42results about How to "Robustness" patented technology

The invention discloses a micro-grid network topology flexible configuration method based on matrix topology, and relates to the field of micro-grids, and the method comprises the steps: constructing a heterogeneous database of a micro-grid network and equipment; mapping a micro-grid network and equipment into a graph structure, and constructing a graph neural network model; setting mapping strategies of different heterogeneous hardware units according to heterogeneous characteristics of the different heterogeneous hardware units, and constructing a scheduling model of the heterogeneous hardware units by adopting a reinforcement learning algorithm; inputting the preprocessed data into a graph neural network model to obtain an incidence relation between a micro-grid network topology structure and equipment parameters, and taking the incidence relation as a constraint condition; inputting the preprocessed data into a scheduling model to obtain a scheduling strategy of the heterogeneous hardware unit, and taking the scheduling strategy as an optimization target; constructing a multi-objective function of micro-grid network topology optimization; and solving the multi-objective function of the microgrid network topology optimization by adopting an intelligent algorithm, and generating a microgrid network topology scheme. According to the invention, the robustness of the topology scheme is improved.

The invention provides a distributed steer-by-wire vehicle fault-tolerant control method. The method comprises the steps that a distributed steer-by-wire vehicle dynamics model, a two-degree-of-freedom reference model and a steering motor failure model are established; in the design of the upper fault-tolerant controller, a self-adaptive terminal sliding mode control method is adopted, the method has the characteristics of high robustness and self-adaptive parameter change, and the robustness of the system can be considered while the target tracking performance is ensured. In the design of a lower-layer torque and rotation angle optimal distribution reconstruction controller, a failure factor is introduced, two more objective functions are established on the basis of minimizing the tire load rate, and errors between longitudinal and lateral forces and expected target values and errors between yawing moment and additional yawing moment are reduced while the optimized objective functions are ensured to have solutions. According to the method, the uncertainty of the system during the steering failure is fully considered, and the robustness of the system can be considered while the target tracking performance is ensured, so that the control stability of the distributed steer-by-wire vehicle during the steering failure is improved.

The invention belongs to the technical field of path planning and robot motion planning, and particularly relates to a multi-strategy fusion improved RRT * sampling planning method and device and a storage medium, and the method comprises the steps: constructing a dual-search tree through a starting point and a target point in a three-dimensional working space, and selecting a to-be-expanded search tree through a random mode; halton-Bridge mixed sampling is adopted to generate sampling points, so that the global coverage uniformity is improved, and the narrow channel sampling capability is enhanced; calculating the dynamic step length of the exponential decay based on the local obstacle proportion; adopting a hierarchical guidance expansion strategy, preferentially expanding towards a target, introducing an improved artificial potential field to guide expansion when the expansion fails, and taking the bottom with random expansion; performing RRT * neighborhood father node reselection and reconnection optimization on the new node; greedy pruning, self-adaptive interpolation and cubic B-spline smoothing are carried out on a path after the double trees are connected, and an executable smoothing track is output; according to the scheme, planning efficiency, success rate and path quality are improved.

The invention relates to the technical field of power distribution cabinet internal harmonic control, in particular to a power distribution cabinet harmonic suppression method based on multi-source data fusion. By extracting amplitude and phase information at each harmonic frequency point, branch equivalent admittance is calculated, and an admittance distribution state is formed; judging an abnormal branch with admittance deviation at each frequency point, and establishing a corresponding relation between the adjustable electrical parameters and the admittance deviation; in the operation process of the power distribution cabinet, the admittance distribution state and the abnormal branch set are periodically updated, the equivalent admittance of the target branch is dynamically adjusted according to the corresponding relation, and the distribution of the harmonic current among the branches is changed; through the method, the harmonic peak value and the multi-node superposition effect in the power distribution cabinet can be actively reduced, online operation under the condition of multi-frequency and multi-node load fluctuation is adapted, harmonic flow direction reconstruction is achieved on the premise that complex filtering or hardware devices are not added, and the electric energy quality and the system stability of the power distribution cabinet are improved.

The invention discloses a forklift path tracking control method based on MPC and ARC cooperation, and belongs to the technical field of industrial vehicle automatic control. Comprising the steps that an upper-layer MPC is designed based on a forklift kinematics model, and an expected steering angle and an expected speed instruction are obtained through solving; a lower-layer ARC is designed based on the pump control steering system dynamics model to track an expected steering angle, and a PID controller is designed to track an expected speed; a time delay compensation mechanism is integrated in the upper-layer MPC, and a high-frequency execution strategy is adopted. According to the method, interference of the steering system is accurately observed and compensated, parameter uncertainty is adaptively processed through parameters, nonlinear uncertainty is processed through robust items, a time delay compensation mechanism is integrated in MPC, all control rates are display expressions, online solution is not needed, and the method is suitable for large-scale popularization and application. The influence of steering resistance dramatic change, external interference, unmodeled dynamics and system time delay on the control precision is effectively reduced, and high-precision and low-time-delay path tracking performance is achieved.

This invention belongs to the field of medical signal processing technology, specifically a method and prediction system for constructing a three-dimensional tumor motion trajectory prediction model. It involves collecting temporal data of point clouds from the human chest and abdomen, as well as temporal location information of the tumor, and establishing a unified spatial coordinate system. The temporal data of the point clouds are divided into n time periods of total duration T1, and the temporal location information of the tumor is divided into n time periods of total duration T2, resulting in n training samples. The point clouds of the training samples are further divided into several pseudo-voxel sub-blocks according to the human projection plane, and K pseudo-voxel sub-blocks are selected based on motion saliency to obtain the temporal feature vector of the tumor motion. The temporal feature vector of the tumor three-dimensional motion trajectory prediction is used as input, and the corresponding tumor temporal location information is used as output to train an attention-enhanced dilated convolutional prediction network. This invention enables high-precision, non-invasive prediction of the three-dimensional tumor motion trajectory.

This application discloses an auxiliary classification system and method based on multi-view spatiotemporal interaction and difference compensation, relating to the field of auxiliary classification. The method includes: acquiring time-series data from multiple brain regions and calculating Pearson correlation matrices and partial correlation matrices; constructing first and second-dimensional time matrices based on features extracted from the time-series data; generating multiple spatial connectivity matrices based on the Pearson correlation matrix and partial correlation matrix, thereby constructing first and second-dimensional spatial matrices; performing bidirectional cross-attention interaction on the time and spatial matrices of the two dimensions respectively to obtain corresponding graph structure representations and node representations; generating compensation terms based on the differences between the graph structure representations and node representations of the two dimensions; enhancing the graph structure representation and node representation of the second dimension using the compensation terms to obtain fused features; and inputting the fused features into a graph convolutional network to output classification results. This invention improves the accuracy of auxiliary classification through two-dimensional spatiotemporal interaction and difference compensation.

This invention discloses a method and apparatus for active discharge of a DC bus. The method includes: acquiring the bus voltage signal of the DC bus in the y-th sampling period and calculating the actual discharge energy of the DC bus in the y-th sampling period; dynamically adjusting the reference discharge energy based on the actual discharge energy of the DC bus and the target total discharge time; and dynamically adjusting the d-axis current in the (y+1)-th sampling period based on the actual discharge energy of the DC bus in the y-th sampling period and the reference discharge energy. In this invention, the target total discharge time is preset, making the discharge duration of the DC bus controllable and eliminating the need for manual calibration, thus achieving the effect of reducing the bus voltage to a safe threshold within a specified time. Based on the closed-loop feedback of the actual discharge energy and the dynamic adjustment of the reference discharge energy, the d-axis current is dynamically adjusted, thereby achieving dynamic adjustment of the actual discharge energy in each sampling period and improving discharge safety.

This invention belongs to the field of Internet technology and relates to a distributed task monitoring method and system based on state transition and targeted push. It includes receiving subscription requests initiated by clients, adding clients to communication groups associated with the task status corresponding to a tab, and maintaining the group affiliation information of each client in real time; based on the group affiliation information, when a task status transitions from its original state to a new state, locating the original state group and the new state group, pushing a removal notification to the original state group and a move-in notification to the new state group; when a task in the execution state experiences a progress update, locating the execution state group and pushing a progress update message only to that group; and using the removal notification, move-in notification, and progress update message as outputs to trigger clients to perform removal, addition, and progress refresh operations on the corresponding tasks in their local task lists, respectively. This solves problems such as list synchronization issues and numerous invalid pushes, achieving real-time list consistency and accurate push notifications.

The invention relates to an elastic prediction control method and device based on interval target self-optimization. The method comprises the steps that process output at the current moment and a process output sequence in a past preset time period are acquired; determining a process stability index for representing process stability according to the process output sequence; judging whether the process stability index is smaller than an index threshold value or not; if it is determined that the process stability index is not smaller than the index threshold value, a first objective function of a set value control strategy is solved; if it is determined that the process stability index is smaller than the index threshold value, whether process output at the current moment is within the range of the target interval or not is further judged; if it is determined that the process output at the current moment is within the range of the target interval, solving a second target function of the reference trajectory self-optimization strategy; and if it is determined that the process output at the current moment is out of the range of the target interval, the first target function is solved, so that the system robustness and the device operation benefit are both considered.

The invention relates to a permanent magnet synchronous motor model prediction control method and system, belongs to the technical field of permanent magnet synchronous motor driving, and solves the problem that parameter robustness and dynamic performance cannot be considered when parameter mismatch occurs in permanent magnet synchronous motor model prediction control in the prior art. Comprising the steps of obtaining sampling values of stator current, stator voltage and electrical angular velocity at a current sampling moment and two previous moments and a direct current bus voltage at the current sampling moment, further updating incremental current prediction errors and time sequence sampling data of incremental voltage, and further obtaining a first sample vector and a second sample vector at the current sampling moment; and then a small-batch stochastic gradient descent method is adopted to obtain an identification value of a parameter mismatch coefficient at the current sampling moment, then stator current prediction values of different voltage vectors at the next moment are obtained, then an optimal voltage vector is obtained, and permanent magnet synchronous motor model prediction control at the next moment is carried out based on the optimal voltage vector.

The application discloses an AFS, ARS and DYC cooperative control method of a distributed electric drive vehicle based on DDPG. The method utilizes a cooperative control method and a deep reinforcement learning method, designs and develops a vehicle stability control method based on a centroid side slip angle and a yaw angular velocity, takes driving signals provided by a driver and measured or estimated vehicle state information as input, and judges the stability of the vehicle. When the active safety system needs to intervene, a DDPG-AAD model trained offline by a strategy deep gradient algorithm is used to coordinately control AFS, ARS and DYC systems, precise and independent control of each wheel is realized, and thus the active safety driving stability of the vehicle is improved.

The invention discloses a curved surface self-adaptive grinding and polishing method based on a two-degree-of-freedom force control grinding and polishing robot. The method comprises the steps that a surface image of a workpiece to be machined is acquired, and local curvature information is extracted; calculating a feed-forward compensation amount according to the local curvature information, and correcting a preset expected position and a preset expected speed based on the feed-forward compensation amount to obtain a corrected expected position and a corrected expected speed; on the basis of the position deviation between the corrected expected position and the actual position and the speed deviation between the corrected expected speed and the actual speed, inertial parameters and damping parameters of a position-based impedance model are adjusted in real time through a fuzzy adaptive algorithm; and generating a control instruction according to the adjusted impedance model, and driving the two-degree-of-freedom linear motion module to perform compensation motion. According to the method, through cooperation of curvature feedforward compensation and fuzzy variable impedance control, force control lag at the curvature sudden change position is reduced, and force control stability and machining consistency of complex curved surface grinding and polishing are improved.

The application discloses a small sample geothermal resource prediction method and device, system and storage medium, comprising: learning the Bayesian prior knowledge with generalization ability from limited source data by meta-learning, migrating to a new small sample target area through a rapid adaptation mechanism, and utilizing multi-scale differential similarity calculation and mixed student distribution to realize high-precision point prediction and reliable uncertainty interval estimation. By adopting the technical scheme of the application, under the objective constraints of small sample, high cost and strong regional heterogeneity in geothermal resource exploration, high-precision, high-robustness and energy-based uncertainty reserve prediction can be realized.

The invention discloses oil seal air tightness detection equipment for an electric vehicle driving motor. The oil seal air tightness detection equipment comprises a rack, a conveying module and an air tightness detection module. The conveying module is matched with a jacking limiting assembly through a roller conveying line, and automatic feeding, accurate positioning and rigid locking of the motor are achieved. The air tightness detection module comprises an air tightness instrument, an air tightness pressing module and an automatic point inspection module. A pressing air cylinder drives an air tightness pressing head to conduct sealing detection on the motor oil seal. The equipment innovatively adopts a jacking-support changing structure, stable stress is provided by a supporting plate, and detection micro-motion is eliminated; an automatic point inspection function is integrated, a zero leakage comparison tool can be periodically called to carry out on-line self-calibration on the system, and precision drift is effectively early warned. The motor oil seal airtightness detection device realizes full automation of the detection process, has the characteristics of high precision, high stability and intelligent self-maintenance, and ensures the consistency and long-term reliability of motor oil seal airtightness detection.

The invention provides an all-vanadium redox flow battery pump fault detection method based on a self-adaptive GCN, and the method comprises the following steps: carrying out the operation experiment of an all-vanadium redox flow battery in a plurality of operation states, collecting a battery voltage signal and an electrolyte flow signal in the experiment process, and marking the corresponding battery state as a training set; performing feature extraction and feature fusion on the battery voltage signal and the electrolyte flow signal to obtain a multi-modal node feature matrix of the all-vanadium redox flow battery; the Euclidean distance between the nodes is calculated to serve as the similarity between the nodes, a similarity matrix is constructed, neighbors with the highest similarity are screened out for each node from the similarity matrix, and an adjacent matrix is constructed; constructing a graph convolutional neural network model based on the adjacent matrix, and introducing an adaptive smoothing factor used for dynamically adjusting the smoothing strength of neighborhood features during volume accumulation; and inputting to-be-detected battery operation data into the trained graph convolutional neural network model to obtain a battery pump fault diagnosis result.

The invention relates to a fault identification method and device for a random working condition battery pack. The method comprises the following steps: constructing a battery pack state model and a battery monomer deviation model; respectively performing state prediction on the battery pack state model and the battery monomer deviation model by using a filter to obtain a first prediction result of the battery pack at the next moment and a first prediction result of the battery monomer at the next moment; inputting the obtained battery state parameters and the vehicle state parameters into a trained sequential network prediction model to obtain a second prediction result of the battery pack at the next moment; inputting the second prediction result of the battery pack and the first prediction result of the single battery into a trained physical enhancement encoder to obtain a second prediction matrix of the single battery; and calculating a residual value of the second prediction matrix of the single batteries, and comparing the residual value with a preset threshold value to perform fault identification on the battery pack. According to the scheme, the fault of the battery pack under the random working condition can be accurately identified.

The invention discloses a roadside parking scene analysis method and system for a night scene, and relates to the field of intelligent parking management.The method comprises the steps that firstly, connection of two different modal data of an image and an event is constructed through an image motion feature extraction model; then, a content feature extraction model is utilized to reduce the difference of data between two different domains of a daytime scene and a night scene, the requirements of model efficiency, robustness, accuracy and the like in a roadside scene are considered, and meanwhile, a semantic segmentation model structure is adopted to realize a scene analysis task of the night scene; the classification and mask output of different targets in the roadside scene is realized, the semantic segmentation precision of the night scene is improved, and the traffic management of the roadside parking scene is facilitated.

The present invention proposes a distributionally robust optimization method based on Wasserstein metric and kernel density estimation. First, the probability distribution of wind power prediction error is derived by kernel density estimation method. Then, the probability distribution fuzzy set of wind power prediction error is constructed by integrating Wasserstein metric and kernel density estimation. Finally, according to the probability distribution fuzzy set of wind power prediction error and comprehensive consideration of power system operation constraints, a distributionally robust unit combination model is established with the goal of minimizing the total operating cost of the power system under the worst wind power output. The present invention solves the unit combination optimization problem considering unstable wind power generation in the field of power dispatching, can effectively reduce the conservatism of wind power uncertainty characterization, further improve the economy of power system operation, and provide dispatchers of power systems with a dispatching strategy that can effectively handle wind power uncertainty and takes both robustness and economy into account.

The invention discloses a two-stage distributed robust unit combination optimization scheduling method, system and device and a medium, and the method comprises the steps: obtaining historical wind and light prediction error data, and constructing a data drive fuzzy set based on the historical wind and light prediction error data; based on the data-driven fuzzy set, in combination with topological information, unit information, load information and wind and light output prediction information of a power system, establishing a two-stage distribution robust unit combinatorial optimization scheduling model based on the data-driven fuzzy set; and solving the two-stage distributed robust unit combinatorial optimization scheduling model to obtain a two-stage scheduling scheme of the power system, and the method, the system, the equipment and the medium are high in reliability of optimization scheduling.

The application provides a multi-modal electric power form recognition method and device. In the execution of the method, first, the image of the electric power form to be recognized is obtained, then the image of the electric power form to be recognized is preprocessed to obtain a preprocessed image, then the CTPN neural network model is used to extract the text line position information of the preprocessed image to obtain the position information of the text line of the preprocessed image; and based on the position information of the text line, the CRNN model is used to recognize the preprocessed image to obtain the recognition result of the preprocessed image; finally, the multi-modal feature information extraction algorithm is used to extract the recognition result of the preprocessed image, and the multi-modal feature information is taken as the recognition result of the electric power form to be recognized. The application can automatically recognize the multi-modal feature information of the electric power form with the nature of the table in multiple scenes, solve the problem of low work efficiency of manual extraction, improve the electric power form recognition efficiency, and has strong robustness.

The present disclosure provides a terminal initialization method, device, electronic equipment and storage medium. The terminal initialization method comprises: in response to an initialization event of a terminal, acquiring image data of the terminal from an initial time to an end time and inertial data measured in an inertial coordinate system, the inertial coordinate system being fixed relative to the position of the terminal; establishing a linear constraint equation of a feature point according to a constraint equation of the coordinates of the feature point in a normalized plane and the coordinates of the feature point in a camera coordinate system, and a representation of the coordinates of the feature point in the camera coordinate system converted into the inertial system at the same time in the form of inertial data; projecting the linear constraint equation to a left null space of the corresponding feature point for dimension reduction; superimposing the linear constraint equations of each feature point to obtain an integrated constraint equation; and solving an initialization parameter of the terminal at the initial time according to the integrated constraint equation. The method of the present disclosure can improve the solving speed compared with the existing tight coupling initialization method, and takes into account robustness and efficiency.

The invention discloses a probability constraint-considered day-ahead power generation plan compiling method and system, computer equipment and a storage medium. The method comprises the steps of constructing a unit model required by day-ahead power generation plan compiling; generating a plurality of output prediction scenes with different confidence levels based on a new energy output prediction algorithm; taking the lowest total operation cost of a unit model required by a day-ahead power generation plan compilation scheme as an objective function, solving the objective function in an output prediction scene with the highest confidence level, and obtaining a corresponding unit commitment optimization result; solving the objective function in an output prediction scene with a second high confidence level to obtain a corresponding unit commitment optimization result; and so on, unit commitment optimization results under all output prediction scenes are obtained. And integrating unit commitment optimization results under all output prediction scenes by adopting a weighted average method to obtain a final day-ahead power generation plan scheme.

This invention provides a method and system for automatically picking up the first arrival time of ultrasonic waves, relating to the field of ultrasonic signal processing. The method involves: acquiring ultrasonic signals and constructing a three-channel input tensor; constructing a mask target sequence and a heatmap target sequence based on the actual first arrival time; constructing and training an encoder-decoder network, employing a residual attention mechanism at skip connections to output modulated features in residual form; inputting the decoder output into the mask branch and the heatmap branch respectively for collaborative learning; fusing the dual-branch outputs during the inference stage, and determining the first arrival time position based on a joint scoring function. The system includes an ultrasonic signal acquisition and preprocessing module, a target label construction module, a picking network construction and training module, and a first arrival time position determination module, used to execute the above method. This invention achieves automatic, high-precision picking of the first arrival time of ultrasonic waves in complex noise environments.

The invention discloses a complex sound scene target voice selectivity enhancement method and system based on eye movement guidance, and the method comprises the steps: synchronously collecting a gazing direction and a multi-channel audio signal through a miniature eye movement tracking module and an annular quaternary microphone array, and constructing a dynamic sound source positioning mechanism based on eye movement guidance; based on visual confidence score adaptive switching or fusion beam forming and deep learning enhancement strategies, voice intelligibility evaluation and bone conduction tactile feedback closed loops are introduced, and children are actively guided to adjust gazing behaviors. The system comprises an eye movement tracking module, a sound source positioning module, an adaptive enhancement module, a feedback prompt module and the like, and is integrally light, low in delay and high in comfort. According to the invention, through multi-mode fusion and closed-loop interaction, the selective perception ability and communication efficiency of hearing-impaired children to the voice of the target speaker in a complex sound scene are improved.

The application discloses a steel stamp character recognition method based on a two-stage response feature adaptive technology. YOLO11 The method first obtains multi-scale features of an input image through a feature extraction backbone network; then, the multi-scale features are processed through a multi-level texture feature modulation module; the multi-level texture feature modulation module enhances the saliency of target features and suppresses background interference through a dual-domain cross attention enhancement module, and adaptively extracts key response features; after the enhanced features are fused, the features are input into a dynamic alignment adaptive detection head; the dynamic alignment adaptive detection head adaptively processes shape changes and scale changes of the target through a spatial deformation modulation convolution and a dynamic activation function. The application has the beneficial effects that: in combination with the multi-level texture feature modulation module and the dynamic alignment adaptive detection head, a whole-process dynamic optimization from feature extraction to prediction output is realized, and the accuracy and robustness of industrial visual target recognition are significantly improved.

The application belongs to the technical field of composite machining, and discloses a composite workpiece element automatic extraction method based on three-dimensional point cloud semantic segmentation, which comprises the following steps: obtaining a two-dimensional image and a three-dimensional point cloud of a composite workpiece element; respectively performing data preprocessing on the two-dimensional image and the three-dimensional point cloud of the composite workpiece element to obtain a co-configuration image set composed of a three-dimensional point set and a two-dimensional image; extracting the composite workpiece element in the two-dimensional image based on a convolutional neural network and obtaining a two-dimensional feature mapping set; establishing the corresponding relationship between the two-dimensional image and the three-dimensional point cloud of the composite workpiece element by using a multi-view aggregation model based on deep learning; and performing semantic segmentation on the three-dimensional point cloud to obtain corresponding point cloud sets of different composite workpiece elements. The application can realize the automatic denoising segmentation and extraction of a point cloud file scanned by a laser after the point cloud file is imported into an algorithm, and can segment the workpiece element according to the characteristics of the composite material, mark the belonging element into the belonging information of each point, and facilitate the extraction and calling.

This invention discloses a power system dispatch optimization method and system based on deep reinforcement learning, belonging to the field of smart grid optimization and dispatch technology. It includes: receiving and synchronizing real-time operational data, meteorological data, equipment health indicators, and renewable energy output data; constructing a power grid diagram and generating a node time-series matrix, encoding it as a multi-temporal embedding vector; generating short-term output prediction values ​​and uncertainty indicators based on meteorological and renewable energy output data, converting them into compensation factors; calculating risk scores based on equipment health indicators and meteorological data and mapping them to dynamic weights; inputting the multi-temporal embedding vectors, compensation factors, and dynamic weights into a deep reinforcement learning model to generate a dispatch strategy and performing feasibility verification; if the verification passes, the strategy is executed and the results are returned for online model updates. This invention effectively improves the security and robustness of power system dispatch by integrating multi-source data and a risk perception mechanism.

The invention discloses a color image encryption method based on deep learning and optical chaotic signals, and the method comprises the steps: firstly carrying out the feature extraction of a color image through a deep learning model, and obtaining feature maps of RGB channels; secondly, the side emitting laser emits light with a set wavelength, an optical chaos phenomenon is caused through a chaos system, and a chaos optical signal is output. Then, the three channel feature maps are respectively coded into holograms through a spatial light modulator, a chaotic light signal is used as an illumination light source for irradiation, an encrypted chaotic light signal is generated through phase modulation and wavelength division multiplexing, and the encrypted chaotic light signal is separated into three independent channel signals through a beam splitter according to RGB wavelengths. And finally, performing inverse modulation decryption on the channel signal, restoring the optical information, converting the optical information into an electric signal, performing image restoration on the electric signal based on a random phase mask, and decrypting the image. According to the method, the security problem of encrypted information in the image encryption process can be effectively solved, and the encryption is more targeted and robust.

The present invention provides a log anomaly detection model training method, a log anomaly detection method and a system, which belong to the field of big data platform anomaly detection technology. The log anomaly detection model is trained, and the log anomaly detection model includes an anomaly detection branch and a self-knowledge distillation branch; the input of the self-knowledge distillation branch is the feature map of each independent convolution layer, and a transverse convolution layer is used to perform convolution on the convolution feature map. The transverse layer does not change the dimension of the feature map, and the optimized feature map is connected through a maximum pooling operation; two layers of fully connected layers with activation functions are used to classify log events. The present invention fully learns log information representation through a dual-branch structure of self-knowledge distillation, extracts the spatiotemporal logic between logs, enhances the generalization ability of the model, and makes the model more robust to complex and unstable data sets in the real world, while ensuring the log anomaly detection ability of the model and the accuracy of anomaly detection.