11 results about "Modular decomposition" patented technology

The invention provides a neural network model co-evolution method based on model modularization and model merging, and belongs to the field of artificial intelligence. Comprising the following steps: (1) carrying out model modular decomposition based on gradient search; and through common optimization of a weight retention rate index and a performance index on a field pre-training set, extracting a function-related sparse module from the pre-training model. And (2) the module only updates the specific weight on the downstream task, so that knowledge updating is carried out on the specific field. And (3) knowledge fusion based on model merging. A sparse task vector is directly obtained by subtracting the weights of the fine-tuned modules from the weights of the pre-training models, and then a sparse weight updating matrix among the multiple modules is added back to the pre-training weights, so that a multi-task global model is obtained. According to the method, the mapping relation between neural network parameters and functions is defined, the model co-evolution effect is improved, and parameter conflicts in multi-task learning are relieved.

According to one embodiment, a method for performing a modular decomposition according to a decomposition parameter is described, comprising: representing the number as a sum of a first input number portion and a second input number portion modulo the modulo, calculating a reduced first input number portion by subtracting one from the first input number portion if it is not zero, calculating a division result of the first portion by integer division of the reduced first input number portion by the decomposition parameter, calculating a division result of the second portion by integer division of the second input number portion by the decomposition parameter, calculating a candidate value for the low-bit component, and, if the candidate value for the low-bit component is greater than or equal to the decomposition parameter,Determining the high-bit component by adding one to the sum of the division result of the first part and the division result of the second part.

The invention discloses a federated learning-based satellite internet model security data sharing method, which comprises the following steps of: firstly, constructing a self-adaptive ground mobile terminal clustering strategy based on a dynamic graph attention mechanism, and establishing a space-time sensing ring topology by analyzing a node movement rule and communication characteristics to lay a network foundation for decentration calculation; secondly, a modular decomposition confusion encryption scheme is developed, homomorphic encryption of gradient parameters is realized by using a redundant modular component superposition and random number injection technology, the ciphertext size is compressed, and dependence of a third-party center is avoided through distributed private key management. And finally, creating a multi-sub-ring parallel computing architecture, splitting a modular decomposition task into independent computing units, and realizing local model combination based on a proximity communication principle, so that the bottleneck of a traditional server is eliminated, and the aggregation delay is also reduced. According to the whole scheme, through collaborative design of a topology layer, an encryption layer and a calculation layer, feasibility and efficiency balance of cross-domain collaborative training is achieved under the condition that data privacy is guaranteed.

The application relates to a parallelized fault tree minimal cut set solving method based on bit set minimal cut set representation, and relates to the field of fault tree analysis. In order to solve the problems of redundant cut set data representation, insufficient parallelization degree, low calculation and storage efficiency and difficulty in supporting efficient solving of large-scale complex systems in the prior art, the method generates a standardized equivalent logical structure concise form fault tree by structurally simplifying and modularly decomposing a fault tree model; an intermediate event and a module solving queue are established according to a topological relationship; a multi-core parallel computing environment is established, a bit set is used to represent a cut set and a two-dimensional data array is formed; a bit-by-bit or broadcast operation is performed for an and gate, and row splicing and merging are performed for an or gate, so that parallel calculation of logical gate outputs is realized; and then non-minimal cut sets are screened out through parallel simplification, and a top event minimal cut set result is output. The application is suitable for fault tree analysis and fast solving of minimal cut sets of complex systems such as nuclear power, aerospace, chemical industry and rail transit.

The invention relates to the technical field of power distribution network design, and discloses a power distribution network modular design method and system application. The method comprises the following steps: firstly, acquiring initial network parameters and load distribution data of the power distribution network, and dividing the power distribution network into a plurality of independent module units according to the initial network parameters; constructing a power flow feature operator model, and performing feature coding processing on the module units by using a predefined feature function; obtaining a pre-trained initial optimization model, wherein the model is obtained by comparing and learning feature coding results through an integrated training framework; based on the load distribution data, performing multi-objective optimization training on the initial optimization model to obtain a trained power distribution network design model; and performing modular design on a to-be-designed power distribution network scene by applying the model to obtain an optimized design result. According to the method, by means of modular decomposition, feature coding, model pre-training and multi-objective optimization, different power distribution network scenes are adapted, efficient and accurate design is assisted to be achieved, and design flexibility and expansibility are improved.

The invention discloses a tourism service integration platform based on dynamic bidding and intelligent matching and an implementation method, and belongs to the technical field of online tourism service and electronic commerce. According to the platform, through cooperation of a user terminal, a service provider terminal, a cloud server and an integrated operation management background, precise analysis and modular decomposition of user requirements are realized by adopting a mixed natural language processing technology, and bidirectional dynamic bidding matching is realized in combination with a reinforcement learning driven multi-objective optimization algorithm; and a block chain intelligent contract and an Internet of Things monitoring technology are deeply fused, so that intelligent analysis, modular decomposition and bidirectional dynamic bidding matching of tourism demands are converted into full-link service monitoring and automatic settlement. According to the system, the technical bottlenecks of a traditional tourism platform in the aspects of demand understanding, resource configuration, customization flexibility and service performance transparency are effectively solved, and tourism service integration which is highly personalized, efficient in resource configuration and credible in the whole process is realized.

The application provides a modular decomposition method for a Transformer neural network model, modular training is performed on a randomly initialized model, and structured decomposition is implemented by applying a mask after the modular training, so that flexible on-demand reuse is realized; specifically comprising the following steps: S1, modular training; given a neural network model, firstly, randomly initialize all parameters of the model; a neuron recognizer is included, which is used to identify neurons related to a specific function; S2, structured decomposition; the modular training model is subjected to modular decomposition; S3, on-demand reuse; on-demand reuse is realized by structurally removing neurons, and memory and computing overheads are minimized. The application can realize structured decomposition of a Transformer structure model and flexible on-demand reuse of functions, and has high expandability.

The invention discloses a parallel fault tree minimum cut set solving method based on bit set minimum cut set representation, and relates to the field of fault tree analysis. In order to solve the problems that in the prior art, cut set data representation is redundant, the parallelization degree is insufficient, the calculation and storage efficiency is low, and efficient solving of a large-scale complex system is difficult to support, the method comprises the steps that structural simplification and modular decomposition are conducted on a fault tree model, and a standardized fault tree with an equivalent logic structure and a concise form is generated; establishing an intermediate event and module solving queue according to the topological relation; a multi-core parallel computing environment is established, a bit set mode is adopted to represent a cut set, and a two-dimensional data array is formed; executing bitwise OR broadcast operation for the AND gate, executing row splicing and merging for the OR gate, and realizing parallel calculation of logic gate output; non-minimum cut sets are screened out through parallel simplification, and a top event minimum cut set result is output. The method is suitable for fault tree analysis and minimum cut set rapid solution work of complex systems such as nuclear power, aerospace, chemical engineering, rail transit and the like.

The invention relates to the technical field of life cycle evaluation, and discloses a photovoltaic supply chain LCA model modular deconstruction and precise traceability reconstruction method, which comprises the steps that an LCA model deconstruction engine disassembles a background model into standardized process modules; the supply chain network mapping module converts the data into an attribute directed topological graph; the traceability engine constructs a material source tree through reverse search, and instantiated nodes are generated by fusing actual measurement data of suppliers; and dynamically assembling the foreground and background separated technical matrix and the environment intervention matrix, solving a scale vector, calculating an environment influence score, and executing hotspot analysis. According to the invention, by establishing the material source tree reconstruction model based on the graph theory logic, dynamic decoupling and high-precision recombination of general background data and enterprise live-action data are realized, and the technical problems of difficulty in upstream data traceability and rough evaluation granularity in a photovoltaic complex industrial chain are solved; and the accuracy and the visualization level of environmental impact evaluation are improved.

This invention belongs to the field of software automated refactoring technology, specifically involving a method and system for detecting concern overload odors using topic modeling and mining, and for searching refactoring solutions using genetic algorithms. The method includes: parsing and preprocessing the source code of a microservice system to obtain a code dependency matrix for quality assessment; using topic modeling techniques to mine concerns in the system and detect concern overload odors, while converting overloaded services into a set of refactoring units; and using a multi-objective genetic algorithm to search a list of refactoring solutions and using Euclidean distance to select the optimal refactoring solution. This invention discovers microservices with concern overload odors by mining semantic topics in the software system and provides a modular decomposition and refactoring method for overloaded services, thereby helping architects quickly discover and refactor concern overload odors within the system, improving the efficiency of software maintenance and evolution.

The invention discloses a partial differential equation set solving method and system based on a coupling matrix and a medium, and relates to the technical field of nuclear reactor numerical calculation. A global Jacobian matrix and a local Jacobian matrix are divided into clear sub-matrix blocks according to physical field variables and degrees of freedom thereof, and a coupling matrix is combined to explicitly describe an interaction relationship between physical fields, so that modular decomposition of computational logic is realized; the calculation program of each sub-matrix block can be independently configured and developed and corresponds to the self-action of a specific physical field or the coupling action with other fields; when a physical field is newly added, a developer only needs to focus on a sub-matrix block calculation program of the physical field and define an interaction relationship with other fields by updating a coupling matrix, so that function extension can be realized without touching or reconstructing assembly logic and codes of other existing physical fields; the coupling degree between codes is reduced, calculation modules of different physical fields can be combined and reused, and the development period is shortened.