12 results about "Optimal methods" patented technology

In recent years, reinforcement learning has a significant effect in multiple fields, but a traditional method has limited adaptability in a dynamic environment or a multi-task situation. The invention provides a multi-task reinforcement learning method named as a priority weighted soft modularization (SM-PHT) method, which is used for enhancing the adaptability and the generalization ability of an intelligent agent, and relates to a multi-task reinforcement learning method named as SM-PHT (SM-PHT), which is used for enhancing the adaptability and the generalization ability of the intelligent agent. SM-PHT integrates three key mechanisms: priority weighted knowledge distillation, and student network robustness is improved through weighted fusion of multi-model knowledge; the hierarchical cache is used for respectively storing low-level experience and high-level parameter optimization efficiency; and task embedding: capturing environment characteristics to enrich task representation so as to promote knowledge transfer. In the Meta-World MT10 challenge, the success rate of SM-PHT is twice that of the existing optimal method, and the average reward is improved by 30%; in the MT50 challenge, the success rate and the average reward are both improved by about 10%. Results prove that the stability and generalization of the mechanical arm in complex tasks highlight the application potential in the field of actual multi-task mechanical arm operation.

The invention discloses a key point rotating target detection method and system based on fusion Gaussian measurement. The method comprises the following steps: firstly, outputting a multi-scale feature map through a feature extraction module and a feature fusion module; secondly, on the basis of the multi-scale feature map, a detection head module predicts the center point position of a target, and an initial key point set is regressed by using deformable convolution; gaussian modeling and quality evaluation are carried out on the initial key point set to obtain a final key point set, a loss function fused with Gaussian measurement is constructed in the training stage, key point distribution is fitted through maximum likelihood estimation, and KL divergence is used as distribution distance measurement to obtain a final key point set; evaluating the quality of the key points from five dimensions of classification, positioning, direction, correlation and Gaussian distribution; and finally, converting the final key point set into a rotation bounding box, carrying out post-processing by adopting a rotation box confidence coefficient propagation module, and outputting a final rotation target detection result. The method is superior to an existing optimal method in public data sets, and the detection precision is remarkably improved.

The invention discloses a mechanical arm scanning track planning method based on an improved ant colony algorithm, which comprises the following steps: step 1, dividing a to-be-measured workpiece into n primitives, setting n measuring points on the to-be-measured workpiece, and setting a complete set T of the measuring points as follows: T = {T1, T2, T3,..., Tm}; 2, solving a set coverage problem by using an improved greedy algorithm to obtain an optimal selection method of a scanning point position set; 3, solving a scanning track by using an improved ant colony algorithm and a combined algorithm based on a BFS strategy; step 4, optimizing the scanning track by using a particle swarm algorithm to obtain an overall scanning track with the shortest scanning time; and step 5, obtaining an overall scanning track with the shortest scanning time by using a particle swarm algorithm. According to the method, a scanning track is solved by using an improved ant colony algorithm and a BFS-based strategy, a larger area is selected to continue to carry out strategy path search, and a smaller area is solved by using the ant colony algorithm to obtain a loop. And opening the loop, combining the loop to an original path, and solving the whole path through the BFS to obtain an optimal path.

The application discloses an intelligent question and answer system and method based on problem decomposition and community semantic search. The method first decomposes a complex problem into multiple single-hop sub-problems through an original cyclic problem decomposition technology, then retrieves the community in the knowledge graph based on a vector to obtain the information required to answer each sub-problem and uses a large language model to answer the sub-problems, and finally uses an LLM to combine the answers to all sub-problems to generate the final answer to the original complex problem. Compared with other methods in the industry, the method significantly improves the information integration capability, avoids information fragmentation through community-level search and summary generation, and provides high-quality and logically consistent answers. In handling complex queries, the method outperforms the industry's general method, with a maximum improvement of 29%. At the same time, the method significantly improves the efficiency while ensuring better question and answer results. Under the premise of better results than the existing optimal method GraphRAG, the token consumption is only 3.2% of that of GraphRAG.

The application discloses a social network link prediction method based on a space-time feature perception time sequence diagram network, which divides key features in a time sequence diagram into two categories of time sequence features and structure features, for the time sequence features, through adaptive fusion of a continuous time method and a discrete method, long sequence dependence and recent dependence are effectively captured and weighed, for the graph structure features, through co-occurrence neighbor coding, graph structure information is effectively explicitly coded, and a hash-based method is used to improve co-occurrence neighbor retrieval efficiency. Such a method not only exceeds previous methods in prediction accuracy by fully capturing time sequence features and structure features, for example, the average accuracy is improved by 14.18%, 15.89% and 32.48% respectively on USLegis, UNtrade and Unvote data sets compared with previous optimal methods, and better trade-off is achieved in inference efficiency. And it is simple to realize and easy to reproduce.

The invention discloses a social network link prediction method based on a spatial-temporal feature perception time sequence diagram network, and the method comprises the steps: dividing key features in a time sequence diagram into time sequence features and structural features, and carrying out the adaptive fusion of a continuous time method and an internal discrete method for the time sequence features, long-sequence dependence and recent dependence are effectively captured and balanced; for graph structure features, explicit coding is effectively carried out on graph structure information through co-occurrence neighbor coding, and meanwhile the co-occurrence neighbor retrieval efficiency is improved through a hash-based method. According to the method, the time sequence features and the structural features are fully captured, the prediction precision exceeds that of a preorder method, for example, the average precision of a USLegis data set, a UNtrade data set and a Unvote data set is improved by 14.18%, 15.89% and 32.48% compared with that of a preorder optimal method, and better balance is achieved in reasoning efficiency. The method is easy to implement and easy to reproduce.

The invention discloses a polynomial kernel-based support vector regression model reverse inversion simulation parameter optimal value method, and relates to the technical field of numerical simulation parameter calibration. According to the method, a data set is constructed by designing a simulation parameter value combination, a prediction model is constructed by adopting polynomial kernel function support vector regression (Poly-SVR), hyper-parameter optimization is realized by combining 5-fold cross validation and grid search, and finally, reverse mapping of a response quantity and a simulation parameter is completed by utilizing an L-BFGS-B algorithm. The method solves the problems that a traditional response surface method is low in calibration precision and high in overfitting risk in a small sample scene, achieves efficient and accurate inversion of simulation parameters, and is suitable for various engineering application scenes where the simulation parameters need to be reversely determined through the actual measurement response quantity.

The embodiment of the invention provides an optimization selection method and device for waste slag field design, electronic equipment and a storage medium, and relates to the technical field of geotechnical engineering.The method comprises the steps that the attribute value of a target analysis factor in each to-be-evaluated waste slag field design is obtained, the target analysis factors comprise one or more of an ecological land occupation rate, a maximum slag stacking height, a slag stacking coefficient and a stacking height coefficient; for each to-be-evaluated abandoned slag yard design, according to a pre-configured scoring rule, respectively mapping an attribute value of each target analysis factor in the abandoned slag yard design into a single factor score for representing the value rationality of the attribute value, determining a comprehensive score of the abandoned slag yard design according to the abandoned slag yard design and the single factor score corresponding to each target analysis factor; and determining an optimal abandoned slag yard design according to the comprehensive score of each abandoned slag yard design to be evaluated. By applying the embodiment of the invention, the optimal waste slag field design can be determined from the plurality of to-be-evaluated waste slag field designs.

The application provides a sandstone and conglomerate OVT domain partitioned gather optimization stacking offset optimization method, which comprises the following steps: step 1, determining the fan body burial depth H and the fan body plan distribution map of the target layer in a research area; step 2, reasonably optimizing a real drilling well reconstruction time-depth relationship function formula A in the research area; step 3, calculating a fan body dip angle alpha; step 4, determining the maximum effective offset l2 of the reflection flat section of the target layer in the research area; step 5, calculating the minimum effective offset l1 of the target layer fan body in the research area; and step 6, obtaining the optimal gather stacking offset range l of the target layer fan body in the research area by comprehensively combining the step 4 and the step 5. The sandstone and conglomerate OVT domain partitioned gather optimization stacking offset optimization method can calculate the optimal offset range of the OVT domain partitioned gather optimization stacking of the target layer, and the obtained seismic data volume can better reflect the sandstone and conglomerate fan body distribution range and the effective reservoir development area.

The invention belongs to the field of industrial quality management, and provides a turbine anti-mixing inspection method which comprises the following steps: acquiring a turbine image to be detected, extracting a two-dimensional comparison value and a three-dimensional analysis value, judging whether data are matched or not through a comprehensive matching degree formula, and binding characteristic parameters with a two-dimensional code; if not, judging whether the materials are mixed or not by using a two-dimensional deviation analysis model; if yes, starting a reverse tracking mechanism, determining an actual model by combining cosine similarity, and checking the cause of material mixing; mixing cases of the same type are obtained, type difference degree groups are calculated, distinguishing methods are formulated according to frequency sorting, and the qualified cases are optimal methods and are summarized into a mixing distinguishing library; constructing a feature digital library containing multiple sub-libraries, generating a mixing risk optimization model by using federal learning training, and dynamically updating to iteratively optimize an optimization distinguishing method; according to the scheme, the turbine detection accuracy and the mixing risk prevention and control capability are improved.

The invention discloses a small sample increment training method and device for PCB defect detection and a medium. The consideration of increment expansibility is embedded in the initial optimization stage of basic characterization. Comprising the steps of setting a small sample class incremental learning task and constructing a standard small sample class incremental learning normal form; based on a random scene selection strategy, a dynamic relation projection module constructs a self-propelled prototype refining scheme, the classification model training effect of PCB defect small sample increment is optimized, and the continuous learning ability of the model is improved; and training the model, and evaluating the performance according to a result obtained by training. The invention strictly follows small sample increment (FSCIL) settings for evaluation. Experimental results on a standard data set show that a subsequent continuous learning process can be effectively supported through basic characterization enhanced by the core technology points, and the model is remarkably superior to an existing optimal method in recognition performance and anti-forgetting ability.

The invention provides a target point error sensitivity index-based attitude adjustment mechanism position control axis optimization method, which specifically comprises the following steps of: firstly, deducing an attitude deviation equation of an aircraft large component according to a coordinate system conversion principle; thirdly, constructing a mapping relation between the motion error of the positioner and the pose deviation of the large component, so as to determine Jacobian vectors corresponding to all axes of the positioner, and obtaining a transfer matrix between the motion error of the positioner and the positioning error of the large component of the aircraft in combination with the spatial position deviation of the target point; and finally, optimizing the combination of the position control shafts according to the maximum singular value of the positioning error transfer matrix of the large part of the aircraft. According to the method, the weighted deviation of the target point is taken as an optimization target, and the position control axis combination is optimized in real time based on the positioning error transfer matrix of the large aircraft component, so that the positioning precision of the large aircraft component is improved.