical data time sequence signals into various intrinsic mode functions, redundant information is removed, and effective features are obtained; the feature learning task is optimized, so that deep learning model training and error correction can be more effectively carried out subsequently.

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training an action selection system using reinforcement learning techniques. In one aspect, a method includes, at each of a plurality of iterations: obtaining a batch of experiences, each experience tuple including: a first observation, an action, a second observation, and a reward; for each experience tuple, determining a state value for the second observation, including: processing the first observation using a policy neural network to generate an action score for each action in a set of possible actions; sampling a plurality of actions from the set of possible actions according to the action scores; processing the second observation using a Q neural network to generate a Q value for each sampled action; and determining the state value for the second observation; and using the state value to determine an update to current values of Q neural network parameters.

The utility model discloses a hip pushing training device, which relates to the field of fitness equipment and adopts the technical scheme that the hip pushing training device comprises a main supporting unit serving as a fixed supporting structure of the training device; the back supporting unit is rotationally connected with the main supporting unit and can support the back of the user; the back supporting unit comprises a back frame rotationally connected with the main supporting unit; the back supporting piece is arranged on the side, facing the user, of the back frame, and the back supporting piece is made of an elastic material and makes contact with the back of the human body. The hip pushing device has the advantages that the structure of the scheme provides good movement guide for hip pushing movement, and by combining the back frame and the hip supporting frame, a user can better complete the hip pushing movement under the condition that the user obtains effective supporting. By means of the structural combination of the scheme, when a user does hip pushing exercise, the shanks and the thighs are close to the right angle in the hip pushing training process, and the better training effect is provided for the target hip and hip muscle groups.

The utility model discloses an upper limb training device based on torque sensing, which comprises a bottom frame, a supporting seat and a display screen, the supporting seat and the display screen are arranged on the bottom frame, two sides of the supporting seat are provided with torque turntables, and a torque sensor electrically connected with the torque turntables is arranged in the supporting seat. The signal output end of the torque sensor is electrically connected with the signal input end of the display screen; a connecting rod and a rotating handle covering the connecting rod are arranged on the outer side of the torsion rotating disc, a bent portion is arranged at the other end, away from the torsion rotating disc, of the rotating handle, a sliding part connected with the end of the connecting rod is arranged on the inner side of the bent portion, an elastic element is arranged between the sliding part and the rotating handle, and training operation is facilitated.

The application discloses a game virtual role behavior decision-making method based on deep reinforcement learning, comprising the following steps: S1, obtaining the current environment state information of a game virtual role, and obtaining standardized environment state features after preprocessing the environment state information; S2, performing hierarchical reinforcement decision-making based on an intention-risk double-layer decoupling architecture, and generating a candidate behavior action set matching the current environment state; and S3, performing robustness screening on the candidate behavior action set through opponent counterfactual simulation, and the application relates to the technical field of game artificial intelligence. The game virtual role behavior decision-making method based on deep reinforcement learning performs robustness screening on the candidate behavior action through opponent counterfactual simulation, ensures that the selected behavior not only conforms to the current environment state, but also exhibits good stability and benefits under various possible opponent strategies, and greatly improves the ability of the game virtual role to cope with complex situations.

The application relates to an ASM-oriented radio frequency fingerprint identification method and system, and relates to the field of radio frequency fingerprint identification, wherein the method comprises the following steps: acquiring an ASM signal emitted by a ship and extracting a multivariate time sequence of the ASM signal; extracting time domain local features and frequency domain structural features of the multivariate time sequence, splicing the time domain local features and the frequency domain structural features to obtain first spliced features, fusing the first spliced features to obtain fused features; performing feature enhancement and spatial information fusion on the fused features to obtain multi-scale fused features; extracting global statistical information and key detail information of the multi-scale fused features, splicing the global statistical information and the key detail information to obtain second spliced features; and performing identification on the second spliced features to output a radio frequency fingerprint identification result. The application can effectively identify a radio frequency fingerprint, and further judge whether an ASM signal is tampered with.

The utility model discloses a kind of auxiliary up and down stairs sky rail self-adapting gravity center following device, including walking track and nose component, walking track is along stair arrangement, and is installed in the top of stair by sky rail hanger, nose component is used to generate power and move along walking track, nose component is connected with patient by sling and provides dynamic weight loss effect for patient;Measurement grating is arranged on the side of stair, measurement grating is used to detect the position of patient;A plurality of RFID chip modules are arranged on walking track, a plurality of RFID chip modules and the multiple steps of stair one-to-one correspondence, card reader is arranged on nose component, card reader is used to read the information of RFID chip module;External control system controls nose component movement according to the signal output by measurement grating and the information read by card reader;The utility model can adaptively assist cooperation patient training, achieve synchronization with patient's walking, make patient occupy initiative in the process of climbing stair training and can train alone, make training reasonable, effective and reliable.

The utility model discloses perfusion equipment for stem cell culture, which relates to the technical field of perfusion equipment and comprises a base, a mounting block is arranged on one side of the top surface of the base, a control panel is hinged on one side of the top surface of the mounting block, a culture tank is inserted in the middle of the top surface of the base, and a sealing cover is screwed on the top surface of the culture tank in a threaded manner. A plurality of liquid inlets are formed in the top surface of the sealing cover along the axial side at equal intervals, the liquid inlets are connected with pipelines through connecting assemblies, and culture solution bags are arranged at the tail ends of the pipelines. According to the stem cell culture device, the base is stably matched with the culture tank, so that the culture tank is convenient to mount and fix, the operation stability of the device is improved, and an efficient stem cell culture environment can be realized; through linkage cooperation of the control panel and the speed regulator, the flow rate of the culture solution can be accurately adjusted, and the control precision of the culture process is improved; finally, the problem that pipelines of existing culture equipment are inconvenient to disassemble and assemble is solved, and the stem cell culture effect and efficiency are improved.

The invention provides a bioreactor having a controllable observation function and used for culturing biological samples, the bioreactor comprises a shell, the shell is provided with a first chamber, a second chamber and a control module, and the first chamber and the second chamber are both provided with cabin doors; the first chamber is internally provided with a motion platform used for bearing a cell culture plate or a micro-fluidic chip; the environment control module is used for regulating and controlling internal environment parameters of the first chamber; the imaging module is used for observing a biological sample in the bearing cell culture plate or the micro-fluidic chip; an injection assembly is arranged in the second chamber and is used for conveying or extracting liquid required by the cell culture plate or the micro-fluidic chip; the control module is in signal connection with the motion platform, the environment control module, the imaging module and the injection assembly. Seamless integration of culture, perfusion, stimulation and in-situ observation is achieved, the device can be adapted to various culture carriers, flow velocity accurate adjustment and long-time unattended monitoring are supported, the pollution risk is remarkably reduced, and experiment repeatability and efficiency are improved.

The utility model relates to the technical field of tibia bone marrow infusion training, and discloses a tibia bone marrow infusion training device capable of accurately positioning and determining puncture depth, which comprises a base and a fixing frame, a puncture mechanism is arranged on the upper surface of the fixing frame, the puncture mechanism comprises a U-shaped frame, a connecting rod, a gear, a toothed plate and a moving frame, the back face of the U-shaped frame and the front face of the fixing frame are fixedly installed. According to the tibia bone marrow infusion training device capable of accurately positioning and determining the puncture depth, through the use of the fixing frame, the U-shaped frame, the connecting rod, the gear and the toothed plate, the position of the guide plate can be driven to ascend and descend, the effect of adapting to tibia models at different positions is achieved, and through the use of the moving frame, the guide plate and the scale plate, the puncture depth is accurately determined. The position of the guide plate can abut against the surface of the outer skin model, so that the problem of how to know the puncture depth while ensuring the puncture precision is solved, the effect during puncture training can be better, and the training effect in practice is improved.

This invention relates to the field of traffic big data analysis technology, and specifically to a method for identifying intercity travel modes based on sparse spatiotemporal big data of origin and destination spatial locations. The method includes the following steps: Step S1, constructing an intercity travel dataset and extracting sparse trajectories; Step S2, discriminative feature engineering for sparse trajectories: constructing a feature system based on the spatiotemporal context of origin and destination, including: origin and destination spatial location features, origin and destination time features, travel cost features, and the number of trajectory points and dwell time features; Step S3, based on the intercity travel dataset obtained in Step S1, implementing small-sample training set labeling based on spatial matching rules; Step S4, constructing and training a machine learning traffic mode classification model; Step S5, predicting the traffic mode for intercity trips labeled "traffic mode undetermined" in Step S3. This invention effectively overcomes the limitations on recognition accuracy caused by trajectory data sparsity and inaccurate intermediate point time information, possesses universality, and can achieve refined classification and recognition.

The present application relates to the technical field of rehabilitation equipment, and particularly relates to a dysphagia rehabilitation training device and a control method thereof. The device comprises: a gas pump and a training assembly connected through a gas pipe, wherein the training assembly comprises a plurality of air bags; a first pressure sensor is arranged on the surface of the training assembly, the first pressure sensor is used for collecting a first pressure, and the first pressure represents the tongue pressure of a user; a second pressure sensor is used for collecting a second pressure, and the second pressure represents the pressure in the air bag; a solenoid valve is used for controlling the flow and on-off of the gas in the gas pipe; and a controller is connected with the gas pump, the solenoid valve, the first pressure sensor and the second pressure sensor, and is used for realizing the control of the pressure in any air bag. The device can realize the partition training of the tongue of the user, and improves the rehabilitation effect.

The application discloses a kind of offshore wind turbine fault identification method and system, storage medium, computer equipment, the method includes: obtaining the historical vibration time series signal of offshore wind turbine gearbox multiple operating states and pre-processing training sample, on the basis of traditional one-dimensional convolutional neural network, integration multi-scale convolution module, SE attention mechanism and residual network structure, construct new network model.Training time, model receives sample by input layer, middle layer extracts feature, weighting, output layer generates probability distribution and outputs state, after reaching standard, it predicts the operating state under the new signal of real-time acquisition.By multi-scale convolution, the features of different time scales are extracted, the adaptive weighting of feature channels is realized by combining the SE attention mechanism module, and the stability of deep network training is enhanced by using the residual structure, thereby realizing the accurate identification of complex vibration signals, effectively improving the diagnostic accuracy and robustness, and providing a scientific basis for the operation and maintenance of offshore wind turbines.

The application discloses a three-dimensional model light ray tracing method based on meta learning, and belongs to the technical field of three-dimensional model image processing, and comprises the following steps: step one, constructing a general model for light ray tracing; step two, training a previous model by using a stereo matching convolutional neural network; step three, jointly training the previous model to obtain a new model; and step four, finding customized parameters and a customized model of the new model meeting a specific light ray tracing effect.The application trains an old model by using a stereo matching convolutional neural network based on the construction of a general model, trains general parameters meeting a specific light tracing effect, obtains new potential general parameters through the convergence of a loss function, adapts to a specific light tracing effect, effectively trains the old model by removing 90% of previous data and 98% of previous data, reduces the training data capacity, improves the accuracy of new and old data, and thus solves the problems of long training time, large training data capacity and low light ray tracing effect accuracy of the light ray tracing technology.

The application discloses an infrared and visible light visual information fusion method based on gradient transformation prior, and the method comprises the following steps: acquiring a registered infrared image and a registered visible light source image, wherein the registered infrared image and the registered visible light source image represent a consistent geometric transformation relationship in space; constructing a gradient transformation prior network; inputting the registered infrared image and the registered visible light source image into the gradient transformation prior network for feature extraction image fusion processing to obtain a reconstructed fusion image. By using the application, the detailed texture information in the visible light and the thermal radiation target information in the infrared image can be accurately extracted, and the fusion quality of the image is improved. The application can be widely applied to the field of image fusion technology as the infrared and visible light visual information fusion method based on the gradient transformation prior.

The invention discloses a classification method based on a photon pulse graph neural network and a linear calculation implementation method. The photon pulse graph neural network comprises D network layers and a decision output layer, the D network layers are connected in sequence, the output end of the last network layer is connected with the input end of the decision output layer, and the network structure of each network layer adopts a pulse neural network structure. The linear calculation implementation method comprises the following steps: selecting a weight matrix of M network layers of a trained photon pulse graph neural network; and loading the weight matrix of each network layer to an MZI network, loading an input signal required for linear calculation of each network layer to a light source, and injecting the light source loaded with the input signal into the corresponding MZI network to complete linear calculation of the input signal and output the input signal. According to the invention, the capture sensitivity of the network to the time sequence dependence characteristics can be obviously improved, and the method has higher processing speed, lower power consumption and stronger expandability.

This invention discloses a training method and system for quadruped robots based on deep reinforcement learning. The AMP training framework significantly enhances the robot's adaptability to complex environments. By distinguishing the robot's current motion sample distribution from that of a reference sample, the training aims to maximize the probability of misjudgment and the reward for reinforcement learning, enabling the robot to learn motion patterns that meet actual needs. When the environment changes, the robot can adjust its strategy promptly based on new information without remodeling, solving the problem of traditional methods being sensitive to environmental changes. Furthermore, this framework introduces imitation learning into reinforcement learning, providing effective guidance for robot learning, reducing the blindness of random exploration, accelerating the learning speed of effective strategies, and improving training efficiency. Simultaneously, the integrated hierarchical LSTM Actor network can extract and aggregate motion features at multiple time scales, allowing the robot to generate actions that consider both short-term gait and long-term trends, resulting in more natural and fluid movements. It also fully utilizes existing sample information, improving sample utilization and reducing training costs.

This application relates to the field of resource allocation technology, specifically to a resource allocation method, device, and system based on a serverless architecture. The method includes: obtaining the number of concurrent requests and request queuing latency within an observation window at each read time; obtaining the latency degradation risk level at each read time by analyzing the significance of the correlation between the number of concurrent requests and request queuing latency within the observation window, and the average difference between the actual queuing latency value and the fitted queuing latency value; analyzing the abnormal fluctuation degree of queuing latency degradation within the observation window at each read time, obtaining the abnormal fluctuation risk level at each read time, and further obtaining the risk characteristic value at each read time; using a neural network model to predict the number of concurrent requests at the next read time; dynamically adjusting the number of Pods based on the predicted number of concurrent requests; and allocating resources to concurrent request tasks. This application can improve resource allocation efficiency.