25results about How to "Improve search accuracy" patented technology

The invention relates to the technical field of knowledge management and intelligent retrieval, and discloses a knowledge enhancement method and device based on context adaptive retrieval, equipment and a storage medium, which are used for improving the accuracy of a retrieval result. The knowledge enhancement method based on context adaptive retrieval comprises the following steps: acquiring post portraits and task context information, and generating context representation vectors; calculating a knowledge granularity weight vector according to the context representation vector; determining a target retrieval hierarchy based on the knowledge granularity weight vector, and performing retrieval operation on the target retrieval hierarchy to obtain a candidate knowledge set; performing fusion processing on the candidate knowledge set according to the knowledge granularity weight vector and the retrieval similarity to obtain a target knowledge result; and sending the target knowledge result to a service system.

This invention relates to the field of multi-target search using swarm robots, proposing a multi-source localization method based on virtual targets and a virtual population using particle swarm optimization (PSO) algorithm. This method utilizes a swarm of robots with certain communication and sensing capabilities to achieve multi-source target search and localization. The method equally divides the search area into multiple units, each with a virtual target at its center. Only one group of robots traverses all virtual targets and uses the PSO algorithm to search the corresponding regions of each virtual target, ultimately covering the entire search space. Common grouping methods in this field suffer from the difficulty in determining the number of groups, and their performance decreases as the number of source targets increases. The method proposed in this invention requires only one group to complete multi-source target localization, thus greatly reducing the number of robots needed and reducing physical costs. Furthermore, this method can be adapted to large-area passive scenarios.

The application discloses a mobile device path planning method and an electronic device. The mobile device path planning method comprises the following steps: constructing a fitness function of mobile device path planning; initializing a particle swarm population, and a particle position is used for encoding a candidate path composed of multiple intermediate path point coordinates; performing multiple iterations of optimization, in each iteration, after each velocity and position update is completed, performing a dimension-by-dimension optimization operation on the entire population, the dimension-by-dimension optimization operation searches the path point coordinates of the candidate path encoded by all particle positions in the population dimension by dimension, and optimizes the corresponding path point coordinates in the global optimal position; after the iterations are completed, determining an optimal path based on the path encoded by the global optimal position. The application overcomes the defect that the effective local feature is lost due to simple elimination of low fitness particles in the existing PSO algorithm; meanwhile, repeated calculation of full path evaluation is avoided, and the real-time performance and high efficiency of algorithm operation are ensured while the optimization precision is effectively improved.

This invention discloses a method and system for injecting knowledge into large-scale models for vertical industries, belonging to the field of large-scale model knowledge injection technology. The method involves: collecting professional knowledge data from vertical industries, performing denoising, structured transformation, and terminology standardization; dividing the data into knowledge modules and associating them with vectorized indexes and metadata tags to form a vertical industry knowledge base; parsing the semantic feature vector of the input question using a dynamic routing classifier and routing it to the target knowledge module in the vertical industry knowledge base; employing a hybrid retrieval strategy to recall relevant document sets from the target knowledge module and deduplicate them to generate enhanced retrieval results; based on this, concatenating the enhanced retrieval results with the input question using industry-specific prompt templates; and inputting the LoRA-tuned large-scale model to generate accurate answers conforming to industry standards. This invention solves the problem of insufficient knowledge adaptability of large-scale models in vertical industries, improving the accuracy and applicability of large-scale models in vertical industries.

The application discloses a kind of low-carbon scheduling method, system, equipment and medium of power system, belong to power grid low-carbon scheduling technical field, method includes initialization parameter and constructs collaborative space;By particle swarm module optimization reinforcement learning hyperparameter;Reinforcement learning module interactive experience data generation;Extract priority sample and elite particle realize two-way experience sharing;Converge after output scheduling strategy.System includes initialization module, collaborative space construction module, hyperparameter optimization module, interactive storage module, two-way experience sharing module and strategy output module.The application is coupled by constructing dynamic collaborative optimization space and two-way experience sharing mechanism, particle swarm and deep reinforcement learning.Feedback guides particle swarm to realize adaptive disturbance, avoid local optimum trap.At the same time, spontaneous optimization network parameter, significantly enhance the adaptability and generalization ability of algorithm.The scheme breaks through single algorithm decision bottleneck, improves low-carbon scheduling efficiency and power system reliability.

The application relates to a vector coding learning method and device for neighbor graph vector retrieval. After an original vector is acquired, a neighbor graph of the original vector is constructed; then, coding model parameters are initialized to complete preparation work; in a training process, the original vector is converted into transition coding through the coding model; the transition coding corresponds to replacing the original vector in the neighbor graph of the original vector to obtain a transition coding neighbor graph; through performing neighbor search on the transition coding neighbor graph, routing data is acquired and screened to adjust the coding model parameters, so that the coding model is adapted to the search characteristics of the neighbor graph; iterative training is performed until a training termination condition is met, and finally, compressed coding is output. In the vector coding learning process, a search process based on the neighbor graph is introduced, the retrieval performance of the compressed coding under the neighbor graph is directly optimized, and the search precision and efficiency can be improved while the memory overhead is reduced.

The invention discloses a throughput optimization method and system for a deep space communication system, and relates to the technical field of deep space communication, and the method comprises the following steps: constructing an optimization problem; under a plurality of constraint conditions, randomly generating an initial population comprising a plurality of particles, and obtaining a fitness value of each particle; sorting each particle in the initial population according to the fitness values from large to small, and taking the first J particles as primary satellites of different sub-populations; performing multi-round iteration on the plurality of sub-populations; and after multiple rounds of iteration are completed, outputting optimal particles to obtain an optimal solution and a corresponding maximum throughput. According to the method, an initial population is divided into a plurality of sub-populations, and each sub-population is independently evolved by taking a primary satellite as a core; on the basis, each sub-population can carry out annular reconnaissance search in parallel, so that individuals can explore in multiple directions, and global search and local search strategies are further introduced, so that the population is balanced between global exploration and local development.

This invention relates to the field of 3D layout technology, specifically to a center cube layout method based on an intelligent heuristic algorithm, comprising the following steps: S1, determining the size parameters of the space to be laid out, the number of objects to be laid out, and the size parameters of each object to be laid out, setting layout constraints and preset goals; S2, constructing a mathematical model of the center cube layout problem, defining the layout state, envelope box, center cube, available space, and envelope box fill rate; S3, designing an intelligent heuristic algorithm, which is designed according to the available space calculation strategy, the available space cube selection strategy, and the object placement strategy; S4, based on the intelligent heuristic algorithm designed in S3, combined with the parameters of S1 and the mathematical model of S2, solving the center cube layout, outputting a layout scheme that satisfies the layout constraints and preset goals, thereby improving the engineering feasibility of the layout scheme.

Embodiments of the present application disclose a training method of a retrieval model, a retrieval method and an electronic device. The training method comprises: fusing prior knowledge of historical wafer defect cases in a database into a multi-modal feature vector of a current training sample through a cross-attention module to generate a multi-modal enhanced query vector of the current training sample; splicing enhanced query vectors of different modes of the current training sample into a multi-modal sequence of the current training sample according to a time sequence of a process flow through a vector splicing module; inputting the multi-modal sequence of the current training sample into a sequence processing module to extract fusion features of different modes of the current training sample; calculating a distribution distance of the fusion features of different modes of the current training sample in a same feature space to construct a global alignment loss function, and optimizing parameters of the retrieval model through a back propagation algorithm. Embodiments of the present application improve the retrieval accuracy of historical wafer defect cases.

The application discloses an image retrieval method based on multi-direction global features. First, the integrated deep feature map of an image is extracted by using a deep residual network (ResNet) model and an iterative multi-layer fusion model. Second, the spatial and frequency clues of the integrated deep feature map are captured by using a Gabor filter to obtain a multi-direction feature map. Finally, a matching feature vector is obtained by using an L2 normalization and whitening method. The final retrieval result is obtained by using the cosine similarity between the matching feature vector of a current image to be retrieved and the matching feature vector of each sample image in a retrieval sample database, so that the image retrieval is realized. The method integrates deep feature maps of different levels to obtain an integrated deep feature map with distinguishing characteristics. The integrated deep feature map learned is combined with direction clues by simulating the direction selection mechanism of a human being, so that an image representation with higher discrimination is obtained, and the image retrieval performance is significantly improved.

The invention provides a method for searching a target high-quality semiconductor material based on lightweight and heavy neural networks and a related device, and belongs to the technical field of high-quality semiconductor material positioning. The method comprises the following steps: acquiring an original microscopic image of an uncompressed target high-quality semiconductor material; inputting an original microscopic image of an uncompressed target high-quality semiconductor material into the trained lightweight neural network to obtain a potential target area; inputting the potential target area into the trained heavy-weight neural network to obtain the size or thickness of the target high-quality semiconductor material; and determining coordinates of the target high-quality semiconductor material according to the size or thickness of the target high-quality semiconductor material. According to the method, the problems of low speed and low precision of searching the target high-quality semiconductor material are solved.

This invention discloses a method for predicting the unsteady acoustic quality of electric drives based on intelligent sensors. Specifically, it includes: acquiring noise signals and extracting sound pressure level, loudness, sharpness, speech intelligibility, and priority speech interference level to form a feature vector sequence and dataset; constructing a neural network to determine nodes and form network parameter vectors; establishing an information entropy budget allocation model, mapping each feature to an information budget variable vector, setting constraints to construct an objective function containing budget constraint terms and prediction error terms, and embedding the budget variables into the feature representation; constructing a particle swarm, uniformly encoding the information budget vector and network parameter vector into particle state vectors and then initializing them; introducing an information budget modulation factor during iteration to weight the velocity update component, achieving co-evolution of the two; updating the optimal particle based on fitness, outputting the optimal information budget and network parameters to construct a prediction model; and inputting the feature vector sequence to output the prediction result. This invention assigns independent budgets to each acoustic feature, dynamically co-optimizes, and achieves high accuracy and fast response.

This invention discloses a method and system for joint power and bandwidth resource allocation for low-Earth orbit (LEO) satellites. The method includes: Step S1: Establishing an optimization model with the objective of minimizing unmet system capacity; obtaining multiple beam configurations of the LEO satellite communication system and the real-time service requirements of user terminals and inputting them into the model; solving the model using a preset improved differential evolution algorithm to obtain the optimal power and bandwidth allocation scheme for each individual satellite; Step S2: Scheduling the beam resources of the LEO satellites according to the power and bandwidth allocation scheme. This invention effectively solves the problems of slow convergence and easy getting trapped in local optima in existing algorithms under strong constraints and high-dynamic satellite scenarios, significantly improving the resource utilization and service quality of the system.

The invention provides a method for suppressing low-frequency oscillation of a photovoltaic grid-connected system through a TCSC, and the method comprises the steps: firstly constructing a potential energy function containing a photovoltaic power grid branch mode from the perspective of network mode energy, achieving the online identification of key links of the system according to online data, considering the scene of weaker system damping after photovoltaic grid connection, and employing the TCSC in combination with the mode energy, thereby achieving the low-frequency oscillation suppression of the photovoltaic grid-connected system. Therefore, oscillation on-line suppression with the purposes of improving the small interference stability of the system and rapidly calming the oscillation energy is realized.

This invention discloses a rapid bridge damping ratio identification method based on a stationary inspection vehicle, belonging to the field of bridge structure vibration testing technology. The method includes: parking the inspection vehicle at the mid-span of the bridge under inspection; exciting the bridge vibration using a passing vehicle and then driving away; collecting the acceleration response of the inspection vehicle in the attenuation segment; establishing a dynamic model of the attenuation segment of the vehicle-bridge coupled system; constructing a two-degree-of-freedom joint state model, internalizing the vehicle-bridge interaction force as a linear combination of system state variables; extracting the effective attenuation segment through signal preprocessing and introducing a dual anti-noise mechanism of adaptive signal-to-noise ratio weights and Huber weights; and employing a profiled two-stage parameter identification strategy, first locking the frequency and modal mass, and then finely scanning and identifying the bridge's first-order damping ratio. This invention requires only a single inspection vehicle and an acceleration sensor, eliminating the need to deploy sensors on the bridge structure. It has strong anti-interference capabilities, high identification accuracy, and is suitable for rapid inspection of small-to-medium span bridges.

The invention provides a credit policy question and answer generation method and device, electronic equipment and a storage medium, and relates to the technical field of artificial intelligence. Credit consultation information input by a user is received and combined with business process node information of the user, and corresponding consultation intention information and business context are obtained through analysis; on the basis of a pre-constructed structured credit domain knowledge base, relevant credit policy information is retrieved according to the consultation intention information and the service context, and then the retrieved credit policy information is screened and sorted to determine a core policy basis; and finally, a policy interpretation answer is generated based on the core policy basis by using a natural language generation model, so that the problems of model illusion, serious noise interference and insufficient service adaptability in the prior art can be solved, and the technical effect of accurately matching a specific credit service scene of the user is achieved.

The invention relates to the technical field of electronic commerce, artificial intelligence and information retrieval, and discloses a point shopping mall multi-modal search method based on a vectorization search architecture. According to the core scheme, the method comprises the following steps: receiving multi-mode search input of user texts, images, audios, videos and the like, realizing efficient transmission and fusion processing through RTC and ITC technologies, and generating unified semantic representation; in combination with LLM and ELCTRA algorithms, deep semantic understanding is carried out, and a user query intention vector is generated; the method comprises the following steps: constructing a commodity multi-modal feature vector library (encoding through an ELCTRA / ResNet-50 / WaveNet embedded model group) and a user portrait vector library, and constructing a dynamically updated vector index based on an HNSW algorithm; a candidate commodity set is retrieved by using an ANN algorithm, and personalized sorting is realized through an L2R sorting model (fusing multi-dimensional factors such as user portrait matching degree and commodity correlation); collecting user behavior data, and regularly finely adjusting model parameters and updating indexes through incremental learning. According to the method, the bottleneck of traditional keyword search is broken through, millisecond-level high-concurrency retrieval is realized, the search accuracy and personalized experience are improved, meanwhile, data support is provided for shopping mall operation, the problem of data drift is effectively relieved, and the method is suitable for high-concurrency and multi-mode search scenes of integral shopping malls.

The present application relates to a kind of sound image archives two-dimension fusion intelligent retrieval system and method, including archive data preprocessing module, photo and sound image fusion face recognition library, two-dimension fusion retrieval model and double-engine collaborative retrieval scheduling module, photo and sound image fusion face recognition library establishes the association index of face feature and business attribute;Two-dimension fusion retrieval model dynamically adjusts the matching weight of photo and business attribute, sets similarity threshold, filters out the archives of two-dimension comprehensive matching degree standard;In double-engine collaborative retrieval scheduling module, first by face recognition retrieval engine according to face feature matching candidate archives, output candidate archive set, then by attribute screening engine according to user setting condition to candidate archive carries out business attribute matching, the archives of matching success are output, obtain target archive.The present application has the advantages that: by constructing fusion face recognition library, two-dimension fusion retrieval model and double-engine collaborative retrieval scheduling module, realize fast and accurate retrieval.

A method for measuring the bulk density of topsoil based on the sound wave penetration characteristics belongs to the field of intelligent agricultural sensing technology. In the sound wave penetration characteristic acquisition system of this invention, a digital power amplifier board is electrically connected to a subwoofer, a DC power supply, and a computer terminal; a data acquisition card is electrically connected to a microphone and the computer terminal; and a robotic arm is electrically connected to the DC power supply and the computer terminal. The robotic arm, through five servo motors in coordinated control, achieves automatic grasping, precise positioning, and vertical insertion of the ring cutter tube, ensuring consistent sample placement. The computer terminal is equipped with a soil bulk density measurement model based on an adaptive mutant particle swarm optimization variational mode decomposition model and an adaptive mutant particle swarm optimization BP neural network. This system can automatically acquire real-time soil penetration sound waves and realize soil bulk density inversion based on sound wave penetration characteristics. The soil bulk density measurement model of this invention has high timeliness and high accuracy.

The present application relates to the technical field of production planning and operation optimization, in particular to a cigarette equipment capacity matching optimization method based on an improved NSGA-Ⅱ algorithm, comprising: obtaining basic data of matching of cigarette production resources and demand; calculating constraint parameters of each product according to the basic data; constructing a cigarette equipment capacity matching multi-objective optimization model based on the constraint parameters; solving the cigarette equipment capacity matching multi-objective optimization model based on the improved NSGA-Ⅱ algorithm to obtain an optimal capacity matching scheme. The embodiment of the present application constructs a multi-objective optimization model, covers capacity gap, surplus rate balance and overtime, and improves the accuracy and efficiency of the algorithm in a complex production environment through an improved initial population generation strategy and a self-adaptive crossover and mutation mechanism. The method can accurately match capacity and demand, reduce waste, overtime and uneven load, significantly improve production efficiency and resource utilization, and optimize cigarette production management.

The application relates to an efficient neural network architecture search method and device for a 3D target detection algorithm, which comprises the following steps: based on a pre-trained super network model, at least one neural network sub-architecture is sampled, single-time and zero-time evaluation strategies thereof are generated, the performance of each neural network sub-architecture is evaluated respectively, and an optimal neural network sub-architecture is obtained; meanwhile, the optimal neural network sub-architecture is independently trained by using an independent evaluation strategy, and the performance predictor of the at least one neural network sub-architecture is obtained; delay data of the at least one neural network sub-architecture is acquired to train a delay predictor; and based on the performance predictor and the delay predictor, the optimal neural network architecture is searched by combining a Pareto evolution algorithm. Thus, the problems of large calculation resource consumption, low network architecture search precision and high delay in the search process are solved, the network architecture search efficiency and precision are improved by training the corresponding performance predictor by using multiple evaluation strategies.

The invention relates to the technical field of resistance heating furnace temperature control, and discloses an improved resistance heating furnace temperature control method based on a sparrow search algorithm, and the method comprises the steps: dividing the global temperature of a resistance heating furnace into a plurality of temperature intervals, global knowledge sharing among subgroups is realized by using a distributed subgroup optimization algorithm and a collaborative learning mechanism, and an adaptive fuzzy rule base is generated; the temperature deviation and the temperature deviation change rate serve as input variables of a fuzzy controller, and PID parameters are generated after the input variables are processed through an adaptive fuzzy rule base; iterative optimization is conducted on the PID parameters through an improved sparrow search algorithm, the optimal temperature control parameters of the resistance heating furnace are obtained, and closed-loop self-adaptive control over the temperature of the resistance heating furnace is achieved based on the optimal temperature control parameters. According to the invention, an intelligent control algorithm combining the improved sparrow search algorithm and the adaptive fuzzy PID is utilized, so that the temperature control precision, the adaptivity and the stability of the resistance heating furnace are improved.

The photovoltaic maximum power point tracking method based on the adaptive vulture search algorithm includes the following steps: Step 1: Based on the BES algorithm, a Gaussian mixture adaptive walk strategy, a progressive dive adaptive switching strategy, and a vulture flock size adjustment mechanism are introduced to construct the ABES algorithm; Step 2: The ABES algorithm constructed in Step 1 is used for photovoltaic maximum power point tracking; Step 3: Based on Step 2, a photovoltaic power generation system simulation model is built to verify the tracking performance of the ABES algorithm constructed in Step 1 under different scenarios. This invention introduces a Gaussian mixture adaptive walk strategy, a progressive dive adaptive switching strategy, and a vulture flock size adjustment mechanism into the BES algorithm, and applies them to photovoltaic maximum power point tracking. This method can successfully track the global maximum power point under local shading conditions, and the tracking speed is faster and the accuracy is higher.

The invention relates to the field of medical data processing, in particular to a medical content intelligent cooperative processing system and method. The system comprises a medical content intelligent cooperative processing system which is provided with a data acquisition module for acquiring medical field unstructured original data, a preprocessing module for format unification and quality filtering, and an intelligent processing module for performing semantic understanding and multi-dimensional tagging processing on the original data. The collaborative optimization module performs manual verification on a multi-dimensional tagging result and feeds back the result, and the data analysis and generation module performs content analysis and auxiliary creation based on the verified tagging data; the invention further comprises a processing method suitable for the system. The processing method comprises the steps of data acquisition, preprocessing, intelligent tagging, collaborative optimization, value output and the like. According to the method, the effects of effectively processing the unstructured original data in the medical field, realizing intelligent cooperative processing of medical contents, mining medical research hotspots and content gaps, generating medical literature review or science popularization manuscript first draft and the like are achieved.

The application provides a COA-based BP neural network mechanical model of a magneto-rheological damper and an application method, obtains a plurality of groups of 7-dimensional time sequence characteristic data of the magneto-rheological damper under different frequency conditions as training data; a basic mechanical model is constructed based on a BP neural network and the training data, the improved crayfish algorithm and the training data are used to iteratively update the weight and threshold of the basic mechanical model until the update condition is met to obtain the BP neural network mechanical model of the magneto-rheological damper, and finally, the obtained BP neural network mechanical model of the magneto-rheological damper can be adapted to the magneto-rheological damper to accurately predict the damping force at the current moment.