27results about How to "Avoid "forgetting" patented technology

The invention discloses a load prediction method and system based on error division multiple tasks, a medium and equipment, and the method comprises the steps: combining an error division task with a continuous learning mechanism, firstly calculating a sample relative error in a training process, classifying low-error data into a data set corresponding to a learned load feature mode, and carrying out the calculation of a data set corresponding to the learned load feature mode; the high-error data is delimited as a new task to be learned; measuring model parameter importance through a Fisher information matrix of EWC, introducing an L2 regularization item to constrain core parameter updating, and avoiding forgetting a learned old mode; and circularly executing task division and model training, so that a single model gradually masters various load characteristic modes in continuous learning. A division basis does not need to be set manually, a multi-mode learning process is simplified, and prediction accuracy and robustness in a complex load scene are improved.

The application provides a traffic classification method and system based on federal semi-supervised learning. The method comprises the following steps: constructing an unlabeled traffic dataset and a labeled traffic dataset; a center server decomposes a global model into supervised learning parameters and unsupervised learning parameters and initializes the parameters; the parameters and an auxiliary agent are sent to each client; each client performs unsupervised training based on the supervised learning parameters, the unsupervised learning parameters and the auxiliary agent by using a local unlabeled traffic dataset, uploads the difference between the unsupervised learning parameters to the center server; the center server aggregates and updates each unsupervised learning parameter; supervised training is performed by using a local labeled traffic dataset, and the difference between the supervised learning parameters and the difference between the unsupervised learning parameters are sent to each client; a new auxiliary agent is obtained based on nearest neighbor search, and the new auxiliary agent is sent to each client when a set sending condition is met; the two steps are iteratively executed until a stop condition is met.

The utility model provides a multifunctional tuning fork device for ear-nose-throat head and neck surgery, and relates to the technical field of medical instruments, the multifunctional tuning fork device mainly comprises a tuning fork, a connecting rod assembly and a hammer, the tuning fork is Y-shaped, the connecting rod assembly is fixedly arranged at the non-bifurcated end, the hammer is arranged at the tail end of the connecting rod assembly, and the connecting rod assembly is connected with the hammer. The connecting rod assembly comprises a connecting rod, a first handle and a second handle which are arranged in sequence, one end of the connecting rod is fixedly connected with the tuning fork, the opposite end of the connecting rod is detachably connected with one end of the first handle, the opposite end of the first handle is detachably connected with one end of the second handle, and a sound hammer is fixedly arranged at the opposite end of the second handle; according to the tuning fork device, when the tuning fork device is not used, the second handle and the first handle are connected together, so that forgetting to carry the tuning fork can be prevented, when the tuning fork device is used, the second handle and the first handle are detached, the tuning fork and the hammer are used in an impact mode, and a doctor can use the tuning fork device conveniently.

A method and system for identifying wireless communication devices based on radio frequency fingerprinting is disclosed. The receiver extracts radio frequency fingerprint features from the wireless signals transmitted by the wireless communication device to identify its identity. The method includes: Step 1. Pre-training a feature extraction network and a classification head using known device label data; Step 2. Jointly estimating the number of known and unknown device categories in the mixed signal through feature space stripping and effective rank estimation; Step 3. An adaptation phase, freezing the anchor points of the known class classification head and constraining the anchor points of the unknown class to actively guide isolation; Step 4. Obtaining pseudo-labels by performing semantic geometric registration on unlabeled samples based on balanced entropy regularized optimal transmission; Step 5. Adaptively training the feature extraction network based on progressive manifold anchoring and reliable pseudo-replay; Step 6. Outputting the identity of the device to be identified using the trained feature extraction network and a unified classifier. This invention solves the stability-plasticity dilemma of known class forgetting and unknown class confusion, achieving accurate device identification and number estimation in dynamic environments.

The utility model relates to the field of grounding wire, concretely to a contact net pre -buried channel flexible grounding bow -shaped grounding wire structure, including steel wire rope, bow -shaped end and L -shaped end, the bow -shaped end and L -shaped end are detachable respectively connected in the both ends of steel wire rope, the bow -shaped end and L -shaped end are opposite respectively with steel wire rope rotation, the steel wire rope is flexible steel wire rope, the welding surface of bow -shaped end and L -shaped end is connected with pre -buried channel and contact net welding respectively, the bow -shaped end includes bow -shaped welding section and bow -shaped bolt section, the bow -shaped welding section is connected with pre -buried channel welding, the bow -shaped bolt section is detachable with steel wire rope connects, the utility model can make contact net and pre -buried channel grounding installation more convenient, avoid simultaneously two lining trolley to rise top after grounding welding welding slag burn or damage two lining trolley on already brushed good release agent.

The utility model relates to the field of electronic cigarette discloses an air switch air channel structure of preventing condensate, including zinc alloy shell, PCB assembly, air adjusting device, POD oil storehouse base and data line storage mechanism, data line storage mechanism includes fixed box, the fixed box in top end fixed setting has micro motor, the micro motor output fixedly covers and sets up one no. In the utility model, the data line does not need to be carried additionally, the problem that traditional data line is lost or forgotten can be effectively avoided, after charging, the micro motor can be started to wind the data line neatly back to the outer end of the first binding roller.

The application discloses an intelligent network operation and maintenance management method based on user behavior analysis and belongs to the technical field of network operation and maintenance management. The method obtains original behavior log data of a user terminal, performs double alignment of a behavior baseline and an abnormal deviation degree to obtain a user behavior time sequence feature sequence; a user behavior analysis model including an input gating layer, a behavior memory evolution layer and an output mapping layer is constructed, the input gating layer splits the feature sequence into a steady-state behavior component and a transient behavior component; the behavior memory evolution layer differentiates and stores and fuses the two components through a competition mechanism to generate a behavior evolution state field; the output mapping layer performs nonlinear decoupling on the behavior evolution state field to output a user behavior abnormal risk measure and a behavior prediction path; and a potential risk node and an expected behavior trajectory are determined according to the user behavior abnormal risk measure and the behavior prediction path, and a dynamic access control strategy is generated. The method can improve the abnormal behavior recognition accuracy and the timeliness of dynamic adjustment of access control.

This invention belongs to the field of artificial intelligence technology, specifically relating to a cross-platform data interaction and sharing method and system based on artificial intelligence. It includes: constructing a health-aware edge cloud collaborative architecture to establish a dynamic health baseline for nodes; constructing a modified reward function with node health as a stability constraint, and using a multi-agent reinforcement learning model to train and obtain the optimal Nash equilibrium policy; decoding the optimal policy into an executable cross-platform data interaction planning graph through a policy interpreter network, and using game theory to coordinate resource conflicts; based on the planning graph, performing context-aware agent re-encryption and hierarchical homomorphic encryption, and establishing a power token chain to achieve fine-grained permission tracking; and performing closed-loop optimization of the policy and system self-evolution through multi-dimensional quality scoring, multi-layer source analysis, and elastic weight solidification technology; achieving synergistic gains in efficiency, security, and long-term stability of cross-platform data interaction.

This application belongs to the field of artificial intelligence and healthcare technology, specifically disclosing a multi-memory large model enhancement method and system based on adaptive resonance theory. The method includes: obtaining memory units from the plot memory module based on a new medical dialogue event; processing the new medical dialogue event through a large model based on multi-source memory; multi-source memory is constructed based on information stored in memory units, working memory modules, and semantic memory modules; the plot memory module is used to: determine candidate plot prototypes by analyzing attribute similarity; determine whether there is resonance between the new medical dialogue event and the candidate plot prototype; if there is resonance, use historical dialogue events in the event set of the resonant plot prototype as memory units and update the resonant plot prototype; if there is no resonance, create a new plot prototype. This application enables effective memory management and improves the stability and accuracy of large model reasoning in medical dialogue scenarios.

The invention discloses a multi-modal intelligent osteoarthritis detection method based on loop optimization, a storage medium and electronic equipment. The multi-modal intelligent osteoarthritis detection method comprises the following steps of multi-modal data acquisition and preprocessing, modal-by-modal alternate training, test set division and uncertainty-based multi-modal fusion diagnosis; and finally, evaluating and optimizing the detection model. According to the method, a weighted aggregation mechanism based on uncertainty is introduced. In clinical scenarios, some modalities may have noise or missing. The model can automatically calculate the entropy value (uncertainty) of each modal prediction result, allocate a higher weight for a high-quality modal and allocate a low weight for a low-quality modal, so that misdiagnosis caused by poor quality of a single data source is avoided, and the diagnosis stability in a complex clinical environment is remarkably improved.

The utility model provides a multifunctional IC support structure for a mobile phone, which comprises a support body, a mobile phone shell is arranged on one side of the support body, a clamping structure is arranged in the support body, an IC card is arranged in the support body, the clamping structure comprises a plurality of elastic clamping blocks, the IC card corresponds to the elastic clamping blocks, a limit structure is arranged on the support body, and the limit structure is arranged on the support body. The limiting structure comprises a plurality of limiting clamping blocks, the limiting clamping blocks correspond to the IC card, a partition plate is clamped and fixed to the side, close to the mobile phone shell, of the support body, and a heat dissipation structure is arranged on the support body. According to the utility model, the mobile phone shell is installed on one side of the support body, the support body can be installed on a mobile phone through the mobile phone shell, an IC card can be placed in the support body, the IC card is prevented from being forgotten and lost, the clamping structure is installed in the support body, and the limiting structure is installed on the support body. The IC card can be limited and fixed through the clamping structure and the limiting structure, and the IC card is prevented from falling off.

The utility model discloses a support foot pipe, including foot pipe, base, ground plug and foot pad, the upper end of base is fixedly connected with the lower end of foot pipe, the base is equipped with the axial threaded hole, one end of ground plug is plane and the other end is sharp, and the local of ground plug outer circumferential surface is equipped with the external thread section, ground plug can be with any one end upwards mode through the external thread section and be screwed into the threaded hole of base, the foot pad is detachably installed in the lower end of base, and the middle part of foot pad is equipped with the through hole that ground plug sharp part passes out. The foot pipe provided by the utility model is simple in structure, so processing and use are very convenient, and durable, no matter whether foot pipe is in ground plug support mode or foot pad support mode, the two parts of ground plug and foot pad are always installed on the foot pipe simultaneously, thereby can fundamentally avoid the loss or forget of ground plug and foot pad.

This invention relates to a method, apparatus, device, and medium for generating intelligent flight operations. The method includes the following steps: S1. Calibrate script settings before flight begins; S2. After flight begins, collect real-time response data of the aircraft under pilot control and define its features; S3. Standardize the collected data; S4. Construct a model based on the standardized data; S5. Calculate the prediction difference using a loss function on the model; S6. Train and optimize model parameters based on the difference. This invention trains individual models based on actual pilot control data, adaptively fitting multi-dimensional parameters such as operation amount, operation speed, and response latency. Compared with the fixed mapping of the average standard, this significantly reduces individual understanding bias, causing the command-control bias to converge to a set threshold after calibration, thus improving the consistency and predictability of human-machine interaction.

The utility model relates to the technical field of camera lenses, in particular to a reflective single lens reflex camera lens, which comprises a lens barrel, the lens barrel comprises a butt joint barrel connected with an external camera, an inner barrel is fixedly mounted at one end of the butt joint barrel, a focusing ring is rotatably attached to the outer side of the inner barrel, one end of the focusing ring is fixedly connected with a movable barrel, and the movable barrel is fixedly connected with a camera. The movable cylinder is connected with the inner cylinder in a screwed mode, and lens assemblies in the prior art are detachably installed and fixed in the butt joint cylinder and the movable cylinder. According to the utility model, the outer side of the lens cone of the lens is slidably sleeved with the first annular cover and the second annular cover, and after focusing of the focusing ring is completed, the second annular cover can be slid to enable the second annular cover and the first annular cover to be sleeved and protected on the outer side of the focusing ring, so that focal length change caused by the fact that fingers of a user mistakenly touch the focusing ring is effectively prevented; the lens cover is rotatably and slidably clamped at the open end of the light shield through the connecting column, so that the independent design of a traditional lens cover is omitted, and the lens cover is effectively prevented from being forgotten or lost after being disassembled.

The application relates to a model training method, which comprises the following steps: obtaining a plurality of song triplets; each song triplet comprises two sample songs collected by a sample user and one sample song not collected by a sample user; inputting the plurality of song triplets into a trained music understanding large model to extract first audio features of the sample songs in the plurality of song triplets, and determining a first feature similarity matrix based on the first audio features; inputting the plurality of song triplets into a to-be-trained song recommendation model to extract second audio features of the sample songs in the plurality of song triplets, and determining a second feature similarity matrix based on the second audio features; determining a relationship distillation loss according to the first feature similarity matrix and the second feature similarity matrix, and determining a triple loss according to the second audio features; training the song recommendation model based on the relationship distillation loss and the triple loss; and using the trained song recommendation model for song recommendation. The method can improve the accuracy of song recommendation.

The present application relates to a cross-view target geolocation method and device based on cross-task knowledge transfer. The method comprises: constructing a pre-training feature extraction module, including a first feature map extraction module and a second feature map extraction module with the same structure and weight sharing; based on a first loss function, inputting a reference image and a query image into the first feature map extraction module and the second feature map extraction module respectively for comparative learning training, to obtain trained first feature map extraction module and second feature map extraction module; constructing a cross-view target geolocation model, and transferring the trained first feature map extraction module and the second feature map extraction module to the cross-view target geolocation model; based on a second loss function, inputting the reference image and the query image with the query target into the cross-view target geolocation model for training, to obtain a trained cross-view target geolocation model. The present application can improve the target positioning accuracy and generalization ability in the cross-view scene.

The invention relates to the technical field of glass seal welding process optimization and defect prediction, in particular to a glass seal welding process optimization and defect prediction method based on deep learning. The method comprises the following steps: firstly, acquiring multi-source sensor electric signal data of a welding furnace in real time, and synchronizing and preprocessing to form a standard time sequence data sequence; and then, inputting the data sequence into a pre-trained multi-task deep learning model, and synchronously realizing dynamic process optimization and early defect prediction by the model through a shared feature extraction network and two parallel task branch networks. After the process is finished, the system associates an actual quality result with process data to form an incremental sample, and performs online fine adjustment on the model based on an intelligent trigger mechanism and an anti-forgetting algorithm, so that the system can adapt to changes of equipment and materials. According to the invention, the transformation of the glass seal welding process from fixed parameter control to real-time closed-loop intelligent optimization is realized, and the product yield, the process stability and the production intelligence level are effectively improved.

This invention discloses a wind turbine fault diagnosis method based on contrastive continuous learning, belonging to the field of wind turbine fault diagnosis. The main steps of this method include: acquiring raw data of the wind turbine's operating status and preprocessing it as training data; establishing a CNN-BiLSTM-based neural network model, initializing it, and inputting the training data into the model for multiple rounds of training; selecting the best-performing model as the wind turbine fault diagnosis model and putting it into use. Finally, in actual diagnosis, if the fault type is a known fault and the output confidence level is higher than a preset threshold, the diagnosis result is deemed valid; if the output confidence level is lower than the threshold, it is determined to be a new fault sample, and the model is retrained using contrastive continuous learning and experience replay mechanisms to ensure rapid learning of new fault modes and prevent the loss of old knowledge. After training, the updated model is redeployed to the real-time fault diagnosis system to dynamically adapt to new fault modes. This invention, through a closed-loop mechanism of diagnosis-detection-update, utilizes the feature representation capabilities of contrastive learning and the knowledge transfer mechanism of knowledge distillation to achieve rapid learning of the fault diagnosis model when facing new fault modes and retention of old knowledge, ensuring the continuity, accuracy, and stability of the fault diagnosis function.

The invention discloses a reactive power metering method suitable for a nonlinear load, and belongs to the technical field of power grid metering, and the method comprises the steps: obtaining all-condition electrical data, and obtaining a load harmonic feature vector data set based on the all-condition electrical data; constructing a load classification model based on the load harmonic feature vector data set; acquiring a real-time feature vector based on the current operation data; obtaining a load type based on the real-time feature vector and a load classification model; inputting the real-time feature vectors into the exclusive load correction model to obtain equivalent circuit parameters; obtaining a virtual optimal circuit structure based on equivalent circuit parameters, and obtaining a high-precision reactive power value; acquiring a universal reactive value based on the real-time feature vector and a universal reactive correction model; and obtaining a nonlinear load reactive power calculation result based on the high-precision reactive power value and the general reactive power value. According to the reactive power metering method suitable for the non-linear load disclosed by the invention, the accuracy and the stability of the reactive power metering of the non-linear load are realized.

The application provides a vehicle control method and device, a vehicle, a chip and a storage medium, and relates to the field of vehicle control.The method comprises the following steps: in response to the fact that a vehicle receives a power-off command and identifies that an object in the vehicle leaves the vehicle, acquiring statistical characteristics of signal strengths of at least one broadcast signal monitored in a target period; wherein the target period is determined according to a first time and a time when the vehicle door is closed last time; the first time is a time when the driving speed of the vehicle reaches a speed threshold; determining whether there is a target device left in the vehicle cabin according to the statistical characteristics of the broadcast signals; and if there is, controlling the vehicle to give a reminder. Thus, the method does not require that an electronic device has a wireless charging function or that the electronic device establishes a connection with a vehicle Bluetooth module as a prerequisite, but only needs to acquire the statistical characteristics of the signal strengths of the broadcast signals monitored by the vehicle in the target period, so that it can effectively determine whether there is a left device in the vehicle, and the applicability of the method is improved.

The invention relates to the technical field of medical instruments, in particular to an intelligent organ transfer box which comprises a transfer box body and a box cover arranged at the top of the transfer box body, a damping cylinder part is arranged in the transfer box body, and a hemispherical self-adjusting mechanism used for containing organs is arranged on the inner side of the damping cylinder part. A damping mechanism used for reducing the oscillation amplitude of organs is arranged in the damping cylinder part, an automatic water circulation refrigeration mechanism used for reducing the temperature in the water tank is arranged on the outer side of the damping cylinder part, a hemispherical sliding cavity is designed, self-adaptive angle inclination is achieved, the organs are kept in a horizontal state within a certain amplitude, and the shaking influence is reduced; a two-stage damping spring damping mechanism is matched with a semispherical air bag structure to relieve the influence of impact and vibration on organs in all directions, an automatic water circulation refrigeration mechanism is matched with a temperature control system, precise temperature control is achieved, a display screen and a positioning communication system are matched, the position and state of the transfer box are mastered in real time, and the transfer efficiency is improved.

The invention relates to the technical field of load prediction, and provides an integrated energy system load prediction method and system based on multi-task migration, and the method comprises the steps: obtaining a load data set, carrying out the parameter updating of a basic load prediction model, calculating the influence coefficient of each model parameter, and migrating the basic load prediction model parameters to a new load prediction model; acquiring load data at the current moment in the operation process of the new integrated energy system, calculating an accumulative error at the current moment, and if the accumulative error is greater than a first threshold value, updating parameters of the new load prediction model according to the influence coefficient, and performing load prediction by adopting the new load prediction model; and putting the load data at the current moment into the new integrated energy system data set, calculating the distribution difference between the load data set and the new integrated energy system data set, if the distribution difference is greater than a second threshold value, adding the data in the new integrated energy system data set into the load data set, and updating the basic load prediction model parameters again. And the new load prediction model is ensured to learn the knowledge of the basic load prediction model.

This utility model relates to the field of mine car counting and discloses a mine car counting lifting device linked to an image acquisition device. The device includes a lower mine car track, an upper mine car track, and a lifting device. The lower mine car track is located below the upper mine car track, and the lifting device is located on one side of the lower mine car track. It also includes a camera mounted on top of the lifting device, a computer mounted on the side of the lifting device opposite to the lower mine car track (connected to the lifting device), an industrial control computer mounted on top of the computer (connected to the lifting device), a detection gate mounted on top of the upper mine car track, and photoelectric switches and ultrasonic level sensors installed on the detection gate. The photoelectric switches and ultrasonic level sensors are connected to the computer. An image acquisition structure mounted on top of the lifting device, through the linkage of the lifting device with image acquisition, photoelectric detection, and level sensors, achieves automated and intelligent mine car counting.

The invention relates to the technical field of artificial intelligence and machine learning, and discloses an open environment-oriented new class adaptive adjustment method, which comprises the following steps: 1) when a new sample arrives, calculating an energy score and a consistency score; 2) utilizing the score quantile of a recent sample, and updating an energy and consistency threshold value through index moving average, so that the threshold value drifts adaptively along with the change of data distribution; 3) according to comparison between the sample scores and a threshold value, dividing the sample scores into a strong new class, a weak new class or a known class; 4) performing weighted accumulation on the new class evidence in the window according to time decay, when an accumulated value exceeds a preset threshold, formally judging that a new class appears, and immediately expanding the dimension of an output layer and randomly initializing a newly added weight so as to support the output of the new class; (5) an online updating stage: adopting a relatively low fixed learning rate for a known class, adopting an attenuation learning rate from high to low for a new class, and considering both old knowledge stability and new knowledge rapid absorption; according to the method, the continuous learning and generalization ability in the open environment is remarkably improved.

The utility model provides an education learning machine, which relates to the technical field of learning machines, and comprises a machine body, the inner side wall of the machine body is fixedly connected with a recording assembly, and the recording assembly is composed of a limiting sliding block, a recording plate, a pulling plate, a magnetic suction plate, a white plate, a first magnetic suction block and a second magnetic suction block. According to the educational learning machine, through the arrangement of the recording assembly, when the educational learning machine is used, a pulling plate is pulled to drive a recording plate to slide out, meanwhile, a limiting sliding block limits the sliding position of the recording plate, when a first magnetic attraction block and a second magnetic attraction block are attracted, the position of the recording plate is stabilized, the magnetic attraction plates can attract a whiteboard pen, and a whiteboard can conveniently record learning content; according to the method and the device, the previous learning content can be conveniently and quickly known during subsequent learning, the effect of improving the learning efficiency is achieved, and the conditions that the previous learning content is forgotten, not reviewed in time and not recorded by a paper book during subsequent learning, so that the subsequent learning cannot be performed, and the learning efficiency is reduced are avoided.

The utility model discloses an ear thermometer seat capable of sterilizing an ear thermometer cap, which comprises a frame seat and a sterilizing component, the frame seat is provided with a seat groove, the seat groove is provided with a sterilizing cavity, and an ear thermometer is buckled in the seat groove. A towel conveying channel, a winding cavity and an unwinding cavity are formed in the frame base, the towel conveying channel is arranged at the opening of the disinfection cavity, and the two ends of the towel conveying channel are formed in the side wall of the winding cavity and the side wall of the unwinding cavity respectively. The disinfection assembly comprises long cloth, a winding roller and an unwinding roller, and the long cloth is divided into a plurality of disinfection wet tissues. The winding cavity is internally provided with a rotating shaft, the rotating shaft is driven by a motor to rotate, the rotating shaft drives the winding roller to rotate, the long cloth moves in the towel conveying channel, the used disinfection wet tissue is wound to the winding roller from the disinfection cavity, the unused disinfection wet tissue is unwound to the disinfection cavity from the unwinding roller, and after the ear thermometer is buckled, the disinfection wet tissue in the disinfection cavity is attached to the outer wall of the ear thermometer cap. The purposes of placing the ear thermometer and disinfecting the ear thermometer cap are achieved, the step of manual cleaning and disinfection is omitted, the workload of medical workers is reduced, and the efficiency of measuring the body temperature is improved.