825results about How to "Easy alignment" patented technology

The invention provides a diffusion model LoRA fine tuning optimization method and system based on CLIP loss and perception loss, and the method comprises the steps: 1, dynamically adjusting the weights of the CLIP loss and the perception loss in combination with the CLIP loss and the perception loss in a LoRA fine tuning process; 2, calculating the semantic similarity between the denoised intermediate image and the target text by using a CLIP model, and optimizing the noise prediction capability of a diffusion model according to the similarity difference; step 3, calculating the difference between the intermediate image and the target image in the feature space by using perception loss, optimizing the noise prediction capability of the diffusion model, and improving the visual quality and detail fidelity of the generated image; and step 4, adjusting whether to start CLIP loss and perception loss or not according to the training progress. By introducing the CLIP loss, the model can better align the image and the text in the fine tuning training process, so that the generated image better conforms to the description of the text prompt.

The invention discloses a zero sample image classification method and device, and belongs to the field of image recognition. According to the method, a category description document is generated through a large language model, global and local feature information of an image and the document is obtained through a visual encoder and a text encoder, then decoupling is performed through a visual semantic decomposition module and a text semantic decomposition module, and visual semantic embedding and text semantic embedding are generated; through joint training of the modules, optimizing total loss composed of local-to-semantic mapping variance loss, multi-semantic diversity loss, decomposition semantic alignment loss and local semantic alignment loss; and the training is used for image category prediction in the reasoning stage after completion of training. According to the invention, the difficulty of zero sample image classification can be reduced, and the classification accuracy and efficiency are improved.

The invention discloses a solid-state battery cell manufacturing method and device based on single-sided rubber frame forming, and relates to the technical field of solid-state battery manufacturing. The method comprises the following steps: manufacturing a rubber frame on one surface of a first pole piece positioned at a forming station; the second pole pieces and the first pole pieces with the rubber frames are sequentially and alternately conveyed to a lamination station, and the first pole pieces and the second pole pieces are sequentially and alternately laminated, so that the second pole pieces are embedded into the rubber frames of the first pole pieces; wherein the electrodes of the first pole piece and the second pole piece are opposite, and the thickness of the rubber frame is greater than or equal to that of the second pole piece. According to the invention, two times of rubber frame forming on one pole piece are not needed, so that the connectivity between the pole pieces is improved, and the conditions that the solid electrolyte layer and the pole pieces deviate relatively and the two adjacent pole pieces are short-circuited are avoided.

The invention relates to the technical field of hydrogen fuel cells, in particular to an air-cooled reactor automatic loading line for hydrogen fuel cell stack production, which comprises a rack, a storage rack, a material moving mechanism and two conveyor belts, the material moving mechanism comprises a support frame, a driving assembly and a pressing assembly, and electrode plates and bipolar plates are alternately conveyed to the material moving mechanism through the two conveyor belts. The electrode plate or the bipolar plate is supported through the sliding rollers at the bottoms of the overturning frames, and then the two overturning frames are synchronously driven to move through the driving assembly, so that the electrode plate or the bipolar plate naturally falls between the two overturning frames. The auxiliary electrode plate or the bipolar plate can stably fall on the storage rack in cooperation with downward pressing of the downward pressing assembly; the process does not need manual intervention, and the accuracy and regularity of stacking are ensured. The automatic stacking reduces the dependence on manual operation, reduces the labor cost, and further reduces the rework cost caused by unqualified materials due to the reduction of the surface damage of the electrode plates and the bipolar plates.

The invention belongs to the technical field of computer vision, and particularly relates to a medical image classification method based on priori knowledge enhanced masks and alignment modeling. Comprising the following steps: acquiring an image text pair data set and preprocessing the image text pair data set to obtain a preprocessed image text pair data set; performing preliminary training on a visual encoder and a text encoder according to the preprocessed image text pair data set to obtain a preliminarily trained visual encoder and a preliminarily trained text encoder; training the visual encoder for the second time in combination with the text encoder to obtain a trained visual encoder; classifying the medical images by adopting the trained visual encoder; the model provided by the invention can be better generalized to various downstream medical image recognition tasks, and the accuracy of the downstream medical image recognition tasks is improved.

The invention discloses a cross-modal foreground distillation method for multi-view three-dimensional target detection, and belongs to the technical field of automatic driving perception. According to the method, a laser radar or a multi-modal three-dimensional target detection model is used as a teacher model, a multi-view aerial view BEV perception three-dimensional target detection model is used as a student model, and the model performance is improved through multi-scale feature distillation; a space attention mechanism and a foreground mask module are introduced, and the student model can simulate feature responses of the teacher model in different space positions under multiple scales, focus on a foreground target and suppress background noise; the adaptive module ensures the consistency of the multi-scale features in feature dimensions and spatial resolution, and finally, the difference between the features and regression output is measured through loss functions such as mean square error and KL divergence, so that efficient cross-modal and multi-scale foreground feature distillation is realized. Finally, the multi-view three-dimensional target detection precision and efficiency are improved, and meanwhile, the distillation requirements of multi-scale features are fully considered.

The invention discloses a multi-modal emotion prompting method and system for an online auxiliary voice dialogue scene, and relates to the technical field of artificial intelligence, voice signals in the voice dialogue scene are input into a trained multi-modal emotion prompting large model, and emotion labels and voice emotion prompting information are output; the training process of the multi-modal emotion prompt large model comprises the following steps: acquiring original voice data of a real dialogue scene to construct a training data set for emotion recognition; performing feature extraction on the original voice data in the training data set to obtain an original voice feature vector; inputting the original voice feature vector into a sparse bridging Transform module, and outputting a fusion feature; inputting the fusion features into a large language model, generating coherent voice emotion prompt information through text decoding, and associating with corresponding emotion tags; according to the multi-mode emotion prompting method and system, accurate analysis and prompting of voice emotion in an online auxiliary voice dialogue scene are realized.

The invention belongs to the field of computer vision, and particularly relates to a consistent identity picture generation method based on a diffusion model, which comprises the following steps of: generating optimal paired data by adopting a multi-modal large language model, the paired data comprising text data and corresponding image data; carrying out encoding processing on the text data by adopting a text encoder to obtain a text embedding vector; processing the image data by adopting an image encoder to obtain an image embedding vector; inputting the text embedding vector and the image embedding vector into an MLP module for fusion to obtain a fusion embedding vector; stacking all the fusion embedding vectors, and processing the stacked fusion embedding vectors by adopting a cross attention layer to obtain deep-level features; performing multi-stage adjustment on the deep feature to obtain a consistent identity picture; according to the invention, a coarse-to-fine multi-stage fine adjustment method is adopted, and the visual quality of the generated image is gradually optimized.

The invention discloses a high-temperature superconducting Tokamak cold shield assembly assembling system and method. The high-temperature superconducting Tokamak cold shield assembly assembling system comprises a positioning assembly, a welding assembly, a supporting frame, supporting legs, a locking mechanism and a driving mechanism. The invention relates to the technical field of welding tools. According to the high-temperature superconducting Tokamak cold shield assembly assembling system and method, the supporting frame is arranged to be installed around the lower-layer high-temperature superconducting Tokamak cold shield, the positioning assembly and the welding assembly can be placed on the supporting frame in sequence, and the positioning assembly can abut against the lower-layer high-temperature superconducting Tokamak cold shield by half; the high-temperature superconducting Tokamak cold shield suspended on the upper layer to be put down is positioned and guided, it is guaranteed that the high-temperature superconducting Tokamak cold shield can be accurately in butt joint with the high-temperature superconducting Tokamak cold shield on the lower layer, and then the problem that butt joint dislocation of the heavy high-temperature superconducting Tokamak cold shield is difficult to adjust is solved. And after the positioning assembly is taken down, the welding assembly is replaced, the welding assembly can be driven by the driving mechanism to rotate around the high-temperature superconducting Tokamak cold shield along the supporting frame for welding, and manpower is saved.

The invention provides an easy-to-assemble backheating device for a solid elastic card, which comprises a sleeve, a plurality of solid elastic card material plates stacked in the sleeve and a pressure head for applying stress to the solid elastic card material plates, and the sleeve is composed of a plurality of parts and is used for inspecting or assembling and disassembling the plurality of solid elastic card material plates when being unfolded; each solid elastic clamping material plate is provided with a flow cavity after being stacked, a heat-conducting medium exchanges heat with the solid elastic clamping material plates through the flow cavities when heat or cold is generated, and the heat-conducting medium flows into a heat exchanger through a liquid through hole in the pressure head for heat exchange. The sleeve is designed to be a plurality of detachable parts or folding parts, during assembly, a plurality of solid elastic clamping material plates can be conveniently placed in, alignment is convenient, in the loading and unloading process, even if a certain solid elastic clamping material plate goes wrong, opening and inspection are convenient, and therefore the assembly efficiency is improved, and the fault solving efficiency is improved.

The invention discloses an image-text retrieval method based on information enhancement and multi-modal global local feature alignment, and relates to the technical field of image-text retrieval, and the method comprises the following steps: 1, obtaining image information and corresponding text description data; 2, respectively carrying out feature extraction on the picture data and the text data to obtain picture global features, local features, text word features and sentence features; 3, fusing the picture local features and the picture global features to obtain local features with global information, and realizing information enhancement; 4, putting the image features and the text features into a cross attention mechanism model for coarse and fine granularity feature fusion to obtain image-text global fusion features and image-text local fusion features, and realizing coarse and fine granularity alignment of the image-text; and 5, carrying out similarity calculation on the image-text local fusion feature and the image-text local feature to realize image-text retrieval. Through information enhancement and feature fusion, alignment of coarse granularity and fine granularity is ensured, and alignment of images and texts is improved.

The invention relates to the field of insulating sheath installation, in particular to an electrified installation transformer pile head insulating sheath installation device which comprises a first installation rod, a second installation rod and a connecting rod, an insulating sheath is installed in the installation rod, a wire sheath is installed on the insulating sheath, and wire grooves are formed in the same horizontal line of the insulating sheath and the wire sheath. A linear motor is installed in the installation sleeve, the output end of the linear motor is connected with the top of the insulation sheath, an empty groove is formed in the installation sleeve, a first spring is installed in the empty groove and connected with a clamping block, a top block is installed on the top of the insulation sheath, and the side, close to the clamping block, of the top block is provided with an inclined face matched with the side face of the clamping block. A first mounting disc is mounted on the outer side of the mounting sleeve, a driving structure is arranged in the first mounting disc, and the driving structure is used for driving the mounting sleeve to rotate so as to adjust the mounting direction of the insulating sheath and the wire sheath, manual mounting is not needed, the risk of safety accidents is reduced, and the life safety of operators is guaranteed.

The invention discloses a three-dimensional model package arranging method, device and equipment, a medium and a product, and relates to the technical field of three-dimensional (3D) printing.The method comprises the steps that space parameters of a to-be-printed model and a printing area are determined, the to-be-printed model comprises i three-dimensional models, and i is a positive integer larger than 1; based on the model parameters of the i three-dimensional models, the arrangement sequence of the i three-dimensional models is determined, and the model parameters are the contour parameters of the minimum enclosing rectangles of the three-dimensional models; and the i three-dimensional models are placed in the printing area based on the space parameters of the printing area and the placing sequence of the i three-dimensional models by utilizing a preset model printing rule. According to the method, the arrangement uniformity of the printing models can be improved while the package arrangement speed is ensured, the to-be-printed models are arranged in the printing area as many as possible, excessive analysis on the to-be-printed models is avoided, the calculation amount of a processor is saved, package arrangement efficiency and package arrangement quality are both considered, and the package arrangement experience of a user is improved.

The invention belongs to the technical field of computer vision and image processing, and particularly discloses a video defogging method and system based on enhanced alignment, and the method comprises the steps: carrying out the color mapping of a plurality of foggy images through a color lookup table, and obtaining a plurality of enhanced images; based on the enhanced image corresponding to the foggy target frame and the enhanced image corresponding to each foggy reference frame, performing space-time optical flow estimation to obtain a three-dimensional optical flow; performing image transformation on the foggy reference frame based on the three-dimensional optical flow to obtain an alignment frame; and based on the alignment frame, the foggy target frame and the enhanced image corresponding to the foggy target frame, performing splicing and feature fusion to obtain a defogged image corresponding to the foggy target frame. According to the method, details of the foggy image are enhanced, a more accurate alignment result is obtained, more inter-frame information is provided through accurate alignment frames, the network can be effectively helped to learn a better defogging capability, the utilization capability of the deep neural network on the inter-frame information can be improved, and a better video defogging effect is achieved.

The invention discloses a remote catching net catching device and belongs to the technical field of catching, the height of the device is smaller than the outer diameter, the device is launched in the radial direction of the device, and the device comprises an intermediate control module, an external launching module and a catching net; the multiple external injection modules are located on the periphery of the middle control module, the adjacent external injection modules are detachably connected, and the external injection modules are detachably connected with the middle control module; the catching net is located in an annular cavity formed by splicing a plurality of external shooting modules, and one end of the catching net is connected with the external shooting modules; the intermediate control module is used for receiving an operation instruction, and according to the instruction of the operation instruction, the intermediate control module delays or instantly emits the external emission module to the outside so as to ensure that the capturing net is horizontally unfolded above the front part of the target object; the external shooting module pulls the catching net to move downwards so that the catching net can fall on the target object. According to the device, the catching net is approximately in a horizontal opening state and does forward parabolic motion, it is guaranteed that the catching net can fall down from the position above a target object, the effective catching area is increased, and the catching efficiency is improved.

The invention discloses a model training method and a data alignment method, and the method comprises the steps: training a time sequence feature extractor through a time sequence data set, inputting the time sequence data set into the trained time sequence feature extractor, and obtaining time sequence feature vector mapping; obtaining text feature vector mapping corresponding to the time sequence data set by using a text data label of the time sequence data set; constructing a first loss function by using time sequence feature vector mapping and text feature vector mapping, and adjusting model parameters by taking minimization of the first loss function as a target to obtain a trained data alignment model; the time sequence features of the time sequence data are extracted by using the time sequence feature extractor, so that the data alignment model can realize alignment of the time sequence data and the text data, and data of different sources and types are effectively fused.

The utility model relates to the technical field of workpiece positioning equipment, and discloses a positioning mechanism for glass production, which comprises a bearing table, a plurality of racks are arranged on the bearing table, the plurality of racks are arranged along the circumference of the center point of the bearing table in an array manner, and the upper side surface of each rack is provided with a clamping unit for positioning and clamping glass. The lower surface of the bearing table is provided with a driving unit for simultaneously driving the racks to get close to or away from each other; the clamping unit comprises a fixing base arranged on the upper side face of the rack, and a movable moving plate and a fixed plate with the fixed position are arranged on the upper side face of the fixing base. According to the glass clamping device, the racks are arranged, the clamping units are arranged on the racks, the driving unit drives the racks to move oppositely at the same time, the movable plate drives the elastic pressing wheel to press glass on the bearing table when moving, and the center position of the glass is aligned with the center position of the bearing table conveniently; and the glass is fixed immediately after alignment, so that the secondary processing efficiency of the glass is improved.

The invention relates to the technical field of household appliances, and provides a door body assembly and refrigeration equipment. The door body assembly comprises a box door rotationally connected to a box body and suitable for opening or closing a storage space of the box body; the door plate is arranged on the outer side of the box door in a sliding manner; and the driving sliding door mechanism is arranged on the box door and is configured to drive the door plate to move relative to the width direction of the box door in the opening or closing process of the box door. According to the door body assembly, the driving sliding door mechanism is arranged on the box door, and the driving sliding door mechanism is connected with the box door and the door plate so as to drive the door plate to slide relative to the width direction of the box door in the rotating opening or closing process of the box door relative to the box body; the door plate can avoid obstacles such as walls and objects on the side face in the closing or opening process of the box door, the situation that rotation of the box door is affected due to interference between the door plate and the obstacles is avoided, and the use requirements of users in different scenes are met.

The invention relates to the technical field of inertial navigation, in particular to a group affine property-based rotation modulation inertial navigation system backtracking alignment method, which is characterized in that on the basis of rotation inertial navigation multi-position alignment, forward and reverse Kalman filtering can be respectively combined with an alignment state equation meeting group affine property based on the Lie group theory, so that the rotation inertial navigation multi-position alignment can be realized. Carrying out repeated backtracking initial alignment on any large misalignment angle, and obtaining an alignment result with relatively high precision in a relatively short time; the method has the capability of quickly realizing initial alignment in a maximum misalignment angle environment, and the speed of the method is far higher than that of a traditional alignment technology; within the same alignment time, the precision performance of the method is more excellent; compared with a current Lie group alignment method, the method has the advantages that the speed and the precision are obviously improved under the condition that the calculation performance is relatively excellent; within limited alignment time, the higher the computing power of the equipment is, the better the alignment effect is, and the method is particularly suitable for the equipment needing to be quickly started.

The invention provides a large language model alignment method, which comprises the following steps: acquiring a pre-trained large language model, and initializing the pre-trained large language model to enter a text completion task state; performing model feature retrieval on the target data set, and determining at least one target large language model; constructing a context alignment prompt word according to the target large language model, and inputting the context alignment prompt word into a pre-trained large language model; obtaining a target session text, updating the target session text into the context alignment prompt word, and reasoning to obtain a session response text; the method comprises the following steps of: adding a session response text of a current session round into a context alignment prompt word; and obtaining a target session text of the next dialogue round, inputting the target session text and the context alignment prompt word into the large language model to obtain a session response text of the next dialogue round, and setting the next dialogue round as the current dialogue round. According to the method, efficient alignment of the pre-trained large language model can be realized.

The present invention relates to the field of hydrogen fuel cell technology, and specifically to an automatic stacking line for air-cooled stacks used in the production of hydrogen fuel cell stacks, including a frame, a storage rack, a material transfer mechanism and two conveyor belts, the material transfer mechanism including a support frame, a drive assembly and a pressure assembly, the electrode plates and bipolar plates are alternately transported to the material transfer mechanism by two conveyor belts, the electrode plates or bipolar plates are supported by sliding rollers at the bottom of the flip rack, and then the two flip racks are synchronously driven to move by the drive assembly, so that the electrode plates or bipolar plates fall naturally from between the two flip racks. With the downward pressure of the pressure assembly, the auxiliary electrode plate or bipolar plate can fall stably on the storage rack; the above process does not require manual intervention, ensuring the accuracy and neatness of the stacking. Automated stacking reduces dependence on manual operation and reduces labor costs, and due to the reduction in surface damage to the electrode plates and bipolar plates, the rework costs caused by unqualified materials are further reduced.

The invention provides a laser confocal ocular surface image stitching method based on deep learning. Efficient and accurate image stitching is realized through unsupervised learning. Firstly, an ocular surface image is preprocessed, and a data set is constructed; then, a multi-scale feature pyramid is constructed by using a convolutional neural network, image features are automatically extracted and pre-aligned, and image alignment is optimized through global motion aggregation and local motion correction technologies. Thirdly, constructing a content fusion generative adversarial network which comprises a generator and a discriminator to smooth the fused image and generate a spliced image conforming to human eye perception; a staged training strategy is adopted, image alignment is optimized firstly, and then image smooth fusion is achieved. And finally, automatically finishing image alignment and fusion by using the trained network to obtain a seamless large-view confocal ocular surface image. According to the method, the splicing precision and robustness are improved, the method is suitable for image splicing in different environments, and the splicing success probability is remarkably improved.

The invention discloses a display panel defect detection method, device and equipment based on attention potential energy, a medium and a product. The method comprises the following steps: acquiring an image set of a multi-scale channel associated with a to-be-detected panel; performing quality assignment on each image pixel in the image set of the multi-scale channel to obtain a point mass body set of the multi-scale channel; constructing an attention potential energy function, and obtaining an attention potential energy vector set of the multi-scale channel according to the point mass body set of the multi-scale channel; and obtaining a defect detection result of the to-be-detected panel based on the attention potential energy value and the attention potential energy range associated with the attention potential energy vector set of the multi-scale channel. According to the invention, the defect detection result and the human eyes have extremely high alignment, so that the distortion of the detection result is reduced.

The invention discloses a large traffic model intelligent traffic event identification method based on cross-modal retrieval. The method comprises the following steps: inputting an image of a traffic event to be identified and text description thereof into a traffic event identification network model; a visual feature extraction module is used for extracting visual features in the image; the text feature extraction module is used for extracting text features in the text description; the visual selective filtering module is used for selectively filtering the irrelevant visual extraction features to obtain visual filtering features; the text selective filtering module is used for selectively filtering unnecessary text extraction features to obtain text filtering features; the cross-modal selective alignment module is used for performing cross-modal alignment on the visual and text extraction features to obtain visual alignment features and text alignment features; and the calculation module is used for fusing the visual filtering features and the visual alignment features as well as the text filtering features and the text alignment features to determine an identification result. According to the invention, the accuracy and efficiency of retrieval identification are improved.

The invention discloses a voice-driven digital human construction method and device based on enhanced deformable convolution and space-time motion compensation. Firstly, Speech2Lip is used to generate a preliminary mouth shape synchronization video according to an input audio. Then, mouth shape features in the generated video and phoneme features of the audio are extracted, and two time sequences are formed; next, DTW (Dynamic Time Warping) is applied to perform time alignment on the audio and the mouth shape sequence, an optimal time path is found, and the mouth shape action is corrected, so that the mouth shape action is more accurately matched with the audio pronunciation time sequence; and adjusting the mouth shape animation according to the alignment result, generating a corrected mouth shape synchronous video, and ensuring that the mouth shape action is consistent with the pronunciation height of the audio. And finally, synthesizing the optimized mouth shape synchronous video with the audio to generate a final high-quality digital human video. Compared with the prior art, the method has the advantages that the generated digital human video driven by voice can be optimized, and the alignment accuracy of the mouth shape and the audio in the video is improved.

The invention relates to the field of thread cutting, in particular to a thread machining device for a stud. The technical problems that due to the fact that an existing thread rolling machine cannot conveniently conduct thread rolling on the two ends of a workpiece at the same time, the working efficiency is low, the workpiece cannot be conveniently aligned before thread rolling, the workpiece is prone to deviation in the thread rolling process, the thread rolling precision is low, and the thread rolling effect is poor are solved. A thread machining device for a stud comprises a bottom frame, two mounting frames are fixedly connected to the bottom frame, side plates are fixedly connected to the two mounting frames, and guide rails are fixedly connected to the two mounting frames. The two electric thread rolling machines of each set move in the direction close to each other, the electric thread rolling machines drive the racks to move, the racks drive the gears to rotate, the material moving disc rotates the lowermost workpiece to move to the position between the electric thread rolling machines, the electric thread rolling machines conduct thread rolling on the two ends of the workpiece correspondingly, and thread rolling can be conveniently conducted on the two ends of the workpiece at the same time; the working efficiency is improved.

The utility model discloses an automobile part overturning welding table, which relates to the technical field of overturning welding tables and comprises a bottom component, a moving component and a welding component. The bottom assembly comprises a bottom plate, and two sliding grooves are formed in one side of the bottom plate. The moving assembly comprises two sets of moving blocks, the welding assembly comprises a connecting plate, a controller is arranged on one side of the connecting plate, a groove is formed in one side of the connecting plate, a connecting rod is arranged in the groove, and a sliding block is arranged on the outer surface wall of the connecting rod. The welding station can be in contact with each welding point of the component conveniently, frequent manual adjustment is not needed, the welding efficiency is greatly improved, the welding gun can be ensured to be optimally aligned with a welding seam during welding by accurately controlling the overturning angle and position, the welding defect is reduced, and the welding quality and the overall performance of the component are improved.

The utility model belongs to the technical field of electric leakage detection, particularly relates to a fire-fighting electric leakage detection device for electrical equipment, and provides the following scheme aiming at the problems that the adjustment process of a probe is relatively tedious, the probe is inconvenient to use, and the whole device is inconvenient to take and move to carry out electric leakage detection in a relatively large area. An electric leakage sensor is fixedly installed at a pipe opening in the top of the hollow pipe, a plurality of warning lamps are fixedly installed at the upper end of the outer wall of the hollow pipe in an annular array mode, and the multiple warning lamps are electrically connected with the electric leakage sensor through wires. When the electric leakage detection device is used, a walking stick type structure is adopted, a user can conveniently take and use the electric leakage detection device, the electric leakage detection device can move freely in the detection process, use is convenient, and the electric leakage detection efficiency of a large area can be improved; and the risk of direct contact between the user and the to-be-detected equipment can also be reduced, so that the safety protection on the user is improved.

The invention discloses a competency extraction and evaluation method based on employee occupational behavior data driving, which comprises the following steps: firstly collecting employee occupational behavior data and extracting behavior characteristics related to competency to obtain a characteristic set, then constructing a competency factor model, defining a hierarchical structure of competency factors, and performing fine evaluation by a fine-grained feedback model. The method comprises the following steps of: selecting a total competency factor set of a hierarchical structure, optimizing by applying a self-reflection mechanism and a reinforcement learning algorithm, dynamically adjusting the total competency factor set of the hierarchical structure, and finally evaluating the competency of each employee in real time based on a real-time competency factor model and a fine-grained feedback model. According to the invention, the accuracy and consistency of employee competency evaluation results are ensured, and the problems of inaccurate employee competency evaluation results and poor evaluation efficiency at present are solved.